root/llama.cpp
1
0
Fork 0
mirror of https://github.com/ggerganov/llama.cpp.git synced 2024-11-14 06:49:54 +00:00
Code Issues Actions 14 Packages Projects Releases Wiki Activity
9a697d842b
llama.cpp/ggml-quants.c

9828 lines
390 KiB
C
Raw Normal View History

ggml : quantization refactoring (#3833) * ggml : factor all quantization code in ggml-quants ggml-ci * ggml-quants : fix Zig and Swift builds + quantize tool ggml-ci * quantize : --pure option for disabling k-quant mixtures --------- Co-authored-by: cebtenzzre <cebtenzzre@gmail.com>
2023-10-29 16:32:28 +00:00
#include "ggml-quants.h"
ggml : move FP16 <-> FP32 code to ggml-impl.h (#3861) * ggml : move FP16 <-> FP32 stuff to ggml-impl.h ggml-ci * tests : fix ARM build * ggml : explicitly initialize deprecated type traits * ggml : add math.h to ggml-impl.h * ggml : remove duplicate static assert macros * ggml : prefix lookup tables with ggml_ ggml-ci * ggml-impl : move extern "C" to start of file
2023-10-30 17:19:15 +00:00
#include "ggml-impl.h"
ggml : add SOTA 2,3,4,5,6 bit k-quantizations (#1684) * Starting to add k-quantization to ggml I think it is better to have quantization separate from ggml. For now just adding the k-quants there, but it would be better to also factor out the existing ggml quantizations. * Adding Q3_K and Q8_K (de)-quantization * Q3_K now working on CUDA and AVX2/scalar CUDA is not ideal - ~50% slower than Q4_0 for single token prediction, about the same in batch mode (perplexity). CPU single token is ~55 ms (on Ryzen 7950X). * Some improvement for Q3_K on CUDA It is now ~22.5 ms/token on my GPU, so ~30% slower than Q4_0. * Some more CUDA optimizations for Q3_K Single token is now 20.5 ms/token (~20% slower than Q4_0). Perplexity is on par with Q4_0. * Adding Q4_K - scalar, AVX2, CUDA Performance is the same or perhaps very slightly better than Q4_0 on the CPU. On the GPU, single token prediction is ~10% better than Q4_0, batch mode (perplexity is about the same). * Adding Q6_K - scalar, AVX2, CUDA Performance is ~40% lower compared to Q4_K on the CPU. This is to be expected, considering that we are memory bound on the CPU and the 6-bit model is ~44% larger than the 4-bit. On the GPU, single token prediction is ~6% lower than Q4_0, batch mode (perplexity) is even closer (but still slower). * Adding Q5_K - scalar, AVX2, CUDA Performance is ~20% lower compared to Q4_K on the CPU. This is to be expected, considering that we are memory bound on the CPU and the 5-bit model is ~22% larger than the 4-bit. On the GPU, single token prediction is about the same as Q4_0 for both, single token and batch prediction. * Per convention, all QX_K quantizations use Q5_K for output.weight * Adding quantization mixes * Quantization mixes: didn't quite get what I wanted in the last commit * Q4_K dot product for ARM_NEON * Q6_K dot product for ARM_NEON * Q5_K dot product for ARM_NEON * Adding Q3_K dot for ARM_NEON It is 22% slower than Q4_K, despite the smaller model size. On x86_64, where we are memory bound, the Q3_K model is quite a bit faster than Q4_K. * A very slightly faster ARM_NEON Q3_K dot * Adding Q2_K - just CUDA for now Token prediction is pretty good - about 15.5 ms on a RTX 4080. Perplexity is about the same as Q4_K. * Adding scalar and AVX2 Q2_K dot * Adding ARM_NEON Q2_K dot About the same performance as Q4_K. * A slightly faster ARM_NEON Q2_K dot Single token prediction is now ~36 ms on M2 Max. The code is much simpler too. * Fixed bug in Q2_K CUDA dot product kernel Stranegly enough, for the few prompts I tried with the 7B model the responses looked perfectly reasonable. Only realized something is not quite right when I tried the larger models and started getting nonse back. In any case, Q2_K single token evaluation time on an RTX 4080 in a Ryzen7950X box iusing CUDA and model fully loaded on the GPU are ~15.5 ms for 7B, ~25.4 ms for 13B, and ~55.8 ms for 30B. The max number of layers that fit in VRAM for The 65B is 32. With that, we get ~330 ms per token, which is not that much faster than just running on the CPU (~470 ms per token). * Don't print zeros/NaNs when no count histogram has been collected * A 10% faster CUDA vector dot kernel for Q3_K Q3_K is now running at ~18.5 ms / token on CUDA, so the gap to Q4_0 is only 10%. It seems memory acccess pattern is more important for performance than the amount of computation the kernel does. * A slightly daster Q4_K AVX2 dot product For perplexity, where we are less memory bound, time per pass drops by ~5%. Barely measurable difference for single token prediction. * A slightly faster ARM_NEON A4_K dot product * Minor * Fix quantization error test We cannot possibly be expecting rmse < 0.002 for 2- and 3-bit quantization variants. * Fix docker build I have been sloppy with vector reinterpret casts on ARM_NEON. It seems clang is very forgiving in that regard. * Added forgotten ggml.o dependence on k_quants.h to the Makefile * Had unintentionally committed the Makefile with -Ofast enabled * ggml : rename k_quants -> ggml-quants-k, use lowercase in code --------- Co-authored-by: Iwan Kawrakow <iwan.kawrakow@gmail.com> Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
2023-06-05 19:56:18 +00:00
#include <math.h>
#include <string.h>
#include <assert.h>
ggml : quantization refactoring (#3833) * ggml : factor all quantization code in ggml-quants ggml-ci * ggml-quants : fix Zig and Swift builds + quantize tool ggml-ci * quantize : --pure option for disabling k-quant mixtures --------- Co-authored-by: cebtenzzre <cebtenzzre@gmail.com>
2023-10-29 16:32:28 +00:00
#include <float.h>
2-bit quantizations (#4897) * imatrix: load * imatrix: WIP * imatrix: Add Q2_K quantization * imatrix: also guard against Q2_K_S quantization without importance matrix * imatrix: guard even more against low-bit quantization misuse --------- Co-authored-by: Iwan Kawrakow <iwan.kawrakow@gmail.com>
2024-01-14 07:45:56 +00:00
#include <stdlib.h> // for qsort
#include <stdio.h> // for GGML_ASSERT
ggml : add SOTA 2,3,4,5,6 bit k-quantizations (#1684) * Starting to add k-quantization to ggml I think it is better to have quantization separate from ggml. For now just adding the k-quants there, but it would be better to also factor out the existing ggml quantizations. * Adding Q3_K and Q8_K (de)-quantization * Q3_K now working on CUDA and AVX2/scalar CUDA is not ideal - ~50% slower than Q4_0 for single token prediction, about the same in batch mode (perplexity). CPU single token is ~55 ms (on Ryzen 7950X). * Some improvement for Q3_K on CUDA It is now ~22.5 ms/token on my GPU, so ~30% slower than Q4_0. * Some more CUDA optimizations for Q3_K Single token is now 20.5 ms/token (~20% slower than Q4_0). Perplexity is on par with Q4_0. * Adding Q4_K - scalar, AVX2, CUDA Performance is the same or perhaps very slightly better than Q4_0 on the CPU. On the GPU, single token prediction is ~10% better than Q4_0, batch mode (perplexity is about the same). * Adding Q6_K - scalar, AVX2, CUDA Performance is ~40% lower compared to Q4_K on the CPU. This is to be expected, considering that we are memory bound on the CPU and the 6-bit model is ~44% larger than the 4-bit. On the GPU, single token prediction is ~6% lower than Q4_0, batch mode (perplexity) is even closer (but still slower). * Adding Q5_K - scalar, AVX2, CUDA Performance is ~20% lower compared to Q4_K on the CPU. This is to be expected, considering that we are memory bound on the CPU and the 5-bit model is ~22% larger than the 4-bit. On the GPU, single token prediction is about the same as Q4_0 for both, single token and batch prediction. * Per convention, all QX_K quantizations use Q5_K for output.weight * Adding quantization mixes * Quantization mixes: didn't quite get what I wanted in the last commit * Q4_K dot product for ARM_NEON * Q6_K dot product for ARM_NEON * Q5_K dot product for ARM_NEON * Adding Q3_K dot for ARM_NEON It is 22% slower than Q4_K, despite the smaller model size. On x86_64, where we are memory bound, the Q3_K model is quite a bit faster than Q4_K. * A very slightly faster ARM_NEON Q3_K dot * Adding Q2_K - just CUDA for now Token prediction is pretty good - about 15.5 ms on a RTX 4080. Perplexity is about the same as Q4_K. * Adding scalar and AVX2 Q2_K dot * Adding ARM_NEON Q2_K dot About the same performance as Q4_K. * A slightly faster ARM_NEON Q2_K dot Single token prediction is now ~36 ms on M2 Max. The code is much simpler too. * Fixed bug in Q2_K CUDA dot product kernel Stranegly enough, for the few prompts I tried with the 7B model the responses looked perfectly reasonable. Only realized something is not quite right when I tried the larger models and started getting nonse back. In any case, Q2_K single token evaluation time on an RTX 4080 in a Ryzen7950X box iusing CUDA and model fully loaded on the GPU are ~15.5 ms for 7B, ~25.4 ms for 13B, and ~55.8 ms for 30B. The max number of layers that fit in VRAM for The 65B is 32. With that, we get ~330 ms per token, which is not that much faster than just running on the CPU (~470 ms per token). * Don't print zeros/NaNs when no count histogram has been collected * A 10% faster CUDA vector dot kernel for Q3_K Q3_K is now running at ~18.5 ms / token on CUDA, so the gap to Q4_0 is only 10%. It seems memory acccess pattern is more important for performance than the amount of computation the kernel does. * A slightly daster Q4_K AVX2 dot product For perplexity, where we are less memory bound, time per pass drops by ~5%. Barely measurable difference for single token prediction. * A slightly faster ARM_NEON A4_K dot product * Minor * Fix quantization error test We cannot possibly be expecting rmse < 0.002 for 2- and 3-bit quantization variants. * Fix docker build I have been sloppy with vector reinterpret casts on ARM_NEON. It seems clang is very forgiving in that regard. * Added forgotten ggml.o dependence on k_quants.h to the Makefile * Had unintentionally committed the Makefile with -Ofast enabled * ggml : rename k_quants -> ggml-quants-k, use lowercase in code --------- Co-authored-by: Iwan Kawrakow <iwan.kawrakow@gmail.com> Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
2023-06-05 19:56:18 +00:00
#ifdef __ARM_NEON
// if YCM cannot find <arm_neon.h>, make a symbolic link to it, for example:
//
// $ ln -sfn /Library/Developer/CommandLineTools/usr/lib/clang/13.1.6/include/arm_neon.h ./src/
//
#include <arm_neon.h>
#else
#ifdef __wasm_simd128__
#include <wasm_simd128.h>
#else
build : support ppc64le build for make and CMake (#3963) * build: support ppc64le build for make and CMake * build: keep __POWER9_VECTOR__ ifdef and extend with __powerpc64__ Co-authored-by: Georgi Gerganov <ggerganov@gmail.com> --------- Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
2023-11-17 16:11:23 +00:00
#if defined(__POWER9_VECTOR__) || defined(__powerpc64__)
ggml : add SOTA 2,3,4,5,6 bit k-quantizations (#1684) * Starting to add k-quantization to ggml I think it is better to have quantization separate from ggml. For now just adding the k-quants there, but it would be better to also factor out the existing ggml quantizations. * Adding Q3_K and Q8_K (de)-quantization * Q3_K now working on CUDA and AVX2/scalar CUDA is not ideal - ~50% slower than Q4_0 for single token prediction, about the same in batch mode (perplexity). CPU single token is ~55 ms (on Ryzen 7950X). * Some improvement for Q3_K on CUDA It is now ~22.5 ms/token on my GPU, so ~30% slower than Q4_0. * Some more CUDA optimizations for Q3_K Single token is now 20.5 ms/token (~20% slower than Q4_0). Perplexity is on par with Q4_0. * Adding Q4_K - scalar, AVX2, CUDA Performance is the same or perhaps very slightly better than Q4_0 on the CPU. On the GPU, single token prediction is ~10% better than Q4_0, batch mode (perplexity is about the same). * Adding Q6_K - scalar, AVX2, CUDA Performance is ~40% lower compared to Q4_K on the CPU. This is to be expected, considering that we are memory bound on the CPU and the 6-bit model is ~44% larger than the 4-bit. On the GPU, single token prediction is ~6% lower than Q4_0, batch mode (perplexity) is even closer (but still slower). * Adding Q5_K - scalar, AVX2, CUDA Performance is ~20% lower compared to Q4_K on the CPU. This is to be expected, considering that we are memory bound on the CPU and the 5-bit model is ~22% larger than the 4-bit. On the GPU, single token prediction is about the same as Q4_0 for both, single token and batch prediction. * Per convention, all QX_K quantizations use Q5_K for output.weight * Adding quantization mixes * Quantization mixes: didn't quite get what I wanted in the last commit * Q4_K dot product for ARM_NEON * Q6_K dot product for ARM_NEON * Q5_K dot product for ARM_NEON * Adding Q3_K dot for ARM_NEON It is 22% slower than Q4_K, despite the smaller model size. On x86_64, where we are memory bound, the Q3_K model is quite a bit faster than Q4_K. * A very slightly faster ARM_NEON Q3_K dot * Adding Q2_K - just CUDA for now Token prediction is pretty good - about 15.5 ms on a RTX 4080. Perplexity is about the same as Q4_K. * Adding scalar and AVX2 Q2_K dot * Adding ARM_NEON Q2_K dot About the same performance as Q4_K. * A slightly faster ARM_NEON Q2_K dot Single token prediction is now ~36 ms on M2 Max. The code is much simpler too. * Fixed bug in Q2_K CUDA dot product kernel Stranegly enough, for the few prompts I tried with the 7B model the responses looked perfectly reasonable. Only realized something is not quite right when I tried the larger models and started getting nonse back. In any case, Q2_K single token evaluation time on an RTX 4080 in a Ryzen7950X box iusing CUDA and model fully loaded on the GPU are ~15.5 ms for 7B, ~25.4 ms for 13B, and ~55.8 ms for 30B. The max number of layers that fit in VRAM for The 65B is 32. With that, we get ~330 ms per token, which is not that much faster than just running on the CPU (~470 ms per token). * Don't print zeros/NaNs when no count histogram has been collected * A 10% faster CUDA vector dot kernel for Q3_K Q3_K is now running at ~18.5 ms / token on CUDA, so the gap to Q4_0 is only 10%. It seems memory acccess pattern is more important for performance than the amount of computation the kernel does. * A slightly daster Q4_K AVX2 dot product For perplexity, where we are less memory bound, time per pass drops by ~5%. Barely measurable difference for single token prediction. * A slightly faster ARM_NEON A4_K dot product * Minor * Fix quantization error test We cannot possibly be expecting rmse < 0.002 for 2- and 3-bit quantization variants. * Fix docker build I have been sloppy with vector reinterpret casts on ARM_NEON. It seems clang is very forgiving in that regard. * Added forgotten ggml.o dependence on k_quants.h to the Makefile * Had unintentionally committed the Makefile with -Ofast enabled * ggml : rename k_quants -> ggml-quants-k, use lowercase in code --------- Co-authored-by: Iwan Kawrakow <iwan.kawrakow@gmail.com> Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
2023-06-05 19:56:18 +00:00
#include <altivec.h>
#undef bool
#define bool _Bool
#else
#if defined(_MSC_VER) || defined(__MINGW32__)
#include <intrin.h>
#else
ggml : sync (im2col, GPU conv, 32-bit arm compat) (#4060) ggml-ci
2023-11-13 14:55:52 +00:00
#if defined(__AVX__) || defined(__AVX2__) || defined(__AVX512F__) || defined(__SSSE3__) || defined(__SSE3__)
#if !defined(__riscv)
ggml : add SOTA 2,3,4,5,6 bit k-quantizations (#1684) * Starting to add k-quantization to ggml I think it is better to have quantization separate from ggml. For now just adding the k-quants there, but it would be better to also factor out the existing ggml quantizations. * Adding Q3_K and Q8_K (de)-quantization * Q3_K now working on CUDA and AVX2/scalar CUDA is not ideal - ~50% slower than Q4_0 for single token prediction, about the same in batch mode (perplexity). CPU single token is ~55 ms (on Ryzen 7950X). * Some improvement for Q3_K on CUDA It is now ~22.5 ms/token on my GPU, so ~30% slower than Q4_0. * Some more CUDA optimizations for Q3_K Single token is now 20.5 ms/token (~20% slower than Q4_0). Perplexity is on par with Q4_0. * Adding Q4_K - scalar, AVX2, CUDA Performance is the same or perhaps very slightly better than Q4_0 on the CPU. On the GPU, single token prediction is ~10% better than Q4_0, batch mode (perplexity is about the same). * Adding Q6_K - scalar, AVX2, CUDA Performance is ~40% lower compared to Q4_K on the CPU. This is to be expected, considering that we are memory bound on the CPU and the 6-bit model is ~44% larger than the 4-bit. On the GPU, single token prediction is ~6% lower than Q4_0, batch mode (perplexity) is even closer (but still slower). * Adding Q5_K - scalar, AVX2, CUDA Performance is ~20% lower compared to Q4_K on the CPU. This is to be expected, considering that we are memory bound on the CPU and the 5-bit model is ~22% larger than the 4-bit. On the GPU, single token prediction is about the same as Q4_0 for both, single token and batch prediction. * Per convention, all QX_K quantizations use Q5_K for output.weight * Adding quantization mixes * Quantization mixes: didn't quite get what I wanted in the last commit * Q4_K dot product for ARM_NEON * Q6_K dot product for ARM_NEON * Q5_K dot product for ARM_NEON * Adding Q3_K dot for ARM_NEON It is 22% slower than Q4_K, despite the smaller model size. On x86_64, where we are memory bound, the Q3_K model is quite a bit faster than Q4_K. * A very slightly faster ARM_NEON Q3_K dot * Adding Q2_K - just CUDA for now Token prediction is pretty good - about 15.5 ms on a RTX 4080. Perplexity is about the same as Q4_K. * Adding scalar and AVX2 Q2_K dot * Adding ARM_NEON Q2_K dot About the same performance as Q4_K. * A slightly faster ARM_NEON Q2_K dot Single token prediction is now ~36 ms on M2 Max. The code is much simpler too. * Fixed bug in Q2_K CUDA dot product kernel Stranegly enough, for the few prompts I tried with the 7B model the responses looked perfectly reasonable. Only realized something is not quite right when I tried the larger models and started getting nonse back. In any case, Q2_K single token evaluation time on an RTX 4080 in a Ryzen7950X box iusing CUDA and model fully loaded on the GPU are ~15.5 ms for 7B, ~25.4 ms for 13B, and ~55.8 ms for 30B. The max number of layers that fit in VRAM for The 65B is 32. With that, we get ~330 ms per token, which is not that much faster than just running on the CPU (~470 ms per token). * Don't print zeros/NaNs when no count histogram has been collected * A 10% faster CUDA vector dot kernel for Q3_K Q3_K is now running at ~18.5 ms / token on CUDA, so the gap to Q4_0 is only 10%. It seems memory acccess pattern is more important for performance than the amount of computation the kernel does. * A slightly daster Q4_K AVX2 dot product For perplexity, where we are less memory bound, time per pass drops by ~5%. Barely measurable difference for single token prediction. * A slightly faster ARM_NEON A4_K dot product * Minor * Fix quantization error test We cannot possibly be expecting rmse < 0.002 for 2- and 3-bit quantization variants. * Fix docker build I have been sloppy with vector reinterpret casts on ARM_NEON. It seems clang is very forgiving in that regard. * Added forgotten ggml.o dependence on k_quants.h to the Makefile * Had unintentionally committed the Makefile with -Ofast enabled * ggml : rename k_quants -> ggml-quants-k, use lowercase in code --------- Co-authored-by: Iwan Kawrakow <iwan.kawrakow@gmail.com> Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
2023-06-05 19:56:18 +00:00
#include <immintrin.h>
#endif
#endif
#endif
#endif
#endif
ggml : sync (im2col, GPU conv, 32-bit arm compat) (#4060) ggml-ci
2023-11-13 14:55:52 +00:00
#endif
ggml : add SOTA 2,3,4,5,6 bit k-quantizations (#1684) * Starting to add k-quantization to ggml I think it is better to have quantization separate from ggml. For now just adding the k-quants there, but it would be better to also factor out the existing ggml quantizations. * Adding Q3_K and Q8_K (de)-quantization * Q3_K now working on CUDA and AVX2/scalar CUDA is not ideal - ~50% slower than Q4_0 for single token prediction, about the same in batch mode (perplexity). CPU single token is ~55 ms (on Ryzen 7950X). * Some improvement for Q3_K on CUDA It is now ~22.5 ms/token on my GPU, so ~30% slower than Q4_0. * Some more CUDA optimizations for Q3_K Single token is now 20.5 ms/token (~20% slower than Q4_0). Perplexity is on par with Q4_0. * Adding Q4_K - scalar, AVX2, CUDA Performance is the same or perhaps very slightly better than Q4_0 on the CPU. On the GPU, single token prediction is ~10% better than Q4_0, batch mode (perplexity is about the same). * Adding Q6_K - scalar, AVX2, CUDA Performance is ~40% lower compared to Q4_K on the CPU. This is to be expected, considering that we are memory bound on the CPU and the 6-bit model is ~44% larger than the 4-bit. On the GPU, single token prediction is ~6% lower than Q4_0, batch mode (perplexity) is even closer (but still slower). * Adding Q5_K - scalar, AVX2, CUDA Performance is ~20% lower compared to Q4_K on the CPU. This is to be expected, considering that we are memory bound on the CPU and the 5-bit model is ~22% larger than the 4-bit. On the GPU, single token prediction is about the same as Q4_0 for both, single token and batch prediction. * Per convention, all QX_K quantizations use Q5_K for output.weight * Adding quantization mixes * Quantization mixes: didn't quite get what I wanted in the last commit * Q4_K dot product for ARM_NEON * Q6_K dot product for ARM_NEON * Q5_K dot product for ARM_NEON * Adding Q3_K dot for ARM_NEON It is 22% slower than Q4_K, despite the smaller model size. On x86_64, where we are memory bound, the Q3_K model is quite a bit faster than Q4_K. * A very slightly faster ARM_NEON Q3_K dot * Adding Q2_K - just CUDA for now Token prediction is pretty good - about 15.5 ms on a RTX 4080. Perplexity is about the same as Q4_K. * Adding scalar and AVX2 Q2_K dot * Adding ARM_NEON Q2_K dot About the same performance as Q4_K. * A slightly faster ARM_NEON Q2_K dot Single token prediction is now ~36 ms on M2 Max. The code is much simpler too. * Fixed bug in Q2_K CUDA dot product kernel Stranegly enough, for the few prompts I tried with the 7B model the responses looked perfectly reasonable. Only realized something is not quite right when I tried the larger models and started getting nonse back. In any case, Q2_K single token evaluation time on an RTX 4080 in a Ryzen7950X box iusing CUDA and model fully loaded on the GPU are ~15.5 ms for 7B, ~25.4 ms for 13B, and ~55.8 ms for 30B. The max number of layers that fit in VRAM for The 65B is 32. With that, we get ~330 ms per token, which is not that much faster than just running on the CPU (~470 ms per token). * Don't print zeros/NaNs when no count histogram has been collected * A 10% faster CUDA vector dot kernel for Q3_K Q3_K is now running at ~18.5 ms / token on CUDA, so the gap to Q4_0 is only 10%. It seems memory acccess pattern is more important for performance than the amount of computation the kernel does. * A slightly daster Q4_K AVX2 dot product For perplexity, where we are less memory bound, time per pass drops by ~5%. Barely measurable difference for single token prediction. * A slightly faster ARM_NEON A4_K dot product * Minor * Fix quantization error test We cannot possibly be expecting rmse < 0.002 for 2- and 3-bit quantization variants. * Fix docker build I have been sloppy with vector reinterpret casts on ARM_NEON. It seems clang is very forgiving in that regard. * Added forgotten ggml.o dependence on k_quants.h to the Makefile * Had unintentionally committed the Makefile with -Ofast enabled * ggml : rename k_quants -> ggml-quants-k, use lowercase in code --------- Co-authored-by: Iwan Kawrakow <iwan.kawrakow@gmail.com> Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
2023-06-05 19:56:18 +00:00
ggml : add RISC-V Vector Support for K-Quants and improved the existing intrinsics (#3453) * Added RVV intrinsics support for Q8 quantize row and also improved the existing dot product function for risc-v. The RVV intrinsics is added for the following quantize row functions quantize_row_q8_0 quantize_row_q8_1 The following dot product functions have also been optimized by using LMUL = 1/2 instead of LMUL = 1 ggml_vec_dot_q4_0_q8_0 ggml_vec_dot_q4_1_q8_1 ggml_vec_dot_q5_0_q8_0 ggml_vec_dot_q5_1_q8_1 And vector initialization in Q5 by temporary array is also replaced by the vid intrinsics Signed-off-by: Ahmad Tameem <ahmad.tameem@10xengineers.ai> * Added RVV intrinsics support for k_quants This adds RISC-V Vector intrinsics support for the following K_quants functions for both QKK = 256 and QKK = 64 ggml_vec_dot_q2_K_q8_K ggml_vec_dot_q3_K_q8_K ggml_vec_dot_q4_K_q8_K ggml_vec_dot_q5_K_q8_K ggml_vec_dot_q6_K_q8_K Signed-off-by: Ahmad Tameem <ahmad.tameem@10xengineers.ai> --------- Signed-off-by: Ahmad Tameem <ahmad.tameem@10xengineers.ai>
2023-10-03 18:38:19 +00:00
#ifdef __riscv_v_intrinsic
#include <riscv_vector.h>
#endif
ggml : add SOTA 2,3,4,5,6 bit k-quantizations (#1684) * Starting to add k-quantization to ggml I think it is better to have quantization separate from ggml. For now just adding the k-quants there, but it would be better to also factor out the existing ggml quantizations. * Adding Q3_K and Q8_K (de)-quantization * Q3_K now working on CUDA and AVX2/scalar CUDA is not ideal - ~50% slower than Q4_0 for single token prediction, about the same in batch mode (perplexity). CPU single token is ~55 ms (on Ryzen 7950X). * Some improvement for Q3_K on CUDA It is now ~22.5 ms/token on my GPU, so ~30% slower than Q4_0. * Some more CUDA optimizations for Q3_K Single token is now 20.5 ms/token (~20% slower than Q4_0). Perplexity is on par with Q4_0. * Adding Q4_K - scalar, AVX2, CUDA Performance is the same or perhaps very slightly better than Q4_0 on the CPU. On the GPU, single token prediction is ~10% better than Q4_0, batch mode (perplexity is about the same). * Adding Q6_K - scalar, AVX2, CUDA Performance is ~40% lower compared to Q4_K on the CPU. This is to be expected, considering that we are memory bound on the CPU and the 6-bit model is ~44% larger than the 4-bit. On the GPU, single token prediction is ~6% lower than Q4_0, batch mode (perplexity) is even closer (but still slower). * Adding Q5_K - scalar, AVX2, CUDA Performance is ~20% lower compared to Q4_K on the CPU. This is to be expected, considering that we are memory bound on the CPU and the 5-bit model is ~22% larger than the 4-bit. On the GPU, single token prediction is about the same as Q4_0 for both, single token and batch prediction. * Per convention, all QX_K quantizations use Q5_K for output.weight * Adding quantization mixes * Quantization mixes: didn't quite get what I wanted in the last commit * Q4_K dot product for ARM_NEON * Q6_K dot product for ARM_NEON * Q5_K dot product for ARM_NEON * Adding Q3_K dot for ARM_NEON It is 22% slower than Q4_K, despite the smaller model size. On x86_64, where we are memory bound, the Q3_K model is quite a bit faster than Q4_K. * A very slightly faster ARM_NEON Q3_K dot * Adding Q2_K - just CUDA for now Token prediction is pretty good - about 15.5 ms on a RTX 4080. Perplexity is about the same as Q4_K. * Adding scalar and AVX2 Q2_K dot * Adding ARM_NEON Q2_K dot About the same performance as Q4_K. * A slightly faster ARM_NEON Q2_K dot Single token prediction is now ~36 ms on M2 Max. The code is much simpler too. * Fixed bug in Q2_K CUDA dot product kernel Stranegly enough, for the few prompts I tried with the 7B model the responses looked perfectly reasonable. Only realized something is not quite right when I tried the larger models and started getting nonse back. In any case, Q2_K single token evaluation time on an RTX 4080 in a Ryzen7950X box iusing CUDA and model fully loaded on the GPU are ~15.5 ms for 7B, ~25.4 ms for 13B, and ~55.8 ms for 30B. The max number of layers that fit in VRAM for The 65B is 32. With that, we get ~330 ms per token, which is not that much faster than just running on the CPU (~470 ms per token). * Don't print zeros/NaNs when no count histogram has been collected * A 10% faster CUDA vector dot kernel for Q3_K Q3_K is now running at ~18.5 ms / token on CUDA, so the gap to Q4_0 is only 10%. It seems memory acccess pattern is more important for performance than the amount of computation the kernel does. * A slightly daster Q4_K AVX2 dot product For perplexity, where we are less memory bound, time per pass drops by ~5%. Barely measurable difference for single token prediction. * A slightly faster ARM_NEON A4_K dot product * Minor * Fix quantization error test We cannot possibly be expecting rmse < 0.002 for 2- and 3-bit quantization variants. * Fix docker build I have been sloppy with vector reinterpret casts on ARM_NEON. It seems clang is very forgiving in that regard. * Added forgotten ggml.o dependence on k_quants.h to the Makefile * Had unintentionally committed the Makefile with -Ofast enabled * ggml : rename k_quants -> ggml-quants-k, use lowercase in code --------- Co-authored-by: Iwan Kawrakow <iwan.kawrakow@gmail.com> Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
2023-06-05 19:56:18 +00:00
#undef MIN
#undef MAX
ggml : sync (im2col, GPU conv, 32-bit arm compat) (#4060) ggml-ci
2023-11-13 14:55:52 +00:00
ggml : add SOTA 2,3,4,5,6 bit k-quantizations (#1684) * Starting to add k-quantization to ggml I think it is better to have quantization separate from ggml. For now just adding the k-quants there, but it would be better to also factor out the existing ggml quantizations. * Adding Q3_K and Q8_K (de)-quantization * Q3_K now working on CUDA and AVX2/scalar CUDA is not ideal - ~50% slower than Q4_0 for single token prediction, about the same in batch mode (perplexity). CPU single token is ~55 ms (on Ryzen 7950X). * Some improvement for Q3_K on CUDA It is now ~22.5 ms/token on my GPU, so ~30% slower than Q4_0. * Some more CUDA optimizations for Q3_K Single token is now 20.5 ms/token (~20% slower than Q4_0). Perplexity is on par with Q4_0. * Adding Q4_K - scalar, AVX2, CUDA Performance is the same or perhaps very slightly better than Q4_0 on the CPU. On the GPU, single token prediction is ~10% better than Q4_0, batch mode (perplexity is about the same). * Adding Q6_K - scalar, AVX2, CUDA Performance is ~40% lower compared to Q4_K on the CPU. This is to be expected, considering that we are memory bound on the CPU and the 6-bit model is ~44% larger than the 4-bit. On the GPU, single token prediction is ~6% lower than Q4_0, batch mode (perplexity) is even closer (but still slower). * Adding Q5_K - scalar, AVX2, CUDA Performance is ~20% lower compared to Q4_K on the CPU. This is to be expected, considering that we are memory bound on the CPU and the 5-bit model is ~22% larger than the 4-bit. On the GPU, single token prediction is about the same as Q4_0 for both, single token and batch prediction. * Per convention, all QX_K quantizations use Q5_K for output.weight * Adding quantization mixes * Quantization mixes: didn't quite get what I wanted in the last commit * Q4_K dot product for ARM_NEON * Q6_K dot product for ARM_NEON * Q5_K dot product for ARM_NEON * Adding Q3_K dot for ARM_NEON It is 22% slower than Q4_K, despite the smaller model size. On x86_64, where we are memory bound, the Q3_K model is quite a bit faster than Q4_K. * A very slightly faster ARM_NEON Q3_K dot * Adding Q2_K - just CUDA for now Token prediction is pretty good - about 15.5 ms on a RTX 4080. Perplexity is about the same as Q4_K. * Adding scalar and AVX2 Q2_K dot * Adding ARM_NEON Q2_K dot About the same performance as Q4_K. * A slightly faster ARM_NEON Q2_K dot Single token prediction is now ~36 ms on M2 Max. The code is much simpler too. * Fixed bug in Q2_K CUDA dot product kernel Stranegly enough, for the few prompts I tried with the 7B model the responses looked perfectly reasonable. Only realized something is not quite right when I tried the larger models and started getting nonse back. In any case, Q2_K single token evaluation time on an RTX 4080 in a Ryzen7950X box iusing CUDA and model fully loaded on the GPU are ~15.5 ms for 7B, ~25.4 ms for 13B, and ~55.8 ms for 30B. The max number of layers that fit in VRAM for The 65B is 32. With that, we get ~330 ms per token, which is not that much faster than just running on the CPU (~470 ms per token). * Don't print zeros/NaNs when no count histogram has been collected * A 10% faster CUDA vector dot kernel for Q3_K Q3_K is now running at ~18.5 ms / token on CUDA, so the gap to Q4_0 is only 10%. It seems memory acccess pattern is more important for performance than the amount of computation the kernel does. * A slightly daster Q4_K AVX2 dot product For perplexity, where we are less memory bound, time per pass drops by ~5%. Barely measurable difference for single token prediction. * A slightly faster ARM_NEON A4_K dot product * Minor * Fix quantization error test We cannot possibly be expecting rmse < 0.002 for 2- and 3-bit quantization variants. * Fix docker build I have been sloppy with vector reinterpret casts on ARM_NEON. It seems clang is very forgiving in that regard. * Added forgotten ggml.o dependence on k_quants.h to the Makefile * Had unintentionally committed the Makefile with -Ofast enabled * ggml : rename k_quants -> ggml-quants-k, use lowercase in code --------- Co-authored-by: Iwan Kawrakow <iwan.kawrakow@gmail.com> Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
2023-06-05 19:56:18 +00:00
#define MIN(a, b) ((a) < (b) ? (a) : (b))
#define MAX(a, b) ((a) > (b) ? (a) : (b))
ggml : workaround for missing _mm256_setr_m128i in GCC < 8 in k_quants.c (#2405)
2023-07-28 18:17:45 +00:00
#define MM256_SET_M128I(a, b) _mm256_insertf128_si256(_mm256_castsi128_si256(b), (a), 1)
ggml : quantization refactoring (#3833) * ggml : factor all quantization code in ggml-quants ggml-ci * ggml-quants : fix Zig and Swift builds + quantize tool ggml-ci * quantize : --pure option for disabling k-quant mixtures --------- Co-authored-by: cebtenzzre <cebtenzzre@gmail.com>
2023-10-29 16:32:28 +00:00
#if defined(__AVX__) || defined(__AVX2__) || defined(__AVX512F__) || defined(__SSSE3__)
// multiply int8_t, add results pairwise twice
static inline __m128i mul_sum_i8_pairs(const __m128i x, const __m128i y) {
// Get absolute values of x vectors
const __m128i ax = _mm_sign_epi8(x, x);
// Sign the values of the y vectors
const __m128i sy = _mm_sign_epi8(y, x);
// Perform multiplication and create 16-bit values
const __m128i dot = _mm_maddubs_epi16(ax, sy);
const __m128i ones = _mm_set1_epi16(1);
return _mm_madd_epi16(ones, dot);
}
ggml : add SOTA 2,3,4,5,6 bit k-quantizations (#1684) * Starting to add k-quantization to ggml I think it is better to have quantization separate from ggml. For now just adding the k-quants there, but it would be better to also factor out the existing ggml quantizations. * Adding Q3_K and Q8_K (de)-quantization * Q3_K now working on CUDA and AVX2/scalar CUDA is not ideal - ~50% slower than Q4_0 for single token prediction, about the same in batch mode (perplexity). CPU single token is ~55 ms (on Ryzen 7950X). * Some improvement for Q3_K on CUDA It is now ~22.5 ms/token on my GPU, so ~30% slower than Q4_0. * Some more CUDA optimizations for Q3_K Single token is now 20.5 ms/token (~20% slower than Q4_0). Perplexity is on par with Q4_0. * Adding Q4_K - scalar, AVX2, CUDA Performance is the same or perhaps very slightly better than Q4_0 on the CPU. On the GPU, single token prediction is ~10% better than Q4_0, batch mode (perplexity is about the same). * Adding Q6_K - scalar, AVX2, CUDA Performance is ~40% lower compared to Q4_K on the CPU. This is to be expected, considering that we are memory bound on the CPU and the 6-bit model is ~44% larger than the 4-bit. On the GPU, single token prediction is ~6% lower than Q4_0, batch mode (perplexity) is even closer (but still slower). * Adding Q5_K - scalar, AVX2, CUDA Performance is ~20% lower compared to Q4_K on the CPU. This is to be expected, considering that we are memory bound on the CPU and the 5-bit model is ~22% larger than the 4-bit. On the GPU, single token prediction is about the same as Q4_0 for both, single token and batch prediction. * Per convention, all QX_K quantizations use Q5_K for output.weight * Adding quantization mixes * Quantization mixes: didn't quite get what I wanted in the last commit * Q4_K dot product for ARM_NEON * Q6_K dot product for ARM_NEON * Q5_K dot product for ARM_NEON * Adding Q3_K dot for ARM_NEON It is 22% slower than Q4_K, despite the smaller model size. On x86_64, where we are memory bound, the Q3_K model is quite a bit faster than Q4_K. * A very slightly faster ARM_NEON Q3_K dot * Adding Q2_K - just CUDA for now Token prediction is pretty good - about 15.5 ms on a RTX 4080. Perplexity is about the same as Q4_K. * Adding scalar and AVX2 Q2_K dot * Adding ARM_NEON Q2_K dot About the same performance as Q4_K. * A slightly faster ARM_NEON Q2_K dot Single token prediction is now ~36 ms on M2 Max. The code is much simpler too. * Fixed bug in Q2_K CUDA dot product kernel Stranegly enough, for the few prompts I tried with the 7B model the responses looked perfectly reasonable. Only realized something is not quite right when I tried the larger models and started getting nonse back. In any case, Q2_K single token evaluation time on an RTX 4080 in a Ryzen7950X box iusing CUDA and model fully loaded on the GPU are ~15.5 ms for 7B, ~25.4 ms for 13B, and ~55.8 ms for 30B. The max number of layers that fit in VRAM for The 65B is 32. With that, we get ~330 ms per token, which is not that much faster than just running on the CPU (~470 ms per token). * Don't print zeros/NaNs when no count histogram has been collected * A 10% faster CUDA vector dot kernel for Q3_K Q3_K is now running at ~18.5 ms / token on CUDA, so the gap to Q4_0 is only 10%. It seems memory acccess pattern is more important for performance than the amount of computation the kernel does. * A slightly daster Q4_K AVX2 dot product For perplexity, where we are less memory bound, time per pass drops by ~5%. Barely measurable difference for single token prediction. * A slightly faster ARM_NEON A4_K dot product * Minor * Fix quantization error test We cannot possibly be expecting rmse < 0.002 for 2- and 3-bit quantization variants. * Fix docker build I have been sloppy with vector reinterpret casts on ARM_NEON. It seems clang is very forgiving in that regard. * Added forgotten ggml.o dependence on k_quants.h to the Makefile * Had unintentionally committed the Makefile with -Ofast enabled * ggml : rename k_quants -> ggml-quants-k, use lowercase in code --------- Co-authored-by: Iwan Kawrakow <iwan.kawrakow@gmail.com> Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
2023-06-05 19:56:18 +00:00
ggml : quantization refactoring (#3833) * ggml : factor all quantization code in ggml-quants ggml-ci * ggml-quants : fix Zig and Swift builds + quantize tool ggml-ci * quantize : --pure option for disabling k-quant mixtures --------- Co-authored-by: cebtenzzre <cebtenzzre@gmail.com>
2023-10-29 16:32:28 +00:00
#if __AVX__ || __AVX2__ || __AVX512F__
// horizontally add 8 floats
static inline float hsum_float_8(const __m256 x) {
__m128 res = _mm256_extractf128_ps(x, 1);
res = _mm_add_ps(res, _mm256_castps256_ps128(x));
res = _mm_add_ps(res, _mm_movehl_ps(res, res));
res = _mm_add_ss(res, _mm_movehdup_ps(res));
return _mm_cvtss_f32(res);
ggml : add SOTA 2,3,4,5,6 bit k-quantizations (#1684) * Starting to add k-quantization to ggml I think it is better to have quantization separate from ggml. For now just adding the k-quants there, but it would be better to also factor out the existing ggml quantizations. * Adding Q3_K and Q8_K (de)-quantization * Q3_K now working on CUDA and AVX2/scalar CUDA is not ideal - ~50% slower than Q4_0 for single token prediction, about the same in batch mode (perplexity). CPU single token is ~55 ms (on Ryzen 7950X). * Some improvement for Q3_K on CUDA It is now ~22.5 ms/token on my GPU, so ~30% slower than Q4_0. * Some more CUDA optimizations for Q3_K Single token is now 20.5 ms/token (~20% slower than Q4_0). Perplexity is on par with Q4_0. * Adding Q4_K - scalar, AVX2, CUDA Performance is the same or perhaps very slightly better than Q4_0 on the CPU. On the GPU, single token prediction is ~10% better than Q4_0, batch mode (perplexity is about the same). * Adding Q6_K - scalar, AVX2, CUDA Performance is ~40% lower compared to Q4_K on the CPU. This is to be expected, considering that we are memory bound on the CPU and the 6-bit model is ~44% larger than the 4-bit. On the GPU, single token prediction is ~6% lower than Q4_0, batch mode (perplexity) is even closer (but still slower). * Adding Q5_K - scalar, AVX2, CUDA Performance is ~20% lower compared to Q4_K on the CPU. This is to be expected, considering that we are memory bound on the CPU and the 5-bit model is ~22% larger than the 4-bit. On the GPU, single token prediction is about the same as Q4_0 for both, single token and batch prediction. * Per convention, all QX_K quantizations use Q5_K for output.weight * Adding quantization mixes * Quantization mixes: didn't quite get what I wanted in the last commit * Q4_K dot product for ARM_NEON * Q6_K dot product for ARM_NEON * Q5_K dot product for ARM_NEON * Adding Q3_K dot for ARM_NEON It is 22% slower than Q4_K, despite the smaller model size. On x86_64, where we are memory bound, the Q3_K model is quite a bit faster than Q4_K. * A very slightly faster ARM_NEON Q3_K dot * Adding Q2_K - just CUDA for now Token prediction is pretty good - about 15.5 ms on a RTX 4080. Perplexity is about the same as Q4_K. * Adding scalar and AVX2 Q2_K dot * Adding ARM_NEON Q2_K dot About the same performance as Q4_K. * A slightly faster ARM_NEON Q2_K dot Single token prediction is now ~36 ms on M2 Max. The code is much simpler too. * Fixed bug in Q2_K CUDA dot product kernel Stranegly enough, for the few prompts I tried with the 7B model the responses looked perfectly reasonable. Only realized something is not quite right when I tried the larger models and started getting nonse back. In any case, Q2_K single token evaluation time on an RTX 4080 in a Ryzen7950X box iusing CUDA and model fully loaded on the GPU are ~15.5 ms for 7B, ~25.4 ms for 13B, and ~55.8 ms for 30B. The max number of layers that fit in VRAM for The 65B is 32. With that, we get ~330 ms per token, which is not that much faster than just running on the CPU (~470 ms per token). * Don't print zeros/NaNs when no count histogram has been collected * A 10% faster CUDA vector dot kernel for Q3_K Q3_K is now running at ~18.5 ms / token on CUDA, so the gap to Q4_0 is only 10%. It seems memory acccess pattern is more important for performance than the amount of computation the kernel does. * A slightly daster Q4_K AVX2 dot product For perplexity, where we are less memory bound, time per pass drops by ~5%. Barely measurable difference for single token prediction. * A slightly faster ARM_NEON A4_K dot product * Minor * Fix quantization error test We cannot possibly be expecting rmse < 0.002 for 2- and 3-bit quantization variants. * Fix docker build I have been sloppy with vector reinterpret casts on ARM_NEON. It seems clang is very forgiving in that regard. * Added forgotten ggml.o dependence on k_quants.h to the Makefile * Had unintentionally committed the Makefile with -Ofast enabled * ggml : rename k_quants -> ggml-quants-k, use lowercase in code --------- Co-authored-by: Iwan Kawrakow <iwan.kawrakow@gmail.com> Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
2023-06-05 19:56:18 +00:00
}
ggml : quantization refactoring (#3833) * ggml : factor all quantization code in ggml-quants ggml-ci * ggml-quants : fix Zig and Swift builds + quantize tool ggml-ci * quantize : --pure option for disabling k-quant mixtures --------- Co-authored-by: cebtenzzre <cebtenzzre@gmail.com>
2023-10-29 16:32:28 +00:00
// horizontally add 8 int32_t
static inline int hsum_i32_8(const __m256i a) {
const __m128i sum128 = _mm_add_epi32(_mm256_castsi256_si128(a), _mm256_extractf128_si256(a, 1));
const __m128i hi64 = _mm_unpackhi_epi64(sum128, sum128);
const __m128i sum64 = _mm_add_epi32(hi64, sum128);
const __m128i hi32 = _mm_shuffle_epi32(sum64, _MM_SHUFFLE(2, 3, 0, 1));
return _mm_cvtsi128_si32(_mm_add_epi32(sum64, hi32));
}
// horizontally add 4 int32_t
static inline int hsum_i32_4(const __m128i a) {
const __m128i hi64 = _mm_unpackhi_epi64(a, a);
const __m128i sum64 = _mm_add_epi32(hi64, a);
const __m128i hi32 = _mm_shuffle_epi32(sum64, _MM_SHUFFLE(2, 3, 0, 1));
return _mm_cvtsi128_si32(_mm_add_epi32(sum64, hi32));
}
#if defined(__AVX2__) || defined(__AVX512F__)
// spread 32 bits to 32 bytes { 0x00, 0xFF }
static inline __m256i bytes_from_bits_32(const uint8_t * x) {
uint32_t x32;
memcpy(&x32, x, sizeof(uint32_t));
const __m256i shuf_mask = _mm256_set_epi64x(
0x0303030303030303, 0x0202020202020202,
0x0101010101010101, 0x0000000000000000);
__m256i bytes = _mm256_shuffle_epi8(_mm256_set1_epi32(x32), shuf_mask);
const __m256i bit_mask = _mm256_set1_epi64x(0x7fbfdfeff7fbfdfe);
bytes = _mm256_or_si256(bytes, bit_mask);
return _mm256_cmpeq_epi8(bytes, _mm256_set1_epi64x(-1));
}
// Unpack 32 4-bit fields into 32 bytes
// The output vector contains 32 bytes, each one in [ 0 .. 15 ] interval
static inline __m256i bytes_from_nibbles_32(const uint8_t * rsi)
{
const __m128i tmp = _mm_loadu_si128((const __m128i *)rsi);
const __m256i bytes = MM256_SET_M128I(_mm_srli_epi16(tmp, 4), tmp);
const __m256i lowMask = _mm256_set1_epi8( 0xF );
return _mm256_and_si256(lowMask, bytes);
}
// add int16_t pairwise and return as float vector
static inline __m256 sum_i16_pairs_float(const __m256i x) {
const __m256i ones = _mm256_set1_epi16(1);
const __m256i summed_pairs = _mm256_madd_epi16(ones, x);
return _mm256_cvtepi32_ps(summed_pairs);
}
static inline __m256 mul_sum_us8_pairs_float(const __m256i ax, const __m256i sy) {
#if __AVXVNNI__
const __m256i zero = _mm256_setzero_si256();
const __m256i summed_pairs = _mm256_dpbusd_epi32(zero, ax, sy);
return _mm256_cvtepi32_ps(summed_pairs);
#else
// Perform multiplication and create 16-bit values
const __m256i dot = _mm256_maddubs_epi16(ax, sy);
return sum_i16_pairs_float(dot);
#endif
}
// multiply int8_t, add results pairwise twice and return as float vector
static inline __m256 mul_sum_i8_pairs_float(const __m256i x, const __m256i y) {
#if __AVXVNNIINT8__
const __m256i zero = _mm256_setzero_si256();
const __m256i summed_pairs = _mm256_dpbssd_epi32(zero, x, y);
return _mm256_cvtepi32_ps(summed_pairs);
#else
// Get absolute values of x vectors
const __m256i ax = _mm256_sign_epi8(x, x);
// Sign the values of the y vectors
const __m256i sy = _mm256_sign_epi8(y, x);
return mul_sum_us8_pairs_float(ax, sy);
#endif
}
static inline __m128i packNibbles( __m256i bytes )
{
// Move bits within 16-bit lanes from 0000_abcd_0000_efgh into 0000_0000_abcd_efgh
#if __AVX512F__
const __m256i bytes_srli_4 = _mm256_srli_epi16(bytes, 4); // 0000_0000_abcd_0000
bytes = _mm256_or_si256(bytes, bytes_srli_4); // 0000_abcd_abcd_efgh
return _mm256_cvtepi16_epi8(bytes); // abcd_efgh
#else
const __m256i lowByte = _mm256_set1_epi16( 0xFF );
__m256i high = _mm256_andnot_si256( lowByte, bytes );
__m256i low = _mm256_and_si256( lowByte, bytes );
high = _mm256_srli_epi16( high, 4 );
bytes = _mm256_or_si256( low, high );
// Compress uint16_t lanes into bytes
__m128i r0 = _mm256_castsi256_si128( bytes );
__m128i r1 = _mm256_extracti128_si256( bytes, 1 );
return _mm_packus_epi16( r0, r1 );
#endif
}
#elif defined(__AVX__)
// spread 32 bits to 32 bytes { 0x00, 0xFF }
static inline __m256i bytes_from_bits_32(const uint8_t * x) {
uint32_t x32;
memcpy(&x32, x, sizeof(uint32_t));
const __m128i shuf_maskl = _mm_set_epi64x(0x0101010101010101, 0x0000000000000000);
const __m128i shuf_maskh = _mm_set_epi64x(0x0303030303030303, 0x0202020202020202);
__m128i bytesl = _mm_shuffle_epi8(_mm_set1_epi32(x32), shuf_maskl);
__m128i bytesh = _mm_shuffle_epi8(_mm_set1_epi32(x32), shuf_maskh);
const __m128i bit_mask = _mm_set1_epi64x(0x7fbfdfeff7fbfdfe);
bytesl = _mm_or_si128(bytesl, bit_mask);
bytesh = _mm_or_si128(bytesh, bit_mask);
bytesl = _mm_cmpeq_epi8(bytesl, _mm_set1_epi64x(-1));
bytesh = _mm_cmpeq_epi8(bytesh, _mm_set1_epi64x(-1));
return MM256_SET_M128I(bytesh, bytesl);
}
// Unpack 32 4-bit fields into 32 bytes
// The output vector contains 32 bytes, each one in [ 0 .. 15 ] interval
static inline __m256i bytes_from_nibbles_32(const uint8_t * rsi)
{
// Load 16 bytes from memory
__m128i tmpl = _mm_loadu_si128((const __m128i *)rsi);
__m128i tmph = _mm_srli_epi16(tmpl, 4);
const __m128i lowMask = _mm_set1_epi8(0xF);
tmpl = _mm_and_si128(lowMask, tmpl);
tmph = _mm_and_si128(lowMask, tmph);
return MM256_SET_M128I(tmph, tmpl);
}
// add int16_t pairwise and return as float vector
static inline __m256 sum_i16_pairs_float(const __m128i xh, const __m128i xl) {
const __m128i ones = _mm_set1_epi16(1);
const __m128i summed_pairsl = _mm_madd_epi16(ones, xl);
const __m128i summed_pairsh = _mm_madd_epi16(ones, xh);
const __m256i summed_pairs = MM256_SET_M128I(summed_pairsh, summed_pairsl);
return _mm256_cvtepi32_ps(summed_pairs);
}
static inline __m256 mul_sum_us8_pairs_float(const __m256i ax, const __m256i sy) {
const __m128i axl = _mm256_castsi256_si128(ax);
const __m128i axh = _mm256_extractf128_si256(ax, 1);
const __m128i syl = _mm256_castsi256_si128(sy);
const __m128i syh = _mm256_extractf128_si256(sy, 1);
// Perform multiplication and create 16-bit values
const __m128i dotl = _mm_maddubs_epi16(axl, syl);
const __m128i doth = _mm_maddubs_epi16(axh, syh);
return sum_i16_pairs_float(doth, dotl);
}
// multiply int8_t, add results pairwise twice and return as float vector
static inline __m256 mul_sum_i8_pairs_float(const __m256i x, const __m256i y) {
const __m128i xl = _mm256_castsi256_si128(x);
const __m128i xh = _mm256_extractf128_si256(x, 1);
const __m128i yl = _mm256_castsi256_si128(y);
const __m128i yh = _mm256_extractf128_si256(y, 1);
// Get absolute values of x vectors
const __m128i axl = _mm_sign_epi8(xl, xl);
const __m128i axh = _mm_sign_epi8(xh, xh);
// Sign the values of the y vectors
const __m128i syl = _mm_sign_epi8(yl, xl);
const __m128i syh = _mm_sign_epi8(yh, xh);
// Perform multiplication and create 16-bit values
const __m128i dotl = _mm_maddubs_epi16(axl, syl);
const __m128i doth = _mm_maddubs_epi16(axh, syh);
return sum_i16_pairs_float(doth, dotl);
}
static inline __m128i packNibbles( __m128i bytes1, __m128i bytes2 )
{
// Move bits within 16-bit lanes from 0000_abcd_0000_efgh into 0000_0000_abcd_efgh
const __m128i lowByte = _mm_set1_epi16( 0xFF );
__m128i high = _mm_andnot_si128( lowByte, bytes1 );
__m128i low = _mm_and_si128( lowByte, bytes1 );
high = _mm_srli_epi16( high, 4 );
bytes1 = _mm_or_si128( low, high );
high = _mm_andnot_si128( lowByte, bytes2 );
low = _mm_and_si128( lowByte, bytes2 );
high = _mm_srli_epi16( high, 4 );
bytes2 = _mm_or_si128( low, high );
return _mm_packus_epi16( bytes1, bytes2);
}
#endif
#elif defined(__SSSE3__)
// horizontally add 4x4 floats
static inline float hsum_float_4x4(const __m128 a, const __m128 b, const __m128 c, const __m128 d) {
__m128 res_0 =_mm_hadd_ps(a, b);
__m128 res_1 =_mm_hadd_ps(c, d);
__m128 res =_mm_hadd_ps(res_0, res_1);
res =_mm_hadd_ps(res, res);
res =_mm_hadd_ps(res, res);
return _mm_cvtss_f32(res);
}
#endif // __AVX__ || __AVX2__ || __AVX512F__
#endif // defined(__AVX__) || defined(__AVX2__) || defined(__AVX512F__) || defined(__SSSE3__)
#if defined(__ARM_NEON)
#if !defined(__aarch64__)
ggml : sync (im2col, GPU conv, 32-bit arm compat) (#4060) ggml-ci
2023-11-13 14:55:52 +00:00
// 64-bit compatibility
// vaddvq_s16
// vpaddq_s16
ggml : fix 32-bit ARM compat for IQ2_XS (whisper/1758) * ggml : fix 32-bit ARM compat * ggml : fix fix * ggml : fix fix fix
2024-01-12 12:02:30 +00:00
// vpaddq_s32
ggml : sync (im2col, GPU conv, 32-bit arm compat) (#4060) ggml-ci
2023-11-13 14:55:52 +00:00
// vaddvq_s32
// vaddvq_f32
// vmaxvq_f32
// vcvtnq_s32_f32
ggml : fix 32-bit ARM compat for IQ2_XS (whisper/1758) * ggml : fix 32-bit ARM compat * ggml : fix fix * ggml : fix fix fix
2024-01-12 12:02:30 +00:00
// vzip1_u8
// vzip2_u8
ggml : sync (im2col, GPU conv, 32-bit arm compat) (#4060) ggml-ci
2023-11-13 14:55:52 +00:00
inline static int32_t vaddvq_s16(int16x8_t v) {
return
(int32_t)vgetq_lane_s16(v, 0) + (int32_t)vgetq_lane_s16(v, 1) +
(int32_t)vgetq_lane_s16(v, 2) + (int32_t)vgetq_lane_s16(v, 3) +
(int32_t)vgetq_lane_s16(v, 4) + (int32_t)vgetq_lane_s16(v, 5) +
(int32_t)vgetq_lane_s16(v, 6) + (int32_t)vgetq_lane_s16(v, 7);
}
inline static int16x8_t vpaddq_s16(int16x8_t a, int16x8_t b) {
int16x4_t a0 = vpadd_s16(vget_low_s16(a), vget_high_s16(a));
int16x4_t b0 = vpadd_s16(vget_low_s16(b), vget_high_s16(b));
return vcombine_s16(a0, b0);
}
ggml : fix 32-bit ARM compat for IQ2_XS (whisper/1758) * ggml : fix 32-bit ARM compat * ggml : fix fix * ggml : fix fix fix
2024-01-12 12:02:30 +00:00
inline static int32x4_t vpaddq_s32(int32x4_t a, int32x4_t b) {
int32x2_t a0 = vpadd_s32(vget_low_s32(a), vget_high_s32(a));
int32x2_t b0 = vpadd_s32(vget_low_s32(b), vget_high_s32(b));
return vcombine_s32(a0, b0);
}
ggml : quantization refactoring (#3833) * ggml : factor all quantization code in ggml-quants ggml-ci * ggml-quants : fix Zig and Swift builds + quantize tool ggml-ci * quantize : --pure option for disabling k-quant mixtures --------- Co-authored-by: cebtenzzre <cebtenzzre@gmail.com>
2023-10-29 16:32:28 +00:00
inline static int32_t vaddvq_s32(int32x4_t v) {
return vgetq_lane_s32(v, 0) + vgetq_lane_s32(v, 1) + vgetq_lane_s32(v, 2) + vgetq_lane_s32(v, 3);
}
inline static float vaddvq_f32(float32x4_t v) {
return vgetq_lane_f32(v, 0) + vgetq_lane_f32(v, 1) + vgetq_lane_f32(v, 2) + vgetq_lane_f32(v, 3);
}
inline static float vmaxvq_f32(float32x4_t v) {
return
MAX(MAX(vgetq_lane_f32(v, 0), vgetq_lane_f32(v, 1)),
MAX(vgetq_lane_f32(v, 2), vgetq_lane_f32(v, 3)));
}
inline static int32x4_t vcvtnq_s32_f32(float32x4_t v) {
int32x4_t res;
res[0] = roundf(vgetq_lane_f32(v, 0));
res[1] = roundf(vgetq_lane_f32(v, 1));
res[2] = roundf(vgetq_lane_f32(v, 2));
res[3] = roundf(vgetq_lane_f32(v, 3));
return res;
}
ggml : fix 32-bit ARM compat for IQ2_XS (whisper/1758) * ggml : fix 32-bit ARM compat * ggml : fix fix * ggml : fix fix fix
2024-01-12 12:02:30 +00:00
inline static uint8x8_t vzip1_u8(uint8x8_t a, uint8x8_t b) {
uint8x8_t res;
res[0] = a[0]; res[1] = b[0];
res[2] = a[1]; res[3] = b[1];
res[4] = a[2]; res[5] = b[2];
res[6] = a[3]; res[7] = b[3];
return res;
}
inline static uint8x8_t vzip2_u8(uint8x8_t a, uint8x8_t b) {
uint8x8_t res;
res[0] = a[4]; res[1] = b[4];
res[2] = a[5]; res[3] = b[5];
res[4] = a[6]; res[5] = b[6];
res[6] = a[7]; res[7] = b[7];
return res;
}
ggml : sync (im2col, GPU conv, 32-bit arm compat) (#4060) ggml-ci
2023-11-13 14:55:52 +00:00
// vld1q_s16_x2
// vld1q_u8_x2
// vld1q_u8_x4
// vld1q_s8_x2
// vld1q_s8_x4
// TODO: double-check these work correctly
typedef struct ggml_int16x8x2_t {
int16x8_t val[2];
} ggml_int16x8x2_t;
inline static ggml_int16x8x2_t ggml_vld1q_s16_x2(const int16_t * ptr) {
ggml_int16x8x2_t res;
res.val[0] = vld1q_s16(ptr + 0);
res.val[1] = vld1q_s16(ptr + 8);
return res;
}
typedef struct ggml_uint8x16x2_t {
uint8x16_t val[2];
} ggml_uint8x16x2_t;
inline static ggml_uint8x16x2_t ggml_vld1q_u8_x2(const uint8_t * ptr) {
ggml_uint8x16x2_t res;
res.val[0] = vld1q_u8(ptr + 0);
res.val[1] = vld1q_u8(ptr + 16);
return res;
}
typedef struct ggml_uint8x16x4_t {
uint8x16_t val[4];
} ggml_uint8x16x4_t;
inline static ggml_uint8x16x4_t ggml_vld1q_u8_x4(const uint8_t * ptr) {
ggml_uint8x16x4_t res;
res.val[0] = vld1q_u8(ptr + 0);
res.val[1] = vld1q_u8(ptr + 16);
res.val[2] = vld1q_u8(ptr + 32);
res.val[3] = vld1q_u8(ptr + 48);
return res;
}
typedef struct ggml_int8x16x2_t {
int8x16_t val[2];
} ggml_int8x16x2_t;
inline static ggml_int8x16x2_t ggml_vld1q_s8_x2(const int8_t * ptr) {
ggml_int8x16x2_t res;
res.val[0] = vld1q_s8(ptr + 0);
res.val[1] = vld1q_s8(ptr + 16);
return res;
}
typedef struct ggml_int8x16x4_t {
int8x16_t val[4];
} ggml_int8x16x4_t;
inline static ggml_int8x16x4_t ggml_vld1q_s8_x4(const int8_t * ptr) {
ggml_int8x16x4_t res;
res.val[0] = vld1q_s8(ptr + 0);
res.val[1] = vld1q_s8(ptr + 16);
res.val[2] = vld1q_s8(ptr + 32);
res.val[3] = vld1q_s8(ptr + 48);
return res;
}
#else
#define ggml_int16x8x2_t int16x8x2_t
#define ggml_uint8x16x2_t uint8x16x2_t
#define ggml_uint8x16x4_t uint8x16x4_t
#define ggml_int8x16x2_t int8x16x2_t
#define ggml_int8x16x4_t int8x16x4_t
#define ggml_vld1q_s16_x2 vld1q_s16_x2
#define ggml_vld1q_u8_x2 vld1q_u8_x2
#define ggml_vld1q_u8_x4 vld1q_u8_x4
#define ggml_vld1q_s8_x2 vld1q_s8_x2
#define ggml_vld1q_s8_x4 vld1q_s8_x4
ggml : quantization refactoring (#3833) * ggml : factor all quantization code in ggml-quants ggml-ci * ggml-quants : fix Zig and Swift builds + quantize tool ggml-ci * quantize : --pure option for disabling k-quant mixtures --------- Co-authored-by: cebtenzzre <cebtenzzre@gmail.com>
2023-10-29 16:32:28 +00:00
#endif
ggml : fix dot product for ARM (#4630) ggml-ci
2023-12-27 09:02:13 +00:00
#if !defined(__ARM_FEATURE_DOTPROD)
ggml : add ggml_vdotq_s32 alias (#4715) ggml-ci
2023-12-31 09:43:31 +00:00
inline static int32x4_t ggml_vdotq_s32(int32x4_t acc, int8x16_t a, int8x16_t b) {
ggml : fix dot product for ARM (#4630) ggml-ci
2023-12-27 09:02:13 +00:00
const int16x8_t p0 = vmull_s8(vget_low_s8 (a), vget_low_s8 (b));
const int16x8_t p1 = vmull_s8(vget_high_s8(a), vget_high_s8(b));
return vaddq_s32(acc, vaddq_s32(vpaddlq_s16(p0), vpaddlq_s16(p1)));
}
ggml : add ggml_vdotq_s32 alias (#4715) ggml-ci
2023-12-31 09:43:31 +00:00
#else
#define ggml_vdotq_s32(a, b, c) vdotq_s32(a, b, c)
ggml : fix dot product for ARM (#4630) ggml-ci
2023-12-27 09:02:13 +00:00
#endif
ggml : quantization refactoring (#3833) * ggml : factor all quantization code in ggml-quants ggml-ci * ggml-quants : fix Zig and Swift builds + quantize tool ggml-ci * quantize : --pure option for disabling k-quant mixtures --------- Co-authored-by: cebtenzzre <cebtenzzre@gmail.com>
2023-10-29 16:32:28 +00:00
#endif
#if defined(__ARM_NEON) || defined(__wasm_simd128__)
#define B1(c,s,n) 0x ## n ## c , 0x ## n ## s
#define B2(c,s,n) B1(c,s,n ## c), B1(c,s,n ## s)
#define B3(c,s,n) B2(c,s,n ## c), B2(c,s,n ## s)
#define B4(c,s,n) B3(c,s,n ## c), B3(c,s,n ## s)
#define B5(c,s,n) B4(c,s,n ## c), B4(c,s,n ## s)
#define B6(c,s,n) B5(c,s,n ## c), B5(c,s,n ## s)
#define B7(c,s,n) B6(c,s,n ## c), B6(c,s,n ## s)
#define B8(c,s ) B7(c,s, c), B7(c,s, s)
// precomputed tables for expanding 8bits to 8 bytes:
static const uint64_t table_b2b_0[1 << 8] = { B8(00, 10) }; // ( b) << 4
static const uint64_t table_b2b_1[1 << 8] = { B8(10, 00) }; // (!b) << 4
#endif
// reference implementation for deterministic creation of model files
void quantize_row_q4_0_reference(const float * restrict x, block_q4_0 * restrict y, int k) {
static const int qk = QK4_0;
assert(k % qk == 0);
const int nb = k / qk;
for (int i = 0; i < nb; i++) {
float amax = 0.0f; // absolute max
float max = 0.0f;
for (int j = 0; j < qk; j++) {
const float v = x[i*qk + j];
if (amax < fabsf(v)) {
amax = fabsf(v);
max = v;
}
ggml : add SOTA 2,3,4,5,6 bit k-quantizations (#1684) * Starting to add k-quantization to ggml I think it is better to have quantization separate from ggml. For now just adding the k-quants there, but it would be better to also factor out the existing ggml quantizations. * Adding Q3_K and Q8_K (de)-quantization * Q3_K now working on CUDA and AVX2/scalar CUDA is not ideal - ~50% slower than Q4_0 for single token prediction, about the same in batch mode (perplexity). CPU single token is ~55 ms (on Ryzen 7950X). * Some improvement for Q3_K on CUDA It is now ~22.5 ms/token on my GPU, so ~30% slower than Q4_0. * Some more CUDA optimizations for Q3_K Single token is now 20.5 ms/token (~20% slower than Q4_0). Perplexity is on par with Q4_0. * Adding Q4_K - scalar, AVX2, CUDA Performance is the same or perhaps very slightly better than Q4_0 on the CPU. On the GPU, single token prediction is ~10% better than Q4_0, batch mode (perplexity is about the same). * Adding Q6_K - scalar, AVX2, CUDA Performance is ~40% lower compared to Q4_K on the CPU. This is to be expected, considering that we are memory bound on the CPU and the 6-bit model is ~44% larger than the 4-bit. On the GPU, single token prediction is ~6% lower than Q4_0, batch mode (perplexity) is even closer (but still slower). * Adding Q5_K - scalar, AVX2, CUDA Performance is ~20% lower compared to Q4_K on the CPU. This is to be expected, considering that we are memory bound on the CPU and the 5-bit model is ~22% larger than the 4-bit. On the GPU, single token prediction is about the same as Q4_0 for both, single token and batch prediction. * Per convention, all QX_K quantizations use Q5_K for output.weight * Adding quantization mixes * Quantization mixes: didn't quite get what I wanted in the last commit * Q4_K dot product for ARM_NEON * Q6_K dot product for ARM_NEON * Q5_K dot product for ARM_NEON * Adding Q3_K dot for ARM_NEON It is 22% slower than Q4_K, despite the smaller model size. On x86_64, where we are memory bound, the Q3_K model is quite a bit faster than Q4_K. * A very slightly faster ARM_NEON Q3_K dot * Adding Q2_K - just CUDA for now Token prediction is pretty good - about 15.5 ms on a RTX 4080. Perplexity is about the same as Q4_K. * Adding scalar and AVX2 Q2_K dot * Adding ARM_NEON Q2_K dot About the same performance as Q4_K. * A slightly faster ARM_NEON Q2_K dot Single token prediction is now ~36 ms on M2 Max. The code is much simpler too. * Fixed bug in Q2_K CUDA dot product kernel Stranegly enough, for the few prompts I tried with the 7B model the responses looked perfectly reasonable. Only realized something is not quite right when I tried the larger models and started getting nonse back. In any case, Q2_K single token evaluation time on an RTX 4080 in a Ryzen7950X box iusing CUDA and model fully loaded on the GPU are ~15.5 ms for 7B, ~25.4 ms for 13B, and ~55.8 ms for 30B. The max number of layers that fit in VRAM for The 65B is 32. With that, we get ~330 ms per token, which is not that much faster than just running on the CPU (~470 ms per token). * Don't print zeros/NaNs when no count histogram has been collected * A 10% faster CUDA vector dot kernel for Q3_K Q3_K is now running at ~18.5 ms / token on CUDA, so the gap to Q4_0 is only 10%. It seems memory acccess pattern is more important for performance than the amount of computation the kernel does. * A slightly daster Q4_K AVX2 dot product For perplexity, where we are less memory bound, time per pass drops by ~5%. Barely measurable difference for single token prediction. * A slightly faster ARM_NEON A4_K dot product * Minor * Fix quantization error test We cannot possibly be expecting rmse < 0.002 for 2- and 3-bit quantization variants. * Fix docker build I have been sloppy with vector reinterpret casts on ARM_NEON. It seems clang is very forgiving in that regard. * Added forgotten ggml.o dependence on k_quants.h to the Makefile * Had unintentionally committed the Makefile with -Ofast enabled * ggml : rename k_quants -> ggml-quants-k, use lowercase in code --------- Co-authored-by: Iwan Kawrakow <iwan.kawrakow@gmail.com> Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
2023-06-05 19:56:18 +00:00
}
ggml : quantization refactoring (#3833) * ggml : factor all quantization code in ggml-quants ggml-ci * ggml-quants : fix Zig and Swift builds + quantize tool ggml-ci * quantize : --pure option for disabling k-quant mixtures --------- Co-authored-by: cebtenzzre <cebtenzzre@gmail.com>
2023-10-29 16:32:28 +00:00
const float d = max / -8;
const float id = d ? 1.0f/d : 0.0f;
ggml : move FP16 <-> FP32 code to ggml-impl.h (#3861) * ggml : move FP16 <-> FP32 stuff to ggml-impl.h ggml-ci * tests : fix ARM build * ggml : explicitly initialize deprecated type traits * ggml : add math.h to ggml-impl.h * ggml : remove duplicate static assert macros * ggml : prefix lookup tables with ggml_ ggml-ci * ggml-impl : move extern "C" to start of file
2023-10-30 17:19:15 +00:00
y[i].d = GGML_FP32_TO_FP16(d);
ggml : quantization refactoring (#3833) * ggml : factor all quantization code in ggml-quants ggml-ci * ggml-quants : fix Zig and Swift builds + quantize tool ggml-ci * quantize : --pure option for disabling k-quant mixtures --------- Co-authored-by: cebtenzzre <cebtenzzre@gmail.com>
2023-10-29 16:32:28 +00:00
for (int j = 0; j < qk/2; ++j) {
const float x0 = x[i*qk + 0 + j]*id;
const float x1 = x[i*qk + qk/2 + j]*id;
const uint8_t xi0 = MIN(15, (int8_t)(x0 + 8.5f));
const uint8_t xi1 = MIN(15, (int8_t)(x1 + 8.5f));
y[i].qs[j] = xi0;
y[i].qs[j] |= xi1 << 4;
ggml : add SOTA 2,3,4,5,6 bit k-quantizations (#1684) * Starting to add k-quantization to ggml I think it is better to have quantization separate from ggml. For now just adding the k-quants there, but it would be better to also factor out the existing ggml quantizations. * Adding Q3_K and Q8_K (de)-quantization * Q3_K now working on CUDA and AVX2/scalar CUDA is not ideal - ~50% slower than Q4_0 for single token prediction, about the same in batch mode (perplexity). CPU single token is ~55 ms (on Ryzen 7950X). * Some improvement for Q3_K on CUDA It is now ~22.5 ms/token on my GPU, so ~30% slower than Q4_0. * Some more CUDA optimizations for Q3_K Single token is now 20.5 ms/token (~20% slower than Q4_0). Perplexity is on par with Q4_0. * Adding Q4_K - scalar, AVX2, CUDA Performance is the same or perhaps very slightly better than Q4_0 on the CPU. On the GPU, single token prediction is ~10% better than Q4_0, batch mode (perplexity is about the same). * Adding Q6_K - scalar, AVX2, CUDA Performance is ~40% lower compared to Q4_K on the CPU. This is to be expected, considering that we are memory bound on the CPU and the 6-bit model is ~44% larger than the 4-bit. On the GPU, single token prediction is ~6% lower than Q4_0, batch mode (perplexity) is even closer (but still slower). * Adding Q5_K - scalar, AVX2, CUDA Performance is ~20% lower compared to Q4_K on the CPU. This is to be expected, considering that we are memory bound on the CPU and the 5-bit model is ~22% larger than the 4-bit. On the GPU, single token prediction is about the same as Q4_0 for both, single token and batch prediction. * Per convention, all QX_K quantizations use Q5_K for output.weight * Adding quantization mixes * Quantization mixes: didn't quite get what I wanted in the last commit * Q4_K dot product for ARM_NEON * Q6_K dot product for ARM_NEON * Q5_K dot product for ARM_NEON * Adding Q3_K dot for ARM_NEON It is 22% slower than Q4_K, despite the smaller model size. On x86_64, where we are memory bound, the Q3_K model is quite a bit faster than Q4_K. * A very slightly faster ARM_NEON Q3_K dot * Adding Q2_K - just CUDA for now Token prediction is pretty good - about 15.5 ms on a RTX 4080. Perplexity is about the same as Q4_K. * Adding scalar and AVX2 Q2_K dot * Adding ARM_NEON Q2_K dot About the same performance as Q4_K. * A slightly faster ARM_NEON Q2_K dot Single token prediction is now ~36 ms on M2 Max. The code is much simpler too. * Fixed bug in Q2_K CUDA dot product kernel Stranegly enough, for the few prompts I tried with the 7B model the responses looked perfectly reasonable. Only realized something is not quite right when I tried the larger models and started getting nonse back. In any case, Q2_K single token evaluation time on an RTX 4080 in a Ryzen7950X box iusing CUDA and model fully loaded on the GPU are ~15.5 ms for 7B, ~25.4 ms for 13B, and ~55.8 ms for 30B. The max number of layers that fit in VRAM for The 65B is 32. With that, we get ~330 ms per token, which is not that much faster than just running on the CPU (~470 ms per token). * Don't print zeros/NaNs when no count histogram has been collected * A 10% faster CUDA vector dot kernel for Q3_K Q3_K is now running at ~18.5 ms / token on CUDA, so the gap to Q4_0 is only 10%. It seems memory acccess pattern is more important for performance than the amount of computation the kernel does. * A slightly daster Q4_K AVX2 dot product For perplexity, where we are less memory bound, time per pass drops by ~5%. Barely measurable difference for single token prediction. * A slightly faster ARM_NEON A4_K dot product * Minor * Fix quantization error test We cannot possibly be expecting rmse < 0.002 for 2- and 3-bit quantization variants. * Fix docker build I have been sloppy with vector reinterpret casts on ARM_NEON. It seems clang is very forgiving in that regard. * Added forgotten ggml.o dependence on k_quants.h to the Makefile * Had unintentionally committed the Makefile with -Ofast enabled * ggml : rename k_quants -> ggml-quants-k, use lowercase in code --------- Co-authored-by: Iwan Kawrakow <iwan.kawrakow@gmail.com> Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
2023-06-05 19:56:18 +00:00
}
}
ggml : quantization refactoring (#3833) * ggml : factor all quantization code in ggml-quants ggml-ci * ggml-quants : fix Zig and Swift builds + quantize tool ggml-ci * quantize : --pure option for disabling k-quant mixtures --------- Co-authored-by: cebtenzzre <cebtenzzre@gmail.com>
2023-10-29 16:32:28 +00:00
}
void quantize_row_q4_0(const float * restrict x, void * restrict y, int k) {
quantize_row_q4_0_reference(x, y, k);
}
ggml : importance matrix support for legacy quants (#4969) * imatrix: adding support for legacy quants * imatrix: guard Q4_0/Q5_0 against ffn_down craziness --------- Co-authored-by: Iwan Kawrakow <iwan.kawrakow@gmail.com>
2024-01-16 17:51:26 +00:00
ggml : quantization refactoring (#3833) * ggml : factor all quantization code in ggml-quants ggml-ci * ggml-quants : fix Zig and Swift builds + quantize tool ggml-ci * quantize : --pure option for disabling k-quant mixtures --------- Co-authored-by: cebtenzzre <cebtenzzre@gmail.com>
2023-10-29 16:32:28 +00:00
void quantize_row_q4_1_reference(const float * restrict x, block_q4_1 * restrict y, int k) {
const int qk = QK4_1;
assert(k % qk == 0);
const int nb = k / qk;
for (int i = 0; i < nb; i++) {
float min = FLT_MAX;
float max = -FLT_MAX;
for (int j = 0; j < qk; j++) {
const float v = x[i*qk + j];
if (v < min) min = v;
if (v > max) max = v;
ggml : add SOTA 2,3,4,5,6 bit k-quantizations (#1684) * Starting to add k-quantization to ggml I think it is better to have quantization separate from ggml. For now just adding the k-quants there, but it would be better to also factor out the existing ggml quantizations. * Adding Q3_K and Q8_K (de)-quantization * Q3_K now working on CUDA and AVX2/scalar CUDA is not ideal - ~50% slower than Q4_0 for single token prediction, about the same in batch mode (perplexity). CPU single token is ~55 ms (on Ryzen 7950X). * Some improvement for Q3_K on CUDA It is now ~22.5 ms/token on my GPU, so ~30% slower than Q4_0. * Some more CUDA optimizations for Q3_K Single token is now 20.5 ms/token (~20% slower than Q4_0). Perplexity is on par with Q4_0. * Adding Q4_K - scalar, AVX2, CUDA Performance is the same or perhaps very slightly better than Q4_0 on the CPU. On the GPU, single token prediction is ~10% better than Q4_0, batch mode (perplexity is about the same). * Adding Q6_K - scalar, AVX2, CUDA Performance is ~40% lower compared to Q4_K on the CPU. This is to be expected, considering that we are memory bound on the CPU and the 6-bit model is ~44% larger than the 4-bit. On the GPU, single token prediction is ~6% lower than Q4_0, batch mode (perplexity) is even closer (but still slower). * Adding Q5_K - scalar, AVX2, CUDA Performance is ~20% lower compared to Q4_K on the CPU. This is to be expected, considering that we are memory bound on the CPU and the 5-bit model is ~22% larger than the 4-bit. On the GPU, single token prediction is about the same as Q4_0 for both, single token and batch prediction. * Per convention, all QX_K quantizations use Q5_K for output.weight * Adding quantization mixes * Quantization mixes: didn't quite get what I wanted in the last commit * Q4_K dot product for ARM_NEON * Q6_K dot product for ARM_NEON * Q5_K dot product for ARM_NEON * Adding Q3_K dot for ARM_NEON It is 22% slower than Q4_K, despite the smaller model size. On x86_64, where we are memory bound, the Q3_K model is quite a bit faster than Q4_K. * A very slightly faster ARM_NEON Q3_K dot * Adding Q2_K - just CUDA for now Token prediction is pretty good - about 15.5 ms on a RTX 4080. Perplexity is about the same as Q4_K. * Adding scalar and AVX2 Q2_K dot * Adding ARM_NEON Q2_K dot About the same performance as Q4_K. * A slightly faster ARM_NEON Q2_K dot Single token prediction is now ~36 ms on M2 Max. The code is much simpler too. * Fixed bug in Q2_K CUDA dot product kernel Stranegly enough, for the few prompts I tried with the 7B model the responses looked perfectly reasonable. Only realized something is not quite right when I tried the larger models and started getting nonse back. In any case, Q2_K single token evaluation time on an RTX 4080 in a Ryzen7950X box iusing CUDA and model fully loaded on the GPU are ~15.5 ms for 7B, ~25.4 ms for 13B, and ~55.8 ms for 30B. The max number of layers that fit in VRAM for The 65B is 32. With that, we get ~330 ms per token, which is not that much faster than just running on the CPU (~470 ms per token). * Don't print zeros/NaNs when no count histogram has been collected * A 10% faster CUDA vector dot kernel for Q3_K Q3_K is now running at ~18.5 ms / token on CUDA, so the gap to Q4_0 is only 10%. It seems memory acccess pattern is more important for performance than the amount of computation the kernel does. * A slightly daster Q4_K AVX2 dot product For perplexity, where we are less memory bound, time per pass drops by ~5%. Barely measurable difference for single token prediction. * A slightly faster ARM_NEON A4_K dot product * Minor * Fix quantization error test We cannot possibly be expecting rmse < 0.002 for 2- and 3-bit quantization variants. * Fix docker build I have been sloppy with vector reinterpret casts on ARM_NEON. It seems clang is very forgiving in that regard. * Added forgotten ggml.o dependence on k_quants.h to the Makefile * Had unintentionally committed the Makefile with -Ofast enabled * ggml : rename k_quants -> ggml-quants-k, use lowercase in code --------- Co-authored-by: Iwan Kawrakow <iwan.kawrakow@gmail.com> Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
2023-06-05 19:56:18 +00:00
}
ggml : quantization refactoring (#3833) * ggml : factor all quantization code in ggml-quants ggml-ci * ggml-quants : fix Zig and Swift builds + quantize tool ggml-ci * quantize : --pure option for disabling k-quant mixtures --------- Co-authored-by: cebtenzzre <cebtenzzre@gmail.com>
2023-10-29 16:32:28 +00:00
const float d = (max - min) / ((1 << 4) - 1);
const float id = d ? 1.0f/d : 0.0f;
ggml : move FP16 <-> FP32 code to ggml-impl.h (#3861) * ggml : move FP16 <-> FP32 stuff to ggml-impl.h ggml-ci * tests : fix ARM build * ggml : explicitly initialize deprecated type traits * ggml : add math.h to ggml-impl.h * ggml : remove duplicate static assert macros * ggml : prefix lookup tables with ggml_ ggml-ci * ggml-impl : move extern "C" to start of file
2023-10-30 17:19:15 +00:00
y[i].d = GGML_FP32_TO_FP16(d);
y[i].m = GGML_FP32_TO_FP16(min);
ggml : quantization refactoring (#3833) * ggml : factor all quantization code in ggml-quants ggml-ci * ggml-quants : fix Zig and Swift builds + quantize tool ggml-ci * quantize : --pure option for disabling k-quant mixtures --------- Co-authored-by: cebtenzzre <cebtenzzre@gmail.com>
2023-10-29 16:32:28 +00:00
for (int j = 0; j < qk/2; ++j) {
const float x0 = (x[i*qk + 0 + j] - min)*id;
const float x1 = (x[i*qk + qk/2 + j] - min)*id;
const uint8_t xi0 = MIN(15, (int8_t)(x0 + 0.5f));
const uint8_t xi1 = MIN(15, (int8_t)(x1 + 0.5f));
y[i].qs[j] = xi0;
y[i].qs[j] |= xi1 << 4;
ggml : add SOTA 2,3,4,5,6 bit k-quantizations (#1684) * Starting to add k-quantization to ggml I think it is better to have quantization separate from ggml. For now just adding the k-quants there, but it would be better to also factor out the existing ggml quantizations. * Adding Q3_K and Q8_K (de)-quantization * Q3_K now working on CUDA and AVX2/scalar CUDA is not ideal - ~50% slower than Q4_0 for single token prediction, about the same in batch mode (perplexity). CPU single token is ~55 ms (on Ryzen 7950X). * Some improvement for Q3_K on CUDA It is now ~22.5 ms/token on my GPU, so ~30% slower than Q4_0. * Some more CUDA optimizations for Q3_K Single token is now 20.5 ms/token (~20% slower than Q4_0). Perplexity is on par with Q4_0. * Adding Q4_K - scalar, AVX2, CUDA Performance is the same or perhaps very slightly better than Q4_0 on the CPU. On the GPU, single token prediction is ~10% better than Q4_0, batch mode (perplexity is about the same). * Adding Q6_K - scalar, AVX2, CUDA Performance is ~40% lower compared to Q4_K on the CPU. This is to be expected, considering that we are memory bound on the CPU and the 6-bit model is ~44% larger than the 4-bit. On the GPU, single token prediction is ~6% lower than Q4_0, batch mode (perplexity) is even closer (but still slower). * Adding Q5_K - scalar, AVX2, CUDA Performance is ~20% lower compared to Q4_K on the CPU. This is to be expected, considering that we are memory bound on the CPU and the 5-bit model is ~22% larger than the 4-bit. On the GPU, single token prediction is about the same as Q4_0 for both, single token and batch prediction. * Per convention, all QX_K quantizations use Q5_K for output.weight * Adding quantization mixes * Quantization mixes: didn't quite get what I wanted in the last commit * Q4_K dot product for ARM_NEON * Q6_K dot product for ARM_NEON * Q5_K dot product for ARM_NEON * Adding Q3_K dot for ARM_NEON It is 22% slower than Q4_K, despite the smaller model size. On x86_64, where we are memory bound, the Q3_K model is quite a bit faster than Q4_K. * A very slightly faster ARM_NEON Q3_K dot * Adding Q2_K - just CUDA for now Token prediction is pretty good - about 15.5 ms on a RTX 4080. Perplexity is about the same as Q4_K. * Adding scalar and AVX2 Q2_K dot * Adding ARM_NEON Q2_K dot About the same performance as Q4_K. * A slightly faster ARM_NEON Q2_K dot Single token prediction is now ~36 ms on M2 Max. The code is much simpler too. * Fixed bug in Q2_K CUDA dot product kernel Stranegly enough, for the few prompts I tried with the 7B model the responses looked perfectly reasonable. Only realized something is not quite right when I tried the larger models and started getting nonse back. In any case, Q2_K single token evaluation time on an RTX 4080 in a Ryzen7950X box iusing CUDA and model fully loaded on the GPU are ~15.5 ms for 7B, ~25.4 ms for 13B, and ~55.8 ms for 30B. The max number of layers that fit in VRAM for The 65B is 32. With that, we get ~330 ms per token, which is not that much faster than just running on the CPU (~470 ms per token). * Don't print zeros/NaNs when no count histogram has been collected * A 10% faster CUDA vector dot kernel for Q3_K Q3_K is now running at ~18.5 ms / token on CUDA, so the gap to Q4_0 is only 10%. It seems memory acccess pattern is more important for performance than the amount of computation the kernel does. * A slightly daster Q4_K AVX2 dot product For perplexity, where we are less memory bound, time per pass drops by ~5%. Barely measurable difference for single token prediction. * A slightly faster ARM_NEON A4_K dot product * Minor * Fix quantization error test We cannot possibly be expecting rmse < 0.002 for 2- and 3-bit quantization variants. * Fix docker build I have been sloppy with vector reinterpret casts on ARM_NEON. It seems clang is very forgiving in that regard. * Added forgotten ggml.o dependence on k_quants.h to the Makefile * Had unintentionally committed the Makefile with -Ofast enabled * ggml : rename k_quants -> ggml-quants-k, use lowercase in code --------- Co-authored-by: Iwan Kawrakow <iwan.kawrakow@gmail.com> Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
2023-06-05 19:56:18 +00:00
}
}
}
ggml : quantization refactoring (#3833) * ggml : factor all quantization code in ggml-quants ggml-ci * ggml-quants : fix Zig and Swift builds + quantize tool ggml-ci * quantize : --pure option for disabling k-quant mixtures --------- Co-authored-by: cebtenzzre <cebtenzzre@gmail.com>
2023-10-29 16:32:28 +00:00
void quantize_row_q4_1(const float * restrict x, void * restrict y, int k) {
quantize_row_q4_1_reference(x, y, k);
}
void quantize_row_q5_0_reference(const float * restrict x, block_q5_0 * restrict y, int k) {
static const int qk = QK5_0;
assert(k % qk == 0);
const int nb = k / qk;
for (int i = 0; i < nb; i++) {
float amax = 0.0f; // absolute max
float max = 0.0f;
for (int j = 0; j < qk; j++) {
const float v = x[i*qk + j];
if (amax < fabsf(v)) {
amax = fabsf(v);
max = v;
ggml : add SOTA 2,3,4,5,6 bit k-quantizations (#1684) * Starting to add k-quantization to ggml I think it is better to have quantization separate from ggml. For now just adding the k-quants there, but it would be better to also factor out the existing ggml quantizations. * Adding Q3_K and Q8_K (de)-quantization * Q3_K now working on CUDA and AVX2/scalar CUDA is not ideal - ~50% slower than Q4_0 for single token prediction, about the same in batch mode (perplexity). CPU single token is ~55 ms (on Ryzen 7950X). * Some improvement for Q3_K on CUDA It is now ~22.5 ms/token on my GPU, so ~30% slower than Q4_0. * Some more CUDA optimizations for Q3_K Single token is now 20.5 ms/token (~20% slower than Q4_0). Perplexity is on par with Q4_0. * Adding Q4_K - scalar, AVX2, CUDA Performance is the same or perhaps very slightly better than Q4_0 on the CPU. On the GPU, single token prediction is ~10% better than Q4_0, batch mode (perplexity is about the same). * Adding Q6_K - scalar, AVX2, CUDA Performance is ~40% lower compared to Q4_K on the CPU. This is to be expected, considering that we are memory bound on the CPU and the 6-bit model is ~44% larger than the 4-bit. On the GPU, single token prediction is ~6% lower than Q4_0, batch mode (perplexity) is even closer (but still slower). * Adding Q5_K - scalar, AVX2, CUDA Performance is ~20% lower compared to Q4_K on the CPU. This is to be expected, considering that we are memory bound on the CPU and the 5-bit model is ~22% larger than the 4-bit. On the GPU, single token prediction is about the same as Q4_0 for both, single token and batch prediction. * Per convention, all QX_K quantizations use Q5_K for output.weight * Adding quantization mixes * Quantization mixes: didn't quite get what I wanted in the last commit * Q4_K dot product for ARM_NEON * Q6_K dot product for ARM_NEON * Q5_K dot product for ARM_NEON * Adding Q3_K dot for ARM_NEON It is 22% slower than Q4_K, despite the smaller model size. On x86_64, where we are memory bound, the Q3_K model is quite a bit faster than Q4_K. * A very slightly faster ARM_NEON Q3_K dot * Adding Q2_K - just CUDA for now Token prediction is pretty good - about 15.5 ms on a RTX 4080. Perplexity is about the same as Q4_K. * Adding scalar and AVX2 Q2_K dot * Adding ARM_NEON Q2_K dot About the same performance as Q4_K. * A slightly faster ARM_NEON Q2_K dot Single token prediction is now ~36 ms on M2 Max. The code is much simpler too. * Fixed bug in Q2_K CUDA dot product kernel Stranegly enough, for the few prompts I tried with the 7B model the responses looked perfectly reasonable. Only realized something is not quite right when I tried the larger models and started getting nonse back. In any case, Q2_K single token evaluation time on an RTX 4080 in a Ryzen7950X box iusing CUDA and model fully loaded on the GPU are ~15.5 ms for 7B, ~25.4 ms for 13B, and ~55.8 ms for 30B. The max number of layers that fit in VRAM for The 65B is 32. With that, we get ~330 ms per token, which is not that much faster than just running on the CPU (~470 ms per token). * Don't print zeros/NaNs when no count histogram has been collected * A 10% faster CUDA vector dot kernel for Q3_K Q3_K is now running at ~18.5 ms / token on CUDA, so the gap to Q4_0 is only 10%. It seems memory acccess pattern is more important for performance than the amount of computation the kernel does. * A slightly daster Q4_K AVX2 dot product For perplexity, where we are less memory bound, time per pass drops by ~5%. Barely measurable difference for single token prediction. * A slightly faster ARM_NEON A4_K dot product * Minor * Fix quantization error test We cannot possibly be expecting rmse < 0.002 for 2- and 3-bit quantization variants. * Fix docker build I have been sloppy with vector reinterpret casts on ARM_NEON. It seems clang is very forgiving in that regard. * Added forgotten ggml.o dependence on k_quants.h to the Makefile * Had unintentionally committed the Makefile with -Ofast enabled * ggml : rename k_quants -> ggml-quants-k, use lowercase in code --------- Co-authored-by: Iwan Kawrakow <iwan.kawrakow@gmail.com> Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
2023-06-05 19:56:18 +00:00
}
}
ggml : quantization refactoring (#3833) * ggml : factor all quantization code in ggml-quants ggml-ci * ggml-quants : fix Zig and Swift builds + quantize tool ggml-ci * quantize : --pure option for disabling k-quant mixtures --------- Co-authored-by: cebtenzzre <cebtenzzre@gmail.com>
2023-10-29 16:32:28 +00:00
const float d = max / -16;
const float id = d ? 1.0f/d : 0.0f;
ggml : move FP16 <-> FP32 code to ggml-impl.h (#3861) * ggml : move FP16 <-> FP32 stuff to ggml-impl.h ggml-ci * tests : fix ARM build * ggml : explicitly initialize deprecated type traits * ggml : add math.h to ggml-impl.h * ggml : remove duplicate static assert macros * ggml : prefix lookup tables with ggml_ ggml-ci * ggml-impl : move extern "C" to start of file
2023-10-30 17:19:15 +00:00
y[i].d = GGML_FP32_TO_FP16(d);
ggml : quantization refactoring (#3833) * ggml : factor all quantization code in ggml-quants ggml-ci * ggml-quants : fix Zig and Swift builds + quantize tool ggml-ci * quantize : --pure option for disabling k-quant mixtures --------- Co-authored-by: cebtenzzre <cebtenzzre@gmail.com>
2023-10-29 16:32:28 +00:00
uint32_t qh = 0;
for (int j = 0; j < qk/2; ++j) {
const float x0 = x[i*qk + 0 + j]*id;
const float x1 = x[i*qk + qk/2 + j]*id;
const uint8_t xi0 = MIN(31, (int8_t)(x0 + 16.5f));
const uint8_t xi1 = MIN(31, (int8_t)(x1 + 16.5f));
y[i].qs[j] = (xi0 & 0x0F) | ((xi1 & 0x0F) << 4);
// get the 5-th bit and store it in qh at the right position
qh |= ((xi0 & 0x10u) >> 4) << (j + 0);
qh |= ((xi1 & 0x10u) >> 4) << (j + qk/2);
ggml : add SOTA 2,3,4,5,6 bit k-quantizations (#1684) * Starting to add k-quantization to ggml I think it is better to have quantization separate from ggml. For now just adding the k-quants there, but it would be better to also factor out the existing ggml quantizations. * Adding Q3_K and Q8_K (de)-quantization * Q3_K now working on CUDA and AVX2/scalar CUDA is not ideal - ~50% slower than Q4_0 for single token prediction, about the same in batch mode (perplexity). CPU single token is ~55 ms (on Ryzen 7950X). * Some improvement for Q3_K on CUDA It is now ~22.5 ms/token on my GPU, so ~30% slower than Q4_0. * Some more CUDA optimizations for Q3_K Single token is now 20.5 ms/token (~20% slower than Q4_0). Perplexity is on par with Q4_0. * Adding Q4_K - scalar, AVX2, CUDA Performance is the same or perhaps very slightly better than Q4_0 on the CPU. On the GPU, single token prediction is ~10% better than Q4_0, batch mode (perplexity is about the same). * Adding Q6_K - scalar, AVX2, CUDA Performance is ~40% lower compared to Q4_K on the CPU. This is to be expected, considering that we are memory bound on the CPU and the 6-bit model is ~44% larger than the 4-bit. On the GPU, single token prediction is ~6% lower than Q4_0, batch mode (perplexity) is even closer (but still slower). * Adding Q5_K - scalar, AVX2, CUDA Performance is ~20% lower compared to Q4_K on the CPU. This is to be expected, considering that we are memory bound on the CPU and the 5-bit model is ~22% larger than the 4-bit. On the GPU, single token prediction is about the same as Q4_0 for both, single token and batch prediction. * Per convention, all QX_K quantizations use Q5_K for output.weight * Adding quantization mixes * Quantization mixes: didn't quite get what I wanted in the last commit * Q4_K dot product for ARM_NEON * Q6_K dot product for ARM_NEON * Q5_K dot product for ARM_NEON * Adding Q3_K dot for ARM_NEON It is 22% slower than Q4_K, despite the smaller model size. On x86_64, where we are memory bound, the Q3_K model is quite a bit faster than Q4_K. * A very slightly faster ARM_NEON Q3_K dot * Adding Q2_K - just CUDA for now Token prediction is pretty good - about 15.5 ms on a RTX 4080. Perplexity is about the same as Q4_K. * Adding scalar and AVX2 Q2_K dot * Adding ARM_NEON Q2_K dot About the same performance as Q4_K. * A slightly faster ARM_NEON Q2_K dot Single token prediction is now ~36 ms on M2 Max. The code is much simpler too. * Fixed bug in Q2_K CUDA dot product kernel Stranegly enough, for the few prompts I tried with the 7B model the responses looked perfectly reasonable. Only realized something is not quite right when I tried the larger models and started getting nonse back. In any case, Q2_K single token evaluation time on an RTX 4080 in a Ryzen7950X box iusing CUDA and model fully loaded on the GPU are ~15.5 ms for 7B, ~25.4 ms for 13B, and ~55.8 ms for 30B. The max number of layers that fit in VRAM for The 65B is 32. With that, we get ~330 ms per token, which is not that much faster than just running on the CPU (~470 ms per token). * Don't print zeros/NaNs when no count histogram has been collected * A 10% faster CUDA vector dot kernel for Q3_K Q3_K is now running at ~18.5 ms / token on CUDA, so the gap to Q4_0 is only 10%. It seems memory acccess pattern is more important for performance than the amount of computation the kernel does. * A slightly daster Q4_K AVX2 dot product For perplexity, where we are less memory bound, time per pass drops by ~5%. Barely measurable difference for single token prediction. * A slightly faster ARM_NEON A4_K dot product * Minor * Fix quantization error test We cannot possibly be expecting rmse < 0.002 for 2- and 3-bit quantization variants. * Fix docker build I have been sloppy with vector reinterpret casts on ARM_NEON. It seems clang is very forgiving in that regard. * Added forgotten ggml.o dependence on k_quants.h to the Makefile * Had unintentionally committed the Makefile with -Ofast enabled * ggml : rename k_quants -> ggml-quants-k, use lowercase in code --------- Co-authored-by: Iwan Kawrakow <iwan.kawrakow@gmail.com> Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
2023-06-05 19:56:18 +00:00
}
ggml : quantization refactoring (#3833) * ggml : factor all quantization code in ggml-quants ggml-ci * ggml-quants : fix Zig and Swift builds + quantize tool ggml-ci * quantize : --pure option for disabling k-quant mixtures --------- Co-authored-by: cebtenzzre <cebtenzzre@gmail.com>
2023-10-29 16:32:28 +00:00
memcpy(&y[i].qh, &qh, sizeof(qh));
ggml : add SOTA 2,3,4,5,6 bit k-quantizations (#1684) * Starting to add k-quantization to ggml I think it is better to have quantization separate from ggml. For now just adding the k-quants there, but it would be better to also factor out the existing ggml quantizations. * Adding Q3_K and Q8_K (de)-quantization * Q3_K now working on CUDA and AVX2/scalar CUDA is not ideal - ~50% slower than Q4_0 for single token prediction, about the same in batch mode (perplexity). CPU single token is ~55 ms (on Ryzen 7950X). * Some improvement for Q3_K on CUDA It is now ~22.5 ms/token on my GPU, so ~30% slower than Q4_0. * Some more CUDA optimizations for Q3_K Single token is now 20.5 ms/token (~20% slower than Q4_0). Perplexity is on par with Q4_0. * Adding Q4_K - scalar, AVX2, CUDA Performance is the same or perhaps very slightly better than Q4_0 on the CPU. On the GPU, single token prediction is ~10% better than Q4_0, batch mode (perplexity is about the same). * Adding Q6_K - scalar, AVX2, CUDA Performance is ~40% lower compared to Q4_K on the CPU. This is to be expected, considering that we are memory bound on the CPU and the 6-bit model is ~44% larger than the 4-bit. On the GPU, single token prediction is ~6% lower than Q4_0, batch mode (perplexity) is even closer (but still slower). * Adding Q5_K - scalar, AVX2, CUDA Performance is ~20% lower compared to Q4_K on the CPU. This is to be expected, considering that we are memory bound on the CPU and the 5-bit model is ~22% larger than the 4-bit. On the GPU, single token prediction is about the same as Q4_0 for both, single token and batch prediction. * Per convention, all QX_K quantizations use Q5_K for output.weight * Adding quantization mixes * Quantization mixes: didn't quite get what I wanted in the last commit * Q4_K dot product for ARM_NEON * Q6_K dot product for ARM_NEON * Q5_K dot product for ARM_NEON * Adding Q3_K dot for ARM_NEON It is 22% slower than Q4_K, despite the smaller model size. On x86_64, where we are memory bound, the Q3_K model is quite a bit faster than Q4_K. * A very slightly faster ARM_NEON Q3_K dot * Adding Q2_K - just CUDA for now Token prediction is pretty good - about 15.5 ms on a RTX 4080. Perplexity is about the same as Q4_K. * Adding scalar and AVX2 Q2_K dot * Adding ARM_NEON Q2_K dot About the same performance as Q4_K. * A slightly faster ARM_NEON Q2_K dot Single token prediction is now ~36 ms on M2 Max. The code is much simpler too. * Fixed bug in Q2_K CUDA dot product kernel Stranegly enough, for the few prompts I tried with the 7B model the responses looked perfectly reasonable. Only realized something is not quite right when I tried the larger models and started getting nonse back. In any case, Q2_K single token evaluation time on an RTX 4080 in a Ryzen7950X box iusing CUDA and model fully loaded on the GPU are ~15.5 ms for 7B, ~25.4 ms for 13B, and ~55.8 ms for 30B. The max number of layers that fit in VRAM for The 65B is 32. With that, we get ~330 ms per token, which is not that much faster than just running on the CPU (~470 ms per token). * Don't print zeros/NaNs when no count histogram has been collected * A 10% faster CUDA vector dot kernel for Q3_K Q3_K is now running at ~18.5 ms / token on CUDA, so the gap to Q4_0 is only 10%. It seems memory acccess pattern is more important for performance than the amount of computation the kernel does. * A slightly daster Q4_K AVX2 dot product For perplexity, where we are less memory bound, time per pass drops by ~5%. Barely measurable difference for single token prediction. * A slightly faster ARM_NEON A4_K dot product * Minor * Fix quantization error test We cannot possibly be expecting rmse < 0.002 for 2- and 3-bit quantization variants. * Fix docker build I have been sloppy with vector reinterpret casts on ARM_NEON. It seems clang is very forgiving in that regard. * Added forgotten ggml.o dependence on k_quants.h to the Makefile * Had unintentionally committed the Makefile with -Ofast enabled * ggml : rename k_quants -> ggml-quants-k, use lowercase in code --------- Co-authored-by: Iwan Kawrakow <iwan.kawrakow@gmail.com> Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
2023-06-05 19:56:18 +00:00
}
}
ggml : quantization refactoring (#3833) * ggml : factor all quantization code in ggml-quants ggml-ci * ggml-quants : fix Zig and Swift builds + quantize tool ggml-ci * quantize : --pure option for disabling k-quant mixtures --------- Co-authored-by: cebtenzzre <cebtenzzre@gmail.com>
2023-10-29 16:32:28 +00:00
void quantize_row_q5_0(const float * restrict x, void * restrict y, int k) {
quantize_row_q5_0_reference(x, y, k);
}
void quantize_row_q5_1_reference(const float * restrict x, block_q5_1 * restrict y, int k) {
const int qk = QK5_1;
assert(k % qk == 0);
const int nb = k / qk;
for (int i = 0; i < nb; i++) {
float min = FLT_MAX;
float max = -FLT_MAX;
for (int j = 0; j < qk; j++) {
const float v = x[i*qk + j];
if (v < min) min = v;
if (v > max) max = v;
}
const float d = (max - min) / ((1 << 5) - 1);
const float id = d ? 1.0f/d : 0.0f;
ggml : move FP16 <-> FP32 code to ggml-impl.h (#3861) * ggml : move FP16 <-> FP32 stuff to ggml-impl.h ggml-ci * tests : fix ARM build * ggml : explicitly initialize deprecated type traits * ggml : add math.h to ggml-impl.h * ggml : remove duplicate static assert macros * ggml : prefix lookup tables with ggml_ ggml-ci * ggml-impl : move extern "C" to start of file
2023-10-30 17:19:15 +00:00
y[i].d = GGML_FP32_TO_FP16(d);
y[i].m = GGML_FP32_TO_FP16(min);
ggml : quantization refactoring (#3833) * ggml : factor all quantization code in ggml-quants ggml-ci * ggml-quants : fix Zig and Swift builds + quantize tool ggml-ci * quantize : --pure option for disabling k-quant mixtures --------- Co-authored-by: cebtenzzre <cebtenzzre@gmail.com>
2023-10-29 16:32:28 +00:00
uint32_t qh = 0;
for (int j = 0; j < qk/2; ++j) {
const float x0 = (x[i*qk + 0 + j] - min)*id;
const float x1 = (x[i*qk + qk/2 + j] - min)*id;
const uint8_t xi0 = (uint8_t)(x0 + 0.5f);
const uint8_t xi1 = (uint8_t)(x1 + 0.5f);
y[i].qs[j] = (xi0 & 0x0F) | ((xi1 & 0x0F) << 4);
// get the 5-th bit and store it in qh at the right position
qh |= ((xi0 & 0x10u) >> 4) << (j + 0);
qh |= ((xi1 & 0x10u) >> 4) << (j + qk/2);
}
memcpy(&y[i].qh, &qh, sizeof(y[i].qh));
ggml : add SOTA 2,3,4,5,6 bit k-quantizations (#1684) * Starting to add k-quantization to ggml I think it is better to have quantization separate from ggml. For now just adding the k-quants there, but it would be better to also factor out the existing ggml quantizations. * Adding Q3_K and Q8_K (de)-quantization * Q3_K now working on CUDA and AVX2/scalar CUDA is not ideal - ~50% slower than Q4_0 for single token prediction, about the same in batch mode (perplexity). CPU single token is ~55 ms (on Ryzen 7950X). * Some improvement for Q3_K on CUDA It is now ~22.5 ms/token on my GPU, so ~30% slower than Q4_0. * Some more CUDA optimizations for Q3_K Single token is now 20.5 ms/token (~20% slower than Q4_0). Perplexity is on par with Q4_0. * Adding Q4_K - scalar, AVX2, CUDA Performance is the same or perhaps very slightly better than Q4_0 on the CPU. On the GPU, single token prediction is ~10% better than Q4_0, batch mode (perplexity is about the same). * Adding Q6_K - scalar, AVX2, CUDA Performance is ~40% lower compared to Q4_K on the CPU. This is to be expected, considering that we are memory bound on the CPU and the 6-bit model is ~44% larger than the 4-bit. On the GPU, single token prediction is ~6% lower than Q4_0, batch mode (perplexity) is even closer (but still slower). * Adding Q5_K - scalar, AVX2, CUDA Performance is ~20% lower compared to Q4_K on the CPU. This is to be expected, considering that we are memory bound on the CPU and the 5-bit model is ~22% larger than the 4-bit. On the GPU, single token prediction is about the same as Q4_0 for both, single token and batch prediction. * Per convention, all QX_K quantizations use Q5_K for output.weight * Adding quantization mixes * Quantization mixes: didn't quite get what I wanted in the last commit * Q4_K dot product for ARM_NEON * Q6_K dot product for ARM_NEON * Q5_K dot product for ARM_NEON * Adding Q3_K dot for ARM_NEON It is 22% slower than Q4_K, despite the smaller model size. On x86_64, where we are memory bound, the Q3_K model is quite a bit faster than Q4_K. * A very slightly faster ARM_NEON Q3_K dot * Adding Q2_K - just CUDA for now Token prediction is pretty good - about 15.5 ms on a RTX 4080. Perplexity is about the same as Q4_K. * Adding scalar and AVX2 Q2_K dot * Adding ARM_NEON Q2_K dot About the same performance as Q4_K. * A slightly faster ARM_NEON Q2_K dot Single token prediction is now ~36 ms on M2 Max. The code is much simpler too. * Fixed bug in Q2_K CUDA dot product kernel Stranegly enough, for the few prompts I tried with the 7B model the responses looked perfectly reasonable. Only realized something is not quite right when I tried the larger models and started getting nonse back. In any case, Q2_K single token evaluation time on an RTX 4080 in a Ryzen7950X box iusing CUDA and model fully loaded on the GPU are ~15.5 ms for 7B, ~25.4 ms for 13B, and ~55.8 ms for 30B. The max number of layers that fit in VRAM for The 65B is 32. With that, we get ~330 ms per token, which is not that much faster than just running on the CPU (~470 ms per token). * Don't print zeros/NaNs when no count histogram has been collected * A 10% faster CUDA vector dot kernel for Q3_K Q3_K is now running at ~18.5 ms / token on CUDA, so the gap to Q4_0 is only 10%. It seems memory acccess pattern is more important for performance than the amount of computation the kernel does. * A slightly daster Q4_K AVX2 dot product For perplexity, where we are less memory bound, time per pass drops by ~5%. Barely measurable difference for single token prediction. * A slightly faster ARM_NEON A4_K dot product * Minor * Fix quantization error test We cannot possibly be expecting rmse < 0.002 for 2- and 3-bit quantization variants. * Fix docker build I have been sloppy with vector reinterpret casts on ARM_NEON. It seems clang is very forgiving in that regard. * Added forgotten ggml.o dependence on k_quants.h to the Makefile * Had unintentionally committed the Makefile with -Ofast enabled * ggml : rename k_quants -> ggml-quants-k, use lowercase in code --------- Co-authored-by: Iwan Kawrakow <iwan.kawrakow@gmail.com> Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
2023-06-05 19:56:18 +00:00
}
ggml : quantization refactoring (#3833) * ggml : factor all quantization code in ggml-quants ggml-ci * ggml-quants : fix Zig and Swift builds + quantize tool ggml-ci * quantize : --pure option for disabling k-quant mixtures --------- Co-authored-by: cebtenzzre <cebtenzzre@gmail.com>
2023-10-29 16:32:28 +00:00
}
void quantize_row_q5_1(const float * restrict x, void * restrict y, int k) {
quantize_row_q5_1_reference(x, y, k);
}
// reference implementation for deterministic creation of model files
void quantize_row_q8_0_reference(const float * restrict x, block_q8_0 * restrict y, int k) {
assert(k % QK8_0 == 0);
const int nb = k / QK8_0;
for (int i = 0; i < nb; i++) {
float amax = 0.0f; // absolute max
for (int j = 0; j < QK8_0; j++) {
const float v = x[i*QK8_0 + j];
amax = MAX(amax, fabsf(v));
}
const float d = amax / ((1 << 7) - 1);
const float id = d ? 1.0f/d : 0.0f;
ggml : move FP16 <-> FP32 code to ggml-impl.h (#3861) * ggml : move FP16 <-> FP32 stuff to ggml-impl.h ggml-ci * tests : fix ARM build * ggml : explicitly initialize deprecated type traits * ggml : add math.h to ggml-impl.h * ggml : remove duplicate static assert macros * ggml : prefix lookup tables with ggml_ ggml-ci * ggml-impl : move extern "C" to start of file
2023-10-30 17:19:15 +00:00
y[i].d = GGML_FP32_TO_FP16(d);
ggml : quantization refactoring (#3833) * ggml : factor all quantization code in ggml-quants ggml-ci * ggml-quants : fix Zig and Swift builds + quantize tool ggml-ci * quantize : --pure option for disabling k-quant mixtures --------- Co-authored-by: cebtenzzre <cebtenzzre@gmail.com>
2023-10-29 16:32:28 +00:00
for (int j = 0; j < QK8_0; ++j) {
const float x0 = x[i*QK8_0 + j]*id;
y[i].qs[j] = roundf(x0);
}
}
}
void quantize_row_q8_0(const float * restrict x, void * restrict vy, int k) {
assert(QK8_0 == 32);
assert(k % QK8_0 == 0);
const int nb = k / QK8_0;
block_q8_0 * restrict y = vy;
#if defined(__ARM_NEON)
for (int i = 0; i < nb; i++) {
float32x4_t srcv [8];
float32x4_t asrcv[8];
float32x4_t amaxv[8];
for (int j = 0; j < 8; j++) srcv[j] = vld1q_f32(x + i*32 + 4*j);
for (int j = 0; j < 8; j++) asrcv[j] = vabsq_f32(srcv[j]);
for (int j = 0; j < 4; j++) amaxv[2*j] = vmaxq_f32(asrcv[2*j], asrcv[2*j+1]);
for (int j = 0; j < 2; j++) amaxv[4*j] = vmaxq_f32(amaxv[4*j], amaxv[4*j+2]);
for (int j = 0; j < 1; j++) amaxv[8*j] = vmaxq_f32(amaxv[8*j], amaxv[8*j+4]);
const float amax = vmaxvq_f32(amaxv[0]);
const float d = amax / ((1 << 7) - 1);
const float id = d ? 1.0f/d : 0.0f;
ggml : move FP16 <-> FP32 code to ggml-impl.h (#3861) * ggml : move FP16 <-> FP32 stuff to ggml-impl.h ggml-ci * tests : fix ARM build * ggml : explicitly initialize deprecated type traits * ggml : add math.h to ggml-impl.h * ggml : remove duplicate static assert macros * ggml : prefix lookup tables with ggml_ ggml-ci * ggml-impl : move extern "C" to start of file
2023-10-30 17:19:15 +00:00
y[i].d = GGML_FP32_TO_FP16(d);
ggml : quantization refactoring (#3833) * ggml : factor all quantization code in ggml-quants ggml-ci * ggml-quants : fix Zig and Swift builds + quantize tool ggml-ci * quantize : --pure option for disabling k-quant mixtures --------- Co-authored-by: cebtenzzre <cebtenzzre@gmail.com>
2023-10-29 16:32:28 +00:00
for (int j = 0; j < 8; j++) {
const float32x4_t v = vmulq_n_f32(srcv[j], id);
const int32x4_t vi = vcvtnq_s32_f32(v);
y[i].qs[4*j + 0] = vgetq_lane_s32(vi, 0);
y[i].qs[4*j + 1] = vgetq_lane_s32(vi, 1);
y[i].qs[4*j + 2] = vgetq_lane_s32(vi, 2);
y[i].qs[4*j + 3] = vgetq_lane_s32(vi, 3);
}
}
#elif defined(__wasm_simd128__)
for (int i = 0; i < nb; i++) {
v128_t srcv [8];
v128_t asrcv[8];
v128_t amaxv[8];
for (int j = 0; j < 8; j++) srcv[j] = wasm_v128_load(x + i*32 + 4*j);
for (int j = 0; j < 8; j++) asrcv[j] = wasm_f32x4_abs(srcv[j]);
for (int j = 0; j < 4; j++) amaxv[2*j] = wasm_f32x4_max(asrcv[2*j], asrcv[2*j+1]);
for (int j = 0; j < 2; j++) amaxv[4*j] = wasm_f32x4_max(amaxv[4*j], amaxv[4*j+2]);
for (int j = 0; j < 1; j++) amaxv[8*j] = wasm_f32x4_max(amaxv[8*j], amaxv[8*j+4]);
const float amax = MAX(MAX(wasm_f32x4_extract_lane(amaxv[0], 0),
wasm_f32x4_extract_lane(amaxv[0], 1)),
MAX(wasm_f32x4_extract_lane(amaxv[0], 2),
wasm_f32x4_extract_lane(amaxv[0], 3)));
const float d = amax / ((1 << 7) - 1);
const float id = d ? 1.0f/d : 0.0f;
ggml : move FP16 <-> FP32 code to ggml-impl.h (#3861) * ggml : move FP16 <-> FP32 stuff to ggml-impl.h ggml-ci * tests : fix ARM build * ggml : explicitly initialize deprecated type traits * ggml : add math.h to ggml-impl.h * ggml : remove duplicate static assert macros * ggml : prefix lookup tables with ggml_ ggml-ci * ggml-impl : move extern "C" to start of file
2023-10-30 17:19:15 +00:00
y[i].d = GGML_FP32_TO_FP16(d);
ggml : quantization refactoring (#3833) * ggml : factor all quantization code in ggml-quants ggml-ci * ggml-quants : fix Zig and Swift builds + quantize tool ggml-ci * quantize : --pure option for disabling k-quant mixtures --------- Co-authored-by: cebtenzzre <cebtenzzre@gmail.com>
2023-10-29 16:32:28 +00:00
for (int j = 0; j < 8; j++) {
const v128_t v = wasm_f32x4_mul(srcv[j], wasm_f32x4_splat(id));
const v128_t vi = wasm_i32x4_trunc_sat_f32x4(v);
y[i].qs[4*j + 0] = wasm_i32x4_extract_lane(vi, 0);
y[i].qs[4*j + 1] = wasm_i32x4_extract_lane(vi, 1);
y[i].qs[4*j + 2] = wasm_i32x4_extract_lane(vi, 2);
y[i].qs[4*j + 3] = wasm_i32x4_extract_lane(vi, 3);
}
}
#elif defined(__AVX2__) || defined(__AVX__)
for (int i = 0; i < nb; i++) {
// Load elements into 4 AVX vectors
__m256 v0 = _mm256_loadu_ps( x );
__m256 v1 = _mm256_loadu_ps( x + 8 );
__m256 v2 = _mm256_loadu_ps( x + 16 );
__m256 v3 = _mm256_loadu_ps( x + 24 );
x += 32;
// Compute max(abs(e)) for the block
const __m256 signBit = _mm256_set1_ps( -0.0f );
__m256 maxAbs = _mm256_andnot_ps( signBit, v0 );
maxAbs = _mm256_max_ps( maxAbs, _mm256_andnot_ps( signBit, v1 ) );
maxAbs = _mm256_max_ps( maxAbs, _mm256_andnot_ps( signBit, v2 ) );
maxAbs = _mm256_max_ps( maxAbs, _mm256_andnot_ps( signBit, v3 ) );
__m128 max4 = _mm_max_ps( _mm256_extractf128_ps( maxAbs, 1 ), _mm256_castps256_ps128( maxAbs ) );
max4 = _mm_max_ps( max4, _mm_movehl_ps( max4, max4 ) );
max4 = _mm_max_ss( max4, _mm_movehdup_ps( max4 ) );
const float maxScalar = _mm_cvtss_f32( max4 );
// Quantize these floats
const float d = maxScalar / 127.f;
ggml : move FP16 <-> FP32 code to ggml-impl.h (#3861) * ggml : move FP16 <-> FP32 stuff to ggml-impl.h ggml-ci * tests : fix ARM build * ggml : explicitly initialize deprecated type traits * ggml : add math.h to ggml-impl.h * ggml : remove duplicate static assert macros * ggml : prefix lookup tables with ggml_ ggml-ci * ggml-impl : move extern "C" to start of file
2023-10-30 17:19:15 +00:00
y[i].d = GGML_FP32_TO_FP16(d);
ggml : quantization refactoring (#3833) * ggml : factor all quantization code in ggml-quants ggml-ci * ggml-quants : fix Zig and Swift builds + quantize tool ggml-ci * quantize : --pure option for disabling k-quant mixtures --------- Co-authored-by: cebtenzzre <cebtenzzre@gmail.com>
2023-10-29 16:32:28 +00:00
const float id = ( maxScalar != 0.0f ) ? 127.f / maxScalar : 0.0f;
const __m256 mul = _mm256_set1_ps( id );
// Apply the multiplier
v0 = _mm256_mul_ps( v0, mul );
v1 = _mm256_mul_ps( v1, mul );
v2 = _mm256_mul_ps( v2, mul );
v3 = _mm256_mul_ps( v3, mul );
// Round to nearest integer
v0 = _mm256_round_ps( v0, _MM_ROUND_NEAREST );
v1 = _mm256_round_ps( v1, _MM_ROUND_NEAREST );
v2 = _mm256_round_ps( v2, _MM_ROUND_NEAREST );
v3 = _mm256_round_ps( v3, _MM_ROUND_NEAREST );
// Convert floats to integers
__m256i i0 = _mm256_cvtps_epi32( v0 );
__m256i i1 = _mm256_cvtps_epi32( v1 );
__m256i i2 = _mm256_cvtps_epi32( v2 );
__m256i i3 = _mm256_cvtps_epi32( v3 );
#if defined(__AVX2__)
// Convert int32 to int16
i0 = _mm256_packs_epi32( i0, i1 ); // 0, 1, 2, 3, 8, 9, 10, 11, 4, 5, 6, 7, 12, 13, 14, 15
i2 = _mm256_packs_epi32( i2, i3 ); // 16, 17, 18, 19, 24, 25, 26, 27, 20, 21, 22, 23, 28, 29, 30, 31
// Convert int16 to int8
i0 = _mm256_packs_epi16( i0, i2 ); // 0, 1, 2, 3, 8, 9, 10, 11, 16, 17, 18, 19, 24, 25, 26, 27, 4, 5, 6, 7, 12, 13, 14, 15, 20, 21, 22, 23, 28, 29, 30, 31
// We got our precious signed bytes, but the order is now wrong
// These AVX2 pack instructions process 16-byte pieces independently
// The following instruction is fixing the order
const __m256i perm = _mm256_setr_epi32( 0, 4, 1, 5, 2, 6, 3, 7 );
i0 = _mm256_permutevar8x32_epi32( i0, perm );
_mm256_storeu_si256((__m256i *)y[i].qs, i0);
#else
// Since we don't have in AVX some necessary functions,
// we split the registers in half and call AVX2 analogs from SSE
__m128i ni0 = _mm256_castsi256_si128( i0 );
__m128i ni1 = _mm256_extractf128_si256( i0, 1);
__m128i ni2 = _mm256_castsi256_si128( i1 );
__m128i ni3 = _mm256_extractf128_si256( i1, 1);
__m128i ni4 = _mm256_castsi256_si128( i2 );
__m128i ni5 = _mm256_extractf128_si256( i2, 1);
__m128i ni6 = _mm256_castsi256_si128( i3 );
__m128i ni7 = _mm256_extractf128_si256( i3, 1);
// Convert int32 to int16
ni0 = _mm_packs_epi32( ni0, ni1 );
ni2 = _mm_packs_epi32( ni2, ni3 );
ni4 = _mm_packs_epi32( ni4, ni5 );
ni6 = _mm_packs_epi32( ni6, ni7 );
// Convert int16 to int8
ni0 = _mm_packs_epi16( ni0, ni2 );
ni4 = _mm_packs_epi16( ni4, ni6 );
_mm_storeu_si128((__m128i *)(y[i].qs + 0), ni0);
_mm_storeu_si128((__m128i *)(y[i].qs + 16), ni4);
#endif
}
#elif defined(__riscv_v_intrinsic)
size_t vl = __riscv_vsetvl_e32m4(QK8_0);
for (int i = 0; i < nb; i++) {
// load elements
vfloat32m4_t v_x = __riscv_vle32_v_f32m4(x+i*QK8_0, vl);
vfloat32m4_t vfabs = __riscv_vfabs_v_f32m4(v_x, vl);
vfloat32m1_t tmp = __riscv_vfmv_v_f_f32m1(0.0f, vl);
vfloat32m1_t vmax = __riscv_vfredmax_vs_f32m4_f32m1(vfabs, tmp, vl);
float amax = __riscv_vfmv_f_s_f32m1_f32(vmax);
const float d = amax / ((1 << 7) - 1);
const float id = d ? 1.0f/d : 0.0f;
ggml : move FP16 <-> FP32 code to ggml-impl.h (#3861) * ggml : move FP16 <-> FP32 stuff to ggml-impl.h ggml-ci * tests : fix ARM build * ggml : explicitly initialize deprecated type traits * ggml : add math.h to ggml-impl.h * ggml : remove duplicate static assert macros * ggml : prefix lookup tables with ggml_ ggml-ci * ggml-impl : move extern "C" to start of file
2023-10-30 17:19:15 +00:00
y[i].d = GGML_FP32_TO_FP16(d);
ggml : quantization refactoring (#3833) * ggml : factor all quantization code in ggml-quants ggml-ci * ggml-quants : fix Zig and Swift builds + quantize tool ggml-ci * quantize : --pure option for disabling k-quant mixtures --------- Co-authored-by: cebtenzzre <cebtenzzre@gmail.com>
2023-10-29 16:32:28 +00:00
vfloat32m4_t x0 = __riscv_vfmul_vf_f32m4(v_x, id, vl);
// convert to integer
vint16m2_t vi = __riscv_vfncvt_x_f_w_i16m2(x0, vl);
vint8m1_t vs = __riscv_vncvt_x_x_w_i8m1(vi, vl);
// store result
__riscv_vse8_v_i8m1(y[i].qs , vs, vl);
}
#else
ggml : fix UNUSED macro (#3762)
2023-11-01 11:50:45 +00:00
GGML_UNUSED(nb);
ggml : quantization refactoring (#3833) * ggml : factor all quantization code in ggml-quants ggml-ci * ggml-quants : fix Zig and Swift builds + quantize tool ggml-ci * quantize : --pure option for disabling k-quant mixtures --------- Co-authored-by: cebtenzzre <cebtenzzre@gmail.com>
2023-10-29 16:32:28 +00:00
// scalar
quantize_row_q8_0_reference(x, y, k);
#endif
}
// reference implementation for deterministic creation of model files
void quantize_row_q8_1_reference(const float * restrict x, block_q8_1 * restrict y, int k) {
assert(QK8_1 == 32);
assert(k % QK8_1 == 0);
const int nb = k / QK8_1;
for (int i = 0; i < nb; i++) {
float amax = 0.0f; // absolute max
for (int j = 0; j < QK8_1; j++) {
const float v = x[i*QK8_1 + j];
amax = MAX(amax, fabsf(v));
}
const float d = amax / ((1 << 7) - 1);
const float id = d ? 1.0f/d : 0.0f;
y[i].d = d;
int sum = 0;
for (int j = 0; j < QK8_1/2; ++j) {
const float v0 = x[i*QK8_1 + j]*id;
const float v1 = x[i*QK8_1 + QK8_1/2 + j]*id;
y[i].qs[ j] = roundf(v0);
y[i].qs[QK8_1/2 + j] = roundf(v1);
sum += y[i].qs[ j];
sum += y[i].qs[QK8_1/2 + j];
}
y[i].s = sum*d;
}
}
void quantize_row_q8_1(const float * restrict x, void * restrict vy, int k) {
assert(k % QK8_1 == 0);
const int nb = k / QK8_1;
block_q8_1 * restrict y = vy;
#if defined(__ARM_NEON)
for (int i = 0; i < nb; i++) {
float32x4_t srcv [8];
float32x4_t asrcv[8];
float32x4_t amaxv[8];
for (int j = 0; j < 8; j++) srcv[j] = vld1q_f32(x + i*32 + 4*j);
for (int j = 0; j < 8; j++) asrcv[j] = vabsq_f32(srcv[j]);
for (int j = 0; j < 4; j++) amaxv[2*j] = vmaxq_f32(asrcv[2*j], asrcv[2*j+1]);
for (int j = 0; j < 2; j++) amaxv[4*j] = vmaxq_f32(amaxv[4*j], amaxv[4*j+2]);
for (int j = 0; j < 1; j++) amaxv[8*j] = vmaxq_f32(amaxv[8*j], amaxv[8*j+4]);
const float amax = vmaxvq_f32(amaxv[0]);
const float d = amax / ((1 << 7) - 1);
const float id = d ? 1.0f/d : 0.0f;
y[i].d = d;
int32x4_t accv = vdupq_n_s32(0);
for (int j = 0; j < 8; j++) {
const float32x4_t v = vmulq_n_f32(srcv[j], id);
const int32x4_t vi = vcvtnq_s32_f32(v);
y[i].qs[4*j + 0] = vgetq_lane_s32(vi, 0);
y[i].qs[4*j + 1] = vgetq_lane_s32(vi, 1);
y[i].qs[4*j + 2] = vgetq_lane_s32(vi, 2);
y[i].qs[4*j + 3] = vgetq_lane_s32(vi, 3);
accv = vaddq_s32(accv, vi);
}
y[i].s = d * vaddvq_s32(accv);
}
#elif defined(__wasm_simd128__)
for (int i = 0; i < nb; i++) {
v128_t srcv [8];
v128_t asrcv[8];
v128_t amaxv[8];
for (int j = 0; j < 8; j++) srcv[j] = wasm_v128_load(x + i*32 + 4*j);
for (int j = 0; j < 8; j++) asrcv[j] = wasm_f32x4_abs(srcv[j]);
for (int j = 0; j < 4; j++) amaxv[2*j] = wasm_f32x4_max(asrcv[2*j], asrcv[2*j+1]);
for (int j = 0; j < 2; j++) amaxv[4*j] = wasm_f32x4_max(amaxv[4*j], amaxv[4*j+2]);
for (int j = 0; j < 1; j++) amaxv[8*j] = wasm_f32x4_max(amaxv[8*j], amaxv[8*j+4]);
const float amax = MAX(MAX(wasm_f32x4_extract_lane(amaxv[0], 0),
wasm_f32x4_extract_lane(amaxv[0], 1)),
MAX(wasm_f32x4_extract_lane(amaxv[0], 2),
wasm_f32x4_extract_lane(amaxv[0], 3)));
const float d = amax / ((1 << 7) - 1);
const float id = d ? 1.0f/d : 0.0f;
y[i].d = d;
v128_t accv = wasm_i32x4_splat(0);
for (int j = 0; j < 8; j++) {
const v128_t v = wasm_f32x4_mul(srcv[j], wasm_f32x4_splat(id));
const v128_t vi = wasm_i32x4_trunc_sat_f32x4(v);
y[i].qs[4*j + 0] = wasm_i32x4_extract_lane(vi, 0);
y[i].qs[4*j + 1] = wasm_i32x4_extract_lane(vi, 1);
y[i].qs[4*j + 2] = wasm_i32x4_extract_lane(vi, 2);
y[i].qs[4*j + 3] = wasm_i32x4_extract_lane(vi, 3);
accv = wasm_i32x4_add(accv, vi);
}
y[i].s = d * (wasm_i32x4_extract_lane(accv, 0) +
wasm_i32x4_extract_lane(accv, 1) +
wasm_i32x4_extract_lane(accv, 2) +
wasm_i32x4_extract_lane(accv, 3));
}
#elif defined(__AVX2__) || defined(__AVX__)
for (int i = 0; i < nb; i++) {
// Load elements into 4 AVX vectors
__m256 v0 = _mm256_loadu_ps( x );
__m256 v1 = _mm256_loadu_ps( x + 8 );
__m256 v2 = _mm256_loadu_ps( x + 16 );
__m256 v3 = _mm256_loadu_ps( x + 24 );
x += 32;
// Compute max(abs(e)) for the block
const __m256 signBit = _mm256_set1_ps( -0.0f );
__m256 maxAbs = _mm256_andnot_ps( signBit, v0 );
maxAbs = _mm256_max_ps( maxAbs, _mm256_andnot_ps( signBit, v1 ) );
maxAbs = _mm256_max_ps( maxAbs, _mm256_andnot_ps( signBit, v2 ) );
maxAbs = _mm256_max_ps( maxAbs, _mm256_andnot_ps( signBit, v3 ) );
__m128 max4 = _mm_max_ps( _mm256_extractf128_ps( maxAbs, 1 ), _mm256_castps256_ps128( maxAbs ) );
max4 = _mm_max_ps( max4, _mm_movehl_ps( max4, max4 ) );
max4 = _mm_max_ss( max4, _mm_movehdup_ps( max4 ) );
const float maxScalar = _mm_cvtss_f32( max4 );
// Quantize these floats
const float d = maxScalar / 127.f;
y[i].d = d;
const float id = ( maxScalar != 0.0f ) ? 127.f / maxScalar : 0.0f;
const __m256 mul = _mm256_set1_ps( id );
// Apply the multiplier
v0 = _mm256_mul_ps( v0, mul );
v1 = _mm256_mul_ps( v1, mul );
v2 = _mm256_mul_ps( v2, mul );
v3 = _mm256_mul_ps( v3, mul );
// Round to nearest integer
v0 = _mm256_round_ps( v0, _MM_ROUND_NEAREST );
v1 = _mm256_round_ps( v1, _MM_ROUND_NEAREST );
v2 = _mm256_round_ps( v2, _MM_ROUND_NEAREST );
v3 = _mm256_round_ps( v3, _MM_ROUND_NEAREST );
// Convert floats to integers
__m256i i0 = _mm256_cvtps_epi32( v0 );
__m256i i1 = _mm256_cvtps_epi32( v1 );
__m256i i2 = _mm256_cvtps_epi32( v2 );
__m256i i3 = _mm256_cvtps_epi32( v3 );
#if defined(__AVX2__)
// Compute the sum of the quants and set y[i].s
y[i].s = d * hsum_i32_8(_mm256_add_epi32(_mm256_add_epi32(i0, i1), _mm256_add_epi32(i2, i3)));
// Convert int32 to int16
i0 = _mm256_packs_epi32( i0, i1 ); // 0, 1, 2, 3, 8, 9, 10, 11, 4, 5, 6, 7, 12, 13, 14, 15
i2 = _mm256_packs_epi32( i2, i3 ); // 16, 17, 18, 19, 24, 25, 26, 27, 20, 21, 22, 23, 28, 29, 30, 31
// Convert int16 to int8
i0 = _mm256_packs_epi16( i0, i2 ); // 0, 1, 2, 3, 8, 9, 10, 11, 16, 17, 18, 19, 24, 25, 26, 27, 4, 5, 6, 7, 12, 13, 14, 15, 20, 21, 22, 23, 28, 29, 30, 31
// We got our precious signed bytes, but the order is now wrong
// These AVX2 pack instructions process 16-byte pieces independently
// The following instruction is fixing the order
const __m256i perm = _mm256_setr_epi32( 0, 4, 1, 5, 2, 6, 3, 7 );
i0 = _mm256_permutevar8x32_epi32( i0, perm );
_mm256_storeu_si256((__m256i *)y[i].qs, i0);
#else
// Since we don't have in AVX some necessary functions,
// we split the registers in half and call AVX2 analogs from SSE
__m128i ni0 = _mm256_castsi256_si128( i0 );
__m128i ni1 = _mm256_extractf128_si256( i0, 1);
__m128i ni2 = _mm256_castsi256_si128( i1 );
__m128i ni3 = _mm256_extractf128_si256( i1, 1);
__m128i ni4 = _mm256_castsi256_si128( i2 );
__m128i ni5 = _mm256_extractf128_si256( i2, 1);
__m128i ni6 = _mm256_castsi256_si128( i3 );
__m128i ni7 = _mm256_extractf128_si256( i3, 1);
// Compute the sum of the quants and set y[i].s
const __m128i s0 = _mm_add_epi32(_mm_add_epi32(ni0, ni1), _mm_add_epi32(ni2, ni3));
const __m128i s1 = _mm_add_epi32(_mm_add_epi32(ni4, ni5), _mm_add_epi32(ni6, ni7));
y[i].s = d * hsum_i32_4(_mm_add_epi32(s0, s1));
// Convert int32 to int16
ni0 = _mm_packs_epi32( ni0, ni1 );
ni2 = _mm_packs_epi32( ni2, ni3 );
ni4 = _mm_packs_epi32( ni4, ni5 );
ni6 = _mm_packs_epi32( ni6, ni7 );
// Convert int16 to int8
ni0 = _mm_packs_epi16( ni0, ni2 );
ni4 = _mm_packs_epi16( ni4, ni6 );
_mm_storeu_si128((__m128i *)(y[i].qs + 0), ni0);
_mm_storeu_si128((__m128i *)(y[i].qs + 16), ni4);
#endif
}
#elif defined(__riscv_v_intrinsic)
size_t vl = __riscv_vsetvl_e32m4(QK8_1);
for (int i = 0; i < nb; i++) {
// load elements
vfloat32m4_t v_x = __riscv_vle32_v_f32m4(x+i*QK8_1, vl);
vfloat32m4_t vfabs = __riscv_vfabs_v_f32m4(v_x, vl);
vfloat32m1_t tmp = __riscv_vfmv_v_f_f32m1(0.0, vl);
vfloat32m1_t vmax = __riscv_vfredmax_vs_f32m4_f32m1(vfabs, tmp, vl);
float amax = __riscv_vfmv_f_s_f32m1_f32(vmax);
const float d = amax / ((1 << 7) - 1);
const float id = d ? 1.0f/d : 0.0f;
y[i].d = d;
vfloat32m4_t x0 = __riscv_vfmul_vf_f32m4(v_x, id, vl);
// convert to integer
vint16m2_t vi = __riscv_vfncvt_x_f_w_i16m2(x0, vl);
vint8m1_t vs = __riscv_vncvt_x_x_w_i8m1(vi, vl);
// store result
__riscv_vse8_v_i8m1(y[i].qs , vs, vl);
// compute sum for y[i].s
vint16m1_t tmp2 = __riscv_vmv_v_x_i16m1(0, vl);
vint16m1_t vwrs = __riscv_vwredsum_vs_i8m1_i16m1(vs, tmp2, vl);
// set y[i].s
int sum = __riscv_vmv_x_s_i16m1_i16(vwrs);
y[i].s = sum*d;
}
#else
ggml : fix UNUSED macro (#3762)
2023-11-01 11:50:45 +00:00
GGML_UNUSED(nb);
ggml : quantization refactoring (#3833) * ggml : factor all quantization code in ggml-quants ggml-ci * ggml-quants : fix Zig and Swift builds + quantize tool ggml-ci * quantize : --pure option for disabling k-quant mixtures --------- Co-authored-by: cebtenzzre <cebtenzzre@gmail.com>
2023-10-29 16:32:28 +00:00
// scalar
quantize_row_q8_1_reference(x, y, k);
#endif
}
void dequantize_row_q4_0(const block_q4_0 * restrict x, float * restrict y, int k) {
static const int qk = QK4_0;
assert(k % qk == 0);
const int nb = k / qk;
for (int i = 0; i < nb; i++) {
ggml : move FP16 <-> FP32 code to ggml-impl.h (#3861) * ggml : move FP16 <-> FP32 stuff to ggml-impl.h ggml-ci * tests : fix ARM build * ggml : explicitly initialize deprecated type traits * ggml : add math.h to ggml-impl.h * ggml : remove duplicate static assert macros * ggml : prefix lookup tables with ggml_ ggml-ci * ggml-impl : move extern "C" to start of file
2023-10-30 17:19:15 +00:00
const float d = GGML_FP16_TO_FP32(x[i].d);
ggml : quantization refactoring (#3833) * ggml : factor all quantization code in ggml-quants ggml-ci * ggml-quants : fix Zig and Swift builds + quantize tool ggml-ci * quantize : --pure option for disabling k-quant mixtures --------- Co-authored-by: cebtenzzre <cebtenzzre@gmail.com>
2023-10-29 16:32:28 +00:00
for (int j = 0; j < qk/2; ++j) {
const int x0 = (x[i].qs[j] & 0x0F) - 8;
const int x1 = (x[i].qs[j] >> 4) - 8;
y[i*qk + j + 0 ] = x0*d;
y[i*qk + j + qk/2] = x1*d;
}
}
}
void dequantize_row_q4_1(const block_q4_1 * restrict x, float * restrict y, int k) {
static const int qk = QK4_1;
assert(k % qk == 0);
const int nb = k / qk;
for (int i = 0; i < nb; i++) {
ggml : move FP16 <-> FP32 code to ggml-impl.h (#3861) * ggml : move FP16 <-> FP32 stuff to ggml-impl.h ggml-ci * tests : fix ARM build * ggml : explicitly initialize deprecated type traits * ggml : add math.h to ggml-impl.h * ggml : remove duplicate static assert macros * ggml : prefix lookup tables with ggml_ ggml-ci * ggml-impl : move extern "C" to start of file
2023-10-30 17:19:15 +00:00
const float d = GGML_FP16_TO_FP32(x[i].d);
const float m = GGML_FP16_TO_FP32(x[i].m);
ggml : quantization refactoring (#3833) * ggml : factor all quantization code in ggml-quants ggml-ci * ggml-quants : fix Zig and Swift builds + quantize tool ggml-ci * quantize : --pure option for disabling k-quant mixtures --------- Co-authored-by: cebtenzzre <cebtenzzre@gmail.com>
2023-10-29 16:32:28 +00:00
for (int j = 0; j < qk/2; ++j) {
const int x0 = (x[i].qs[j] & 0x0F);
const int x1 = (x[i].qs[j] >> 4);
y[i*qk + j + 0 ] = x0*d + m;
y[i*qk + j + qk/2] = x1*d + m;
}
}
}
void dequantize_row_q5_0(const block_q5_0 * restrict x, float * restrict y, int k) {
static const int qk = QK5_0;
assert(k % qk == 0);
const int nb = k / qk;
for (int i = 0; i < nb; i++) {
ggml : move FP16 <-> FP32 code to ggml-impl.h (#3861) * ggml : move FP16 <-> FP32 stuff to ggml-impl.h ggml-ci * tests : fix ARM build * ggml : explicitly initialize deprecated type traits * ggml : add math.h to ggml-impl.h * ggml : remove duplicate static assert macros * ggml : prefix lookup tables with ggml_ ggml-ci * ggml-impl : move extern "C" to start of file
2023-10-30 17:19:15 +00:00
const float d = GGML_FP16_TO_FP32(x[i].d);
ggml : quantization refactoring (#3833) * ggml : factor all quantization code in ggml-quants ggml-ci * ggml-quants : fix Zig and Swift builds + quantize tool ggml-ci * quantize : --pure option for disabling k-quant mixtures --------- Co-authored-by: cebtenzzre <cebtenzzre@gmail.com>
2023-10-29 16:32:28 +00:00
uint32_t qh;
memcpy(&qh, x[i].qh, sizeof(qh));
for (int j = 0; j < qk/2; ++j) {
const uint8_t xh_0 = ((qh >> (j + 0)) << 4) & 0x10;
const uint8_t xh_1 = ((qh >> (j + 12)) ) & 0x10;
const int32_t x0 = ((x[i].qs[j] & 0x0F) | xh_0) - 16;
const int32_t x1 = ((x[i].qs[j] >> 4) | xh_1) - 16;
y[i*qk + j + 0 ] = x0*d;
y[i*qk + j + qk/2] = x1*d;
}
}
}
void dequantize_row_q5_1(const block_q5_1 * restrict x, float * restrict y, int k) {
static const int qk = QK5_1;
assert(k % qk == 0);
const int nb = k / qk;
for (int i = 0; i < nb; i++) {
ggml : move FP16 <-> FP32 code to ggml-impl.h (#3861) * ggml : move FP16 <-> FP32 stuff to ggml-impl.h ggml-ci * tests : fix ARM build * ggml : explicitly initialize deprecated type traits * ggml : add math.h to ggml-impl.h * ggml : remove duplicate static assert macros * ggml : prefix lookup tables with ggml_ ggml-ci * ggml-impl : move extern "C" to start of file
2023-10-30 17:19:15 +00:00
const float d = GGML_FP16_TO_FP32(x[i].d);
const float m = GGML_FP16_TO_FP32(x[i].m);
ggml : quantization refactoring (#3833) * ggml : factor all quantization code in ggml-quants ggml-ci * ggml-quants : fix Zig and Swift builds + quantize tool ggml-ci * quantize : --pure option for disabling k-quant mixtures --------- Co-authored-by: cebtenzzre <cebtenzzre@gmail.com>
2023-10-29 16:32:28 +00:00
uint32_t qh;
memcpy(&qh, x[i].qh, sizeof(qh));
for (int j = 0; j < qk/2; ++j) {
const uint8_t xh_0 = ((qh >> (j + 0)) << 4) & 0x10;
const uint8_t xh_1 = ((qh >> (j + 12)) ) & 0x10;
const int x0 = (x[i].qs[j] & 0x0F) | xh_0;
const int x1 = (x[i].qs[j] >> 4) | xh_1;
y[i*qk + j + 0 ] = x0*d + m;
y[i*qk + j + qk/2] = x1*d + m;
}
}
}
void dequantize_row_q8_0(const block_q8_0 * restrict x, float * restrict y, int k) {
static const int qk = QK8_0;
assert(k % qk == 0);
const int nb = k / qk;
for (int i = 0; i < nb; i++) {
ggml : move FP16 <-> FP32 code to ggml-impl.h (#3861) * ggml : move FP16 <-> FP32 stuff to ggml-impl.h ggml-ci * tests : fix ARM build * ggml : explicitly initialize deprecated type traits * ggml : add math.h to ggml-impl.h * ggml : remove duplicate static assert macros * ggml : prefix lookup tables with ggml_ ggml-ci * ggml-impl : move extern "C" to start of file
2023-10-30 17:19:15 +00:00
const float d = GGML_FP16_TO_FP32(x[i].d);
ggml : quantization refactoring (#3833) * ggml : factor all quantization code in ggml-quants ggml-ci * ggml-quants : fix Zig and Swift builds + quantize tool ggml-ci * quantize : --pure option for disabling k-quant mixtures --------- Co-authored-by: cebtenzzre <cebtenzzre@gmail.com>
2023-10-29 16:32:28 +00:00
for (int j = 0; j < qk; ++j) {
y[i*qk + j] = x[i].qs[j]*d;
}
}
}
//
// 2-6 bit quantization in super-blocks
//
//
// ===================== Helper functions
//
static inline int nearest_int(float fval) {
assert(fval <= 4194303.f);
float val = fval + 12582912.f;
int i; memcpy(&i, &val, sizeof(int));
return (i & 0x007fffff) - 0x00400000;
}
Add ability to use importance matrix for all k-quants (#4930) Co-authored-by: Iwan Kawrakow <iwan.kawrakow@gmail.com>
2024-01-14 14:21:12 +00:00
static float make_qx_quants(int n, int nmax, const float * restrict x, int8_t * restrict L, int rmse_type,
const float * restrict qw) {
ggml : quantization refactoring (#3833) * ggml : factor all quantization code in ggml-quants ggml-ci * ggml-quants : fix Zig and Swift builds + quantize tool ggml-ci * quantize : --pure option for disabling k-quant mixtures --------- Co-authored-by: cebtenzzre <cebtenzzre@gmail.com>
2023-10-29 16:32:28 +00:00
float max = 0;
float amax = 0;
for (int i = 0; i < n; ++i) {
float ax = fabsf(x[i]);
if (ax > amax) { amax = ax; max = x[i]; }
}
if (amax < 1e-30f) { // all zero
ggml : add SOTA 2,3,4,5,6 bit k-quantizations (#1684) * Starting to add k-quantization to ggml I think it is better to have quantization separate from ggml. For now just adding the k-quants there, but it would be better to also factor out the existing ggml quantizations. * Adding Q3_K and Q8_K (de)-quantization * Q3_K now working on CUDA and AVX2/scalar CUDA is not ideal - ~50% slower than Q4_0 for single token prediction, about the same in batch mode (perplexity). CPU single token is ~55 ms (on Ryzen 7950X). * Some improvement for Q3_K on CUDA It is now ~22.5 ms/token on my GPU, so ~30% slower than Q4_0. * Some more CUDA optimizations for Q3_K Single token is now 20.5 ms/token (~20% slower than Q4_0). Perplexity is on par with Q4_0. * Adding Q4_K - scalar, AVX2, CUDA Performance is the same or perhaps very slightly better than Q4_0 on the CPU. On the GPU, single token prediction is ~10% better than Q4_0, batch mode (perplexity is about the same). * Adding Q6_K - scalar, AVX2, CUDA Performance is ~40% lower compared to Q4_K on the CPU. This is to be expected, considering that we are memory bound on the CPU and the 6-bit model is ~44% larger than the 4-bit. On the GPU, single token prediction is ~6% lower than Q4_0, batch mode (perplexity) is even closer (but still slower). * Adding Q5_K - scalar, AVX2, CUDA Performance is ~20% lower compared to Q4_K on the CPU. This is to be expected, considering that we are memory bound on the CPU and the 5-bit model is ~22% larger than the 4-bit. On the GPU, single token prediction is about the same as Q4_0 for both, single token and batch prediction. * Per convention, all QX_K quantizations use Q5_K for output.weight * Adding quantization mixes * Quantization mixes: didn't quite get what I wanted in the last commit * Q4_K dot product for ARM_NEON * Q6_K dot product for ARM_NEON * Q5_K dot product for ARM_NEON * Adding Q3_K dot for ARM_NEON It is 22% slower than Q4_K, despite the smaller model size. On x86_64, where we are memory bound, the Q3_K model is quite a bit faster than Q4_K. * A very slightly faster ARM_NEON Q3_K dot * Adding Q2_K - just CUDA for now Token prediction is pretty good - about 15.5 ms on a RTX 4080. Perplexity is about the same as Q4_K. * Adding scalar and AVX2 Q2_K dot * Adding ARM_NEON Q2_K dot About the same performance as Q4_K. * A slightly faster ARM_NEON Q2_K dot Single token prediction is now ~36 ms on M2 Max. The code is much simpler too. * Fixed bug in Q2_K CUDA dot product kernel Stranegly enough, for the few prompts I tried with the 7B model the responses looked perfectly reasonable. Only realized something is not quite right when I tried the larger models and started getting nonse back. In any case, Q2_K single token evaluation time on an RTX 4080 in a Ryzen7950X box iusing CUDA and model fully loaded on the GPU are ~15.5 ms for 7B, ~25.4 ms for 13B, and ~55.8 ms for 30B. The max number of layers that fit in VRAM for The 65B is 32. With that, we get ~330 ms per token, which is not that much faster than just running on the CPU (~470 ms per token). * Don't print zeros/NaNs when no count histogram has been collected * A 10% faster CUDA vector dot kernel for Q3_K Q3_K is now running at ~18.5 ms / token on CUDA, so the gap to Q4_0 is only 10%. It seems memory acccess pattern is more important for performance than the amount of computation the kernel does. * A slightly daster Q4_K AVX2 dot product For perplexity, where we are less memory bound, time per pass drops by ~5%. Barely measurable difference for single token prediction. * A slightly faster ARM_NEON A4_K dot product * Minor * Fix quantization error test We cannot possibly be expecting rmse < 0.002 for 2- and 3-bit quantization variants. * Fix docker build I have been sloppy with vector reinterpret casts on ARM_NEON. It seems clang is very forgiving in that regard. * Added forgotten ggml.o dependence on k_quants.h to the Makefile * Had unintentionally committed the Makefile with -Ofast enabled * ggml : rename k_quants -> ggml-quants-k, use lowercase in code --------- Co-authored-by: Iwan Kawrakow <iwan.kawrakow@gmail.com> Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
2023-06-05 19:56:18 +00:00
for (int i = 0; i < n; ++i) {
ggml : quantization refactoring (#3833) * ggml : factor all quantization code in ggml-quants ggml-ci * ggml-quants : fix Zig and Swift builds + quantize tool ggml-ci * quantize : --pure option for disabling k-quant mixtures --------- Co-authored-by: cebtenzzre <cebtenzzre@gmail.com>
2023-10-29 16:32:28 +00:00
L[i] = 0;
}
return 0.f;
}
float iscale = -nmax / max;
if (rmse_type == 0) {
for (int i = 0; i < n; ++i) {
int l = nearest_int(iscale * x[i]);
L[i] = nmax + MAX(-nmax, MIN(nmax-1, l));
}
return 1/iscale;
}
bool return_early = false;
if (rmse_type < 0) {
rmse_type = -rmse_type;
return_early = true;
}
float sumlx = 0;
float suml2 = 0;
ggml : add IQ2 to test-backend-ops + refactoring (#4990) * ggml : add IQ2 to test-backend-ops + refactoring ggml-ci * cuda : update supports_op for IQ2 ggml-ci * ci : enable LLAMA_CUBLAS=1 for CUDA nodes ggml-ci * cuda : fix out-of-bounds-access in `mul_mat_vec_q` ggml-ci * tests : avoid creating RNGs for each Q tensor ggml-ci * tests : avoid creating RNGs for each tensor ggml-ci
2024-01-17 16:54:56 +00:00
#ifdef HAVE_BUGGY_APPLE_LINKER
// use 'volatile' to prevent unroll and work around a bug in Apple ld64 1015.7
for (volatile int i = 0; i < n; ++i) {
#else
ggml : quantization refactoring (#3833) * ggml : factor all quantization code in ggml-quants ggml-ci * ggml-quants : fix Zig and Swift builds + quantize tool ggml-ci * quantize : --pure option for disabling k-quant mixtures --------- Co-authored-by: cebtenzzre <cebtenzzre@gmail.com>
2023-10-29 16:32:28 +00:00
for (int i = 0; i < n; ++i) {
ggml : add IQ2 to test-backend-ops + refactoring (#4990) * ggml : add IQ2 to test-backend-ops + refactoring ggml-ci * cuda : update supports_op for IQ2 ggml-ci * ci : enable LLAMA_CUBLAS=1 for CUDA nodes ggml-ci * cuda : fix out-of-bounds-access in `mul_mat_vec_q` ggml-ci * tests : avoid creating RNGs for each Q tensor ggml-ci * tests : avoid creating RNGs for each tensor ggml-ci
2024-01-17 16:54:56 +00:00
#endif
ggml : quantization refactoring (#3833) * ggml : factor all quantization code in ggml-quants ggml-ci * ggml-quants : fix Zig and Swift builds + quantize tool ggml-ci * quantize : --pure option for disabling k-quant mixtures --------- Co-authored-by: cebtenzzre <cebtenzzre@gmail.com>
2023-10-29 16:32:28 +00:00
int l = nearest_int(iscale * x[i]);
l = MAX(-nmax, MIN(nmax-1, l));
L[i] = l + nmax;
Add ability to use importance matrix for all k-quants (#4930) Co-authored-by: Iwan Kawrakow <iwan.kawrakow@gmail.com>
2024-01-14 14:21:12 +00:00
float w = qw ? qw[i] : rmse_type == 1 ? x[i] * x[i] : rmse_type == 2 ? 1 : rmse_type == 3 ? fabsf(x[i]) : sqrtf(fabsf(x[i]));
ggml : quantization refactoring (#3833) * ggml : factor all quantization code in ggml-quants ggml-ci * ggml-quants : fix Zig and Swift builds + quantize tool ggml-ci * quantize : --pure option for disabling k-quant mixtures --------- Co-authored-by: cebtenzzre <cebtenzzre@gmail.com>
2023-10-29 16:32:28 +00:00
sumlx += w*x[i]*l;
suml2 += w*l*l;
}
float scale = sumlx/suml2;
if (return_early) return suml2 > 0 ? 0.5f*(scale + 1/iscale) : 1/iscale;
float best = scale * sumlx;
for (int is = -9; is <= 9; ++is) {
if (is == 0) {
continue;
}
iscale = -(nmax + 0.1f*is) / max;
sumlx = suml2 = 0;
for (int i = 0; i < n; ++i) {
int l = nearest_int(iscale * x[i]);
l = MAX(-nmax, MIN(nmax-1, l));
Add ability to use importance matrix for all k-quants (#4930) Co-authored-by: Iwan Kawrakow <iwan.kawrakow@gmail.com>
2024-01-14 14:21:12 +00:00
float w = qw ? qw[i] : rmse_type == 1 ? x[i] * x[i] : rmse_type == 2 ? 1 : rmse_type == 3 ? fabsf(x[i]) : sqrtf(fabsf(x[i]));
ggml : quantization refactoring (#3833) * ggml : factor all quantization code in ggml-quants ggml-ci * ggml-quants : fix Zig and Swift builds + quantize tool ggml-ci * quantize : --pure option for disabling k-quant mixtures --------- Co-authored-by: cebtenzzre <cebtenzzre@gmail.com>
2023-10-29 16:32:28 +00:00
sumlx += w*x[i]*l;
suml2 += w*l*l;
}
if (suml2 > 0 && sumlx*sumlx > best*suml2) {
for (int i = 0; i < n; ++i) {
int l = nearest_int(iscale * x[i]);
L[i] = nmax + MAX(-nmax, MIN(nmax-1, l));
}
scale = sumlx/suml2; best = scale*sumlx;
}
}
return scale;
}
static float make_q3_quants(int n, int nmax, const float * restrict x, int8_t * restrict L, bool do_rmse) {
float max = 0;
float amax = 0;
for (int i = 0; i < n; ++i) {
float ax = fabsf(x[i]);
if (ax > amax) { amax = ax; max = x[i]; }
}
if (!amax) { // all zero
for (int i = 0; i < n; ++i) { L[i] = 0; }
return 0.f;
}
float iscale = -nmax / max;
if (do_rmse) {
float sumlx = 0;
float suml2 = 0;
for (int i = 0; i < n; ++i) {
int l = nearest_int(iscale * x[i]);
l = MAX(-nmax, MIN(nmax-1, l));
L[i] = l;
float w = x[i]*x[i];
sumlx += w*x[i]*l;
suml2 += w*l*l;
}
for (int itry = 0; itry < 5; ++itry) {
int n_changed = 0;
for (int i = 0; i < n; ++i) {
float w = x[i]*x[i];
float slx = sumlx - w*x[i]*L[i];
if (slx > 0) {
float sl2 = suml2 - w*L[i]*L[i];
int new_l = nearest_int(x[i] * sl2 / slx);
new_l = MAX(-nmax, MIN(nmax-1, new_l));
if (new_l != L[i]) {
slx += w*x[i]*new_l;
sl2 += w*new_l*new_l;
if (sl2 > 0 && slx*slx*suml2 > sumlx*sumlx*sl2) {
L[i] = new_l; sumlx = slx; suml2 = sl2;
++n_changed;
}
}
}
}
if (!n_changed) {
break;
}
}
for (int i = 0; i < n; ++i) {
L[i] += nmax;
}
return sumlx / suml2;
}
for (int i = 0; i < n; ++i) {
int l = nearest_int(iscale * x[i]);
l = MAX(-nmax, MIN(nmax-1, l));
L[i] = l + nmax;
}
return 1/iscale;
}
static float make_qkx1_quants(int n, int nmax, const float * restrict x, uint8_t * restrict L, float * restrict the_min,
int ntry, float alpha) {
float min = x[0];
float max = x[0];
for (int i = 1; i < n; ++i) {
if (x[i] < min) min = x[i];
if (x[i] > max) max = x[i];
}
if (max == min) {
for (int i = 0; i < n; ++i) L[i] = 0;
*the_min = 0;
return 0.f;
}
if (min > 0) min = 0;
float iscale = nmax/(max - min);
float scale = 1/iscale;
for (int itry = 0; itry < ntry; ++itry) {
float sumlx = 0; int suml2 = 0;
bool did_change = false;
for (int i = 0; i < n; ++i) {
int l = nearest_int(iscale*(x[i] - min));
l = MAX(0, MIN(nmax, l));
if (l != L[i]) {
L[i] = l;
did_change = true;
}
sumlx += (x[i] - min)*l;
suml2 += l*l;
}
scale = sumlx/suml2;
float sum = 0;
for (int i = 0; i < n; ++i) {
sum += x[i] - scale*L[i];
}
min = alpha*min + (1 - alpha)*sum/n;
if (min > 0) min = 0;
iscale = 1/scale;
if (!did_change) break;
}
*the_min = -min;
return scale;
}
static float make_qkx2_quants(int n, int nmax, const float * restrict x, const float * restrict weights,
uint8_t * restrict L, float * restrict the_min, uint8_t * restrict Laux,
float rmin, float rdelta, int nstep, bool use_mad) {
float min = x[0];
float max = x[0];
float sum_w = weights[0];
float sum_x = sum_w * x[0];
Fix MacOS Sonoma model quantization (#4052) Co-authored-by: Jared Van Bortel <jared@nomic.ai> Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
2023-11-14 17:34:41 +00:00
#ifdef HAVE_BUGGY_APPLE_LINKER
// use 'volatile' to prevent unroll and work around a bug in Apple ld64 1015.7
for (volatile int i = 1; i < n; ++i) {
#else
ggml : quantization refactoring (#3833) * ggml : factor all quantization code in ggml-quants ggml-ci * ggml-quants : fix Zig and Swift builds + quantize tool ggml-ci * quantize : --pure option for disabling k-quant mixtures --------- Co-authored-by: cebtenzzre <cebtenzzre@gmail.com>
2023-10-29 16:32:28 +00:00
for (int i = 1; i < n; ++i) {
Fix MacOS Sonoma model quantization (#4052) Co-authored-by: Jared Van Bortel <jared@nomic.ai> Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
2023-11-14 17:34:41 +00:00
#endif
ggml : quantization refactoring (#3833) * ggml : factor all quantization code in ggml-quants ggml-ci * ggml-quants : fix Zig and Swift builds + quantize tool ggml-ci * quantize : --pure option for disabling k-quant mixtures --------- Co-authored-by: cebtenzzre <cebtenzzre@gmail.com>
2023-10-29 16:32:28 +00:00
if (x[i] < min) min = x[i];
if (x[i] > max) max = x[i];
float w = weights[i];
sum_w += w;
sum_x += w * x[i];
}
if (min > 0) min = 0;
if (max == min) {
for (int i = 0; i < n; ++i) L[i] = 0;
*the_min = -min;
return 0.f;
}
float iscale = nmax/(max - min);
float scale = 1/iscale;
float best_mad = 0;
for (int i = 0; i < n; ++i) {
int l = nearest_int(iscale*(x[i] - min));
L[i] = MAX(0, MIN(nmax, l));
float diff = scale * L[i] + min - x[i];
diff = use_mad ? fabsf(diff) : diff * diff;
float w = weights[i];
best_mad += w * diff;
}
if (nstep < 1) {
*the_min = -min;
return scale;
}
for (int is = 0; is <= nstep; ++is) {
iscale = (rmin + rdelta*is + nmax)/(max - min);
float sum_l = 0, sum_l2 = 0, sum_xl = 0;
for (int i = 0; i < n; ++i) {
int l = nearest_int(iscale*(x[i] - min));
l = MAX(0, MIN(nmax, l));
Laux[i] = l;
float w = weights[i];
sum_l += w*l;
sum_l2 += w*l*l;
sum_xl += w*l*x[i];
}
float D = sum_w * sum_l2 - sum_l * sum_l;
if (D > 0) {
float this_scale = (sum_w * sum_xl - sum_x * sum_l)/D;
float this_min = (sum_l2 * sum_x - sum_l * sum_xl)/D;
if (this_min > 0) {
this_min = 0;
this_scale = sum_xl / sum_l2;
}
float mad = 0;
for (int i = 0; i < n; ++i) {
float diff = this_scale * Laux[i] + this_min - x[i];
diff = use_mad ? fabsf(diff) : diff * diff;
float w = weights[i];
mad += w * diff;
}
if (mad < best_mad) {
for (int i = 0; i < n; ++i) {
L[i] = Laux[i];
}
best_mad = mad;
scale = this_scale;
min = this_min;
}
}
}
*the_min = -min;
return scale;
}
#if QK_K == 256
static inline void get_scale_min_k4(int j, const uint8_t * restrict q, uint8_t * restrict d, uint8_t * restrict m) {
if (j < 4) {
*d = q[j] & 63; *m = q[j + 4] & 63;
} else {
*d = (q[j+4] & 0xF) | ((q[j-4] >> 6) << 4);
*m = (q[j+4] >> 4) | ((q[j-0] >> 6) << 4);
}
}
#endif
//========================- 2-bit (de)-quantization
void quantize_row_q2_K_reference(const float * restrict x, block_q2_K * restrict y, int k) {
assert(k % QK_K == 0);
const int nb = k / QK_K;
uint8_t L[QK_K];
uint8_t Laux[16];
float weights[16];
float mins[QK_K/16];
float scales[QK_K/16];
const float q4scale = 15.f;
for (int i = 0; i < nb; i++) {
float max_scale = 0; // as we are deducting the min, scales are always positive
float max_min = 0;
for (int j = 0; j < QK_K/16; ++j) {
for (int l = 0; l < 16; ++l) weights[l] = fabsf(x[16*j + l]);
scales[j] = make_qkx2_quants(16, 3, x + 16*j, weights, L + 16*j, &mins[j], Laux, -0.5f, 0.1f, 15, true);
float scale = scales[j];
if (scale > max_scale) {
max_scale = scale;
}
float min = mins[j];
if (min > max_min) {
max_min = min;
}
}
if (max_scale > 0) {
float iscale = q4scale/max_scale;
for (int j = 0; j < QK_K/16; ++j) {
int l = nearest_int(iscale*scales[j]);
y[i].scales[j] = l;
}
ggml : move FP16 <-> FP32 code to ggml-impl.h (#3861) * ggml : move FP16 <-> FP32 stuff to ggml-impl.h ggml-ci * tests : fix ARM build * ggml : explicitly initialize deprecated type traits * ggml : add math.h to ggml-impl.h * ggml : remove duplicate static assert macros * ggml : prefix lookup tables with ggml_ ggml-ci * ggml-impl : move extern "C" to start of file
2023-10-30 17:19:15 +00:00
y[i].d = GGML_FP32_TO_FP16(max_scale/q4scale);
ggml : quantization refactoring (#3833) * ggml : factor all quantization code in ggml-quants ggml-ci * ggml-quants : fix Zig and Swift builds + quantize tool ggml-ci * quantize : --pure option for disabling k-quant mixtures --------- Co-authored-by: cebtenzzre <cebtenzzre@gmail.com>
2023-10-29 16:32:28 +00:00
} else {
for (int j = 0; j < QK_K/16; ++j) y[i].scales[j] = 0;
ggml : move FP16 <-> FP32 code to ggml-impl.h (#3861) * ggml : move FP16 <-> FP32 stuff to ggml-impl.h ggml-ci * tests : fix ARM build * ggml : explicitly initialize deprecated type traits * ggml : add math.h to ggml-impl.h * ggml : remove duplicate static assert macros * ggml : prefix lookup tables with ggml_ ggml-ci * ggml-impl : move extern "C" to start of file
2023-10-30 17:19:15 +00:00
y[i].d = GGML_FP32_TO_FP16(0.f);
ggml : quantization refactoring (#3833) * ggml : factor all quantization code in ggml-quants ggml-ci * ggml-quants : fix Zig and Swift builds + quantize tool ggml-ci * quantize : --pure option for disabling k-quant mixtures --------- Co-authored-by: cebtenzzre <cebtenzzre@gmail.com>
2023-10-29 16:32:28 +00:00
}
if (max_min > 0) {
float iscale = q4scale/max_min;
for (int j = 0; j < QK_K/16; ++j) {
int l = nearest_int(iscale*mins[j]);
y[i].scales[j] |= (l << 4);
}
ggml : move FP16 <-> FP32 code to ggml-impl.h (#3861) * ggml : move FP16 <-> FP32 stuff to ggml-impl.h ggml-ci * tests : fix ARM build * ggml : explicitly initialize deprecated type traits * ggml : add math.h to ggml-impl.h * ggml : remove duplicate static assert macros * ggml : prefix lookup tables with ggml_ ggml-ci * ggml-impl : move extern "C" to start of file
2023-10-30 17:19:15 +00:00
y[i].dmin = GGML_FP32_TO_FP16(max_min/q4scale);
ggml : quantization refactoring (#3833) * ggml : factor all quantization code in ggml-quants ggml-ci * ggml-quants : fix Zig and Swift builds + quantize tool ggml-ci * quantize : --pure option for disabling k-quant mixtures --------- Co-authored-by: cebtenzzre <cebtenzzre@gmail.com>
2023-10-29 16:32:28 +00:00
} else {
ggml : move FP16 <-> FP32 code to ggml-impl.h (#3861) * ggml : move FP16 <-> FP32 stuff to ggml-impl.h ggml-ci * tests : fix ARM build * ggml : explicitly initialize deprecated type traits * ggml : add math.h to ggml-impl.h * ggml : remove duplicate static assert macros * ggml : prefix lookup tables with ggml_ ggml-ci * ggml-impl : move extern "C" to start of file
2023-10-30 17:19:15 +00:00
y[i].dmin = GGML_FP32_TO_FP16(0.f);
ggml : quantization refactoring (#3833) * ggml : factor all quantization code in ggml-quants ggml-ci * ggml-quants : fix Zig and Swift builds + quantize tool ggml-ci * quantize : --pure option for disabling k-quant mixtures --------- Co-authored-by: cebtenzzre <cebtenzzre@gmail.com>
2023-10-29 16:32:28 +00:00
}
for (int j = 0; j < QK_K/16; ++j) {
ggml : move FP16 <-> FP32 code to ggml-impl.h (#3861) * ggml : move FP16 <-> FP32 stuff to ggml-impl.h ggml-ci * tests : fix ARM build * ggml : explicitly initialize deprecated type traits * ggml : add math.h to ggml-impl.h * ggml : remove duplicate static assert macros * ggml : prefix lookup tables with ggml_ ggml-ci * ggml-impl : move extern "C" to start of file
2023-10-30 17:19:15 +00:00
const float d = GGML_FP16_TO_FP32(y[i].d) * (y[i].scales[j] & 0xF);
ggml : quantization refactoring (#3833) * ggml : factor all quantization code in ggml-quants ggml-ci * ggml-quants : fix Zig and Swift builds + quantize tool ggml-ci * quantize : --pure option for disabling k-quant mixtures --------- Co-authored-by: cebtenzzre <cebtenzzre@gmail.com>
2023-10-29 16:32:28 +00:00
if (!d) continue;
ggml : move FP16 <-> FP32 code to ggml-impl.h (#3861) * ggml : move FP16 <-> FP32 stuff to ggml-impl.h ggml-ci * tests : fix ARM build * ggml : explicitly initialize deprecated type traits * ggml : add math.h to ggml-impl.h * ggml : remove duplicate static assert macros * ggml : prefix lookup tables with ggml_ ggml-ci * ggml-impl : move extern "C" to start of file
2023-10-30 17:19:15 +00:00
const float dm = GGML_FP16_TO_FP32(y[i].dmin) * (y[i].scales[j] >> 4);
ggml : quantization refactoring (#3833) * ggml : factor all quantization code in ggml-quants ggml-ci * ggml-quants : fix Zig and Swift builds + quantize tool ggml-ci * quantize : --pure option for disabling k-quant mixtures --------- Co-authored-by: cebtenzzre <cebtenzzre@gmail.com>
2023-10-29 16:32:28 +00:00
for (int ii = 0; ii < 16; ++ii) {
int l = nearest_int((x[16*j + ii] + dm)/d);
l = MAX(0, MIN(3, l));
L[16*j + ii] = l;
}
}
#if QK_K == 256
for (int j = 0; j < QK_K; j += 128) {
for (int l = 0; l < 32; ++l) {
y[i].qs[j/4 + l] = L[j + l] | (L[j + l + 32] << 2) | (L[j + l + 64] << 4) | (L[j + l + 96] << 6);
}
}
#else
for (int l = 0; l < 16; ++l) {
y[i].qs[l] = L[l] | (L[l + 16] << 2) | (L[l + 32] << 4) | (L[l + 48] << 6);
}
#endif
x += QK_K;
}
}
void dequantize_row_q2_K(const block_q2_K * restrict x, float * restrict y, int k) {
assert(k % QK_K == 0);
const int nb = k / QK_K;
for (int i = 0; i < nb; i++) {
ggml : move FP16 <-> FP32 code to ggml-impl.h (#3861) * ggml : move FP16 <-> FP32 stuff to ggml-impl.h ggml-ci * tests : fix ARM build * ggml : explicitly initialize deprecated type traits * ggml : add math.h to ggml-impl.h * ggml : remove duplicate static assert macros * ggml : prefix lookup tables with ggml_ ggml-ci * ggml-impl : move extern "C" to start of file
2023-10-30 17:19:15 +00:00
const float d = GGML_FP16_TO_FP32(x[i].d);
const float min = GGML_FP16_TO_FP32(x[i].dmin);
ggml : quantization refactoring (#3833) * ggml : factor all quantization code in ggml-quants ggml-ci * ggml-quants : fix Zig and Swift builds + quantize tool ggml-ci * quantize : --pure option for disabling k-quant mixtures --------- Co-authored-by: cebtenzzre <cebtenzzre@gmail.com>
2023-10-29 16:32:28 +00:00
const uint8_t * q = x[i].qs;
#if QK_K == 256
int is = 0;
float dl, ml;
for (int n = 0; n < QK_K; n += 128) {
int shift = 0;
for (int j = 0; j < 4; ++j) {
uint8_t sc = x[i].scales[is++];
dl = d * (sc & 0xF); ml = min * (sc >> 4);
for (int l = 0; l < 16; ++l) *y++ = dl * ((int8_t)((q[l] >> shift) & 3)) - ml;
sc = x[i].scales[is++];
dl = d * (sc & 0xF); ml = min * (sc >> 4);
for (int l = 0; l < 16; ++l) *y++ = dl * ((int8_t)((q[l+16] >> shift) & 3)) - ml;
shift += 2;
}
q += 32;
}
#else
float dl1 = d * (x[i].scales[0] & 0xF), ml1 = min * (x[i].scales[0] >> 4);
float dl2 = d * (x[i].scales[1] & 0xF), ml2 = min * (x[i].scales[1] >> 4);
float dl3 = d * (x[i].scales[2] & 0xF), ml3 = min * (x[i].scales[2] >> 4);
float dl4 = d * (x[i].scales[3] & 0xF), ml4 = min * (x[i].scales[3] >> 4);
for (int l = 0; l < 16; ++l) {
y[l+ 0] = dl1 * ((int8_t)((q[l] >> 0) & 3)) - ml1;
y[l+16] = dl2 * ((int8_t)((q[l] >> 2) & 3)) - ml2;
y[l+32] = dl3 * ((int8_t)((q[l] >> 4) & 3)) - ml3;
y[l+48] = dl4 * ((int8_t)((q[l] >> 6) & 3)) - ml4;
}
y += QK_K;
#endif
}
}
void quantize_row_q2_K(const float * restrict x, void * restrict vy, int k) {
quantize_row_q2_K_reference(x, vy, k);
}
size_t ggml_quantize_q2_K(const float * restrict src, void * restrict dst, int n, int k, int64_t * restrict hist) {
(void)hist; // TODO: collect histograms
for (int j = 0; j < n; j += k) {
block_q2_K * restrict y = (block_q2_K *)dst + j/QK_K;
quantize_row_q2_K_reference(src + j, y, k);
}
return (n/QK_K*sizeof(block_q2_K));
}
2-bit quantizations (#4897) * imatrix: load * imatrix: WIP * imatrix: Add Q2_K quantization * imatrix: also guard against Q2_K_S quantization without importance matrix * imatrix: guard even more against low-bit quantization misuse --------- Co-authored-by: Iwan Kawrakow <iwan.kawrakow@gmail.com>
2024-01-14 07:45:56 +00:00
static float make_qkx3_quants(int n, int nmax, const float * restrict x, const float * restrict weights,
uint8_t * restrict L, float * restrict the_min, uint8_t * restrict Laux,
float rmin, float rdelta, int nstep, bool use_mad) {
float min = x[0];
float max = x[0];
float sum_w = weights ? weights[0] : x[0]*x[0];
float sum_x = sum_w * x[0];
ggml : add IQ2 to test-backend-ops + refactoring (#4990) * ggml : add IQ2 to test-backend-ops + refactoring ggml-ci * cuda : update supports_op for IQ2 ggml-ci * ci : enable LLAMA_CUBLAS=1 for CUDA nodes ggml-ci * cuda : fix out-of-bounds-access in `mul_mat_vec_q` ggml-ci * tests : avoid creating RNGs for each Q tensor ggml-ci * tests : avoid creating RNGs for each tensor ggml-ci
2024-01-17 16:54:56 +00:00
#ifdef HAVE_BUGGY_APPLE_LINKER
// use 'volatile' to prevent unroll and work around a bug in Apple ld64 1015.7
for (volatile int i = 1; i < n; ++i) {
#else
2-bit quantizations (#4897) * imatrix: load * imatrix: WIP * imatrix: Add Q2_K quantization * imatrix: also guard against Q2_K_S quantization without importance matrix * imatrix: guard even more against low-bit quantization misuse --------- Co-authored-by: Iwan Kawrakow <iwan.kawrakow@gmail.com>
2024-01-14 07:45:56 +00:00
for (int i = 1; i < n; ++i) {
ggml : add IQ2 to test-backend-ops + refactoring (#4990) * ggml : add IQ2 to test-backend-ops + refactoring ggml-ci * cuda : update supports_op for IQ2 ggml-ci * ci : enable LLAMA_CUBLAS=1 for CUDA nodes ggml-ci * cuda : fix out-of-bounds-access in `mul_mat_vec_q` ggml-ci * tests : avoid creating RNGs for each Q tensor ggml-ci * tests : avoid creating RNGs for each tensor ggml-ci
2024-01-17 16:54:56 +00:00
#endif
2-bit quantizations (#4897) * imatrix: load * imatrix: WIP * imatrix: Add Q2_K quantization * imatrix: also guard against Q2_K_S quantization without importance matrix * imatrix: guard even more against low-bit quantization misuse --------- Co-authored-by: Iwan Kawrakow <iwan.kawrakow@gmail.com>
2024-01-14 07:45:56 +00:00
if (x[i] < min) min = x[i];
if (x[i] > max) max = x[i];
float w = weights ? weights[i] : x[i]*x[i];
sum_w += w;
sum_x += w * x[i];
}
if (min > 0) {
min = 0;
}
if (max <= min) {
ggml : add IQ2 to test-backend-ops + refactoring (#4990) * ggml : add IQ2 to test-backend-ops + refactoring ggml-ci * cuda : update supports_op for IQ2 ggml-ci * ci : enable LLAMA_CUBLAS=1 for CUDA nodes ggml-ci * cuda : fix out-of-bounds-access in `mul_mat_vec_q` ggml-ci * tests : avoid creating RNGs for each Q tensor ggml-ci * tests : avoid creating RNGs for each tensor ggml-ci
2024-01-17 16:54:56 +00:00
memset(L, 0, n);
2-bit quantizations (#4897) * imatrix: load * imatrix: WIP * imatrix: Add Q2_K quantization * imatrix: also guard against Q2_K_S quantization without importance matrix * imatrix: guard even more against low-bit quantization misuse --------- Co-authored-by: Iwan Kawrakow <iwan.kawrakow@gmail.com>
2024-01-14 07:45:56 +00:00
*the_min = -min;
return 0.f;
}
float iscale = nmax/(max - min);
float scale = 1/iscale;
float best_mad = 0;
for (int i = 0; i < n; ++i) {
int l = nearest_int(iscale*(x[i] - min));
L[i] = MAX(0, MIN(nmax, l));
float diff = scale * L[i] + min - x[i];
diff = use_mad ? fabsf(diff) : diff*diff;
float w = weights ? weights[i] : x[i]*x[i];
best_mad += w * diff;
}
if (nstep < 1) {
*the_min = -min;
return scale;
}
for (int is = 0; is <= nstep; ++is) {
iscale = (rmin + rdelta*is + nmax)/(max - min);
float sum_l = 0, sum_l2 = 0, sum_xl = 0;
for (int i = 0; i < n; ++i) {
int l = nearest_int(iscale*(x[i] - min));
l = MAX(0, MIN(nmax, l));
Laux[i] = l;
float w = weights ? weights[i] : x[i]*x[i];
sum_l += w*l;
sum_l2 += w*l*l;
sum_xl += w*l*x[i];
}
float D = sum_w * sum_l2 - sum_l * sum_l;
if (D > 0) {
float this_scale = (sum_w * sum_xl - sum_x * sum_l)/D;
float this_min = (sum_l2 * sum_x - sum_l * sum_xl)/D;
if (this_min > 0) {
this_min = 0;
this_scale = sum_xl / sum_l2;
}
float mad = 0;
for (int i = 0; i < n; ++i) {
float diff = this_scale * Laux[i] + this_min - x[i];
diff = use_mad ? fabsf(diff) : diff*diff;
float w = weights ? weights[i] : x[i]*x[i];
mad += w * diff;
}
if (mad < best_mad) {
for (int i = 0; i < n; ++i) {
L[i] = Laux[i];
}
best_mad = mad;
scale = this_scale;
min = this_min;
}
}
}
*the_min = -min;
return scale;
}
static float make_qp_quants(int n, int nmax, const float * restrict x, uint8_t * restrict L, const float * quant_weights) {
float max = 0;
for (int i = 0; i < n; ++i) {
max = MAX(max, x[i]);
}
if (!max) { // all zero
for (int i = 0; i < n; ++i) { L[i] = 0; }
return 0.f;
}
float iscale = nmax / max;
for (int i = 0; i < n; ++i) {
L[i] = nearest_int(iscale * x[i]);
}
float scale = 1/iscale;
float best_mse = 0;
for (int i = 0; i < n; ++i) {
float diff = x[i] - scale*L[i];
float w = quant_weights[i];
best_mse += w*diff*diff;
}
for (int is = -4; is <= 4; ++is) {
if (is == 0) continue;
float iscale_is = (0.1f*is + nmax)/max;
float scale_is = 1/iscale_is;
float mse = 0;
for (int i = 0; i < n; ++i) {
int l = nearest_int(iscale_is*x[i]);
l = MIN(nmax, l);
float diff = x[i] - scale_is*l;
float w = quant_weights[i];
mse += w*diff*diff;
}
if (mse < best_mse) {
best_mse = mse;
iscale = iscale_is;
}
}
float sumlx = 0;
float suml2 = 0;
for (int i = 0; i < n; ++i) {
int l = nearest_int(iscale * x[i]);
l = MIN(nmax, l);
L[i] = l;
float w = quant_weights[i];
sumlx += w*x[i]*l;
suml2 += w*l*l;
}
for (int itry = 0; itry < 5; ++itry) {
int n_changed = 0;
for (int i = 0; i < n; ++i) {
float w = quant_weights[i];
float slx = sumlx - w*x[i]*L[i];
float sl2 = suml2 - w*L[i]*L[i];
if (slx > 0 && sl2 > 0) {
int new_l = nearest_int(x[i] * sl2 / slx);
new_l = MIN(nmax, new_l);
if (new_l != L[i]) {
slx += w*x[i]*new_l;
sl2 += w*new_l*new_l;
if (slx*slx*suml2 > sumlx*sumlx*sl2) {
L[i] = new_l; sumlx = slx; suml2 = sl2;
++n_changed;
}
}
}
}
if (!n_changed) {
break;
}
}
return sumlx / suml2;
}
static void quantize_row_q2_K_impl(const float * restrict x, block_q2_K * restrict y, int k, const float * restrict quant_weights) {
GGML_ASSERT(quant_weights);
assert(k % QK_K == 0);
const int nb = k / QK_K;
const bool requantize = true;
uint8_t L[QK_K];
uint8_t Laux[16];
float mins[QK_K/16];
float scales[QK_K/16];
float sw[QK_K/16];
float weight[QK_K/16];
uint8_t Ls[QK_K/16], Lm[QK_K/16];
for (int i = 0; i < nb; i++) {
memset(sw, 0, QK_K/16*sizeof(float));
float sumx2 = 0;
for (int j = 0; j < QK_K; ++j) sumx2 += x[j]*x[j];
float sigma2 = sumx2/QK_K;
for (int j = 0; j < QK_K/16; ++j) {
const float * restrict qw = quant_weights + QK_K * i + 16*j;
for (int l = 0; l < 16; ++l) weight[l] = qw[l] * sqrtf(sigma2 + x[16*j + l]*x[16*j + l]);
for (int l = 0; l < 16; ++l) sw[j] += weight[l];
scales[j] = make_qkx3_quants(16, 3, x + 16*j, weight, L + 16*j, &mins[j], Laux, -0.9f, 0.05f, 36, false);
}
float dm = make_qp_quants(QK_K/16, 15, scales, Ls, sw);
float mm = make_qp_quants(QK_K/16, 15, mins, Lm, sw);
y[i].d = GGML_FP32_TO_FP16(dm);
y[i].dmin = GGML_FP32_TO_FP16(mm);
dm = GGML_FP16_TO_FP32(y[i].d);
mm = GGML_FP16_TO_FP32(y[i].dmin);
for (int j = 0; j < QK_K/16; ++j) {
y[i].scales[j] = Ls[j] | (Lm[j] << 4);
}
if (requantize) {
for (int j = 0; j < QK_K/16; ++j) {
const float d = dm * (y[i].scales[j] & 0xF);
if (!d) continue;
const float m = mm * (y[i].scales[j] >> 4);
for (int ii = 0; ii < 16; ++ii) {
int l = nearest_int((x[16*j + ii] + m)/d);
l = MAX(0, MIN(3, l));
L[16*j + ii] = l;
}
}
}
#if QK_K == 256
for (int j = 0; j < QK_K; j += 128) {
for (int l = 0; l < 32; ++l) {
y[i].qs[j/4 + l] = L[j + l] | (L[j + l + 32] << 2) | (L[j + l + 64] << 4) | (L[j + l + 96] << 6);
}
}
#else
for (int l = 0; l < 16; ++l) {
y[i].qs[l] = L[l] | (L[l + 16] << 2) | (L[l + 32] << 4) | (L[l + 48] << 6);
}
#endif
x += QK_K;
}
}
size_t quantize_q2_K(const float * src, void * dst, int nrow, int n_per_row, int64_t * hist, const float * quant_weights) {
(void)hist;
ggml : add IQ2 to test-backend-ops + refactoring (#4990) * ggml : add IQ2 to test-backend-ops + refactoring ggml-ci * cuda : update supports_op for IQ2 ggml-ci * ci : enable LLAMA_CUBLAS=1 for CUDA nodes ggml-ci * cuda : fix out-of-bounds-access in `mul_mat_vec_q` ggml-ci * tests : avoid creating RNGs for each Q tensor ggml-ci * tests : avoid creating RNGs for each tensor ggml-ci
2024-01-17 16:54:56 +00:00
size_t row_size = ggml_row_size(GGML_TYPE_Q2_K, n_per_row);
2-bit quantizations (#4897) * imatrix: load * imatrix: WIP * imatrix: Add Q2_K quantization * imatrix: also guard against Q2_K_S quantization without importance matrix * imatrix: guard even more against low-bit quantization misuse --------- Co-authored-by: Iwan Kawrakow <iwan.kawrakow@gmail.com>
2024-01-14 07:45:56 +00:00
if (!quant_weights) {
quantize_row_q2_K_reference(src, dst, nrow*n_per_row);
}
else {
char * qrow = (char *)dst;
for (int row = 0; row < nrow; ++row) {
quantize_row_q2_K_impl(src, (block_q2_K*)qrow, n_per_row, quant_weights);
src += n_per_row;
qrow += row_size;
}
}
return nrow * row_size;
}
ggml : quantization refactoring (#3833) * ggml : factor all quantization code in ggml-quants ggml-ci * ggml-quants : fix Zig and Swift builds + quantize tool ggml-ci * quantize : --pure option for disabling k-quant mixtures --------- Co-authored-by: cebtenzzre <cebtenzzre@gmail.com>
2023-10-29 16:32:28 +00:00
//========================= 3-bit (de)-quantization
void quantize_row_q3_K_reference(const float * restrict x, block_q3_K * restrict y, int k) {
assert(k % QK_K == 0);
const int nb = k / QK_K;
int8_t L[QK_K];
float scales[QK_K / 16];
for (int i = 0; i < nb; i++) {
float max_scale = 0;
float amax = 0;
for (int j = 0; j < QK_K/16; ++j) {
scales[j] = make_q3_quants(16, 4, x + 16*j, L + 16*j, true);
float scale = fabsf(scales[j]);
if (scale > amax) {
amax = scale; max_scale = scales[j];
}
}
#if QK_K == 256
memset(y[i].scales, 0, 12);
if (max_scale) {
float iscale = -32.f/max_scale;
for (int j = 0; j < QK_K/16; ++j) {
int8_t l = nearest_int(iscale*scales[j]);
l = MAX(-32, MIN(31, l)) + 32;
if (j < 8) {
y[i].scales[j] = l & 0xF;
} else {
y[i].scales[j-8] |= ((l & 0xF) << 4);
}
l >>= 4;
y[i].scales[j%4 + 8] |= (l << (2*(j/4)));
}
ggml : move FP16 <-> FP32 code to ggml-impl.h (#3861) * ggml : move FP16 <-> FP32 stuff to ggml-impl.h ggml-ci * tests : fix ARM build * ggml : explicitly initialize deprecated type traits * ggml : add math.h to ggml-impl.h * ggml : remove duplicate static assert macros * ggml : prefix lookup tables with ggml_ ggml-ci * ggml-impl : move extern "C" to start of file
2023-10-30 17:19:15 +00:00
y[i].d = GGML_FP32_TO_FP16(1/iscale);
ggml : quantization refactoring (#3833) * ggml : factor all quantization code in ggml-quants ggml-ci * ggml-quants : fix Zig and Swift builds + quantize tool ggml-ci * quantize : --pure option for disabling k-quant mixtures --------- Co-authored-by: cebtenzzre <cebtenzzre@gmail.com>
2023-10-29 16:32:28 +00:00
} else {
ggml : move FP16 <-> FP32 code to ggml-impl.h (#3861) * ggml : move FP16 <-> FP32 stuff to ggml-impl.h ggml-ci * tests : fix ARM build * ggml : explicitly initialize deprecated type traits * ggml : add math.h to ggml-impl.h * ggml : remove duplicate static assert macros * ggml : prefix lookup tables with ggml_ ggml-ci * ggml-impl : move extern "C" to start of file
2023-10-30 17:19:15 +00:00
y[i].d = GGML_FP32_TO_FP16(0.f);
ggml : quantization refactoring (#3833) * ggml : factor all quantization code in ggml-quants ggml-ci * ggml-quants : fix Zig and Swift builds + quantize tool ggml-ci * quantize : --pure option for disabling k-quant mixtures --------- Co-authored-by: cebtenzzre <cebtenzzre@gmail.com>
2023-10-29 16:32:28 +00:00
}
int8_t sc;
for (int j = 0; j < QK_K/16; ++j) {
sc = j < 8 ? y[i].scales[j] & 0xF : y[i].scales[j-8] >> 4;
sc = (sc | (((y[i].scales[8 + j%4] >> (2*(j/4))) & 3) << 4)) - 32;
ggml : move FP16 <-> FP32 code to ggml-impl.h (#3861) * ggml : move FP16 <-> FP32 stuff to ggml-impl.h ggml-ci * tests : fix ARM build * ggml : explicitly initialize deprecated type traits * ggml : add math.h to ggml-impl.h * ggml : remove duplicate static assert macros * ggml : prefix lookup tables with ggml_ ggml-ci * ggml-impl : move extern "C" to start of file
2023-10-30 17:19:15 +00:00
float d = GGML_FP16_TO_FP32(y[i].d) * sc;
ggml : quantization refactoring (#3833) * ggml : factor all quantization code in ggml-quants ggml-ci * ggml-quants : fix Zig and Swift builds + quantize tool ggml-ci * quantize : --pure option for disabling k-quant mixtures --------- Co-authored-by: cebtenzzre <cebtenzzre@gmail.com>
2023-10-29 16:32:28 +00:00
if (!d) {
continue;
}
for (int ii = 0; ii < 16; ++ii) {
int l = nearest_int(x[16*j + ii]/d);
l = MAX(-4, MIN(3, l));
L[16*j + ii] = l + 4;
}
}
#else
if (max_scale) {
float iscale = -8.f/max_scale;
for (int j = 0; j < QK_K/16; j+=2) {
int l1 = nearest_int(iscale*scales[j]);
l1 = 8 + MAX(-8, MIN(7, l1));
int l2 = nearest_int(iscale*scales[j+1]);
l2 = 8 + MAX(-8, MIN(7, l2));
y[i].scales[j/2] = l1 | (l2 << 4);
}
ggml : move FP16 <-> FP32 code to ggml-impl.h (#3861) * ggml : move FP16 <-> FP32 stuff to ggml-impl.h ggml-ci * tests : fix ARM build * ggml : explicitly initialize deprecated type traits * ggml : add math.h to ggml-impl.h * ggml : remove duplicate static assert macros * ggml : prefix lookup tables with ggml_ ggml-ci * ggml-impl : move extern "C" to start of file
2023-10-30 17:19:15 +00:00
y[i].d = GGML_FP32_TO_FP16(1/iscale);
ggml : quantization refactoring (#3833) * ggml : factor all quantization code in ggml-quants ggml-ci * ggml-quants : fix Zig and Swift builds + quantize tool ggml-ci * quantize : --pure option for disabling k-quant mixtures --------- Co-authored-by: cebtenzzre <cebtenzzre@gmail.com>
2023-10-29 16:32:28 +00:00
} else {
for (int j = 0; j < QK_K/16; j+=2) {
y[i].scales[j/2] = 0;
}
ggml : move FP16 <-> FP32 code to ggml-impl.h (#3861) * ggml : move FP16 <-> FP32 stuff to ggml-impl.h ggml-ci * tests : fix ARM build * ggml : explicitly initialize deprecated type traits * ggml : add math.h to ggml-impl.h * ggml : remove duplicate static assert macros * ggml : prefix lookup tables with ggml_ ggml-ci * ggml-impl : move extern "C" to start of file
2023-10-30 17:19:15 +00:00
y[i].d = GGML_FP32_TO_FP16(0.f);
ggml : quantization refactoring (#3833) * ggml : factor all quantization code in ggml-quants ggml-ci * ggml-quants : fix Zig and Swift builds + quantize tool ggml-ci * quantize : --pure option for disabling k-quant mixtures --------- Co-authored-by: cebtenzzre <cebtenzzre@gmail.com>
2023-10-29 16:32:28 +00:00
}
for (int j = 0; j < QK_K/16; ++j) {
int s = j%2 == 0 ? y[i].scales[j/2] & 0xF : y[i].scales[j/2] >> 4;
ggml : move FP16 <-> FP32 code to ggml-impl.h (#3861) * ggml : move FP16 <-> FP32 stuff to ggml-impl.h ggml-ci * tests : fix ARM build * ggml : explicitly initialize deprecated type traits * ggml : add math.h to ggml-impl.h * ggml : remove duplicate static assert macros * ggml : prefix lookup tables with ggml_ ggml-ci * ggml-impl : move extern "C" to start of file
2023-10-30 17:19:15 +00:00
float d = GGML_FP16_TO_FP32(y[i].d) * (s - 8);
ggml : quantization refactoring (#3833) * ggml : factor all quantization code in ggml-quants ggml-ci * ggml-quants : fix Zig and Swift builds + quantize tool ggml-ci * quantize : --pure option for disabling k-quant mixtures --------- Co-authored-by: cebtenzzre <cebtenzzre@gmail.com>
2023-10-29 16:32:28 +00:00
if (!d) {
continue;
}
for (int ii = 0; ii < 16; ++ii) {
int l = nearest_int(x[16*j + ii]/d);
l = MAX(-4, MIN(3, l));
L[16*j + ii] = l + 4;
}
}
#endif
memset(y[i].hmask, 0, QK_K/8);
// We put the high-bit for the 1st 8 quants into bit 0, the next 8 into bit 1, etc.
int m = 0;
uint8_t hm = 1;
for (int j = 0; j < QK_K; ++j) {
if (L[j] > 3) {
y[i].hmask[m] |= hm;
L[j] -= 4;
}
if (++m == QK_K/8) {
m = 0; hm <<= 1;
}
}
#if QK_K == 256
for (int j = 0; j < QK_K; j += 128) {
for (int l = 0; l < 32; ++l) {
y[i].qs[j/4 + l] = L[j + l] | (L[j + l + 32] << 2) | (L[j + l + 64] << 4) | (L[j + l + 96] << 6);
}
}
#else
for (int l = 0; l < 16; ++l) {
y[i].qs[l] = L[l] | (L[l + 16] << 2) | (L[l + 32] << 4) | (L[l + 48] << 6);
}
#endif
x += QK_K;
}
}
#if QK_K == 256
void dequantize_row_q3_K(const block_q3_K * restrict x, float * restrict y, int k) {
assert(k % QK_K == 0);
const int nb = k / QK_K;
const uint32_t kmask1 = 0x03030303;
const uint32_t kmask2 = 0x0f0f0f0f;
uint32_t aux[4];
const int8_t * scales = (const int8_t*)aux;
for (int i = 0; i < nb; i++) {
ggml : move FP16 <-> FP32 code to ggml-impl.h (#3861) * ggml : move FP16 <-> FP32 stuff to ggml-impl.h ggml-ci * tests : fix ARM build * ggml : explicitly initialize deprecated type traits * ggml : add math.h to ggml-impl.h * ggml : remove duplicate static assert macros * ggml : prefix lookup tables with ggml_ ggml-ci * ggml-impl : move extern "C" to start of file
2023-10-30 17:19:15 +00:00
const float d_all = GGML_FP16_TO_FP32(x[i].d);
ggml : quantization refactoring (#3833) * ggml : factor all quantization code in ggml-quants ggml-ci * ggml-quants : fix Zig and Swift builds + quantize tool ggml-ci * quantize : --pure option for disabling k-quant mixtures --------- Co-authored-by: cebtenzzre <cebtenzzre@gmail.com>
2023-10-29 16:32:28 +00:00
const uint8_t * restrict q = x[i].qs;
const uint8_t * restrict hm = x[i].hmask;
uint8_t m = 1;
memcpy(aux, x[i].scales, 12);
uint32_t tmp = aux[2];
aux[2] = ((aux[0] >> 4) & kmask2) | (((tmp >> 4) & kmask1) << 4);
aux[3] = ((aux[1] >> 4) & kmask2) | (((tmp >> 6) & kmask1) << 4);
aux[0] = (aux[0] & kmask2) | (((tmp >> 0) & kmask1) << 4);
aux[1] = (aux[1] & kmask2) | (((tmp >> 2) & kmask1) << 4);
int is = 0;
float dl;
for (int n = 0; n < QK_K; n += 128) {
int shift = 0;
for (int j = 0; j < 4; ++j) {
dl = d_all * (scales[is++] - 32);
for (int l = 0; l < 16; ++l) {
*y++ = dl * ((int8_t)((q[l+ 0] >> shift) & 3) - ((hm[l+ 0] & m) ? 0 : 4));
}
dl = d_all * (scales[is++] - 32);
for (int l = 0; l < 16; ++l) {
*y++ = dl * ((int8_t)((q[l+16] >> shift) & 3) - ((hm[l+16] & m) ? 0 : 4));
}
shift += 2;
m <<= 1;
}
q += 32;
}
}
}
#else
void dequantize_row_q3_K(const block_q3_K * restrict x, float * restrict y, int k) {
assert(k % QK_K == 0);
assert(QK_K == 64);
const int nb = k / QK_K;
for (int i = 0; i < nb; i++) {
ggml : move FP16 <-> FP32 code to ggml-impl.h (#3861) * ggml : move FP16 <-> FP32 stuff to ggml-impl.h ggml-ci * tests : fix ARM build * ggml : explicitly initialize deprecated type traits * ggml : add math.h to ggml-impl.h * ggml : remove duplicate static assert macros * ggml : prefix lookup tables with ggml_ ggml-ci * ggml-impl : move extern "C" to start of file
2023-10-30 17:19:15 +00:00
const float d_all = GGML_FP16_TO_FP32(x[i].d);
ggml : quantization refactoring (#3833) * ggml : factor all quantization code in ggml-quants ggml-ci * ggml-quants : fix Zig and Swift builds + quantize tool ggml-ci * quantize : --pure option for disabling k-quant mixtures --------- Co-authored-by: cebtenzzre <cebtenzzre@gmail.com>
2023-10-29 16:32:28 +00:00
const uint8_t * restrict q = x[i].qs;
const uint8_t * restrict hm = x[i].hmask;
const float d1 = d_all * ((x[i].scales[0] & 0xF) - 8);
const float d2 = d_all * ((x[i].scales[0] >> 4) - 8);
const float d3 = d_all * ((x[i].scales[1] & 0xF) - 8);
const float d4 = d_all * ((x[i].scales[1] >> 4) - 8);
for (int l=0; l<8; ++l) {
uint8_t h = hm[l];
y[l+ 0] = d1 * ((int8_t)((q[l+0] >> 0) & 3) - ((h & 0x01) ? 0 : 4));
y[l+ 8] = d1 * ((int8_t)((q[l+8] >> 0) & 3) - ((h & 0x02) ? 0 : 4));
y[l+16] = d2 * ((int8_t)((q[l+0] >> 2) & 3) - ((h & 0x04) ? 0 : 4));
y[l+24] = d2 * ((int8_t)((q[l+8] >> 2) & 3) - ((h & 0x08) ? 0 : 4));
y[l+32] = d3 * ((int8_t)((q[l+0] >> 4) & 3) - ((h & 0x10) ? 0 : 4));
y[l+40] = d3 * ((int8_t)((q[l+8] >> 4) & 3) - ((h & 0x20) ? 0 : 4));
y[l+48] = d4 * ((int8_t)((q[l+0] >> 6) & 3) - ((h & 0x40) ? 0 : 4));
y[l+56] = d4 * ((int8_t)((q[l+8] >> 6) & 3) - ((h & 0x80) ? 0 : 4));
}
y += QK_K;
}
}
#endif
void quantize_row_q3_K(const float * restrict x, void * restrict vy, int k) {
quantize_row_q3_K_reference(x, vy, k);
}
size_t ggml_quantize_q3_K(const float * restrict src, void * restrict dst, int n, int k, int64_t * restrict hist) {
(void)hist; // TODO: collect histograms
for (int j = 0; j < n; j += k) {
block_q3_K * restrict y = (block_q3_K *)dst + j/QK_K;
quantize_row_q3_K_reference(src + j, y, k);
}
return (n/QK_K*sizeof(block_q3_K));
}
Add ability to use importance matrix for all k-quants (#4930) Co-authored-by: Iwan Kawrakow <iwan.kawrakow@gmail.com>
2024-01-14 14:21:12 +00:00
static void quantize_row_q3_K_impl(const float * restrict x, block_q3_K * restrict y, int n_per_row, const float * restrict quant_weights) {
#if QK_K != 256
(void)quant_weights;
quantize_row_q3_K_reference(x, y, n_per_row);
#else
assert(n_per_row % QK_K == 0);
const int nb = n_per_row / QK_K;
int8_t L[QK_K];
float scales[QK_K / 16];
float weight[16];
float sw[QK_K / 16];
int8_t Ls[QK_K / 16];
for (int i = 0; i < nb; i++) {
float sumx2 = 0;
for (int j = 0; j < QK_K; ++j) sumx2 += x[j]*x[j];
float sigma2 = 2*sumx2/QK_K;
for (int j = 0; j < QK_K/16; ++j) {
if (quant_weights) {
const float * qw = quant_weights ? quant_weights + QK_K * i + 16*j : NULL;
for (int l = 0; l < 16; ++l) weight[l] = qw[l] * sqrtf(sigma2 + x[16*j+l]*x[16*j+l]);
} else {
for (int l = 0; l < 16; ++l) weight[l] = x[16*j+l]*x[16*j+l];
}
float sumw = 0;
for (int l = 0; l < 16; ++l) sumw += weight[l];
sw[j] = sumw;
scales[j] = make_qx_quants(16, 4, x + 16*j, L + 16*j, 1, weight);
}
memset(y[i].scales, 0, 12);
float d_block = make_qx_quants(QK_K/16, 32, scales, Ls, 1, sw);
for (int j = 0; j < QK_K/16; ++j) {
int l = Ls[j];
if (j < 8) {
y[i].scales[j] = l & 0xF;
} else {
y[i].scales[j-8] |= ((l & 0xF) << 4);
}
l >>= 4;
y[i].scales[j%4 + 8] |= (l << (2*(j/4)));
}
y[i].d = GGML_FP32_TO_FP16(d_block);
int8_t sc;
for (int j = 0; j < QK_K/16; ++j) {
sc = j < 8 ? y[i].scales[j] & 0xF : y[i].scales[j-8] >> 4;
sc = (sc | (((y[i].scales[8 + j%4] >> (2*(j/4))) & 3) << 4)) - 32;
float d = GGML_FP16_TO_FP32(y[i].d) * sc;
if (!d) {
continue;
}
for (int ii = 0; ii < 16; ++ii) {
int l = nearest_int(x[16*j + ii]/d);
l = MAX(-4, MIN(3, l));
L[16*j + ii] = l + 4;
}
}
memset(y[i].hmask, 0, QK_K/8);
// We put the high-bit for the 1st 8 quants into bit 0, the next 8 into bit 1, etc.
int m = 0;
uint8_t hm = 1;
for (int j = 0; j < QK_K; ++j) {
if (L[j] > 3) {
y[i].hmask[m] |= hm;
L[j] -= 4;
}
if (++m == QK_K/8) {
m = 0; hm <<= 1;
}
}
for (int j = 0; j < QK_K; j += 128) {
for (int l = 0; l < 32; ++l) {
y[i].qs[j/4 + l] = L[j + l] | (L[j + l + 32] << 2) | (L[j + l + 64] << 4) | (L[j + l + 96] << 6);
}
}
x += QK_K;
}
#endif
}
size_t quantize_q3_K(const float * src, void * dst, int nrow, int n_per_row, int64_t * hist, const float * quant_weights) {
(void)hist;
ggml : add IQ2 to test-backend-ops + refactoring (#4990) * ggml : add IQ2 to test-backend-ops + refactoring ggml-ci * cuda : update supports_op for IQ2 ggml-ci * ci : enable LLAMA_CUBLAS=1 for CUDA nodes ggml-ci * cuda : fix out-of-bounds-access in `mul_mat_vec_q` ggml-ci * tests : avoid creating RNGs for each Q tensor ggml-ci * tests : avoid creating RNGs for each tensor ggml-ci
2024-01-17 16:54:56 +00:00
size_t row_size = ggml_row_size(GGML_TYPE_Q3_K, n_per_row);
Add ability to use importance matrix for all k-quants (#4930) Co-authored-by: Iwan Kawrakow <iwan.kawrakow@gmail.com>
2024-01-14 14:21:12 +00:00
if (!quant_weights) {
quantize_row_q3_K_reference(src, dst, nrow*n_per_row);
}
else {
char * qrow = (char *)dst;
for (int row = 0; row < nrow; ++row) {
quantize_row_q3_K_impl(src, (block_q3_K*)qrow, n_per_row, quant_weights);
src += n_per_row;
qrow += row_size;
}
}
return nrow * row_size;
}
ggml : quantization refactoring (#3833) * ggml : factor all quantization code in ggml-quants ggml-ci * ggml-quants : fix Zig and Swift builds + quantize tool ggml-ci * quantize : --pure option for disabling k-quant mixtures --------- Co-authored-by: cebtenzzre <cebtenzzre@gmail.com>
2023-10-29 16:32:28 +00:00
// ====================== 4-bit (de)-quantization
void quantize_row_q4_K_reference(const float * restrict x, block_q4_K * restrict y, int k) {
assert(k % QK_K == 0);
const int nb = k / QK_K;
uint8_t L[QK_K];
uint8_t Laux[32];
float weights[32];
float mins[QK_K/32];
float scales[QK_K/32];
for (int i = 0; i < nb; i++) {
float max_scale = 0; // as we are deducting the min, scales are always positive
float max_min = 0;
for (int j = 0; j < QK_K/32; ++j) {
//scales[j] = make_qkx1_quants(32, 15, x + 32*j, L + 32*j, &mins[j], 9, 0.5f);
float sum_x2 = 0;
for (int l = 0; l < 32; ++l) sum_x2 += x[32*j + l] * x[32*j + l];
float av_x = sqrtf(sum_x2/32);
for (int l = 0; l < 32; ++l) weights[l] = av_x + fabsf(x[32*j + l]);
scales[j] = make_qkx2_quants(32, 15, x + 32*j, weights, L + 32*j, &mins[j], Laux, -1.f, 0.1f, 20, false);
float scale = scales[j];
if (scale > max_scale) {
max_scale = scale;
}
float min = mins[j];
if (min > max_min) {
max_min = min;
}
}
#if QK_K == 256
float inv_scale = max_scale > 0 ? 63.f/max_scale : 0.f;
float inv_min = max_min > 0 ? 63.f/max_min : 0.f;
for (int j = 0; j < QK_K/32; ++j) {
uint8_t ls = nearest_int(inv_scale*scales[j]);
uint8_t lm = nearest_int(inv_min*mins[j]);
ls = MIN(63, ls);
lm = MIN(63, lm);
if (j < 4) {
y[i].scales[j] = ls;
y[i].scales[j+4] = lm;
} else {
y[i].scales[j+4] = (ls & 0xF) | ((lm & 0xF) << 4);
y[i].scales[j-4] |= ((ls >> 4) << 6);
y[i].scales[j-0] |= ((lm >> 4) << 6);
}
}
ggml : move FP16 <-> FP32 code to ggml-impl.h (#3861) * ggml : move FP16 <-> FP32 stuff to ggml-impl.h ggml-ci * tests : fix ARM build * ggml : explicitly initialize deprecated type traits * ggml : add math.h to ggml-impl.h * ggml : remove duplicate static assert macros * ggml : prefix lookup tables with ggml_ ggml-ci * ggml-impl : move extern "C" to start of file
2023-10-30 17:19:15 +00:00
y[i].d = GGML_FP32_TO_FP16(max_scale/63.f);
y[i].dmin = GGML_FP32_TO_FP16(max_min/63.f);
ggml : quantization refactoring (#3833) * ggml : factor all quantization code in ggml-quants ggml-ci * ggml-quants : fix Zig and Swift builds + quantize tool ggml-ci * quantize : --pure option for disabling k-quant mixtures --------- Co-authored-by: cebtenzzre <cebtenzzre@gmail.com>
2023-10-29 16:32:28 +00:00
uint8_t sc, m;
for (int j = 0; j < QK_K/32; ++j) {
get_scale_min_k4(j, y[i].scales, &sc, &m);
ggml : move FP16 <-> FP32 code to ggml-impl.h (#3861) * ggml : move FP16 <-> FP32 stuff to ggml-impl.h ggml-ci * tests : fix ARM build * ggml : explicitly initialize deprecated type traits * ggml : add math.h to ggml-impl.h * ggml : remove duplicate static assert macros * ggml : prefix lookup tables with ggml_ ggml-ci * ggml-impl : move extern "C" to start of file
2023-10-30 17:19:15 +00:00
const float d = GGML_FP16_TO_FP32(y[i].d) * sc;
ggml : quantization refactoring (#3833) * ggml : factor all quantization code in ggml-quants ggml-ci * ggml-quants : fix Zig and Swift builds + quantize tool ggml-ci * quantize : --pure option for disabling k-quant mixtures --------- Co-authored-by: cebtenzzre <cebtenzzre@gmail.com>
2023-10-29 16:32:28 +00:00
if (!d) continue;
ggml : move FP16 <-> FP32 code to ggml-impl.h (#3861) * ggml : move FP16 <-> FP32 stuff to ggml-impl.h ggml-ci * tests : fix ARM build * ggml : explicitly initialize deprecated type traits * ggml : add math.h to ggml-impl.h * ggml : remove duplicate static assert macros * ggml : prefix lookup tables with ggml_ ggml-ci * ggml-impl : move extern "C" to start of file
2023-10-30 17:19:15 +00:00
const float dm = GGML_FP16_TO_FP32(y[i].dmin) * m;
ggml : quantization refactoring (#3833) * ggml : factor all quantization code in ggml-quants ggml-ci * ggml-quants : fix Zig and Swift builds + quantize tool ggml-ci * quantize : --pure option for disabling k-quant mixtures --------- Co-authored-by: cebtenzzre <cebtenzzre@gmail.com>
2023-10-29 16:32:28 +00:00
for (int ii = 0; ii < 32; ++ii) {
int l = nearest_int((x[32*j + ii] + dm)/d);
l = MAX(0, MIN(15, l));
L[32*j + ii] = l;
}
}
#else
const float s_factor = 15.f;
float inv_scale = max_scale > 0 ? s_factor/max_scale : 0.f;
float inv_min = max_min > 0 ? s_factor/max_min : 0.f;
int d1 = nearest_int(inv_scale*scales[0]);
int m1 = nearest_int(inv_min*mins[0]);
int d2 = nearest_int(inv_scale*scales[1]);
int m2 = nearest_int(inv_min*mins[1]);
y[i].scales[0] = d1 | (m1 << 4);
y[i].scales[1] = d2 | (m2 << 4);
ggml : move FP16 <-> FP32 code to ggml-impl.h (#3861) * ggml : move FP16 <-> FP32 stuff to ggml-impl.h ggml-ci * tests : fix ARM build * ggml : explicitly initialize deprecated type traits * ggml : add math.h to ggml-impl.h * ggml : remove duplicate static assert macros * ggml : prefix lookup tables with ggml_ ggml-ci * ggml-impl : move extern "C" to start of file
2023-10-30 17:19:15 +00:00
y[i].d[0] = GGML_FP32_TO_FP16(max_scale/s_factor);
y[i].d[1] = GGML_FP32_TO_FP16(max_min/s_factor);
ggml : quantization refactoring (#3833) * ggml : factor all quantization code in ggml-quants ggml-ci * ggml-quants : fix Zig and Swift builds + quantize tool ggml-ci * quantize : --pure option for disabling k-quant mixtures --------- Co-authored-by: cebtenzzre <cebtenzzre@gmail.com>
2023-10-29 16:32:28 +00:00
float sumlx = 0;
int suml2 = 0;
for (int j = 0; j < QK_K/32; ++j) {
const uint8_t sd = y[i].scales[j] & 0xF;
const uint8_t sm = y[i].scales[j] >> 4;
ggml : move FP16 <-> FP32 code to ggml-impl.h (#3861) * ggml : move FP16 <-> FP32 stuff to ggml-impl.h ggml-ci * tests : fix ARM build * ggml : explicitly initialize deprecated type traits * ggml : add math.h to ggml-impl.h * ggml : remove duplicate static assert macros * ggml : prefix lookup tables with ggml_ ggml-ci * ggml-impl : move extern "C" to start of file
2023-10-30 17:19:15 +00:00
const float d = GGML_FP16_TO_FP32(y[i].d[0]) * sd;
ggml : quantization refactoring (#3833) * ggml : factor all quantization code in ggml-quants ggml-ci * ggml-quants : fix Zig and Swift builds + quantize tool ggml-ci * quantize : --pure option for disabling k-quant mixtures --------- Co-authored-by: cebtenzzre <cebtenzzre@gmail.com>
2023-10-29 16:32:28 +00:00
if (!d) continue;
ggml : move FP16 <-> FP32 code to ggml-impl.h (#3861) * ggml : move FP16 <-> FP32 stuff to ggml-impl.h ggml-ci * tests : fix ARM build * ggml : explicitly initialize deprecated type traits * ggml : add math.h to ggml-impl.h * ggml : remove duplicate static assert macros * ggml : prefix lookup tables with ggml_ ggml-ci * ggml-impl : move extern "C" to start of file
2023-10-30 17:19:15 +00:00
const float m = GGML_FP16_TO_FP32(y[i].d[1]) * sm;
ggml : quantization refactoring (#3833) * ggml : factor all quantization code in ggml-quants ggml-ci * ggml-quants : fix Zig and Swift builds + quantize tool ggml-ci * quantize : --pure option for disabling k-quant mixtures --------- Co-authored-by: cebtenzzre <cebtenzzre@gmail.com>
2023-10-29 16:32:28 +00:00
for (int ii = 0; ii < 32; ++ii) {
int l = nearest_int((x[32*j + ii] + m)/d);
l = MAX(0, MIN(15, l));
L[32*j + ii] = l;
sumlx += (x[32*j + ii] + m)*l*sd;
suml2 += l*l*sd*sd;
}
}
if (suml2) {
ggml : move FP16 <-> FP32 code to ggml-impl.h (#3861) * ggml : move FP16 <-> FP32 stuff to ggml-impl.h ggml-ci * tests : fix ARM build * ggml : explicitly initialize deprecated type traits * ggml : add math.h to ggml-impl.h * ggml : remove duplicate static assert macros * ggml : prefix lookup tables with ggml_ ggml-ci * ggml-impl : move extern "C" to start of file
2023-10-30 17:19:15 +00:00
y[i].d[0] = GGML_FP32_TO_FP16(sumlx/suml2);
ggml : quantization refactoring (#3833) * ggml : factor all quantization code in ggml-quants ggml-ci * ggml-quants : fix Zig and Swift builds + quantize tool ggml-ci * quantize : --pure option for disabling k-quant mixtures --------- Co-authored-by: cebtenzzre <cebtenzzre@gmail.com>
2023-10-29 16:32:28 +00:00
}
#endif
uint8_t * q = y[i].qs;
for (int j = 0; j < QK_K; j += 64) {
for (int l = 0; l < 32; ++l) q[l] = L[j + l] | (L[j + l + 32] << 4);
q += 32;
}
x += QK_K;
}
}
void dequantize_row_q4_K(const block_q4_K * restrict x, float * restrict y, int k) {
assert(k % QK_K == 0);
const int nb = k / QK_K;
for (int i = 0; i < nb; i++) {
const uint8_t * q = x[i].qs;
#if QK_K == 256
ggml : move FP16 <-> FP32 code to ggml-impl.h (#3861) * ggml : move FP16 <-> FP32 stuff to ggml-impl.h ggml-ci * tests : fix ARM build * ggml : explicitly initialize deprecated type traits * ggml : add math.h to ggml-impl.h * ggml : remove duplicate static assert macros * ggml : prefix lookup tables with ggml_ ggml-ci * ggml-impl : move extern "C" to start of file
2023-10-30 17:19:15 +00:00
const float d = GGML_FP16_TO_FP32(x[i].d);
const float min = GGML_FP16_TO_FP32(x[i].dmin);
ggml : quantization refactoring (#3833) * ggml : factor all quantization code in ggml-quants ggml-ci * ggml-quants : fix Zig and Swift builds + quantize tool ggml-ci * quantize : --pure option for disabling k-quant mixtures --------- Co-authored-by: cebtenzzre <cebtenzzre@gmail.com>
2023-10-29 16:32:28 +00:00
int is = 0;
uint8_t sc, m;
for (int j = 0; j < QK_K; j += 64) {
get_scale_min_k4(is + 0, x[i].scales, &sc, &m);
const float d1 = d * sc; const float m1 = min * m;
get_scale_min_k4(is + 1, x[i].scales, &sc, &m);
const float d2 = d * sc; const float m2 = min * m;
for (int l = 0; l < 32; ++l) *y++ = d1 * (q[l] & 0xF) - m1;
for (int l = 0; l < 32; ++l) *y++ = d2 * (q[l] >> 4) - m2;
q += 32; is += 2;
}
#else
ggml : move FP16 <-> FP32 code to ggml-impl.h (#3861) * ggml : move FP16 <-> FP32 stuff to ggml-impl.h ggml-ci * tests : fix ARM build * ggml : explicitly initialize deprecated type traits * ggml : add math.h to ggml-impl.h * ggml : remove duplicate static assert macros * ggml : prefix lookup tables with ggml_ ggml-ci * ggml-impl : move extern "C" to start of file
2023-10-30 17:19:15 +00:00
const float dall = GGML_FP16_TO_FP32(x[i].d[0]);
const float mall = GGML_FP16_TO_FP32(x[i].d[1]);
ggml : quantization refactoring (#3833) * ggml : factor all quantization code in ggml-quants ggml-ci * ggml-quants : fix Zig and Swift builds + quantize tool ggml-ci * quantize : --pure option for disabling k-quant mixtures --------- Co-authored-by: cebtenzzre <cebtenzzre@gmail.com>
2023-10-29 16:32:28 +00:00
const float d1 = dall * (x[i].scales[0] & 0xF), m1 = mall * (x[i].scales[0] >> 4);
const float d2 = dall * (x[i].scales[1] & 0xF), m2 = mall * (x[i].scales[1] >> 4);
for (int l = 0; l < 32; ++l) {
y[l+ 0] = d1 * (q[l] & 0xF) - m1;
y[l+32] = d2 * (q[l] >> 4) - m2;
}
y += QK_K;
#endif
}
}
void quantize_row_q4_K(const float * restrict x, void * restrict vy, int k) {
assert(k % QK_K == 0);
block_q4_K * restrict y = vy;
quantize_row_q4_K_reference(x, y, k);
}
size_t ggml_quantize_q4_K(const float * restrict src, void * restrict dst, int n, int k, int64_t * restrict hist) {
assert(k % QK_K == 0);
(void)hist; // TODO: collect histograms
for (int j = 0; j < n; j += k) {
block_q4_K * restrict y = (block_q4_K *)dst + j/QK_K;
quantize_row_q4_K_reference(src + j, y, k);
}
return (n/QK_K*sizeof(block_q4_K));
}
Add ability to use importance matrix for all k-quants (#4930) Co-authored-by: Iwan Kawrakow <iwan.kawrakow@gmail.com>
2024-01-14 14:21:12 +00:00
static void quantize_row_q4_K_impl(const float * restrict x, block_q4_K * restrict y, int n_per_row, const float * quant_weights) {
#if QK_K != 256
(void)quant_weights;
quantize_row_q4_K_reference(x, y, n_per_row);
#else
assert(n_per_row % QK_K == 0);
const int nb = n_per_row / QK_K;
uint8_t L[QK_K];
uint8_t Laux[32];
Slight quantization improvement for Q4_K and Q5_K (#5361) * Q4_K: slightly better quantization * Q5_K: slightly better quantization --------- Co-authored-by: Iwan Kawrakow <iwan.kawrakow@gmail.com>
2024-02-06 15:28:02 +00:00
uint8_t Ls[QK_K/32];
uint8_t Lm[QK_K/32];
Add ability to use importance matrix for all k-quants (#4930) Co-authored-by: Iwan Kawrakow <iwan.kawrakow@gmail.com>
2024-01-14 14:21:12 +00:00
float weights[32];
Slight quantization improvement for Q4_K and Q5_K (#5361) * Q4_K: slightly better quantization * Q5_K: slightly better quantization --------- Co-authored-by: Iwan Kawrakow <iwan.kawrakow@gmail.com>
2024-02-06 15:28:02 +00:00
float sw[QK_K/32];
float mins[QK_K/32];
float scales[QK_K/32];
Add ability to use importance matrix for all k-quants (#4930) Co-authored-by: Iwan Kawrakow <iwan.kawrakow@gmail.com>
2024-01-14 14:21:12 +00:00
for (int i = 0; i < nb; i++) {
float sum_x2 = 0;
for (int l = 0; l < QK_K; ++l) sum_x2 += x[l] * x[l];
Slight quantization improvement for Q4_K and Q5_K (#5361) * Q4_K: slightly better quantization * Q5_K: slightly better quantization --------- Co-authored-by: Iwan Kawrakow <iwan.kawrakow@gmail.com>
2024-02-06 15:28:02 +00:00
float sigma2 = 2*sum_x2/QK_K;
Add ability to use importance matrix for all k-quants (#4930) Co-authored-by: Iwan Kawrakow <iwan.kawrakow@gmail.com>
2024-01-14 14:21:12 +00:00
float av_x = sqrtf(sigma2);
for (int j = 0; j < QK_K/32; ++j) {
if (quant_weights) {
const float * qw = quant_weights + QK_K*i + 32*j;
for (int l = 0; l < 32; ++l) weights[l] = qw[l] * sqrtf(sigma2 + x[32*j + l]*x[32*j + l]);
} else {
for (int l = 0; l < 32; ++l) weights[l] = av_x + fabsf(x[32*j + l]);
}
Slight quantization improvement for Q4_K and Q5_K (#5361) * Q4_K: slightly better quantization * Q5_K: slightly better quantization --------- Co-authored-by: Iwan Kawrakow <iwan.kawrakow@gmail.com>
2024-02-06 15:28:02 +00:00
float sumw = 0;
for (int l = 0; l < 32; ++l) sumw += weights[l];
sw[j] = sumw;
Add ability to use importance matrix for all k-quants (#4930) Co-authored-by: Iwan Kawrakow <iwan.kawrakow@gmail.com>
2024-01-14 14:21:12 +00:00
scales[j] = make_qkx3_quants(32, 15, x + 32*j, weights, L + 32*j, &mins[j], Laux, -0.9f, 0.05f, 36, false);
}
Slight quantization improvement for Q4_K and Q5_K (#5361) * Q4_K: slightly better quantization * Q5_K: slightly better quantization --------- Co-authored-by: Iwan Kawrakow <iwan.kawrakow@gmail.com>
2024-02-06 15:28:02 +00:00
float d_block = make_qp_quants(QK_K/32, 63, scales, Ls, sw);
float m_block = make_qp_quants(QK_K/32, 63, mins, Lm, sw);
Add ability to use importance matrix for all k-quants (#4930) Co-authored-by: Iwan Kawrakow <iwan.kawrakow@gmail.com>
2024-01-14 14:21:12 +00:00
for (int j = 0; j < QK_K/32; ++j) {
Slight quantization improvement for Q4_K and Q5_K (#5361) * Q4_K: slightly better quantization * Q5_K: slightly better quantization --------- Co-authored-by: Iwan Kawrakow <iwan.kawrakow@gmail.com>
2024-02-06 15:28:02 +00:00
uint8_t ls = Ls[j];
uint8_t lm = Lm[j];
Add ability to use importance matrix for all k-quants (#4930) Co-authored-by: Iwan Kawrakow <iwan.kawrakow@gmail.com>
2024-01-14 14:21:12 +00:00
if (j < 4) {
y[i].scales[j] = ls;
y[i].scales[j+4] = lm;
} else {
y[i].scales[j+4] = (ls & 0xF) | ((lm & 0xF) << 4);
y[i].scales[j-4] |= ((ls >> 4) << 6);
y[i].scales[j-0] |= ((lm >> 4) << 6);
}
}
Slight quantization improvement for Q4_K and Q5_K (#5361) * Q4_K: slightly better quantization * Q5_K: slightly better quantization --------- Co-authored-by: Iwan Kawrakow <iwan.kawrakow@gmail.com>
2024-02-06 15:28:02 +00:00
y[i].d = GGML_FP32_TO_FP16(d_block);
y[i].dmin = GGML_FP32_TO_FP16(m_block);
Add ability to use importance matrix for all k-quants (#4930) Co-authored-by: Iwan Kawrakow <iwan.kawrakow@gmail.com>
2024-01-14 14:21:12 +00:00
uint8_t sc, m;
for (int j = 0; j < QK_K/32; ++j) {
get_scale_min_k4(j, y[i].scales, &sc, &m);
const float d = GGML_FP16_TO_FP32(y[i].d) * sc;
if (!d) continue;
const float dm = GGML_FP16_TO_FP32(y[i].dmin) * m;
for (int ii = 0; ii < 32; ++ii) {
int l = nearest_int((x[32*j + ii] + dm)/d);
l = MAX(0, MIN(15, l));
L[32*j + ii] = l;
}
}
uint8_t * q = y[i].qs;
for (int j = 0; j < QK_K; j += 64) {
for (int l = 0; l < 32; ++l) q[l] = L[j + l] | (L[j + l + 32] << 4);
q += 32;
}
x += QK_K;
}
#endif
}
size_t quantize_q4_K(const float * src, void * dst, int nrow, int n_per_row, int64_t * hist, const float * quant_weights) {
(void)hist;
ggml : add IQ2 to test-backend-ops + refactoring (#4990) * ggml : add IQ2 to test-backend-ops + refactoring ggml-ci * cuda : update supports_op for IQ2 ggml-ci * ci : enable LLAMA_CUBLAS=1 for CUDA nodes ggml-ci * cuda : fix out-of-bounds-access in `mul_mat_vec_q` ggml-ci * tests : avoid creating RNGs for each Q tensor ggml-ci * tests : avoid creating RNGs for each tensor ggml-ci
2024-01-17 16:54:56 +00:00
size_t row_size = ggml_row_size(GGML_TYPE_Q4_K, n_per_row);
Add ability to use importance matrix for all k-quants (#4930) Co-authored-by: Iwan Kawrakow <iwan.kawrakow@gmail.com>
2024-01-14 14:21:12 +00:00
if (!quant_weights) {
quantize_row_q4_K_reference(src, dst, nrow*n_per_row);
}
else {
char * qrow = (char *)dst;
for (int row = 0; row < nrow; ++row) {
quantize_row_q4_K_impl(src, (block_q4_K*)qrow, n_per_row, quant_weights);
src += n_per_row;
qrow += row_size;
}
}
return nrow * row_size;
}
ggml : quantization refactoring (#3833) * ggml : factor all quantization code in ggml-quants ggml-ci * ggml-quants : fix Zig and Swift builds + quantize tool ggml-ci * quantize : --pure option for disabling k-quant mixtures --------- Co-authored-by: cebtenzzre <cebtenzzre@gmail.com>
2023-10-29 16:32:28 +00:00
// ====================== 5-bit (de)-quantization
void quantize_row_q5_K_reference(const float * restrict x, block_q5_K * restrict y, int k) {
assert(k % QK_K == 0);
const int nb = k / QK_K;
#if QK_K == 256
uint8_t L[QK_K];
float mins[QK_K/32];
float scales[QK_K/32];
float weights[32];
uint8_t Laux[32];
#else
int8_t L[QK_K];
float scales[QK_K/16];
#endif
for (int i = 0; i < nb; i++) {
#if QK_K == 256
float max_scale = 0; // as we are deducting the min, scales are always positive
float max_min = 0;
for (int j = 0; j < QK_K/32; ++j) {
//scales[j] = make_qkx1_quants(32, 31, x + 32*j, L + 32*j, &mins[j], 9, 0.5f);
float sum_x2 = 0;
for (int l = 0; l < 32; ++l) sum_x2 += x[32*j + l] * x[32*j + l];
float av_x = sqrtf(sum_x2/32);
for (int l = 0; l < 32; ++l) weights[l] = av_x + fabsf(x[32*j + l]);
scales[j] = make_qkx2_quants(32, 31, x + 32*j, weights, L + 32*j, &mins[j], Laux, -0.5f, 0.1f, 15, false);
float scale = scales[j];
if (scale > max_scale) {
max_scale = scale;
}
float min = mins[j];
if (min > max_min) {
max_min = min;
}
}
float inv_scale = max_scale > 0 ? 63.f/max_scale : 0.f;
float inv_min = max_min > 0 ? 63.f/max_min : 0.f;
for (int j = 0; j < QK_K/32; ++j) {
uint8_t ls = nearest_int(inv_scale*scales[j]);
uint8_t lm = nearest_int(inv_min*mins[j]);
ls = MIN(63, ls);
lm = MIN(63, lm);
if (j < 4) {
y[i].scales[j] = ls;
y[i].scales[j+4] = lm;
} else {
y[i].scales[j+4] = (ls & 0xF) | ((lm & 0xF) << 4);
y[i].scales[j-4] |= ((ls >> 4) << 6);
y[i].scales[j-0] |= ((lm >> 4) << 6);
}
}
ggml : move FP16 <-> FP32 code to ggml-impl.h (#3861) * ggml : move FP16 <-> FP32 stuff to ggml-impl.h ggml-ci * tests : fix ARM build * ggml : explicitly initialize deprecated type traits * ggml : add math.h to ggml-impl.h * ggml : remove duplicate static assert macros * ggml : prefix lookup tables with ggml_ ggml-ci * ggml-impl : move extern "C" to start of file
2023-10-30 17:19:15 +00:00
y[i].d = GGML_FP32_TO_FP16(max_scale/63.f);
y[i].dmin = GGML_FP32_TO_FP16(max_min/63.f);
ggml : quantization refactoring (#3833) * ggml : factor all quantization code in ggml-quants ggml-ci * ggml-quants : fix Zig and Swift builds + quantize tool ggml-ci * quantize : --pure option for disabling k-quant mixtures --------- Co-authored-by: cebtenzzre <cebtenzzre@gmail.com>
2023-10-29 16:32:28 +00:00
uint8_t sc, m;
for (int j = 0; j < QK_K/32; ++j) {
get_scale_min_k4(j, y[i].scales, &sc, &m);
ggml : move FP16 <-> FP32 code to ggml-impl.h (#3861) * ggml : move FP16 <-> FP32 stuff to ggml-impl.h ggml-ci * tests : fix ARM build * ggml : explicitly initialize deprecated type traits * ggml : add math.h to ggml-impl.h * ggml : remove duplicate static assert macros * ggml : prefix lookup tables with ggml_ ggml-ci * ggml-impl : move extern "C" to start of file
2023-10-30 17:19:15 +00:00
const float d = GGML_FP16_TO_FP32(y[i].d) * sc;
ggml : quantization refactoring (#3833) * ggml : factor all quantization code in ggml-quants ggml-ci * ggml-quants : fix Zig and Swift builds + quantize tool ggml-ci * quantize : --pure option for disabling k-quant mixtures --------- Co-authored-by: cebtenzzre <cebtenzzre@gmail.com>
2023-10-29 16:32:28 +00:00
if (!d) continue;
ggml : move FP16 <-> FP32 code to ggml-impl.h (#3861) * ggml : move FP16 <-> FP32 stuff to ggml-impl.h ggml-ci * tests : fix ARM build * ggml : explicitly initialize deprecated type traits * ggml : add math.h to ggml-impl.h * ggml : remove duplicate static assert macros * ggml : prefix lookup tables with ggml_ ggml-ci * ggml-impl : move extern "C" to start of file
2023-10-30 17:19:15 +00:00
const float dm = GGML_FP16_TO_FP32(y[i].dmin) * m;
ggml : quantization refactoring (#3833) * ggml : factor all quantization code in ggml-quants ggml-ci * ggml-quants : fix Zig and Swift builds + quantize tool ggml-ci * quantize : --pure option for disabling k-quant mixtures --------- Co-authored-by: cebtenzzre <cebtenzzre@gmail.com>
2023-10-29 16:32:28 +00:00
for (int ii = 0; ii < 32; ++ii) {
int l = nearest_int((x[32*j + ii] + dm)/d);
l = MAX(0, MIN(31, l));
L[32*j + ii] = l;
}
}
uint8_t * restrict qh = y[i].qh;
uint8_t * restrict ql = y[i].qs;
memset(qh, 0, QK_K/8);
uint8_t m1 = 1, m2 = 2;
for (int n = 0; n < QK_K; n += 64) {
for (int j = 0; j < 32; ++j) {
int l1 = L[n + j];
if (l1 > 15) {
l1 -= 16; qh[j] |= m1;
}
int l2 = L[n + j + 32];
if (l2 > 15) {
l2 -= 16; qh[j] |= m2;
}
ql[j] = l1 | (l2 << 4);
}
m1 <<= 2; m2 <<= 2;
ql += 32;
}
#else
float max_scale = 0, amax = 0;
for (int j = 0; j < QK_K/16; ++j) {
Add ability to use importance matrix for all k-quants (#4930) Co-authored-by: Iwan Kawrakow <iwan.kawrakow@gmail.com>
2024-01-14 14:21:12 +00:00
scales[j] = make_qx_quants(16, 16, x + 16*j, L + 16*j, 1, NULL);
ggml : quantization refactoring (#3833) * ggml : factor all quantization code in ggml-quants ggml-ci * ggml-quants : fix Zig and Swift builds + quantize tool ggml-ci * quantize : --pure option for disabling k-quant mixtures --------- Co-authored-by: cebtenzzre <cebtenzzre@gmail.com>
2023-10-29 16:32:28 +00:00
float abs_scale = fabsf(scales[j]);
if (abs_scale > amax) {
amax = abs_scale;
max_scale = scales[j];
}
}
float iscale = -128.f/max_scale;
for (int j = 0; j < QK_K/16; ++j) {
int l = nearest_int(iscale*scales[j]);
y[i].scales[j] = MAX(-128, MIN(127, l));
}
ggml : move FP16 <-> FP32 code to ggml-impl.h (#3861) * ggml : move FP16 <-> FP32 stuff to ggml-impl.h ggml-ci * tests : fix ARM build * ggml : explicitly initialize deprecated type traits * ggml : add math.h to ggml-impl.h * ggml : remove duplicate static assert macros * ggml : prefix lookup tables with ggml_ ggml-ci * ggml-impl : move extern "C" to start of file
2023-10-30 17:19:15 +00:00
y[i].d = GGML_FP32_TO_FP16(1/iscale);
ggml : quantization refactoring (#3833) * ggml : factor all quantization code in ggml-quants ggml-ci * ggml-quants : fix Zig and Swift builds + quantize tool ggml-ci * quantize : --pure option for disabling k-quant mixtures --------- Co-authored-by: cebtenzzre <cebtenzzre@gmail.com>
2023-10-29 16:32:28 +00:00
for (int j = 0; j < QK_K/16; ++j) {
ggml : move FP16 <-> FP32 code to ggml-impl.h (#3861) * ggml : move FP16 <-> FP32 stuff to ggml-impl.h ggml-ci * tests : fix ARM build * ggml : explicitly initialize deprecated type traits * ggml : add math.h to ggml-impl.h * ggml : remove duplicate static assert macros * ggml : prefix lookup tables with ggml_ ggml-ci * ggml-impl : move extern "C" to start of file
2023-10-30 17:19:15 +00:00
const float d = GGML_FP16_TO_FP32(y[i].d) * y[i].scales[j];
ggml : quantization refactoring (#3833) * ggml : factor all quantization code in ggml-quants ggml-ci * ggml-quants : fix Zig and Swift builds + quantize tool ggml-ci * quantize : --pure option for disabling k-quant mixtures --------- Co-authored-by: cebtenzzre <cebtenzzre@gmail.com>
2023-10-29 16:32:28 +00:00
if (!d) continue;
for (int ii = 0; ii < 16; ++ii) {
int l = nearest_int(x[16*j + ii]/d);
l = MAX(-16, MIN(15, l));
L[16*j + ii] = l + 16;
}
}
uint8_t * restrict qh = y[i].qh;
uint8_t * restrict ql = y[i].qs;
memset(qh, 0, QK_K/8);
for (int j = 0; j < 32; ++j) {
int jm = j%8;
int is = j/8;
int l1 = L[j];
if (l1 > 15) {
l1 -= 16; qh[jm] |= (1 << is);
}
int l2 = L[j + 32];
if (l2 > 15) {
l2 -= 16; qh[jm] |= (1 << (4 + is));
}
ql[j] = l1 | (l2 << 4);
}
#endif
x += QK_K;
}
}
void dequantize_row_q5_K(const block_q5_K * restrict x, float * restrict y, int k) {
assert(k % QK_K == 0);
const int nb = k / QK_K;
for (int i = 0; i < nb; i++) {
const uint8_t * ql = x[i].qs;
const uint8_t * qh = x[i].qh;
#if QK_K == 256
ggml : move FP16 <-> FP32 code to ggml-impl.h (#3861) * ggml : move FP16 <-> FP32 stuff to ggml-impl.h ggml-ci * tests : fix ARM build * ggml : explicitly initialize deprecated type traits * ggml : add math.h to ggml-impl.h * ggml : remove duplicate static assert macros * ggml : prefix lookup tables with ggml_ ggml-ci * ggml-impl : move extern "C" to start of file
2023-10-30 17:19:15 +00:00
const float d = GGML_FP16_TO_FP32(x[i].d);
const float min = GGML_FP16_TO_FP32(x[i].dmin);
ggml : quantization refactoring (#3833) * ggml : factor all quantization code in ggml-quants ggml-ci * ggml-quants : fix Zig and Swift builds + quantize tool ggml-ci * quantize : --pure option for disabling k-quant mixtures --------- Co-authored-by: cebtenzzre <cebtenzzre@gmail.com>
2023-10-29 16:32:28 +00:00
int is = 0;
uint8_t sc, m;
uint8_t u1 = 1, u2 = 2;
for (int j = 0; j < QK_K; j += 64) {
get_scale_min_k4(is + 0, x[i].scales, &sc, &m);
const float d1 = d * sc; const float m1 = min * m;
get_scale_min_k4(is + 1, x[i].scales, &sc, &m);
const float d2 = d * sc; const float m2 = min * m;
for (int l = 0; l < 32; ++l) *y++ = d1 * ((ql[l] & 0xF) + (qh[l] & u1 ? 16 : 0)) - m1;
for (int l = 0; l < 32; ++l) *y++ = d2 * ((ql[l] >> 4) + (qh[l] & u2 ? 16 : 0)) - m2;
ql += 32; is += 2;
u1 <<= 2; u2 <<= 2;
}
#else
ggml : move FP16 <-> FP32 code to ggml-impl.h (#3861) * ggml : move FP16 <-> FP32 stuff to ggml-impl.h ggml-ci * tests : fix ARM build * ggml : explicitly initialize deprecated type traits * ggml : add math.h to ggml-impl.h * ggml : remove duplicate static assert macros * ggml : prefix lookup tables with ggml_ ggml-ci * ggml-impl : move extern "C" to start of file
2023-10-30 17:19:15 +00:00
float d = GGML_FP16_TO_FP32(x[i].d);
ggml : quantization refactoring (#3833) * ggml : factor all quantization code in ggml-quants ggml-ci * ggml-quants : fix Zig and Swift builds + quantize tool ggml-ci * quantize : --pure option for disabling k-quant mixtures --------- Co-authored-by: cebtenzzre <cebtenzzre@gmail.com>
2023-10-29 16:32:28 +00:00
const int8_t * restrict s = x[i].scales;
for (int l = 0; l < 8; ++l) {
y[l+ 0] = d * s[0] * ((ql[l+ 0] & 0xF) - (qh[l] & 0x01 ? 0 : 16));
y[l+ 8] = d * s[0] * ((ql[l+ 8] & 0xF) - (qh[l] & 0x02 ? 0 : 16));
y[l+16] = d * s[1] * ((ql[l+16] & 0xF) - (qh[l] & 0x04 ? 0 : 16));
y[l+24] = d * s[1] * ((ql[l+24] & 0xF) - (qh[l] & 0x08 ? 0 : 16));
y[l+32] = d * s[2] * ((ql[l+ 0] >> 4) - (qh[l] & 0x10 ? 0 : 16));
y[l+40] = d * s[2] * ((ql[l+ 8] >> 4) - (qh[l] & 0x20 ? 0 : 16));
y[l+48] = d * s[3] * ((ql[l+16] >> 4) - (qh[l] & 0x40 ? 0 : 16));
y[l+56] = d * s[3] * ((ql[l+24] >> 4) - (qh[l] & 0x80 ? 0 : 16));
}
y += QK_K;
#endif
}
}
void quantize_row_q5_K(const float * restrict x, void * restrict vy, int k) {
assert(k % QK_K == 0);
block_q5_K * restrict y = vy;
quantize_row_q5_K_reference(x, y, k);
}
size_t ggml_quantize_q5_K(const float * restrict src, void * restrict dst, int n, int k, int64_t * restrict hist) {
assert(k % QK_K == 0);
(void)hist; // TODO: collect histograms
for (int j = 0; j < n; j += k) {
block_q5_K * restrict y = (block_q5_K *)dst + j/QK_K;
quantize_row_q5_K_reference(src + j, y, k);
}
return (n/QK_K*sizeof(block_q5_K));
}
Add ability to use importance matrix for all k-quants (#4930) Co-authored-by: Iwan Kawrakow <iwan.kawrakow@gmail.com>
2024-01-14 14:21:12 +00:00
static void quantize_row_q5_K_impl(const float * restrict x, block_q5_K * restrict y, int n_per_row, const float * quant_weights) {
#if QK_K != 256
(void)quant_weights;
quantize_row_q5_K_reference(x, y, n_per_row);
#else
assert(n_per_row % QK_K == 0);
const int nb = n_per_row / QK_K;
uint8_t L[QK_K];
uint8_t Laux[32];
Slight quantization improvement for Q4_K and Q5_K (#5361) * Q4_K: slightly better quantization * Q5_K: slightly better quantization --------- Co-authored-by: Iwan Kawrakow <iwan.kawrakow@gmail.com>
2024-02-06 15:28:02 +00:00
uint8_t Ls[QK_K/32];
uint8_t Lm[QK_K/32];
float mins[QK_K/32];
float scales[QK_K/32];
float sw[QK_K/32];
float weights[32];
Add ability to use importance matrix for all k-quants (#4930) Co-authored-by: Iwan Kawrakow <iwan.kawrakow@gmail.com>
2024-01-14 14:21:12 +00:00
for (int i = 0; i < nb; i++) {
float sum_x2 = 0;
for (int l = 0; l < QK_K; ++l) sum_x2 += x[l] * x[l];
Slight quantization improvement for Q4_K and Q5_K (#5361) * Q4_K: slightly better quantization * Q5_K: slightly better quantization --------- Co-authored-by: Iwan Kawrakow <iwan.kawrakow@gmail.com>
2024-02-06 15:28:02 +00:00
float sigma2 = 2*sum_x2/QK_K;
Add ability to use importance matrix for all k-quants (#4930) Co-authored-by: Iwan Kawrakow <iwan.kawrakow@gmail.com>
2024-01-14 14:21:12 +00:00
float av_x = sqrtf(sigma2);
for (int j = 0; j < QK_K/32; ++j) {
if (quant_weights) {
const float * qw = quant_weights + QK_K*i + 32*j;
for (int l = 0; l < 32; ++l) weights[l] = qw[l] * sqrtf(sigma2 + x[32*j + l]*x[32*j + l]);
} else {
for (int l = 0; l < 32; ++l) weights[l] = av_x + fabsf(x[32*j + l]);
}
Slight quantization improvement for Q4_K and Q5_K (#5361) * Q4_K: slightly better quantization * Q5_K: slightly better quantization --------- Co-authored-by: Iwan Kawrakow <iwan.kawrakow@gmail.com>
2024-02-06 15:28:02 +00:00
float sumw = 0;
for (int l = 0; l < 32; ++l) sumw += weights[l];
sw[j] = sumw;
Add ability to use importance matrix for all k-quants (#4930) Co-authored-by: Iwan Kawrakow <iwan.kawrakow@gmail.com>
2024-01-14 14:21:12 +00:00
scales[j] = make_qkx3_quants(32, 31, x + 32*j, weights, L + 32*j, &mins[j], Laux, -0.9f, 0.05f, 36, false);
}
Slight quantization improvement for Q4_K and Q5_K (#5361) * Q4_K: slightly better quantization * Q5_K: slightly better quantization --------- Co-authored-by: Iwan Kawrakow <iwan.kawrakow@gmail.com>
2024-02-06 15:28:02 +00:00
float d_block = make_qp_quants(QK_K/32, 63, scales, Ls, sw);
float m_block = make_qp_quants(QK_K/32, 63, mins, Lm, sw);
Add ability to use importance matrix for all k-quants (#4930) Co-authored-by: Iwan Kawrakow <iwan.kawrakow@gmail.com>
2024-01-14 14:21:12 +00:00
for (int j = 0; j < QK_K/32; ++j) {
Slight quantization improvement for Q4_K and Q5_K (#5361) * Q4_K: slightly better quantization * Q5_K: slightly better quantization --------- Co-authored-by: Iwan Kawrakow <iwan.kawrakow@gmail.com>
2024-02-06 15:28:02 +00:00
uint8_t ls = Ls[j];
uint8_t lm = Lm[j];
Add ability to use importance matrix for all k-quants (#4930) Co-authored-by: Iwan Kawrakow <iwan.kawrakow@gmail.com>
2024-01-14 14:21:12 +00:00
ls = MIN(63, ls);
lm = MIN(63, lm);
if (j < 4) {
y[i].scales[j] = ls;
y[i].scales[j+4] = lm;
} else {
y[i].scales[j+4] = (ls & 0xF) | ((lm & 0xF) << 4);
y[i].scales[j-4] |= ((ls >> 4) << 6);
y[i].scales[j-0] |= ((lm >> 4) << 6);
}
}
Slight quantization improvement for Q4_K and Q5_K (#5361) * Q4_K: slightly better quantization * Q5_K: slightly better quantization --------- Co-authored-by: Iwan Kawrakow <iwan.kawrakow@gmail.com>
2024-02-06 15:28:02 +00:00
y[i].d = GGML_FP32_TO_FP16(d_block);
y[i].dmin = GGML_FP32_TO_FP16(m_block);
Add ability to use importance matrix for all k-quants (#4930) Co-authored-by: Iwan Kawrakow <iwan.kawrakow@gmail.com>
2024-01-14 14:21:12 +00:00
uint8_t sc, m;
for (int j = 0; j < QK_K/32; ++j) {
get_scale_min_k4(j, y[i].scales, &sc, &m);
const float d = GGML_FP16_TO_FP32(y[i].d) * sc;
if (!d) continue;
const float dm = GGML_FP16_TO_FP32(y[i].dmin) * m;
for (int ii = 0; ii < 32; ++ii) {
int l = nearest_int((x[32*j + ii] + dm)/d);
l = MAX(0, MIN(31, l));
L[32*j + ii] = l;
}
}
uint8_t * restrict qh = y[i].qh;
uint8_t * restrict ql = y[i].qs;
memset(qh, 0, QK_K/8);
uint8_t m1 = 1, m2 = 2;
for (int n = 0; n < QK_K; n += 64) {
for (int j = 0; j < 32; ++j) {
int l1 = L[n + j];
if (l1 > 15) {
l1 -= 16; qh[j] |= m1;
}
int l2 = L[n + j + 32];
if (l2 > 15) {
l2 -= 16; qh[j] |= m2;
}
ql[j] = l1 | (l2 << 4);
}
m1 <<= 2; m2 <<= 2;
ql += 32;
}
x += QK_K;
}
#endif
}
size_t quantize_q5_K(const float * src, void * dst, int nrow, int n_per_row, int64_t * hist, const float * quant_weights) {
(void)hist;
ggml : add IQ2 to test-backend-ops + refactoring (#4990) * ggml : add IQ2 to test-backend-ops + refactoring ggml-ci * cuda : update supports_op for IQ2 ggml-ci * ci : enable LLAMA_CUBLAS=1 for CUDA nodes ggml-ci * cuda : fix out-of-bounds-access in `mul_mat_vec_q` ggml-ci * tests : avoid creating RNGs for each Q tensor ggml-ci * tests : avoid creating RNGs for each tensor ggml-ci
2024-01-17 16:54:56 +00:00
size_t row_size = ggml_row_size(GGML_TYPE_Q5_K, n_per_row);
Add ability to use importance matrix for all k-quants (#4930) Co-authored-by: Iwan Kawrakow <iwan.kawrakow@gmail.com>
2024-01-14 14:21:12 +00:00
if (!quant_weights) {
quantize_row_q5_K_reference(src, dst, nrow*n_per_row);
}
else {
char * qrow = (char *)dst;
for (int row = 0; row < nrow; ++row) {
quantize_row_q5_K_impl(src, (block_q5_K*)qrow, n_per_row, quant_weights);
src += n_per_row;
qrow += row_size;
}
}
return nrow * row_size;
}
ggml : quantization refactoring (#3833) * ggml : factor all quantization code in ggml-quants ggml-ci * ggml-quants : fix Zig and Swift builds + quantize tool ggml-ci * quantize : --pure option for disabling k-quant mixtures --------- Co-authored-by: cebtenzzre <cebtenzzre@gmail.com>
2023-10-29 16:32:28 +00:00
// ====================== 6-bit (de)-quantization
void quantize_row_q6_K_reference(const float * restrict x, block_q6_K * restrict y, int k) {
assert(k % QK_K == 0);
const int nb = k / QK_K;
int8_t L[QK_K];
float scales[QK_K/16];
for (int i = 0; i < nb; i++) {
float max_scale = 0;
float max_abs_scale = 0;
for (int ib = 0; ib < QK_K/16; ++ib) {
Add ability to use importance matrix for all k-quants (#4930) Co-authored-by: Iwan Kawrakow <iwan.kawrakow@gmail.com>
2024-01-14 14:21:12 +00:00
const float scale = make_qx_quants(16, 32, x + 16*ib, L + 16*ib, 1, NULL);
ggml : quantization refactoring (#3833) * ggml : factor all quantization code in ggml-quants ggml-ci * ggml-quants : fix Zig and Swift builds + quantize tool ggml-ci * quantize : --pure option for disabling k-quant mixtures --------- Co-authored-by: cebtenzzre <cebtenzzre@gmail.com>
2023-10-29 16:32:28 +00:00
scales[ib] = scale;
const float abs_scale = fabsf(scale);
if (abs_scale > max_abs_scale) {
max_abs_scale = abs_scale;
max_scale = scale;
ggml : add SOTA 2,3,4,5,6 bit k-quantizations (#1684) * Starting to add k-quantization to ggml I think it is better to have quantization separate from ggml. For now just adding the k-quants there, but it would be better to also factor out the existing ggml quantizations. * Adding Q3_K and Q8_K (de)-quantization * Q3_K now working on CUDA and AVX2/scalar CUDA is not ideal - ~50% slower than Q4_0 for single token prediction, about the same in batch mode (perplexity). CPU single token is ~55 ms (on Ryzen 7950X). * Some improvement for Q3_K on CUDA It is now ~22.5 ms/token on my GPU, so ~30% slower than Q4_0. * Some more CUDA optimizations for Q3_K Single token is now 20.5 ms/token (~20% slower than Q4_0). Perplexity is on par with Q4_0. * Adding Q4_K - scalar, AVX2, CUDA Performance is the same or perhaps very slightly better than Q4_0 on the CPU. On the GPU, single token prediction is ~10% better than Q4_0, batch mode (perplexity is about the same). * Adding Q6_K - scalar, AVX2, CUDA Performance is ~40% lower compared to Q4_K on the CPU. This is to be expected, considering that we are memory bound on the CPU and the 6-bit model is ~44% larger than the 4-bit. On the GPU, single token prediction is ~6% lower than Q4_0, batch mode (perplexity) is even closer (but still slower). * Adding Q5_K - scalar, AVX2, CUDA Performance is ~20% lower compared to Q4_K on the CPU. This is to be expected, considering that we are memory bound on the CPU and the 5-bit model is ~22% larger than the 4-bit. On the GPU, single token prediction is about the same as Q4_0 for both, single token and batch prediction. * Per convention, all QX_K quantizations use Q5_K for output.weight * Adding quantization mixes * Quantization mixes: didn't quite get what I wanted in the last commit * Q4_K dot product for ARM_NEON * Q6_K dot product for ARM_NEON * Q5_K dot product for ARM_NEON * Adding Q3_K dot for ARM_NEON It is 22% slower than Q4_K, despite the smaller model size. On x86_64, where we are memory bound, the Q3_K model is quite a bit faster than Q4_K. * A very slightly faster ARM_NEON Q3_K dot * Adding Q2_K - just CUDA for now Token prediction is pretty good - about 15.5 ms on a RTX 4080. Perplexity is about the same as Q4_K. * Adding scalar and AVX2 Q2_K dot * Adding ARM_NEON Q2_K dot About the same performance as Q4_K. * A slightly faster ARM_NEON Q2_K dot Single token prediction is now ~36 ms on M2 Max. The code is much simpler too. * Fixed bug in Q2_K CUDA dot product kernel Stranegly enough, for the few prompts I tried with the 7B model the responses looked perfectly reasonable. Only realized something is not quite right when I tried the larger models and started getting nonse back. In any case, Q2_K single token evaluation time on an RTX 4080 in a Ryzen7950X box iusing CUDA and model fully loaded on the GPU are ~15.5 ms for 7B, ~25.4 ms for 13B, and ~55.8 ms for 30B. The max number of layers that fit in VRAM for The 65B is 32. With that, we get ~330 ms per token, which is not that much faster than just running on the CPU (~470 ms per token). * Don't print zeros/NaNs when no count histogram has been collected * A 10% faster CUDA vector dot kernel for Q3_K Q3_K is now running at ~18.5 ms / token on CUDA, so the gap to Q4_0 is only 10%. It seems memory acccess pattern is more important for performance than the amount of computation the kernel does. * A slightly daster Q4_K AVX2 dot product For perplexity, where we are less memory bound, time per pass drops by ~5%. Barely measurable difference for single token prediction. * A slightly faster ARM_NEON A4_K dot product * Minor * Fix quantization error test We cannot possibly be expecting rmse < 0.002 for 2- and 3-bit quantization variants. * Fix docker build I have been sloppy with vector reinterpret casts on ARM_NEON. It seems clang is very forgiving in that regard. * Added forgotten ggml.o dependence on k_quants.h to the Makefile * Had unintentionally committed the Makefile with -Ofast enabled * ggml : rename k_quants -> ggml-quants-k, use lowercase in code --------- Co-authored-by: Iwan Kawrakow <iwan.kawrakow@gmail.com> Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
2023-06-05 19:56:18 +00:00
}
ggml : quantization refactoring (#3833) * ggml : factor all quantization code in ggml-quants ggml-ci * ggml-quants : fix Zig and Swift builds + quantize tool ggml-ci * quantize : --pure option for disabling k-quant mixtures --------- Co-authored-by: cebtenzzre <cebtenzzre@gmail.com>
2023-10-29 16:32:28 +00:00
ggml : add SOTA 2,3,4,5,6 bit k-quantizations (#1684) * Starting to add k-quantization to ggml I think it is better to have quantization separate from ggml. For now just adding the k-quants there, but it would be better to also factor out the existing ggml quantizations. * Adding Q3_K and Q8_K (de)-quantization * Q3_K now working on CUDA and AVX2/scalar CUDA is not ideal - ~50% slower than Q4_0 for single token prediction, about the same in batch mode (perplexity). CPU single token is ~55 ms (on Ryzen 7950X). * Some improvement for Q3_K on CUDA It is now ~22.5 ms/token on my GPU, so ~30% slower than Q4_0. * Some more CUDA optimizations for Q3_K Single token is now 20.5 ms/token (~20% slower than Q4_0). Perplexity is on par with Q4_0. * Adding Q4_K - scalar, AVX2, CUDA Performance is the same or perhaps very slightly better than Q4_0 on the CPU. On the GPU, single token prediction is ~10% better than Q4_0, batch mode (perplexity is about the same). * Adding Q6_K - scalar, AVX2, CUDA Performance is ~40% lower compared to Q4_K on the CPU. This is to be expected, considering that we are memory bound on the CPU and the 6-bit model is ~44% larger than the 4-bit. On the GPU, single token prediction is ~6% lower than Q4_0, batch mode (perplexity) is even closer (but still slower). * Adding Q5_K - scalar, AVX2, CUDA Performance is ~20% lower compared to Q4_K on the CPU. This is to be expected, considering that we are memory bound on the CPU and the 5-bit model is ~22% larger than the 4-bit. On the GPU, single token prediction is about the same as Q4_0 for both, single token and batch prediction. * Per convention, all QX_K quantizations use Q5_K for output.weight * Adding quantization mixes * Quantization mixes: didn't quite get what I wanted in the last commit * Q4_K dot product for ARM_NEON * Q6_K dot product for ARM_NEON * Q5_K dot product for ARM_NEON * Adding Q3_K dot for ARM_NEON It is 22% slower than Q4_K, despite the smaller model size. On x86_64, where we are memory bound, the Q3_K model is quite a bit faster than Q4_K. * A very slightly faster ARM_NEON Q3_K dot * Adding Q2_K - just CUDA for now Token prediction is pretty good - about 15.5 ms on a RTX 4080. Perplexity is about the same as Q4_K. * Adding scalar and AVX2 Q2_K dot * Adding ARM_NEON Q2_K dot About the same performance as Q4_K. * A slightly faster ARM_NEON Q2_K dot Single token prediction is now ~36 ms on M2 Max. The code is much simpler too. * Fixed bug in Q2_K CUDA dot product kernel Stranegly enough, for the few prompts I tried with the 7B model the responses looked perfectly reasonable. Only realized something is not quite right when I tried the larger models and started getting nonse back. In any case, Q2_K single token evaluation time on an RTX 4080 in a Ryzen7950X box iusing CUDA and model fully loaded on the GPU are ~15.5 ms for 7B, ~25.4 ms for 13B, and ~55.8 ms for 30B. The max number of layers that fit in VRAM for The 65B is 32. With that, we get ~330 ms per token, which is not that much faster than just running on the CPU (~470 ms per token). * Don't print zeros/NaNs when no count histogram has been collected * A 10% faster CUDA vector dot kernel for Q3_K Q3_K is now running at ~18.5 ms / token on CUDA, so the gap to Q4_0 is only 10%. It seems memory acccess pattern is more important for performance than the amount of computation the kernel does. * A slightly daster Q4_K AVX2 dot product For perplexity, where we are less memory bound, time per pass drops by ~5%. Barely measurable difference for single token prediction. * A slightly faster ARM_NEON A4_K dot product * Minor * Fix quantization error test We cannot possibly be expecting rmse < 0.002 for 2- and 3-bit quantization variants. * Fix docker build I have been sloppy with vector reinterpret casts on ARM_NEON. It seems clang is very forgiving in that regard. * Added forgotten ggml.o dependence on k_quants.h to the Makefile * Had unintentionally committed the Makefile with -Ofast enabled * ggml : rename k_quants -> ggml-quants-k, use lowercase in code --------- Co-authored-by: Iwan Kawrakow <iwan.kawrakow@gmail.com> Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
2023-06-05 19:56:18 +00:00
}
ggml : quantization refactoring (#3833) * ggml : factor all quantization code in ggml-quants ggml-ci * ggml-quants : fix Zig and Swift builds + quantize tool ggml-ci * quantize : --pure option for disabling k-quant mixtures --------- Co-authored-by: cebtenzzre <cebtenzzre@gmail.com>
2023-10-29 16:32:28 +00:00
if (!max_abs_scale) {
memset(&y[i], 0, sizeof(block_q6_K));
ggml : move FP16 <-> FP32 code to ggml-impl.h (#3861) * ggml : move FP16 <-> FP32 stuff to ggml-impl.h ggml-ci * tests : fix ARM build * ggml : explicitly initialize deprecated type traits * ggml : add math.h to ggml-impl.h * ggml : remove duplicate static assert macros * ggml : prefix lookup tables with ggml_ ggml-ci * ggml-impl : move extern "C" to start of file
2023-10-30 17:19:15 +00:00
y[i].d = GGML_FP32_TO_FP16(0.f);
ggml : quantization refactoring (#3833) * ggml : factor all quantization code in ggml-quants ggml-ci * ggml-quants : fix Zig and Swift builds + quantize tool ggml-ci * quantize : --pure option for disabling k-quant mixtures --------- Co-authored-by: cebtenzzre <cebtenzzre@gmail.com>
2023-10-29 16:32:28 +00:00
x += QK_K;
continue;
ggml : add SOTA 2,3,4,5,6 bit k-quantizations (#1684) * Starting to add k-quantization to ggml I think it is better to have quantization separate from ggml. For now just adding the k-quants there, but it would be better to also factor out the existing ggml quantizations. * Adding Q3_K and Q8_K (de)-quantization * Q3_K now working on CUDA and AVX2/scalar CUDA is not ideal - ~50% slower than Q4_0 for single token prediction, about the same in batch mode (perplexity). CPU single token is ~55 ms (on Ryzen 7950X). * Some improvement for Q3_K on CUDA It is now ~22.5 ms/token on my GPU, so ~30% slower than Q4_0. * Some more CUDA optimizations for Q3_K Single token is now 20.5 ms/token (~20% slower than Q4_0). Perplexity is on par with Q4_0. * Adding Q4_K - scalar, AVX2, CUDA Performance is the same or perhaps very slightly better than Q4_0 on the CPU. On the GPU, single token prediction is ~10% better than Q4_0, batch mode (perplexity is about the same). * Adding Q6_K - scalar, AVX2, CUDA Performance is ~40% lower compared to Q4_K on the CPU. This is to be expected, considering that we are memory bound on the CPU and the 6-bit model is ~44% larger than the 4-bit. On the GPU, single token prediction is ~6% lower than Q4_0, batch mode (perplexity) is even closer (but still slower). * Adding Q5_K - scalar, AVX2, CUDA Performance is ~20% lower compared to Q4_K on the CPU. This is to be expected, considering that we are memory bound on the CPU and the 5-bit model is ~22% larger than the 4-bit. On the GPU, single token prediction is about the same as Q4_0 for both, single token and batch prediction. * Per convention, all QX_K quantizations use Q5_K for output.weight * Adding quantization mixes * Quantization mixes: didn't quite get what I wanted in the last commit * Q4_K dot product for ARM_NEON * Q6_K dot product for ARM_NEON * Q5_K dot product for ARM_NEON * Adding Q3_K dot for ARM_NEON It is 22% slower than Q4_K, despite the smaller model size. On x86_64, where we are memory bound, the Q3_K model is quite a bit faster than Q4_K. * A very slightly faster ARM_NEON Q3_K dot * Adding Q2_K - just CUDA for now Token prediction is pretty good - about 15.5 ms on a RTX 4080. Perplexity is about the same as Q4_K. * Adding scalar and AVX2 Q2_K dot * Adding ARM_NEON Q2_K dot About the same performance as Q4_K. * A slightly faster ARM_NEON Q2_K dot Single token prediction is now ~36 ms on M2 Max. The code is much simpler too. * Fixed bug in Q2_K CUDA dot product kernel Stranegly enough, for the few prompts I tried with the 7B model the responses looked perfectly reasonable. Only realized something is not quite right when I tried the larger models and started getting nonse back. In any case, Q2_K single token evaluation time on an RTX 4080 in a Ryzen7950X box iusing CUDA and model fully loaded on the GPU are ~15.5 ms for 7B, ~25.4 ms for 13B, and ~55.8 ms for 30B. The max number of layers that fit in VRAM for The 65B is 32. With that, we get ~330 ms per token, which is not that much faster than just running on the CPU (~470 ms per token). * Don't print zeros/NaNs when no count histogram has been collected * A 10% faster CUDA vector dot kernel for Q3_K Q3_K is now running at ~18.5 ms / token on CUDA, so the gap to Q4_0 is only 10%. It seems memory acccess pattern is more important for performance than the amount of computation the kernel does. * A slightly daster Q4_K AVX2 dot product For perplexity, where we are less memory bound, time per pass drops by ~5%. Barely measurable difference for single token prediction. * A slightly faster ARM_NEON A4_K dot product * Minor * Fix quantization error test We cannot possibly be expecting rmse < 0.002 for 2- and 3-bit quantization variants. * Fix docker build I have been sloppy with vector reinterpret casts on ARM_NEON. It seems clang is very forgiving in that regard. * Added forgotten ggml.o dependence on k_quants.h to the Makefile * Had unintentionally committed the Makefile with -Ofast enabled * ggml : rename k_quants -> ggml-quants-k, use lowercase in code --------- Co-authored-by: Iwan Kawrakow <iwan.kawrakow@gmail.com> Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
2023-06-05 19:56:18 +00:00
}
ggml : quantization refactoring (#3833) * ggml : factor all quantization code in ggml-quants ggml-ci * ggml-quants : fix Zig and Swift builds + quantize tool ggml-ci * quantize : --pure option for disabling k-quant mixtures --------- Co-authored-by: cebtenzzre <cebtenzzre@gmail.com>
2023-10-29 16:32:28 +00:00
float iscale = -128.f/max_scale;
ggml : move FP16 <-> FP32 code to ggml-impl.h (#3861) * ggml : move FP16 <-> FP32 stuff to ggml-impl.h ggml-ci * tests : fix ARM build * ggml : explicitly initialize deprecated type traits * ggml : add math.h to ggml-impl.h * ggml : remove duplicate static assert macros * ggml : prefix lookup tables with ggml_ ggml-ci * ggml-impl : move extern "C" to start of file
2023-10-30 17:19:15 +00:00
y[i].d = GGML_FP32_TO_FP16(1/iscale);
ggml : quantization refactoring (#3833) * ggml : factor all quantization code in ggml-quants ggml-ci * ggml-quants : fix Zig and Swift builds + quantize tool ggml-ci * quantize : --pure option for disabling k-quant mixtures --------- Co-authored-by: cebtenzzre <cebtenzzre@gmail.com>
2023-10-29 16:32:28 +00:00
for (int ib = 0; ib < QK_K/16; ++ib) {
y[i].scales[ib] = MIN(127, nearest_int(iscale*scales[ib]));
Quantization imrovements for k_quants (#2707) * Improve LLaMA-2 2-, 3- and 4-bit quantization * Q3_K_S: use Q5_K for 1st 2 layers of attention.wv and feed_forward.w2 * Q4_K_S: use Q6_K for 1st 2 layers of attention.wv and feed_forward.w2 * Q2_K and Q3_K_M: use Q5_K instead of Q4_K for 1st 2 layers of attention.wv and feed_forward.w2 This leads to a slight model sized increase as follows: Q2_K : 2.684G vs 2.670G Q3_K_S: 2.775G vs 2.745G Q3_K_M: 3.071G vs 3.057G Q4_K_S: 3.592G vs 3.563G LLaMA-2 PPL for context 512 changes as follows: Q2_K : 6.6691 vs 6.8201 Q3_K_S: 6.2129 vs 6.2584 Q3_K_M: 6.0387 vs 6.1371 Q4_K_S: 5.9138 vs 6.0041 There are improvements for LLaMA-1 as well, but they are way smaller than the above. * Minor 4-bit quantization improvement For the same model size as previus commit, we get PPL = 5.9069 vs 5.9138. * Some more fine tuning * Adding make_qkx2_quants With it, we get PPL = 5.8828 for L2-7B Q4_K_S. * Another minor improvement * Q2_K improvement Smaller model, lower perplexity. 7B: file size = 2.632G, PPL = 6.3772 vs original 2.670G PPL = 6.8201 12B: file size = 5.056G, PPL = 5.4577 vs original 5.130G PPL = 5.7178 It is mostly Q3_K except for tok_embeddings, attention.wq, attention.wk, which are Q2_K * Iterating * Revert Q5_K back to make_qkx1_quants * Better Q6_K * make_qkx2_quants is better for Q5_K after all * Fix after rebasing on master * Fix for changed tensor names --------- Co-authored-by: Iwan Kawrakow <iwan.kawrakow@gmail.com>
2023-08-22 16:14:09 +00:00
}
ggml : quantization refactoring (#3833) * ggml : factor all quantization code in ggml-quants ggml-ci * ggml-quants : fix Zig and Swift builds + quantize tool ggml-ci * quantize : --pure option for disabling k-quant mixtures --------- Co-authored-by: cebtenzzre <cebtenzzre@gmail.com>
2023-10-29 16:32:28 +00:00
for (int j = 0; j < QK_K/16; ++j) {
ggml : move FP16 <-> FP32 code to ggml-impl.h (#3861) * ggml : move FP16 <-> FP32 stuff to ggml-impl.h ggml-ci * tests : fix ARM build * ggml : explicitly initialize deprecated type traits * ggml : add math.h to ggml-impl.h * ggml : remove duplicate static assert macros * ggml : prefix lookup tables with ggml_ ggml-ci * ggml-impl : move extern "C" to start of file
2023-10-30 17:19:15 +00:00
float d = GGML_FP16_TO_FP32(y[i].d) * y[i].scales[j];
ggml : quantization refactoring (#3833) * ggml : factor all quantization code in ggml-quants ggml-ci * ggml-quants : fix Zig and Swift builds + quantize tool ggml-ci * quantize : --pure option for disabling k-quant mixtures --------- Co-authored-by: cebtenzzre <cebtenzzre@gmail.com>
2023-10-29 16:32:28 +00:00
if (!d) {
continue;
Quantization imrovements for k_quants (#2707) * Improve LLaMA-2 2-, 3- and 4-bit quantization * Q3_K_S: use Q5_K for 1st 2 layers of attention.wv and feed_forward.w2 * Q4_K_S: use Q6_K for 1st 2 layers of attention.wv and feed_forward.w2 * Q2_K and Q3_K_M: use Q5_K instead of Q4_K for 1st 2 layers of attention.wv and feed_forward.w2 This leads to a slight model sized increase as follows: Q2_K : 2.684G vs 2.670G Q3_K_S: 2.775G vs 2.745G Q3_K_M: 3.071G vs 3.057G Q4_K_S: 3.592G vs 3.563G LLaMA-2 PPL for context 512 changes as follows: Q2_K : 6.6691 vs 6.8201 Q3_K_S: 6.2129 vs 6.2584 Q3_K_M: 6.0387 vs 6.1371 Q4_K_S: 5.9138 vs 6.0041 There are improvements for LLaMA-1 as well, but they are way smaller than the above. * Minor 4-bit quantization improvement For the same model size as previus commit, we get PPL = 5.9069 vs 5.9138. * Some more fine tuning * Adding make_qkx2_quants With it, we get PPL = 5.8828 for L2-7B Q4_K_S. * Another minor improvement * Q2_K improvement Smaller model, lower perplexity. 7B: file size = 2.632G, PPL = 6.3772 vs original 2.670G PPL = 6.8201 12B: file size = 5.056G, PPL = 5.4577 vs original 5.130G PPL = 5.7178 It is mostly Q3_K except for tok_embeddings, attention.wq, attention.wk, which are Q2_K * Iterating * Revert Q5_K back to make_qkx1_quants * Better Q6_K * make_qkx2_quants is better for Q5_K after all * Fix after rebasing on master * Fix for changed tensor names --------- Co-authored-by: Iwan Kawrakow <iwan.kawrakow@gmail.com>
2023-08-22 16:14:09 +00:00
}
ggml : quantization refactoring (#3833) * ggml : factor all quantization code in ggml-quants ggml-ci * ggml-quants : fix Zig and Swift builds + quantize tool ggml-ci * quantize : --pure option for disabling k-quant mixtures --------- Co-authored-by: cebtenzzre <cebtenzzre@gmail.com>
2023-10-29 16:32:28 +00:00
for (int ii = 0; ii < 16; ++ii) {
int l = nearest_int(x[16*j + ii]/d);
l = MAX(-32, MIN(31, l));
L[16*j + ii] = l + 32;
Quantization imrovements for k_quants (#2707) * Improve LLaMA-2 2-, 3- and 4-bit quantization * Q3_K_S: use Q5_K for 1st 2 layers of attention.wv and feed_forward.w2 * Q4_K_S: use Q6_K for 1st 2 layers of attention.wv and feed_forward.w2 * Q2_K and Q3_K_M: use Q5_K instead of Q4_K for 1st 2 layers of attention.wv and feed_forward.w2 This leads to a slight model sized increase as follows: Q2_K : 2.684G vs 2.670G Q3_K_S: 2.775G vs 2.745G Q3_K_M: 3.071G vs 3.057G Q4_K_S: 3.592G vs 3.563G LLaMA-2 PPL for context 512 changes as follows: Q2_K : 6.6691 vs 6.8201 Q3_K_S: 6.2129 vs 6.2584 Q3_K_M: 6.0387 vs 6.1371 Q4_K_S: 5.9138 vs 6.0041 There are improvements for LLaMA-1 as well, but they are way smaller than the above. * Minor 4-bit quantization improvement For the same model size as previus commit, we get PPL = 5.9069 vs 5.9138. * Some more fine tuning * Adding make_qkx2_quants With it, we get PPL = 5.8828 for L2-7B Q4_K_S. * Another minor improvement * Q2_K improvement Smaller model, lower perplexity. 7B: file size = 2.632G, PPL = 6.3772 vs original 2.670G PPL = 6.8201 12B: file size = 5.056G, PPL = 5.4577 vs original 5.130G PPL = 5.7178 It is mostly Q3_K except for tok_embeddings, attention.wq, attention.wk, which are Q2_K * Iterating * Revert Q5_K back to make_qkx1_quants * Better Q6_K * make_qkx2_quants is better for Q5_K after all * Fix after rebasing on master * Fix for changed tensor names --------- Co-authored-by: Iwan Kawrakow <iwan.kawrakow@gmail.com>
2023-08-22 16:14:09 +00:00
}
ggml : quantization refactoring (#3833) * ggml : factor all quantization code in ggml-quants ggml-ci * ggml-quants : fix Zig and Swift builds + quantize tool ggml-ci * quantize : --pure option for disabling k-quant mixtures --------- Co-authored-by: cebtenzzre <cebtenzzre@gmail.com>
2023-10-29 16:32:28 +00:00
}
uint8_t * restrict ql = y[i].ql;
uint8_t * restrict qh = y[i].qh;
#if QK_K == 256
for (int j = 0; j < QK_K; j += 128) {
for (int l = 0; l < 32; ++l) {
const uint8_t q1 = L[j + l + 0] & 0xF;
const uint8_t q2 = L[j + l + 32] & 0xF;
const uint8_t q3 = L[j + l + 64] & 0xF;
const uint8_t q4 = L[j + l + 96] & 0xF;
ql[l+ 0] = q1 | (q3 << 4);
ql[l+32] = q2 | (q4 << 4);
qh[l] = (L[j + l] >> 4) | ((L[j + l + 32] >> 4) << 2) | ((L[j + l + 64] >> 4) << 4) | ((L[j + l + 96] >> 4) << 6);
Quantization imrovements for k_quants (#2707) * Improve LLaMA-2 2-, 3- and 4-bit quantization * Q3_K_S: use Q5_K for 1st 2 layers of attention.wv and feed_forward.w2 * Q4_K_S: use Q6_K for 1st 2 layers of attention.wv and feed_forward.w2 * Q2_K and Q3_K_M: use Q5_K instead of Q4_K for 1st 2 layers of attention.wv and feed_forward.w2 This leads to a slight model sized increase as follows: Q2_K : 2.684G vs 2.670G Q3_K_S: 2.775G vs 2.745G Q3_K_M: 3.071G vs 3.057G Q4_K_S: 3.592G vs 3.563G LLaMA-2 PPL for context 512 changes as follows: Q2_K : 6.6691 vs 6.8201 Q3_K_S: 6.2129 vs 6.2584 Q3_K_M: 6.0387 vs 6.1371 Q4_K_S: 5.9138 vs 6.0041 There are improvements for LLaMA-1 as well, but they are way smaller than the above. * Minor 4-bit quantization improvement For the same model size as previus commit, we get PPL = 5.9069 vs 5.9138. * Some more fine tuning * Adding make_qkx2_quants With it, we get PPL = 5.8828 for L2-7B Q4_K_S. * Another minor improvement * Q2_K improvement Smaller model, lower perplexity. 7B: file size = 2.632G, PPL = 6.3772 vs original 2.670G PPL = 6.8201 12B: file size = 5.056G, PPL = 5.4577 vs original 5.130G PPL = 5.7178 It is mostly Q3_K except for tok_embeddings, attention.wq, attention.wk, which are Q2_K * Iterating * Revert Q5_K back to make_qkx1_quants * Better Q6_K * make_qkx2_quants is better for Q5_K after all * Fix after rebasing on master * Fix for changed tensor names --------- Co-authored-by: Iwan Kawrakow <iwan.kawrakow@gmail.com>
2023-08-22 16:14:09 +00:00
}
ggml : quantization refactoring (#3833) * ggml : factor all quantization code in ggml-quants ggml-ci * ggml-quants : fix Zig and Swift builds + quantize tool ggml-ci * quantize : --pure option for disabling k-quant mixtures --------- Co-authored-by: cebtenzzre <cebtenzzre@gmail.com>
2023-10-29 16:32:28 +00:00
ql += 64;
qh += 32;
}
#else
for (int l = 0; l < 32; ++l) {
const uint8_t q1 = L[l + 0] & 0xF;
const uint8_t q2 = L[l + 32] & 0xF;
ql[l] = q1 | (q2 << 4);
}
for (int l = 0; l < 16; ++l) {
qh[l] = (L[l] >> 4) | ((L[l + 16] >> 4) << 2) | ((L[l + 32] >> 4) << 4) | ((L[l + 48] >> 4) << 6);
Quantization imrovements for k_quants (#2707) * Improve LLaMA-2 2-, 3- and 4-bit quantization * Q3_K_S: use Q5_K for 1st 2 layers of attention.wv and feed_forward.w2 * Q4_K_S: use Q6_K for 1st 2 layers of attention.wv and feed_forward.w2 * Q2_K and Q3_K_M: use Q5_K instead of Q4_K for 1st 2 layers of attention.wv and feed_forward.w2 This leads to a slight model sized increase as follows: Q2_K : 2.684G vs 2.670G Q3_K_S: 2.775G vs 2.745G Q3_K_M: 3.071G vs 3.057G Q4_K_S: 3.592G vs 3.563G LLaMA-2 PPL for context 512 changes as follows: Q2_K : 6.6691 vs 6.8201 Q3_K_S: 6.2129 vs 6.2584 Q3_K_M: 6.0387 vs 6.1371 Q4_K_S: 5.9138 vs 6.0041 There are improvements for LLaMA-1 as well, but they are way smaller than the above. * Minor 4-bit quantization improvement For the same model size as previus commit, we get PPL = 5.9069 vs 5.9138. * Some more fine tuning * Adding make_qkx2_quants With it, we get PPL = 5.8828 for L2-7B Q4_K_S. * Another minor improvement * Q2_K improvement Smaller model, lower perplexity. 7B: file size = 2.632G, PPL = 6.3772 vs original 2.670G PPL = 6.8201 12B: file size = 5.056G, PPL = 5.4577 vs original 5.130G PPL = 5.7178 It is mostly Q3_K except for tok_embeddings, attention.wq, attention.wk, which are Q2_K * Iterating * Revert Q5_K back to make_qkx1_quants * Better Q6_K * make_qkx2_quants is better for Q5_K after all * Fix after rebasing on master * Fix for changed tensor names --------- Co-authored-by: Iwan Kawrakow <iwan.kawrakow@gmail.com>
2023-08-22 16:14:09 +00:00
}
ggml : quantization refactoring (#3833) * ggml : factor all quantization code in ggml-quants ggml-ci * ggml-quants : fix Zig and Swift builds + quantize tool ggml-ci * quantize : --pure option for disabling k-quant mixtures --------- Co-authored-by: cebtenzzre <cebtenzzre@gmail.com>
2023-10-29 16:32:28 +00:00
#endif
x += QK_K;
Quantization imrovements for k_quants (#2707) * Improve LLaMA-2 2-, 3- and 4-bit quantization * Q3_K_S: use Q5_K for 1st 2 layers of attention.wv and feed_forward.w2 * Q4_K_S: use Q6_K for 1st 2 layers of attention.wv and feed_forward.w2 * Q2_K and Q3_K_M: use Q5_K instead of Q4_K for 1st 2 layers of attention.wv and feed_forward.w2 This leads to a slight model sized increase as follows: Q2_K : 2.684G vs 2.670G Q3_K_S: 2.775G vs 2.745G Q3_K_M: 3.071G vs 3.057G Q4_K_S: 3.592G vs 3.563G LLaMA-2 PPL for context 512 changes as follows: Q2_K : 6.6691 vs 6.8201 Q3_K_S: 6.2129 vs 6.2584 Q3_K_M: 6.0387 vs 6.1371 Q4_K_S: 5.9138 vs 6.0041 There are improvements for LLaMA-1 as well, but they are way smaller than the above. * Minor 4-bit quantization improvement For the same model size as previus commit, we get PPL = 5.9069 vs 5.9138. * Some more fine tuning * Adding make_qkx2_quants With it, we get PPL = 5.8828 for L2-7B Q4_K_S. * Another minor improvement * Q2_K improvement Smaller model, lower perplexity. 7B: file size = 2.632G, PPL = 6.3772 vs original 2.670G PPL = 6.8201 12B: file size = 5.056G, PPL = 5.4577 vs original 5.130G PPL = 5.7178 It is mostly Q3_K except for tok_embeddings, attention.wq, attention.wk, which are Q2_K * Iterating * Revert Q5_K back to make_qkx1_quants * Better Q6_K * make_qkx2_quants is better for Q5_K after all * Fix after rebasing on master * Fix for changed tensor names --------- Co-authored-by: Iwan Kawrakow <iwan.kawrakow@gmail.com>
2023-08-22 16:14:09 +00:00
}
}
ggml : quantization refactoring (#3833) * ggml : factor all quantization code in ggml-quants ggml-ci * ggml-quants : fix Zig and Swift builds + quantize tool ggml-ci * quantize : --pure option for disabling k-quant mixtures --------- Co-authored-by: cebtenzzre <cebtenzzre@gmail.com>
2023-10-29 16:32:28 +00:00
void dequantize_row_q6_K(const block_q6_K * restrict x, float * restrict y, int k) {
assert(k % QK_K == 0);
const int nb = k / QK_K;
for (int i = 0; i < nb; i++) {
ggml : move FP16 <-> FP32 code to ggml-impl.h (#3861) * ggml : move FP16 <-> FP32 stuff to ggml-impl.h ggml-ci * tests : fix ARM build * ggml : explicitly initialize deprecated type traits * ggml : add math.h to ggml-impl.h * ggml : remove duplicate static assert macros * ggml : prefix lookup tables with ggml_ ggml-ci * ggml-impl : move extern "C" to start of file
2023-10-30 17:19:15 +00:00
const float d = GGML_FP16_TO_FP32(x[i].d);
ggml : quantization refactoring (#3833) * ggml : factor all quantization code in ggml-quants ggml-ci * ggml-quants : fix Zig and Swift builds + quantize tool ggml-ci * quantize : --pure option for disabling k-quant mixtures --------- Co-authored-by: cebtenzzre <cebtenzzre@gmail.com>
2023-10-29 16:32:28 +00:00
const uint8_t * restrict ql = x[i].ql;
const uint8_t * restrict qh = x[i].qh;
const int8_t * restrict sc = x[i].scales;
k-quants : support for super-block size of 64 (#2001) * k_quants: WIP super-blocks with 64 weights * k_quants: WIP super-blocks with 64 weights Q6_K scalar and AVX2 works * k_quants: WIP super-blocks with 64 weights Q4_K scalar and AVX2 works * k_quants: WIP super-blocks with 64 weights Q2_K scalar and AVX2 works. Q2_K is way too slow (it is actually slower than the scalar implementation) * k_quants: WIP super-blocks with 64 weights Q3_K scalar and AVX2 works. * k_quants: WIP super-blocks with 64 weights Q5_K scalar and AVX2 works, and with that all k_quants are done on AVX2 and scalar * k_quants: WIP super-blocks with 64 weights Q6_K working on CUDA. Cannot make it run quite as gast as with super-blocks with 256 weigths: 8% slower on 4080, 20% slower on the 1660 (but there we fit 1 less layer on the GPU because pf the larger model size), so some fraction of these 20% is due to that, * k_quants: WIP super-blocks with 64 weights Q4_K working on CUDA. ~10% slower on GTX-1660, 16% slower on 4080. * k_quants: WIP super-blocks with 64 weights Q2_K working on CUDA. ~3% slower on GTX-1660, 10% slower on 4080. * k_quants: WIP super-blocks with 64 weights Q3_K working on CUDA. * k_quants: WIP super-blocks with 64 weights Q5_K working on CUDA, and with this CUDA is done. * k_quants: WIP super-blocks with 64 weights Q6_K working on ARM_NEON * k_quants: WIP super-blocks with 64 weights Q4_K working on ARM_NEON, but quite a bit slower than 256 weights * k_quants: WIP super-blocks with 64 weights Q2_K working on ARM_NEON, but quite a bit slower than 256 weights * k_quants: WIP super-blocks with 64 weights Q3_K working on ARM_NEON, but quite a bit slower than 256 weights. * k_quants: WIP super-blocks with 64 weights Q5_K working on ARM_NEON, but quite a bit slower than 256 weights. With that, we have full support for ARM_NEON, although performance is not quite there. * k_quants: WIP super-blocks with 64 weights Slightly more efficient Q3_K and Q5_K * k_quants: WIP super-blocks with 64 weights Another small improvement for Q3_K and Q5_K on ARM_NEON * k_quants: WIP super-blocks with 64 weights Yet another speedup for Q5_K on ARM_NEON. We are now within 10% of the QK_K = 256 version. * k_quants: WIP super-blocks with 64 weights * We are able to pass preprocessor macros to the Metal compiler * Q6_K works and is actually slightly more efficient than the QK_K = 256 version (25.2 ms vs 25.8 ms) * k_quants: WIP super-blocks with 64 weights Q4_K works on Metal and is actually slightly faster than QK_K = 256 (21.95 ms vs 24.0 ms). * k_quants: WIP super-blocks with 64 weights Q2_K works on Metal and is very slightly faster than QK_K = 256 (23.8 ms vs 24.2 ms). * k_quants: WIP super-blocks with 64 weights Q3_K works on Metal and is slightly faster than QK_K = 256 (26.6 ms vs 28.3 ms). * k_quants: WIP super-blocks with 64 weights Q5_K works on Metal and is slightly faster than QK_K = 256 (23.7 ms vs 26.3 ms). * k_quants: call them _K, not _k, also on Metal * k_quants: correctly define QK_K in llama.cpp * Fixed bug in q4_K quantization added with the 64-block addition * Simplify via lambda * k_quants: swicth Q3_K to 4-bit scales when QK_K = 64 Otherwise there isn't much benefit from this quantization type. There is some very slight loss in accuracy, but we reduce size by ~7%. E.g., for OpenLLaMA-3B, Q3_K_S perplexity is 8.6131 with 8-bit scales and 8.6352 with 4-bit, while file size decreases from 1.53G to 1.44G. * k_quants: switch Q4_K to 4-bit scales when QK_K = 64 Here the loss in accuracy is greater than for Q3_K, but the Q4_K points still move further to the left on the perplexity vs size curve. * k_quants: forgot to add the Metal changes in last commit * k_quants: change Q5_K to be type 0 when QK_K = 64 Still needs AVX2 implementation * k_quants: AVX2 implementation for new 64-weight Q5_K * k_quants: 10% faster ARM_NEON Q5_K dot product * k_quants: fixed issue caused by merging with master --------- Co-authored-by: Iwan Kawrakow <iwan.kawrakow@gmail.com>
2023-06-26 16:43:07 +00:00
#if QK_K == 256
ggml : quantization refactoring (#3833) * ggml : factor all quantization code in ggml-quants ggml-ci * ggml-quants : fix Zig and Swift builds + quantize tool ggml-ci * quantize : --pure option for disabling k-quant mixtures --------- Co-authored-by: cebtenzzre <cebtenzzre@gmail.com>
2023-10-29 16:32:28 +00:00
for (int n = 0; n < QK_K; n += 128) {
for (int l = 0; l < 32; ++l) {
int is = l/16;
const int8_t q1 = (int8_t)((ql[l + 0] & 0xF) | (((qh[l] >> 0) & 3) << 4)) - 32;
const int8_t q2 = (int8_t)((ql[l + 32] & 0xF) | (((qh[l] >> 2) & 3) << 4)) - 32;
const int8_t q3 = (int8_t)((ql[l + 0] >> 4) | (((qh[l] >> 4) & 3) << 4)) - 32;
const int8_t q4 = (int8_t)((ql[l + 32] >> 4) | (((qh[l] >> 6) & 3) << 4)) - 32;
y[l + 0] = d * sc[is + 0] * q1;
y[l + 32] = d * sc[is + 2] * q2;
y[l + 64] = d * sc[is + 4] * q3;
y[l + 96] = d * sc[is + 6] * q4;
}
y += 128;
ql += 64;
qh += 32;
sc += 8;
}
#else
for (int l = 0; l < 16; ++l) {
const int8_t q1 = (int8_t)((ql[l+ 0] & 0xF) | (((qh[l] >> 0) & 3) << 4)) - 32;
const int8_t q2 = (int8_t)((ql[l+16] & 0xF) | (((qh[l] >> 2) & 3) << 4)) - 32;
const int8_t q3 = (int8_t)((ql[l+ 0] >> 4) | (((qh[l] >> 4) & 3) << 4)) - 32;
const int8_t q4 = (int8_t)((ql[l+16] >> 4) | (((qh[l] >> 6) & 3) << 4)) - 32;
y[l+ 0] = d * sc[0] * q1;
y[l+16] = d * sc[1] * q2;
y[l+32] = d * sc[2] * q3;
y[l+48] = d * sc[3] * q4;
}
y += 64;
#endif
ggml : add SOTA 2,3,4,5,6 bit k-quantizations (#1684) * Starting to add k-quantization to ggml I think it is better to have quantization separate from ggml. For now just adding the k-quants there, but it would be better to also factor out the existing ggml quantizations. * Adding Q3_K and Q8_K (de)-quantization * Q3_K now working on CUDA and AVX2/scalar CUDA is not ideal - ~50% slower than Q4_0 for single token prediction, about the same in batch mode (perplexity). CPU single token is ~55 ms (on Ryzen 7950X). * Some improvement for Q3_K on CUDA It is now ~22.5 ms/token on my GPU, so ~30% slower than Q4_0. * Some more CUDA optimizations for Q3_K Single token is now 20.5 ms/token (~20% slower than Q4_0). Perplexity is on par with Q4_0. * Adding Q4_K - scalar, AVX2, CUDA Performance is the same or perhaps very slightly better than Q4_0 on the CPU. On the GPU, single token prediction is ~10% better than Q4_0, batch mode (perplexity is about the same). * Adding Q6_K - scalar, AVX2, CUDA Performance is ~40% lower compared to Q4_K on the CPU. This is to be expected, considering that we are memory bound on the CPU and the 6-bit model is ~44% larger than the 4-bit. On the GPU, single token prediction is ~6% lower than Q4_0, batch mode (perplexity) is even closer (but still slower). * Adding Q5_K - scalar, AVX2, CUDA Performance is ~20% lower compared to Q4_K on the CPU. This is to be expected, considering that we are memory bound on the CPU and the 5-bit model is ~22% larger than the 4-bit. On the GPU, single token prediction is about the same as Q4_0 for both, single token and batch prediction. * Per convention, all QX_K quantizations use Q5_K for output.weight * Adding quantization mixes * Quantization mixes: didn't quite get what I wanted in the last commit * Q4_K dot product for ARM_NEON * Q6_K dot product for ARM_NEON * Q5_K dot product for ARM_NEON * Adding Q3_K dot for ARM_NEON It is 22% slower than Q4_K, despite the smaller model size. On x86_64, where we are memory bound, the Q3_K model is quite a bit faster than Q4_K. * A very slightly faster ARM_NEON Q3_K dot * Adding Q2_K - just CUDA for now Token prediction is pretty good - about 15.5 ms on a RTX 4080. Perplexity is about the same as Q4_K. * Adding scalar and AVX2 Q2_K dot * Adding ARM_NEON Q2_K dot About the same performance as Q4_K. * A slightly faster ARM_NEON Q2_K dot Single token prediction is now ~36 ms on M2 Max. The code is much simpler too. * Fixed bug in Q2_K CUDA dot product kernel Stranegly enough, for the few prompts I tried with the 7B model the responses looked perfectly reasonable. Only realized something is not quite right when I tried the larger models and started getting nonse back. In any case, Q2_K single token evaluation time on an RTX 4080 in a Ryzen7950X box iusing CUDA and model fully loaded on the GPU are ~15.5 ms for 7B, ~25.4 ms for 13B, and ~55.8 ms for 30B. The max number of layers that fit in VRAM for The 65B is 32. With that, we get ~330 ms per token, which is not that much faster than just running on the CPU (~470 ms per token). * Don't print zeros/NaNs when no count histogram has been collected * A 10% faster CUDA vector dot kernel for Q3_K Q3_K is now running at ~18.5 ms / token on CUDA, so the gap to Q4_0 is only 10%. It seems memory acccess pattern is more important for performance than the amount of computation the kernel does. * A slightly daster Q4_K AVX2 dot product For perplexity, where we are less memory bound, time per pass drops by ~5%. Barely measurable difference for single token prediction. * A slightly faster ARM_NEON A4_K dot product * Minor * Fix quantization error test We cannot possibly be expecting rmse < 0.002 for 2- and 3-bit quantization variants. * Fix docker build I have been sloppy with vector reinterpret casts on ARM_NEON. It seems clang is very forgiving in that regard. * Added forgotten ggml.o dependence on k_quants.h to the Makefile * Had unintentionally committed the Makefile with -Ofast enabled * ggml : rename k_quants -> ggml-quants-k, use lowercase in code --------- Co-authored-by: Iwan Kawrakow <iwan.kawrakow@gmail.com> Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
2023-06-05 19:56:18 +00:00
}
}
ggml : quantization refactoring (#3833) * ggml : factor all quantization code in ggml-quants ggml-ci * ggml-quants : fix Zig and Swift builds + quantize tool ggml-ci * quantize : --pure option for disabling k-quant mixtures --------- Co-authored-by: cebtenzzre <cebtenzzre@gmail.com>
2023-10-29 16:32:28 +00:00
void quantize_row_q6_K(const float * restrict x, void * restrict vy, int k) {
assert(k % QK_K == 0);
block_q6_K * restrict y = vy;
quantize_row_q6_K_reference(x, y, k);
}
ggml : add SOTA 2,3,4,5,6 bit k-quantizations (#1684) * Starting to add k-quantization to ggml I think it is better to have quantization separate from ggml. For now just adding the k-quants there, but it would be better to also factor out the existing ggml quantizations. * Adding Q3_K and Q8_K (de)-quantization * Q3_K now working on CUDA and AVX2/scalar CUDA is not ideal - ~50% slower than Q4_0 for single token prediction, about the same in batch mode (perplexity). CPU single token is ~55 ms (on Ryzen 7950X). * Some improvement for Q3_K on CUDA It is now ~22.5 ms/token on my GPU, so ~30% slower than Q4_0. * Some more CUDA optimizations for Q3_K Single token is now 20.5 ms/token (~20% slower than Q4_0). Perplexity is on par with Q4_0. * Adding Q4_K - scalar, AVX2, CUDA Performance is the same or perhaps very slightly better than Q4_0 on the CPU. On the GPU, single token prediction is ~10% better than Q4_0, batch mode (perplexity is about the same). * Adding Q6_K - scalar, AVX2, CUDA Performance is ~40% lower compared to Q4_K on the CPU. This is to be expected, considering that we are memory bound on the CPU and the 6-bit model is ~44% larger than the 4-bit. On the GPU, single token prediction is ~6% lower than Q4_0, batch mode (perplexity) is even closer (but still slower). * Adding Q5_K - scalar, AVX2, CUDA Performance is ~20% lower compared to Q4_K on the CPU. This is to be expected, considering that we are memory bound on the CPU and the 5-bit model is ~22% larger than the 4-bit. On the GPU, single token prediction is about the same as Q4_0 for both, single token and batch prediction. * Per convention, all QX_K quantizations use Q5_K for output.weight * Adding quantization mixes * Quantization mixes: didn't quite get what I wanted in the last commit * Q4_K dot product for ARM_NEON * Q6_K dot product for ARM_NEON * Q5_K dot product for ARM_NEON * Adding Q3_K dot for ARM_NEON It is 22% slower than Q4_K, despite the smaller model size. On x86_64, where we are memory bound, the Q3_K model is quite a bit faster than Q4_K. * A very slightly faster ARM_NEON Q3_K dot * Adding Q2_K - just CUDA for now Token prediction is pretty good - about 15.5 ms on a RTX 4080. Perplexity is about the same as Q4_K. * Adding scalar and AVX2 Q2_K dot * Adding ARM_NEON Q2_K dot About the same performance as Q4_K. * A slightly faster ARM_NEON Q2_K dot Single token prediction is now ~36 ms on M2 Max. The code is much simpler too. * Fixed bug in Q2_K CUDA dot product kernel Stranegly enough, for the few prompts I tried with the 7B model the responses looked perfectly reasonable. Only realized something is not quite right when I tried the larger models and started getting nonse back. In any case, Q2_K single token evaluation time on an RTX 4080 in a Ryzen7950X box iusing CUDA and model fully loaded on the GPU are ~15.5 ms for 7B, ~25.4 ms for 13B, and ~55.8 ms for 30B. The max number of layers that fit in VRAM for The 65B is 32. With that, we get ~330 ms per token, which is not that much faster than just running on the CPU (~470 ms per token). * Don't print zeros/NaNs when no count histogram has been collected * A 10% faster CUDA vector dot kernel for Q3_K Q3_K is now running at ~18.5 ms / token on CUDA, so the gap to Q4_0 is only 10%. It seems memory acccess pattern is more important for performance than the amount of computation the kernel does. * A slightly daster Q4_K AVX2 dot product For perplexity, where we are less memory bound, time per pass drops by ~5%. Barely measurable difference for single token prediction. * A slightly faster ARM_NEON A4_K dot product * Minor * Fix quantization error test We cannot possibly be expecting rmse < 0.002 for 2- and 3-bit quantization variants. * Fix docker build I have been sloppy with vector reinterpret casts on ARM_NEON. It seems clang is very forgiving in that regard. * Added forgotten ggml.o dependence on k_quants.h to the Makefile * Had unintentionally committed the Makefile with -Ofast enabled * ggml : rename k_quants -> ggml-quants-k, use lowercase in code --------- Co-authored-by: Iwan Kawrakow <iwan.kawrakow@gmail.com> Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
2023-06-05 19:56:18 +00:00
ggml : quantization refactoring (#3833) * ggml : factor all quantization code in ggml-quants ggml-ci * ggml-quants : fix Zig and Swift builds + quantize tool ggml-ci * quantize : --pure option for disabling k-quant mixtures --------- Co-authored-by: cebtenzzre <cebtenzzre@gmail.com>
2023-10-29 16:32:28 +00:00
size_t ggml_quantize_q6_K(const float * src, void * dst, int n, int k, int64_t * hist) {
ggml : add SOTA 2,3,4,5,6 bit k-quantizations (#1684) * Starting to add k-quantization to ggml I think it is better to have quantization separate from ggml. For now just adding the k-quants there, but it would be better to also factor out the existing ggml quantizations. * Adding Q3_K and Q8_K (de)-quantization * Q3_K now working on CUDA and AVX2/scalar CUDA is not ideal - ~50% slower than Q4_0 for single token prediction, about the same in batch mode (perplexity). CPU single token is ~55 ms (on Ryzen 7950X). * Some improvement for Q3_K on CUDA It is now ~22.5 ms/token on my GPU, so ~30% slower than Q4_0. * Some more CUDA optimizations for Q3_K Single token is now 20.5 ms/token (~20% slower than Q4_0). Perplexity is on par with Q4_0. * Adding Q4_K - scalar, AVX2, CUDA Performance is the same or perhaps very slightly better than Q4_0 on the CPU. On the GPU, single token prediction is ~10% better than Q4_0, batch mode (perplexity is about the same). * Adding Q6_K - scalar, AVX2, CUDA Performance is ~40% lower compared to Q4_K on the CPU. This is to be expected, considering that we are memory bound on the CPU and the 6-bit model is ~44% larger than the 4-bit. On the GPU, single token prediction is ~6% lower than Q4_0, batch mode (perplexity) is even closer (but still slower). * Adding Q5_K - scalar, AVX2, CUDA Performance is ~20% lower compared to Q4_K on the CPU. This is to be expected, considering that we are memory bound on the CPU and the 5-bit model is ~22% larger than the 4-bit. On the GPU, single token prediction is about the same as Q4_0 for both, single token and batch prediction. * Per convention, all QX_K quantizations use Q5_K for output.weight * Adding quantization mixes * Quantization mixes: didn't quite get what I wanted in the last commit * Q4_K dot product for ARM_NEON * Q6_K dot product for ARM_NEON * Q5_K dot product for ARM_NEON * Adding Q3_K dot for ARM_NEON It is 22% slower than Q4_K, despite the smaller model size. On x86_64, where we are memory bound, the Q3_K model is quite a bit faster than Q4_K. * A very slightly faster ARM_NEON Q3_K dot * Adding Q2_K - just CUDA for now Token prediction is pretty good - about 15.5 ms on a RTX 4080. Perplexity is about the same as Q4_K. * Adding scalar and AVX2 Q2_K dot * Adding ARM_NEON Q2_K dot About the same performance as Q4_K. * A slightly faster ARM_NEON Q2_K dot Single token prediction is now ~36 ms on M2 Max. The code is much simpler too. * Fixed bug in Q2_K CUDA dot product kernel Stranegly enough, for the few prompts I tried with the 7B model the responses looked perfectly reasonable. Only realized something is not quite right when I tried the larger models and started getting nonse back. In any case, Q2_K single token evaluation time on an RTX 4080 in a Ryzen7950X box iusing CUDA and model fully loaded on the GPU are ~15.5 ms for 7B, ~25.4 ms for 13B, and ~55.8 ms for 30B. The max number of layers that fit in VRAM for The 65B is 32. With that, we get ~330 ms per token, which is not that much faster than just running on the CPU (~470 ms per token). * Don't print zeros/NaNs when no count histogram has been collected * A 10% faster CUDA vector dot kernel for Q3_K Q3_K is now running at ~18.5 ms / token on CUDA, so the gap to Q4_0 is only 10%. It seems memory acccess pattern is more important for performance than the amount of computation the kernel does. * A slightly daster Q4_K AVX2 dot product For perplexity, where we are less memory bound, time per pass drops by ~5%. Barely measurable difference for single token prediction. * A slightly faster ARM_NEON A4_K dot product * Minor * Fix quantization error test We cannot possibly be expecting rmse < 0.002 for 2- and 3-bit quantization variants. * Fix docker build I have been sloppy with vector reinterpret casts on ARM_NEON. It seems clang is very forgiving in that regard. * Added forgotten ggml.o dependence on k_quants.h to the Makefile * Had unintentionally committed the Makefile with -Ofast enabled * ggml : rename k_quants -> ggml-quants-k, use lowercase in code --------- Co-authored-by: Iwan Kawrakow <iwan.kawrakow@gmail.com> Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
2023-06-05 19:56:18 +00:00
assert(k % QK_K == 0);
ggml : quantization refactoring (#3833) * ggml : factor all quantization code in ggml-quants ggml-ci * ggml-quants : fix Zig and Swift builds + quantize tool ggml-ci * quantize : --pure option for disabling k-quant mixtures --------- Co-authored-by: cebtenzzre <cebtenzzre@gmail.com>
2023-10-29 16:32:28 +00:00
(void)hist; // TODO: collect histograms
ggml : add SOTA 2,3,4,5,6 bit k-quantizations (#1684) * Starting to add k-quantization to ggml I think it is better to have quantization separate from ggml. For now just adding the k-quants there, but it would be better to also factor out the existing ggml quantizations. * Adding Q3_K and Q8_K (de)-quantization * Q3_K now working on CUDA and AVX2/scalar CUDA is not ideal - ~50% slower than Q4_0 for single token prediction, about the same in batch mode (perplexity). CPU single token is ~55 ms (on Ryzen 7950X). * Some improvement for Q3_K on CUDA It is now ~22.5 ms/token on my GPU, so ~30% slower than Q4_0. * Some more CUDA optimizations for Q3_K Single token is now 20.5 ms/token (~20% slower than Q4_0). Perplexity is on par with Q4_0. * Adding Q4_K - scalar, AVX2, CUDA Performance is the same or perhaps very slightly better than Q4_0 on the CPU. On the GPU, single token prediction is ~10% better than Q4_0, batch mode (perplexity is about the same). * Adding Q6_K - scalar, AVX2, CUDA Performance is ~40% lower compared to Q4_K on the CPU. This is to be expected, considering that we are memory bound on the CPU and the 6-bit model is ~44% larger than the 4-bit. On the GPU, single token prediction is ~6% lower than Q4_0, batch mode (perplexity) is even closer (but still slower). * Adding Q5_K - scalar, AVX2, CUDA Performance is ~20% lower compared to Q4_K on the CPU. This is to be expected, considering that we are memory bound on the CPU and the 5-bit model is ~22% larger than the 4-bit. On the GPU, single token prediction is about the same as Q4_0 for both, single token and batch prediction. * Per convention, all QX_K quantizations use Q5_K for output.weight * Adding quantization mixes * Quantization mixes: didn't quite get what I wanted in the last commit * Q4_K dot product for ARM_NEON * Q6_K dot product for ARM_NEON * Q5_K dot product for ARM_NEON * Adding Q3_K dot for ARM_NEON It is 22% slower than Q4_K, despite the smaller model size. On x86_64, where we are memory bound, the Q3_K model is quite a bit faster than Q4_K. * A very slightly faster ARM_NEON Q3_K dot * Adding Q2_K - just CUDA for now Token prediction is pretty good - about 15.5 ms on a RTX 4080. Perplexity is about the same as Q4_K. * Adding scalar and AVX2 Q2_K dot * Adding ARM_NEON Q2_K dot About the same performance as Q4_K. * A slightly faster ARM_NEON Q2_K dot Single token prediction is now ~36 ms on M2 Max. The code is much simpler too. * Fixed bug in Q2_K CUDA dot product kernel Stranegly enough, for the few prompts I tried with the 7B model the responses looked perfectly reasonable. Only realized something is not quite right when I tried the larger models and started getting nonse back. In any case, Q2_K single token evaluation time on an RTX 4080 in a Ryzen7950X box iusing CUDA and model fully loaded on the GPU are ~15.5 ms for 7B, ~25.4 ms for 13B, and ~55.8 ms for 30B. The max number of layers that fit in VRAM for The 65B is 32. With that, we get ~330 ms per token, which is not that much faster than just running on the CPU (~470 ms per token). * Don't print zeros/NaNs when no count histogram has been collected * A 10% faster CUDA vector dot kernel for Q3_K Q3_K is now running at ~18.5 ms / token on CUDA, so the gap to Q4_0 is only 10%. It seems memory acccess pattern is more important for performance than the amount of computation the kernel does. * A slightly daster Q4_K AVX2 dot product For perplexity, where we are less memory bound, time per pass drops by ~5%. Barely measurable difference for single token prediction. * A slightly faster ARM_NEON A4_K dot product * Minor * Fix quantization error test We cannot possibly be expecting rmse < 0.002 for 2- and 3-bit quantization variants. * Fix docker build I have been sloppy with vector reinterpret casts on ARM_NEON. It seems clang is very forgiving in that regard. * Added forgotten ggml.o dependence on k_quants.h to the Makefile * Had unintentionally committed the Makefile with -Ofast enabled * ggml : rename k_quants -> ggml-quants-k, use lowercase in code --------- Co-authored-by: Iwan Kawrakow <iwan.kawrakow@gmail.com> Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
2023-06-05 19:56:18 +00:00
ggml : quantization refactoring (#3833) * ggml : factor all quantization code in ggml-quants ggml-ci * ggml-quants : fix Zig and Swift builds + quantize tool ggml-ci * quantize : --pure option for disabling k-quant mixtures --------- Co-authored-by: cebtenzzre <cebtenzzre@gmail.com>
2023-10-29 16:32:28 +00:00
for (int j = 0; j < n; j += k) {
block_q6_K * restrict y = (block_q6_K *)dst + j/QK_K;
quantize_row_q6_K_reference(src + j, y, k);
}
return (n/QK_K*sizeof(block_q6_K));
}
ggml : add SOTA 2,3,4,5,6 bit k-quantizations (#1684) * Starting to add k-quantization to ggml I think it is better to have quantization separate from ggml. For now just adding the k-quants there, but it would be better to also factor out the existing ggml quantizations. * Adding Q3_K and Q8_K (de)-quantization * Q3_K now working on CUDA and AVX2/scalar CUDA is not ideal - ~50% slower than Q4_0 for single token prediction, about the same in batch mode (perplexity). CPU single token is ~55 ms (on Ryzen 7950X). * Some improvement for Q3_K on CUDA It is now ~22.5 ms/token on my GPU, so ~30% slower than Q4_0. * Some more CUDA optimizations for Q3_K Single token is now 20.5 ms/token (~20% slower than Q4_0). Perplexity is on par with Q4_0. * Adding Q4_K - scalar, AVX2, CUDA Performance is the same or perhaps very slightly better than Q4_0 on the CPU. On the GPU, single token prediction is ~10% better than Q4_0, batch mode (perplexity is about the same). * Adding Q6_K - scalar, AVX2, CUDA Performance is ~40% lower compared to Q4_K on the CPU. This is to be expected, considering that we are memory bound on the CPU and the 6-bit model is ~44% larger than the 4-bit. On the GPU, single token prediction is ~6% lower than Q4_0, batch mode (perplexity) is even closer (but still slower). * Adding Q5_K - scalar, AVX2, CUDA Performance is ~20% lower compared to Q4_K on the CPU. This is to be expected, considering that we are memory bound on the CPU and the 5-bit model is ~22% larger than the 4-bit. On the GPU, single token prediction is about the same as Q4_0 for both, single token and batch prediction. * Per convention, all QX_K quantizations use Q5_K for output.weight * Adding quantization mixes * Quantization mixes: didn't quite get what I wanted in the last commit * Q4_K dot product for ARM_NEON * Q6_K dot product for ARM_NEON * Q5_K dot product for ARM_NEON * Adding Q3_K dot for ARM_NEON It is 22% slower than Q4_K, despite the smaller model size. On x86_64, where we are memory bound, the Q3_K model is quite a bit faster than Q4_K. * A very slightly faster ARM_NEON Q3_K dot * Adding Q2_K - just CUDA for now Token prediction is pretty good - about 15.5 ms on a RTX 4080. Perplexity is about the same as Q4_K. * Adding scalar and AVX2 Q2_K dot * Adding ARM_NEON Q2_K dot About the same performance as Q4_K. * A slightly faster ARM_NEON Q2_K dot Single token prediction is now ~36 ms on M2 Max. The code is much simpler too. * Fixed bug in Q2_K CUDA dot product kernel Stranegly enough, for the few prompts I tried with the 7B model the responses looked perfectly reasonable. Only realized something is not quite right when I tried the larger models and started getting nonse back. In any case, Q2_K single token evaluation time on an RTX 4080 in a Ryzen7950X box iusing CUDA and model fully loaded on the GPU are ~15.5 ms for 7B, ~25.4 ms for 13B, and ~55.8 ms for 30B. The max number of layers that fit in VRAM for The 65B is 32. With that, we get ~330 ms per token, which is not that much faster than just running on the CPU (~470 ms per token). * Don't print zeros/NaNs when no count histogram has been collected * A 10% faster CUDA vector dot kernel for Q3_K Q3_K is now running at ~18.5 ms / token on CUDA, so the gap to Q4_0 is only 10%. It seems memory acccess pattern is more important for performance than the amount of computation the kernel does. * A slightly daster Q4_K AVX2 dot product For perplexity, where we are less memory bound, time per pass drops by ~5%. Barely measurable difference for single token prediction. * A slightly faster ARM_NEON A4_K dot product * Minor * Fix quantization error test We cannot possibly be expecting rmse < 0.002 for 2- and 3-bit quantization variants. * Fix docker build I have been sloppy with vector reinterpret casts on ARM_NEON. It seems clang is very forgiving in that regard. * Added forgotten ggml.o dependence on k_quants.h to the Makefile * Had unintentionally committed the Makefile with -Ofast enabled * ggml : rename k_quants -> ggml-quants-k, use lowercase in code --------- Co-authored-by: Iwan Kawrakow <iwan.kawrakow@gmail.com> Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
2023-06-05 19:56:18 +00:00
Add ability to use importance matrix for all k-quants (#4930) Co-authored-by: Iwan Kawrakow <iwan.kawrakow@gmail.com>
2024-01-14 14:21:12 +00:00
static void quantize_row_q6_K_impl(const float * restrict x, block_q6_K * restrict y, int n_per_row, const float * quant_weights) {
#if QK_K != 256
(void)quant_weights;
quantize_row_q6_K_reference(x, y, n_per_row);
#else
assert(n_per_row % QK_K == 0);
const int nb = n_per_row / QK_K;
int8_t L[QK_K];
float scales[QK_K/16];
//float weights[16];
for (int i = 0; i < nb; i++) {
//float sum_x2 = 0;
//for (int j = 0; j < QK_K; ++j) sum_x2 += x[j]*x[j];
//float sigma2 = sum_x2/QK_K;
float max_scale = 0;
float max_abs_scale = 0;
for (int ib = 0; ib < QK_K/16; ++ib) {
float scale;
if (quant_weights) {
const float * qw = quant_weights + QK_K*i + 16*ib;
//for (int j = 0; j < 16; ++j) weights[j] = qw[j] * sqrtf(sigma2 + x[16*ib + j]*x[16*ib + j]);
//scale = make_qx_quants(16, 32, x + 16*ib, L + 16*ib, 1, weights);
scale = make_qx_quants(16, 32, x + 16*ib, L + 16*ib, 1, qw);
} else {
scale = make_qx_quants(16, 32, x + 16*ib, L + 16*ib, 1, NULL);
}
scales[ib] = scale;
const float abs_scale = fabsf(scale);
if (abs_scale > max_abs_scale) {
max_abs_scale = abs_scale;
max_scale = scale;
}
}
if (!max_abs_scale) {
memset(&y[i], 0, sizeof(block_q6_K));
y[i].d = GGML_FP32_TO_FP16(0.f);
x += QK_K;
continue;
}
float iscale = -128.f/max_scale;
y[i].d = GGML_FP32_TO_FP16(1/iscale);
for (int ib = 0; ib < QK_K/16; ++ib) {
y[i].scales[ib] = MIN(127, nearest_int(iscale*scales[ib]));
}
for (int j = 0; j < QK_K/16; ++j) {
float d = GGML_FP16_TO_FP32(y[i].d) * y[i].scales[j];
if (!d) {
continue;
}
for (int ii = 0; ii < 16; ++ii) {
int l = nearest_int(x[16*j + ii]/d);
l = MAX(-32, MIN(31, l));
L[16*j + ii] = l + 32;
}
}
uint8_t * restrict ql = y[i].ql;
uint8_t * restrict qh = y[i].qh;
for (int j = 0; j < QK_K; j += 128) {
for (int l = 0; l < 32; ++l) {
const uint8_t q1 = L[j + l + 0] & 0xF;
const uint8_t q2 = L[j + l + 32] & 0xF;
const uint8_t q3 = L[j + l + 64] & 0xF;
const uint8_t q4 = L[j + l + 96] & 0xF;
ql[l+ 0] = q1 | (q3 << 4);
ql[l+32] = q2 | (q4 << 4);
qh[l] = (L[j + l] >> 4) | ((L[j + l + 32] >> 4) << 2) | ((L[j + l + 64] >> 4) << 4) | ((L[j + l + 96] >> 4) << 6);
}
ql += 64;
qh += 32;
}
x += QK_K;
}
#endif
}
size_t quantize_q6_K(const float * src, void * dst, int nrow, int n_per_row, int64_t * hist, const float * quant_weights) {
(void)hist;
ggml : add IQ2 to test-backend-ops + refactoring (#4990) * ggml : add IQ2 to test-backend-ops + refactoring ggml-ci * cuda : update supports_op for IQ2 ggml-ci * ci : enable LLAMA_CUBLAS=1 for CUDA nodes ggml-ci * cuda : fix out-of-bounds-access in `mul_mat_vec_q` ggml-ci * tests : avoid creating RNGs for each Q tensor ggml-ci * tests : avoid creating RNGs for each tensor ggml-ci
2024-01-17 16:54:56 +00:00
size_t row_size = ggml_row_size(GGML_TYPE_Q6_K, n_per_row);
Add ability to use importance matrix for all k-quants (#4930) Co-authored-by: Iwan Kawrakow <iwan.kawrakow@gmail.com>
2024-01-14 14:21:12 +00:00
if (!quant_weights) {
quantize_row_q6_K_reference(src, dst, nrow*n_per_row);
}
else {
char * qrow = (char *)dst;
for (int row = 0; row < nrow; ++row) {
quantize_row_q6_K_impl(src, (block_q6_K*)qrow, n_per_row, quant_weights);
src += n_per_row;
qrow += row_size;
}
}
return nrow * row_size;
}
ggml : importance matrix support for legacy quants (#4969) * imatrix: adding support for legacy quants * imatrix: guard Q4_0/Q5_0 against ffn_down craziness --------- Co-authored-by: Iwan Kawrakow <iwan.kawrakow@gmail.com>
2024-01-16 17:51:26 +00:00
static void quantize_row_q4_0_impl(const float * restrict x, block_q4_0 * restrict y, int n_per_row, const float * quant_weights) {
static_assert(QK4_0 == 32, "QK4_0 must be 32");
if (!quant_weights) {
quantize_row_q4_0_reference(x, y, n_per_row);
return;
}
float weight[QK4_0];
int8_t L[QK4_0];
float sum_x2 = 0;
for (int j = 0; j < n_per_row; ++j) sum_x2 += x[j]*x[j];
float sigma2 = sum_x2/n_per_row;
const int nb = n_per_row/QK4_0;
for (int ib = 0; ib < nb; ++ib) {
const float * xb = x + QK4_0 * ib;
const float * qw = quant_weights + QK4_0 * ib;
for (int j = 0; j < QK4_0; ++j) weight[j] = qw[j] * sqrtf(sigma2 + xb[j]*xb[j]);
float d = make_qx_quants(QK4_0, 8, xb, L, 1, weight);
y[ib].d = GGML_FP32_TO_FP16(d);
for (int j = 0; j < 16; ++j) {
y[ib].qs[j] = L[j] | (L[j+16] << 4);
}
}
}
size_t quantize_q4_0(const float * src, void * dst, int nrow, int n_per_row, int64_t * hist, const float * quant_weights) {
if (!quant_weights) {
return ggml_quantize_q4_0(src, dst, nrow*n_per_row, n_per_row, hist);
}
ggml : add IQ2 to test-backend-ops + refactoring (#4990) * ggml : add IQ2 to test-backend-ops + refactoring ggml-ci * cuda : update supports_op for IQ2 ggml-ci * ci : enable LLAMA_CUBLAS=1 for CUDA nodes ggml-ci * cuda : fix out-of-bounds-access in `mul_mat_vec_q` ggml-ci * tests : avoid creating RNGs for each Q tensor ggml-ci * tests : avoid creating RNGs for each tensor ggml-ci
2024-01-17 16:54:56 +00:00
size_t row_size = ggml_row_size(GGML_TYPE_Q4_0, n_per_row);
ggml : importance matrix support for legacy quants (#4969) * imatrix: adding support for legacy quants * imatrix: guard Q4_0/Q5_0 against ffn_down craziness --------- Co-authored-by: Iwan Kawrakow <iwan.kawrakow@gmail.com>
2024-01-16 17:51:26 +00:00
char * qrow = (char *)dst;
for (int row = 0; row < nrow; ++row) {
quantize_row_q4_0_impl(src, (block_q4_0*)qrow, n_per_row, quant_weights);
src += n_per_row;
qrow += row_size;
}
return nrow * row_size;
}
static void quantize_row_q4_1_impl(const float * restrict x, block_q4_1 * restrict y, int n_per_row, const float * quant_weights) {
static_assert(QK4_1 == 32, "QK4_1 must be 32");
if (!quant_weights) {
quantize_row_q4_1_reference(x, y, n_per_row);
return;
}
float weight[QK4_1];
uint8_t L[QK4_1], Laux[QK4_1];
float sum_x2 = 0;
for (int j = 0; j < n_per_row; ++j) sum_x2 += x[j]*x[j];
float sigma2 = sum_x2/n_per_row;
const int nb = n_per_row/QK4_1;
for (int ib = 0; ib < nb; ++ib) {
const float * xb = x + QK4_1 * ib;
const float * qw = quant_weights + QK4_1 * ib;
for (int j = 0; j < QK4_1; ++j) weight[j] = qw[j] * sqrtf(sigma2 + xb[j]*xb[j]);
float min;
float d = make_qkx3_quants(QK4_1, 15, xb, weight, L, &min, Laux, -0.9f, 0.05f, 36, false);
y[ib].d = GGML_FP32_TO_FP16(d);
y[ib].m = GGML_FP32_TO_FP16(-min);
for (int j = 0; j < 16; ++j) {
y[ib].qs[j] = L[j] | (L[j+16] << 4);
}
}
}
size_t quantize_q4_1(const float * src, void * dst, int nrow, int n_per_row, int64_t * hist, const float * quant_weights) {
if (!quant_weights) {
return ggml_quantize_q4_1(src, dst, nrow*n_per_row, n_per_row, hist);
}
ggml : add IQ2 to test-backend-ops + refactoring (#4990) * ggml : add IQ2 to test-backend-ops + refactoring ggml-ci * cuda : update supports_op for IQ2 ggml-ci * ci : enable LLAMA_CUBLAS=1 for CUDA nodes ggml-ci * cuda : fix out-of-bounds-access in `mul_mat_vec_q` ggml-ci * tests : avoid creating RNGs for each Q tensor ggml-ci * tests : avoid creating RNGs for each tensor ggml-ci
2024-01-17 16:54:56 +00:00
size_t row_size = ggml_row_size(GGML_TYPE_Q4_1, n_per_row);
ggml : importance matrix support for legacy quants (#4969) * imatrix: adding support for legacy quants * imatrix: guard Q4_0/Q5_0 against ffn_down craziness --------- Co-authored-by: Iwan Kawrakow <iwan.kawrakow@gmail.com>
2024-01-16 17:51:26 +00:00
char * qrow = (char *)dst;
for (int row = 0; row < nrow; ++row) {
quantize_row_q4_1_impl(src, (block_q4_1*)qrow, n_per_row, quant_weights);
src += n_per_row;
qrow += row_size;
}
return nrow * row_size;
}
static void quantize_row_q5_0_impl(const float * restrict x, block_q5_0 * restrict y, int n_per_row, const float * quant_weights) {
static_assert(QK5_0 == 32, "QK5_0 must be 32");
if (!quant_weights) {
quantize_row_q5_0_reference(x, y, n_per_row);
return;
}
float weight[QK5_0];
int8_t L[QK5_0];
float sum_x2 = 0;
for (int j = 0; j < n_per_row; ++j) sum_x2 += x[j]*x[j];
float sigma2 = sum_x2/n_per_row;
const int nb = n_per_row/QK5_0;
for (int ib = 0; ib < nb; ++ib) {
const float * xb = x + QK5_0 * ib;
const float * qw = quant_weights + QK5_0 * ib;
for (int j = 0; j < QK5_0; ++j) weight[j] = qw[j] * sqrtf(sigma2 + xb[j]*xb[j]);
float d = make_qx_quants(QK5_0, 16, xb, L, 1, weight);
y[ib].d = GGML_FP32_TO_FP16(d);
uint32_t qh = 0;
for (int j = 0; j < 16; ++j) {
const uint8_t xi0 = L[j];
const uint8_t xi1 = L[j+16];
y[ib].qs[j] = (xi0 & 0x0F) | ((xi1 & 0x0F) << 4);
// get the 5-th bit and store it in qh at the right position
qh |= ((xi0 & 0x10u) >> 4) << (j + 0);
qh |= ((xi1 & 0x10u) >> 4) << (j + QK5_0/2);
}
memcpy(&y[ib].qh, &qh, sizeof(qh));
}
}
size_t quantize_q5_0(const float * src, void * dst, int nrow, int n_per_row, int64_t * hist, const float * quant_weights) {
if (!quant_weights) {
return ggml_quantize_q5_0(src, dst, nrow*n_per_row, n_per_row, hist);
}
ggml : add IQ2 to test-backend-ops + refactoring (#4990) * ggml : add IQ2 to test-backend-ops + refactoring ggml-ci * cuda : update supports_op for IQ2 ggml-ci * ci : enable LLAMA_CUBLAS=1 for CUDA nodes ggml-ci * cuda : fix out-of-bounds-access in `mul_mat_vec_q` ggml-ci * tests : avoid creating RNGs for each Q tensor ggml-ci * tests : avoid creating RNGs for each tensor ggml-ci
2024-01-17 16:54:56 +00:00
size_t row_size = ggml_row_size(GGML_TYPE_Q5_0, n_per_row);
ggml : importance matrix support for legacy quants (#4969) * imatrix: adding support for legacy quants * imatrix: guard Q4_0/Q5_0 against ffn_down craziness --------- Co-authored-by: Iwan Kawrakow <iwan.kawrakow@gmail.com>
2024-01-16 17:51:26 +00:00
char * qrow = (char *)dst;
for (int row = 0; row < nrow; ++row) {
quantize_row_q5_0_impl(src, (block_q5_0*)qrow, n_per_row, quant_weights);
src += n_per_row;
qrow += row_size;
}
return nrow * row_size;
}
static void quantize_row_q5_1_impl(const float * restrict x, block_q5_1 * restrict y, int n_per_row, const float * quant_weights) {
static_assert(QK5_1 == 32, "QK5_1 must be 32");
if (!quant_weights) {
quantize_row_q5_1_reference(x, y, n_per_row);
return;
}
float weight[QK5_1];
uint8_t L[QK5_1], Laux[QK5_1];
float sum_x2 = 0;
for (int j = 0; j < n_per_row; ++j) sum_x2 += x[j]*x[j];
float sigma2 = sum_x2/n_per_row;
const int nb = n_per_row/QK5_1;
for (int ib = 0; ib < nb; ++ib) {
const float * xb = x + QK5_1 * ib;
const float * qw = quant_weights + QK5_1 * ib;
for (int j = 0; j < QK5_1; ++j) weight[j] = qw[j] * sqrtf(sigma2 + xb[j]*xb[j]);
float min;
float d = make_qkx3_quants(QK5_1, 31, xb, weight, L, &min, Laux, -0.9f, 0.05f, 36, false);
y[ib].d = GGML_FP32_TO_FP16(d);
y[ib].m = GGML_FP32_TO_FP16(-min);
uint32_t qh = 0;
for (int j = 0; j < 16; ++j) {
const uint8_t xi0 = L[j];
const uint8_t xi1 = L[j+16];
y[ib].qs[j] = (xi0 & 0x0F) | ((xi1 & 0x0F) << 4);
// get the 5-th bit and store it in qh at the right position
qh |= ((xi0 & 0x10u) >> 4) << (j + 0);
qh |= ((xi1 & 0x10u) >> 4) << (j + QK5_0/2);
}
memcpy(&y[ib].qh, &qh, sizeof(qh));
}
}
size_t quantize_q5_1(const float * src, void * dst, int nrow, int n_per_row, int64_t * hist, const float * quant_weights) {
if (!quant_weights) {
return ggml_quantize_q5_1(src, dst, nrow*n_per_row, n_per_row, hist);
}
ggml : add IQ2 to test-backend-ops + refactoring (#4990) * ggml : add IQ2 to test-backend-ops + refactoring ggml-ci * cuda : update supports_op for IQ2 ggml-ci * ci : enable LLAMA_CUBLAS=1 for CUDA nodes ggml-ci * cuda : fix out-of-bounds-access in `mul_mat_vec_q` ggml-ci * tests : avoid creating RNGs for each Q tensor ggml-ci * tests : avoid creating RNGs for each tensor ggml-ci
2024-01-17 16:54:56 +00:00
size_t row_size = ggml_row_size(GGML_TYPE_Q5_1, n_per_row);
ggml : importance matrix support for legacy quants (#4969) * imatrix: adding support for legacy quants * imatrix: guard Q4_0/Q5_0 against ffn_down craziness --------- Co-authored-by: Iwan Kawrakow <iwan.kawrakow@gmail.com>
2024-01-16 17:51:26 +00:00
char * qrow = (char *)dst;
for (int row = 0; row < nrow; ++row) {
quantize_row_q5_1_impl(src, (block_q5_1*)qrow, n_per_row, quant_weights);
src += n_per_row;
qrow += row_size;
}
return nrow * row_size;
}
SOTA 2-bit quants (#4773) * iq2_xxs: basics * iq2_xxs: scalar and AVX2 dot products Needed to change Q8_K to have quants in the -127...127 range, else the IQ2_XXS AVX implementation becomes very awkward. The alternative would have been to use Q8_0 instead. Perhaps I'll change later, for now this is what we have. * iq2_xxs: ARM_NEON dot product Somehow strangely slow (112 ms/token). * iq2_xxs: WIP Metal Dequantize works, something is still wrong with the dot product. * iq2_xxs: Metal dot product now works We have PP-512 = 475 t/s TG-128 = 47.3 t/s Not the greatest performance, but not complete garbage either. * iq2_xxs: slighty faster dot product TG-128 is now 48.4 t/s * iq2_xxs: slighty faster dot product TG-128 is now 50.9 t/s * iq2_xxs: even faster Metal dot product TG-128 is now 54.1 t/s. Strangely enough, putting the signs lookup table into shared memory has a bigger impact than the grid values being in shared memory. * iq2_xxs: dequantize CUDA kernel - fix conflict with master * iq2_xxs: quantized CUDA dot product (MMVQ) We get TG-128 = 153.1 t/s * iq2_xxs: slightly faster CUDA dot product TG-128 is now at 155.1 t/s. * iq2_xxs: add to llama ftype enum * iq2_xxs: fix MoE on Metal * Fix missing MMQ ops when on hipBLAS I had put the ggml_supports_mmq call at the wrong place. * Fix bug in qequantize_row_iq2_xxs The 0.25f factor was missing. Great detective work by @ggerganov! * Fixing tests * PR suggestion --------- Co-authored-by: Iwan Kawrakow <iwan.kawrakow@gmail.com>
2024-01-08 15:02:32 +00:00
// ====================== "True" 2-bit (de)-quantization
ggml : SOTA 2-bit quants (add IQ2_XS) (#4856) * iq2_xs: basics * iq2_xs: this should have been in the basics * iq2_xs: CUDA and scalar CPU works * iq2_xs: WIP Metal * iq2_xs: Metal now works * iq2_xs: working, but dog slow, ARM_NEON dot product * iq2_xs: better ARM_NEON dot product We are now at 19.5 t/s for TG-128 and 61 t/s for PP-512 when running on the CPU. * iq2_xs: AVX2 dot product - 19.5 t/s * iq2_xs: faster AVX2 dit product 21.4 t/s for TG-128, 59.2 t/s for PP-512. The latter is 2x compared to the previous version. * iq2_xs: had forgotten to delete iq2-data.h * Add llama enum for IQ2_XS --------- Co-authored-by: Iwan Kawrakow <iwan.kawrakow@gmail.com>
2024-01-11 19:39:39 +00:00
static const uint64_t iq2xxs_grid[256] = {
SOTA 2-bit quants (#4773) * iq2_xxs: basics * iq2_xxs: scalar and AVX2 dot products Needed to change Q8_K to have quants in the -127...127 range, else the IQ2_XXS AVX implementation becomes very awkward. The alternative would have been to use Q8_0 instead. Perhaps I'll change later, for now this is what we have. * iq2_xxs: ARM_NEON dot product Somehow strangely slow (112 ms/token). * iq2_xxs: WIP Metal Dequantize works, something is still wrong with the dot product. * iq2_xxs: Metal dot product now works We have PP-512 = 475 t/s TG-128 = 47.3 t/s Not the greatest performance, but not complete garbage either. * iq2_xxs: slighty faster dot product TG-128 is now 48.4 t/s * iq2_xxs: slighty faster dot product TG-128 is now 50.9 t/s * iq2_xxs: even faster Metal dot product TG-128 is now 54.1 t/s. Strangely enough, putting the signs lookup table into shared memory has a bigger impact than the grid values being in shared memory. * iq2_xxs: dequantize CUDA kernel - fix conflict with master * iq2_xxs: quantized CUDA dot product (MMVQ) We get TG-128 = 153.1 t/s * iq2_xxs: slightly faster CUDA dot product TG-128 is now at 155.1 t/s. * iq2_xxs: add to llama ftype enum * iq2_xxs: fix MoE on Metal * Fix missing MMQ ops when on hipBLAS I had put the ggml_supports_mmq call at the wrong place. * Fix bug in qequantize_row_iq2_xxs The 0.25f factor was missing. Great detective work by @ggerganov! * Fixing tests * PR suggestion --------- Co-authored-by: Iwan Kawrakow <iwan.kawrakow@gmail.com>
2024-01-08 15:02:32 +00:00
0x0808080808080808, 0x080808080808082b, 0x0808080808081919, 0x0808080808082b08,
0x0808080808082b2b, 0x0808080808190819, 0x0808080808191908, 0x08080808082b0808,
0x08080808082b082b, 0x08080808082b2b08, 0x08080808082b2b2b, 0x0808080819080819,
0x0808080819081908, 0x0808080819190808, 0x0808080819192b08, 0x08080808192b0819,
0x08080808192b1908, 0x080808082b080808, 0x080808082b08082b, 0x080808082b082b2b,
0x080808082b2b082b, 0x0808081908080819, 0x0808081908081908, 0x0808081908190808,
0x0808081908191919, 0x0808081919080808, 0x080808192b081908, 0x080808192b192b08,
0x0808082b08080808, 0x0808082b0808082b, 0x0808082b082b082b, 0x0808082b2b08082b,
0x0808190808080819, 0x0808190808081908, 0x0808190808190808, 0x08081908082b0819,
0x08081908082b1908, 0x0808190819080808, 0x080819081908082b, 0x0808190819082b08,
0x08081908192b0808, 0x080819082b080819, 0x080819082b081908, 0x080819082b190808,
0x080819082b2b1908, 0x0808191908080808, 0x080819190808082b, 0x0808191908082b08,
0x08081919082b0808, 0x080819191908192b, 0x08081919192b2b19, 0x080819192b080808,
0x080819192b190819, 0x0808192b08082b19, 0x0808192b08190808, 0x0808192b19080808,
0x0808192b2b081908, 0x0808192b2b2b1908, 0x08082b0808080808, 0x08082b0808081919,
0x08082b0808082b08, 0x08082b0808191908, 0x08082b08082b2b08, 0x08082b0819080819,
0x08082b0819081908, 0x08082b0819190808, 0x08082b081919082b, 0x08082b082b082b08,
0x08082b1908081908, 0x08082b1919080808, 0x08082b2b0808082b, 0x08082b2b08191908,
0x0819080808080819, 0x0819080808081908, 0x0819080808190808, 0x08190808082b0819,
0x0819080819080808, 0x08190808192b0808, 0x081908082b081908, 0x081908082b190808,
0x081908082b191919, 0x0819081908080808, 0x0819081908082b08, 0x08190819082b0808,
0x0819081919190808, 0x0819081919192b2b, 0x081908192b080808, 0x0819082b082b1908,
0x0819082b19081919, 0x0819190808080808, 0x0819190808082b08, 0x08191908082b0808,
0x08191908082b1919, 0x0819190819082b19, 0x081919082b080808, 0x0819191908192b08,
0x08191919192b082b, 0x0819192b08080808, 0x0819192b0819192b, 0x08192b0808080819,
0x08192b0808081908, 0x08192b0808190808, 0x08192b0819080808, 0x08192b082b080819,
0x08192b1908080808, 0x08192b1908081919, 0x08192b192b2b0808, 0x08192b2b19190819,
0x082b080808080808, 0x082b08080808082b, 0x082b080808082b2b, 0x082b080819081908,
0x082b0808192b0819, 0x082b08082b080808, 0x082b08082b08082b, 0x082b0819082b2b19,
0x082b081919082b08, 0x082b082b08080808, 0x082b082b0808082b, 0x082b190808080819,
0x082b190808081908, 0x082b190808190808, 0x082b190819080808, 0x082b19081919192b,
0x082b191908080808, 0x082b191919080819, 0x082b1919192b1908, 0x082b192b2b190808,
0x082b2b0808082b08, 0x082b2b08082b0808, 0x082b2b082b191908, 0x082b2b2b19081908,
0x1908080808080819, 0x1908080808081908, 0x1908080808190808, 0x1908080808192b08,
0x19080808082b0819, 0x19080808082b1908, 0x1908080819080808, 0x1908080819082b08,
0x190808081919192b, 0x19080808192b0808, 0x190808082b080819, 0x190808082b081908,
0x190808082b190808, 0x1908081908080808, 0x19080819082b0808, 0x19080819192b0819,
0x190808192b080808, 0x190808192b081919, 0x1908082b08080819, 0x1908082b08190808,
0x1908082b19082b08, 0x1908082b1919192b, 0x1908082b192b2b08, 0x1908190808080808,
0x1908190808082b08, 0x19081908082b0808, 0x190819082b080808, 0x190819082b192b19,
0x190819190819082b, 0x19081919082b1908, 0x1908192b08080808, 0x19082b0808080819,
0x19082b0808081908, 0x19082b0808190808, 0x19082b0819080808, 0x19082b0819081919,
0x19082b1908080808, 0x19082b1919192b08, 0x19082b19192b0819, 0x19082b192b08082b,
0x19082b2b19081919, 0x19082b2b2b190808, 0x1919080808080808, 0x1919080808082b08,
0x1919080808190819, 0x1919080808192b19, 0x19190808082b0808, 0x191908082b080808,
0x191908082b082b08, 0x1919081908081908, 0x191908191908082b, 0x191908192b2b1908,
0x1919082b2b190819, 0x191919082b190808, 0x191919082b19082b, 0x1919191908082b2b,
0x1919192b08080819, 0x1919192b19191908, 0x19192b0808080808, 0x19192b0808190819,
0x19192b0808192b19, 0x19192b08192b1908, 0x19192b1919080808, 0x19192b2b08082b08,
0x192b080808081908, 0x192b080808190808, 0x192b080819080808, 0x192b0808192b2b08,
0x192b081908080808, 0x192b081919191919, 0x192b082b08192b08, 0x192b082b192b0808,
0x192b190808080808, 0x192b190808081919, 0x192b191908190808, 0x192b19190819082b,
0x192b19192b081908, 0x192b2b081908082b, 0x2b08080808080808, 0x2b0808080808082b,
0x2b08080808082b2b, 0x2b08080819080819, 0x2b0808082b08082b, 0x2b08081908081908,
0x2b08081908192b08, 0x2b08081919080808, 0x2b08082b08190819, 0x2b08190808080819,
0x2b08190808081908, 0x2b08190808190808, 0x2b08190808191919, 0x2b08190819080808,
0x2b081908192b0808, 0x2b08191908080808, 0x2b0819191908192b, 0x2b0819192b191908,
0x2b08192b08082b19, 0x2b08192b19080808, 0x2b08192b192b0808, 0x2b082b080808082b,
0x2b082b1908081908, 0x2b082b2b08190819, 0x2b19080808081908, 0x2b19080808190808,
0x2b190808082b1908, 0x2b19080819080808, 0x2b1908082b2b0819, 0x2b1908190819192b,
0x2b1908192b080808, 0x2b19082b19081919, 0x2b19190808080808, 0x2b191908082b082b,
0x2b19190819081908, 0x2b19191919190819, 0x2b192b082b080819, 0x2b192b19082b0808,
0x2b2b08080808082b, 0x2b2b080819190808, 0x2b2b08082b081919, 0x2b2b081908082b19,
0x2b2b082b08080808, 0x2b2b190808192b08, 0x2b2b2b0819190808, 0x2b2b2b1908081908,
};
ggml : SOTA 2-bit quants (add IQ2_XS) (#4856) * iq2_xs: basics * iq2_xs: this should have been in the basics * iq2_xs: CUDA and scalar CPU works * iq2_xs: WIP Metal * iq2_xs: Metal now works * iq2_xs: working, but dog slow, ARM_NEON dot product * iq2_xs: better ARM_NEON dot product We are now at 19.5 t/s for TG-128 and 61 t/s for PP-512 when running on the CPU. * iq2_xs: AVX2 dot product - 19.5 t/s * iq2_xs: faster AVX2 dit product 21.4 t/s for TG-128, 59.2 t/s for PP-512. The latter is 2x compared to the previous version. * iq2_xs: had forgotten to delete iq2-data.h * Add llama enum for IQ2_XS --------- Co-authored-by: Iwan Kawrakow <iwan.kawrakow@gmail.com>
2024-01-11 19:39:39 +00:00
static const uint64_t iq2xs_grid[512] = {
0x0808080808080808, 0x080808080808082b, 0x0808080808081919, 0x0808080808082b08,
0x0808080808082b2b, 0x0808080808190819, 0x0808080808191908, 0x080808080819192b,
0x0808080808192b19, 0x08080808082b0808, 0x08080808082b082b, 0x08080808082b1919,
0x08080808082b2b08, 0x0808080819080819, 0x0808080819081908, 0x080808081908192b,
0x0808080819082b19, 0x0808080819190808, 0x080808081919082b, 0x0808080819191919,
0x0808080819192b08, 0x08080808192b0819, 0x08080808192b1908, 0x080808082b080808,
0x080808082b08082b, 0x080808082b081919, 0x080808082b082b08, 0x080808082b190819,
0x080808082b191908, 0x080808082b192b19, 0x080808082b2b0808, 0x0808081908080819,
0x0808081908081908, 0x080808190808192b, 0x0808081908082b19, 0x0808081908190808,
0x080808190819082b, 0x0808081908191919, 0x0808081908192b08, 0x0808081908192b2b,
0x08080819082b0819, 0x08080819082b1908, 0x0808081919080808, 0x080808191908082b,
0x0808081919081919, 0x0808081919082b08, 0x0808081919190819, 0x0808081919191908,
0x08080819192b0808, 0x08080819192b2b08, 0x080808192b080819, 0x080808192b081908,
0x080808192b190808, 0x0808082b08080808, 0x0808082b0808082b, 0x0808082b08081919,
0x0808082b08082b08, 0x0808082b08190819, 0x0808082b08191908, 0x0808082b082b0808,
0x0808082b19080819, 0x0808082b19081908, 0x0808082b19190808, 0x0808082b19191919,
0x0808082b2b080808, 0x0808082b2b082b2b, 0x0808190808080819, 0x0808190808081908,
0x080819080808192b, 0x0808190808082b19, 0x0808190808190808, 0x080819080819082b,
0x0808190808191919, 0x0808190808192b08, 0x08081908082b0819, 0x08081908082b1908,
0x0808190819080808, 0x080819081908082b, 0x0808190819081919, 0x0808190819082b08,
0x0808190819190819, 0x0808190819191908, 0x080819081919192b, 0x08081908192b0808,
0x080819082b080819, 0x080819082b081908, 0x080819082b190808, 0x0808191908080808,
0x080819190808082b, 0x0808191908081919, 0x0808191908082b08, 0x0808191908190819,
0x0808191908191908, 0x08081919082b0808, 0x0808191919080819, 0x0808191919081908,
0x0808191919190808, 0x08081919192b0819, 0x080819192b080808, 0x0808192b08080819,
0x0808192b08081908, 0x0808192b08190808, 0x0808192b082b192b, 0x0808192b19080808,
0x0808192b1908082b, 0x0808192b2b081908, 0x08082b0808080808, 0x08082b080808082b,
0x08082b0808081919, 0x08082b0808082b08, 0x08082b0808082b2b, 0x08082b0808190819,
0x08082b0808191908, 0x08082b08082b0808, 0x08082b08082b1919, 0x08082b0819080819,
0x08082b0819081908, 0x08082b0819190808, 0x08082b0819192b08, 0x08082b082b080808,
0x08082b082b2b0808, 0x08082b082b2b2b2b, 0x08082b1908080819, 0x08082b1908081908,
0x08082b1908190808, 0x08082b1919080808, 0x08082b192b080819, 0x08082b192b082b19,
0x08082b2b08080808, 0x08082b2b082b0808, 0x08082b2b082b2b08, 0x08082b2b2b19192b,
0x08082b2b2b2b0808, 0x0819080808080819, 0x0819080808081908, 0x081908080808192b,
0x0819080808082b19, 0x0819080808190808, 0x081908080819082b, 0x0819080808191919,
0x0819080808192b08, 0x08190808082b0819, 0x08190808082b1908, 0x0819080819080808,
0x081908081908082b, 0x0819080819081919, 0x0819080819082b08, 0x0819080819190819,
0x0819080819191908, 0x08190808192b0808, 0x08190808192b2b2b, 0x081908082b080819,
0x081908082b081908, 0x081908082b190808, 0x0819081908080808, 0x081908190808082b,
0x0819081908081919, 0x0819081908082b08, 0x0819081908190819, 0x0819081908191908,
0x08190819082b0808, 0x0819081919080819, 0x0819081919081908, 0x0819081919190808,
0x081908192b080808, 0x081908192b191908, 0x081908192b19192b, 0x0819082b08080819,
0x0819082b08081908, 0x0819082b0808192b, 0x0819082b08190808, 0x0819082b19080808,
0x0819082b192b0808, 0x0819190808080808, 0x081919080808082b, 0x0819190808081919,
0x0819190808082b08, 0x0819190808190819, 0x0819190808191908, 0x08191908082b0808,
0x0819190819080819, 0x0819190819081908, 0x0819190819082b19, 0x0819190819190808,
0x08191908192b1908, 0x081919082b080808, 0x0819191908080819, 0x0819191908081908,
0x0819191908190808, 0x0819191919080808, 0x0819192b08080808, 0x0819192b08191908,
0x0819192b19082b19, 0x08192b0808080819, 0x08192b0808081908, 0x08192b0808190808,
0x08192b080819082b, 0x08192b0819080808, 0x08192b0819191908, 0x08192b082b08192b,
0x08192b1908080808, 0x08192b1908081919, 0x08192b19192b192b, 0x08192b2b19190819,
0x08192b2b2b2b2b19, 0x082b080808080808, 0x082b08080808082b, 0x082b080808081919,
0x082b080808082b08, 0x082b080808082b2b, 0x082b080808190819, 0x082b080808191908,
0x082b0808082b0808, 0x082b080819080819, 0x082b080819081908, 0x082b080819190808,
0x082b08082b080808, 0x082b08082b2b0808, 0x082b081908080819, 0x082b081908081908,
0x082b081908190808, 0x082b081919080808, 0x082b081919082b08, 0x082b0819192b1919,
0x082b082b08080808, 0x082b082b082b082b, 0x082b082b2b080808, 0x082b082b2b2b2b08,
0x082b190808080819, 0x082b190808081908, 0x082b190808190808, 0x082b1908082b2b19,
0x082b190819080808, 0x082b191908080808, 0x082b191919080819, 0x082b19191919082b,
0x082b19192b192b19, 0x082b192b08080819, 0x082b192b08192b2b, 0x082b192b2b2b192b,
0x082b2b0808080808, 0x082b2b0808082b08, 0x082b2b0808082b2b, 0x082b2b08082b0808,
0x082b2b0819191919, 0x082b2b082b082b08, 0x082b2b082b2b082b, 0x082b2b19192b2b08,
0x082b2b192b190808, 0x082b2b2b08082b08, 0x082b2b2b082b0808, 0x082b2b2b2b08082b,
0x082b2b2b2b082b08, 0x082b2b2b2b082b2b, 0x1908080808080819, 0x1908080808081908,
0x190808080808192b, 0x1908080808082b19, 0x1908080808190808, 0x190808080819082b,
0x1908080808191919, 0x1908080808192b08, 0x19080808082b0819, 0x19080808082b1908,
0x1908080819080808, 0x190808081908082b, 0x1908080819081919, 0x1908080819082b08,
0x1908080819082b2b, 0x1908080819190819, 0x1908080819191908, 0x19080808192b0808,
0x19080808192b1919, 0x190808082b080819, 0x190808082b081908, 0x190808082b190808,
0x1908081908080808, 0x190808190808082b, 0x1908081908081919, 0x1908081908082b08,
0x1908081908190819, 0x1908081908191908, 0x19080819082b0808, 0x1908081919080819,
0x1908081919081908, 0x1908081919190808, 0x190808192b080808, 0x190808192b081919,
0x190808192b2b082b, 0x1908082b08080819, 0x1908082b08081908, 0x1908082b08190808,
0x1908082b0819082b, 0x1908082b082b2b19, 0x1908082b19080808, 0x1908190808080808,
0x190819080808082b, 0x1908190808081919, 0x1908190808082b08, 0x1908190808190819,
0x1908190808191908, 0x1908190808192b19, 0x19081908082b0808, 0x1908190819080819,
0x1908190819081908, 0x1908190819190808, 0x190819082b080808, 0x190819082b191908,
0x1908191908080819, 0x1908191908081908, 0x1908191908190808, 0x19081919082b1908,
0x1908191919080808, 0x190819192b192b2b, 0x1908192b08080808, 0x1908192b08082b2b,
0x1908192b19081908, 0x1908192b19190808, 0x19082b0808080819, 0x19082b0808081908,
0x19082b0808190808, 0x19082b0819080808, 0x19082b0819081919, 0x19082b0819191908,
0x19082b08192b082b, 0x19082b1908080808, 0x19082b1908190819, 0x19082b1919081908,
0x19082b1919190808, 0x19082b19192b2b19, 0x19082b2b08081908, 0x1919080808080808,
0x191908080808082b, 0x1919080808081919, 0x1919080808082b08, 0x1919080808190819,
0x1919080808191908, 0x19190808082b0808, 0x19190808082b2b08, 0x1919080819080819,
0x1919080819081908, 0x1919080819190808, 0x191908082b080808, 0x1919081908080819,
0x1919081908081908, 0x1919081908190808, 0x1919081908191919, 0x1919081919080808,
0x191908191908082b, 0x1919082b08080808, 0x1919082b19081908, 0x1919082b2b2b2b2b,
0x1919190808080819, 0x1919190808081908, 0x1919190808190808, 0x19191908082b0819,
0x1919190819080808, 0x19191908192b0808, 0x191919082b080819, 0x191919082b2b0819,
0x1919191908080808, 0x1919191908082b08, 0x191919192b080808, 0x191919192b082b08,
0x1919192b082b0819, 0x1919192b192b2b08, 0x1919192b2b2b0819, 0x19192b0808080808,
0x19192b0808191908, 0x19192b0819080819, 0x19192b0819190808, 0x19192b082b192b19,
0x19192b1908192b2b, 0x19192b1919080808, 0x19192b191908082b, 0x19192b2b2b081919,
0x192b080808080819, 0x192b080808081908, 0x192b080808190808, 0x192b080819080808,
0x192b080819191908, 0x192b0808192b082b, 0x192b08082b08192b, 0x192b08082b2b2b19,
0x192b081908080808, 0x192b082b082b1908, 0x192b082b19082b2b, 0x192b082b2b19082b,
0x192b190808080808, 0x192b19080819192b, 0x192b191908190808, 0x192b191919080808,
0x192b191919081919, 0x192b19192b2b1908, 0x192b2b0808080819, 0x192b2b08192b2b2b,
0x192b2b19082b1919, 0x192b2b2b0808192b, 0x192b2b2b19191908, 0x192b2b2b192b082b,
0x2b08080808080808, 0x2b0808080808082b, 0x2b08080808081919, 0x2b08080808082b08,
0x2b08080808190819, 0x2b08080808191908, 0x2b080808082b0808, 0x2b080808082b2b2b,
0x2b08080819080819, 0x2b08080819081908, 0x2b08080819190808, 0x2b0808082b080808,
0x2b0808082b08082b, 0x2b0808082b2b2b08, 0x2b0808082b2b2b2b, 0x2b08081908080819,
0x2b08081908081908, 0x2b0808190808192b, 0x2b08081908190808, 0x2b08081919080808,
0x2b08081919190819, 0x2b08081919192b19, 0x2b08082b08080808, 0x2b08082b082b0808,
0x2b08082b2b080808, 0x2b08082b2b08082b, 0x2b08082b2b2b0808, 0x2b08082b2b2b2b08,
0x2b08190808080819, 0x2b08190808081908, 0x2b08190808190808, 0x2b0819080819082b,
0x2b08190808191919, 0x2b08190819080808, 0x2b081908192b0808, 0x2b0819082b082b19,
0x2b08191908080808, 0x2b08191919081908, 0x2b0819192b2b1919, 0x2b08192b08192b08,
0x2b08192b192b2b2b, 0x2b082b0808080808, 0x2b082b0808082b08, 0x2b082b08082b1919,
0x2b082b0819192b2b, 0x2b082b082b080808, 0x2b082b082b08082b, 0x2b082b082b2b2b08,
0x2b082b190808192b, 0x2b082b2b082b082b, 0x2b082b2b2b080808, 0x2b082b2b2b082b08,
0x2b082b2b2b19192b, 0x2b082b2b2b2b2b08, 0x2b19080808080819, 0x2b19080808081908,
0x2b19080808190808, 0x2b19080819080808, 0x2b1908081919192b, 0x2b1908082b081908,
0x2b19081908080808, 0x2b190819082b082b, 0x2b190819192b1908, 0x2b19082b1919192b,
0x2b19082b2b082b19, 0x2b19190808080808, 0x2b19190808081919, 0x2b19190819081908,
0x2b19190819190808, 0x2b19190819192b08, 0x2b191919082b2b19, 0x2b1919192b190808,
0x2b1919192b19082b, 0x2b19192b19080819, 0x2b192b0819190819, 0x2b192b082b2b192b,
0x2b192b1919082b19, 0x2b192b2b08191919, 0x2b192b2b192b0808, 0x2b2b080808080808,
0x2b2b08080808082b, 0x2b2b080808082b08, 0x2b2b080808082b2b, 0x2b2b0808082b0808,
0x2b2b0808082b2b2b, 0x2b2b08082b2b0808, 0x2b2b081919190819, 0x2b2b081919192b19,
0x2b2b08192b2b192b, 0x2b2b082b08080808, 0x2b2b082b0808082b, 0x2b2b082b08082b08,
0x2b2b082b082b2b2b, 0x2b2b082b2b080808, 0x2b2b082b2b2b0808, 0x2b2b190819080808,
0x2b2b19082b191919, 0x2b2b192b192b1919, 0x2b2b192b2b192b08, 0x2b2b2b0808082b2b,
0x2b2b2b08082b0808, 0x2b2b2b08082b082b, 0x2b2b2b08082b2b08, 0x2b2b2b082b2b0808,
0x2b2b2b082b2b2b08, 0x2b2b2b1908081908, 0x2b2b2b192b081908, 0x2b2b2b192b08192b,
0x2b2b2b2b082b2b08, 0x2b2b2b2b082b2b2b, 0x2b2b2b2b2b190819, 0x2b2b2b2b2b2b2b2b,
};
SOTA 3-bit quants (#5196) * iq3_xxs: quantize/dequantize RMSE seems a bit high-ish at about half-way between q2_K and q3_K, so need to check more. * iq3_xxs: CUDA dequantize works * iq2_xxs: tuning quantization * iq3_xxs: starting to look better PPL on wiki.test.raw LLaMA-v1-7B: 6.4218 LLaMA-v2-7B: 6.3560 Mistral-7B : 6.0717 This is better than Q3_K_XS, with a 5% reduction in quantized model size. * iq3_xxs: CUDA dot product We have PP-512: 5891 t/s TG-128: 143.9 t/s * iq3_xxs: scalar and AVX2 dot products * iq3_xxs: ARM_NEON and Metal Metal performance is decent, ARM_NEON is pathetic * iq3_xxs: slightly better grid points * Faster iq3_xxs and iq2_xs dot products on CUDA * iq3_xxs: add some quant mix * iq3_xxs: fix failing quantization test Dot product still fails. Is this real? * iq3_xxs: hopefully fix ROCm * iq3_xxs: failing tests This time the dot product accuracy did find an actual bug in the AVX2 implementation. * Add IQ3_XXS to test-backend-ops --------- Co-authored-by: Iwan Kawrakow <iwan.kawrakow@gmail.com>
2024-01-30 13:14:12 +00:00
static const uint32_t iq3xxs_grid[256] = {
0x04040404, 0x04040414, 0x04040424, 0x04040c0c, 0x04040c1c, 0x04040c3e, 0x04041404, 0x04041414,
0x04041c0c, 0x04042414, 0x04043e1c, 0x04043e2c, 0x040c040c, 0x040c041c, 0x040c0c04, 0x040c0c14,
0x040c140c, 0x040c142c, 0x040c1c04, 0x040c1c14, 0x040c240c, 0x040c2c24, 0x040c3e04, 0x04140404,
0x04140414, 0x04140424, 0x04140c0c, 0x04141404, 0x04141414, 0x04141c0c, 0x04141c1c, 0x04141c3e,
0x04142c0c, 0x04142c3e, 0x04143e2c, 0x041c040c, 0x041c043e, 0x041c0c04, 0x041c0c14, 0x041c142c,
0x041c3e04, 0x04240c1c, 0x04241c3e, 0x04242424, 0x04242c3e, 0x04243e1c, 0x04243e2c, 0x042c040c,
0x042c043e, 0x042c1c14, 0x042c2c14, 0x04341c2c, 0x04343424, 0x043e0c04, 0x043e0c24, 0x043e0c34,
0x043e241c, 0x043e340c, 0x0c04040c, 0x0c04041c, 0x0c040c04, 0x0c040c14, 0x0c04140c, 0x0c04141c,
0x0c041c04, 0x0c041c14, 0x0c041c24, 0x0c04243e, 0x0c042c04, 0x0c0c0404, 0x0c0c0414, 0x0c0c0c0c,
0x0c0c1404, 0x0c0c1414, 0x0c14040c, 0x0c14041c, 0x0c140c04, 0x0c140c14, 0x0c14140c, 0x0c141c04,
0x0c143e14, 0x0c1c0404, 0x0c1c0414, 0x0c1c1404, 0x0c1c1c0c, 0x0c1c2434, 0x0c1c3434, 0x0c24040c,
0x0c24042c, 0x0c242c04, 0x0c2c1404, 0x0c2c1424, 0x0c2c2434, 0x0c2c3e0c, 0x0c34042c, 0x0c3e1414,
0x0c3e2404, 0x14040404, 0x14040414, 0x14040c0c, 0x14040c1c, 0x14041404, 0x14041414, 0x14041434,
0x14041c0c, 0x14042414, 0x140c040c, 0x140c041c, 0x140c042c, 0x140c0c04, 0x140c0c14, 0x140c140c,
0x140c1c04, 0x140c341c, 0x140c343e, 0x140c3e04, 0x14140404, 0x14140414, 0x14140c0c, 0x14140c3e,
0x14141404, 0x14141414, 0x14141c3e, 0x14142404, 0x14142c2c, 0x141c040c, 0x141c0c04, 0x141c0c24,
0x141c3e04, 0x141c3e24, 0x14241c2c, 0x14242c1c, 0x142c041c, 0x142c143e, 0x142c240c, 0x142c3e24,
0x143e040c, 0x143e041c, 0x143e0c34, 0x143e242c, 0x1c04040c, 0x1c040c04, 0x1c040c14, 0x1c04140c,
0x1c04141c, 0x1c042c04, 0x1c04342c, 0x1c043e14, 0x1c0c0404, 0x1c0c0414, 0x1c0c1404, 0x1c0c1c0c,
0x1c0c2424, 0x1c0c2434, 0x1c14040c, 0x1c14041c, 0x1c140c04, 0x1c14142c, 0x1c142c14, 0x1c143e14,
0x1c1c0c0c, 0x1c1c1c1c, 0x1c241c04, 0x1c24243e, 0x1c243e14, 0x1c2c0404, 0x1c2c0434, 0x1c2c1414,
0x1c2c2c2c, 0x1c340c24, 0x1c341c34, 0x1c34341c, 0x1c3e1c1c, 0x1c3e3404, 0x24040424, 0x24040c3e,
0x24041c2c, 0x24041c3e, 0x24042c1c, 0x24042c3e, 0x240c3e24, 0x24141404, 0x24141c3e, 0x24142404,
0x24143404, 0x24143434, 0x241c043e, 0x241c242c, 0x24240424, 0x24242c0c, 0x24243424, 0x242c142c,
0x242c241c, 0x242c3e04, 0x243e042c, 0x243e0c04, 0x243e0c14, 0x243e1c04, 0x2c040c14, 0x2c04240c,
0x2c043e04, 0x2c0c0404, 0x2c0c0434, 0x2c0c1434, 0x2c0c2c2c, 0x2c140c24, 0x2c141c14, 0x2c143e14,
0x2c1c0414, 0x2c1c2c1c, 0x2c240c04, 0x2c24141c, 0x2c24143e, 0x2c243e14, 0x2c2c0414, 0x2c2c1c0c,
0x2c342c04, 0x2c3e1424, 0x2c3e2414, 0x34041424, 0x34042424, 0x34042434, 0x34043424, 0x340c140c,
0x340c340c, 0x34140c3e, 0x34143424, 0x341c1c04, 0x341c1c34, 0x34242424, 0x342c042c, 0x342c2c14,
0x34341c1c, 0x343e041c, 0x343e140c, 0x3e04041c, 0x3e04042c, 0x3e04043e, 0x3e040c04, 0x3e041c14,
0x3e042c14, 0x3e0c1434, 0x3e0c2404, 0x3e140c14, 0x3e14242c, 0x3e142c14, 0x3e1c0404, 0x3e1c0c2c,
0x3e1c1c1c, 0x3e1c3404, 0x3e24140c, 0x3e24240c, 0x3e2c0404, 0x3e2c0414, 0x3e2c1424, 0x3e341c04,
};
SOTA 2-bit quants (#4773) * iq2_xxs: basics * iq2_xxs: scalar and AVX2 dot products Needed to change Q8_K to have quants in the -127...127 range, else the IQ2_XXS AVX implementation becomes very awkward. The alternative would have been to use Q8_0 instead. Perhaps I'll change later, for now this is what we have. * iq2_xxs: ARM_NEON dot product Somehow strangely slow (112 ms/token). * iq2_xxs: WIP Metal Dequantize works, something is still wrong with the dot product. * iq2_xxs: Metal dot product now works We have PP-512 = 475 t/s TG-128 = 47.3 t/s Not the greatest performance, but not complete garbage either. * iq2_xxs: slighty faster dot product TG-128 is now 48.4 t/s * iq2_xxs: slighty faster dot product TG-128 is now 50.9 t/s * iq2_xxs: even faster Metal dot product TG-128 is now 54.1 t/s. Strangely enough, putting the signs lookup table into shared memory has a bigger impact than the grid values being in shared memory. * iq2_xxs: dequantize CUDA kernel - fix conflict with master * iq2_xxs: quantized CUDA dot product (MMVQ) We get TG-128 = 153.1 t/s * iq2_xxs: slightly faster CUDA dot product TG-128 is now at 155.1 t/s. * iq2_xxs: add to llama ftype enum * iq2_xxs: fix MoE on Metal * Fix missing MMQ ops when on hipBLAS I had put the ggml_supports_mmq call at the wrong place. * Fix bug in qequantize_row_iq2_xxs The 0.25f factor was missing. Great detective work by @ggerganov! * Fixing tests * PR suggestion --------- Co-authored-by: Iwan Kawrakow <iwan.kawrakow@gmail.com>
2024-01-08 15:02:32 +00:00
static const uint8_t ksigns_iq2xs[128] = {
0, 129, 130, 3, 132, 5, 6, 135, 136, 9, 10, 139, 12, 141, 142, 15,
144, 17, 18, 147, 20, 149, 150, 23, 24, 153, 154, 27, 156, 29, 30, 159,
160, 33, 34, 163, 36, 165, 166, 39, 40, 169, 170, 43, 172, 45, 46, 175,
48, 177, 178, 51, 180, 53, 54, 183, 184, 57, 58, 187, 60, 189, 190, 63,
192, 65, 66, 195, 68, 197, 198, 71, 72, 201, 202, 75, 204, 77, 78, 207,
80, 209, 210, 83, 212, 85, 86, 215, 216, 89, 90, 219, 92, 221, 222, 95,
96, 225, 226, 99, 228, 101, 102, 231, 232, 105, 106, 235, 108, 237, 238, 111,
240, 113, 114, 243, 116, 245, 246, 119, 120, 249, 250, 123, 252, 125, 126, 255,
};
ggml : SOTA 2-bit quants (add IQ2_XS) (#4856) * iq2_xs: basics * iq2_xs: this should have been in the basics * iq2_xs: CUDA and scalar CPU works * iq2_xs: WIP Metal * iq2_xs: Metal now works * iq2_xs: working, but dog slow, ARM_NEON dot product * iq2_xs: better ARM_NEON dot product We are now at 19.5 t/s for TG-128 and 61 t/s for PP-512 when running on the CPU. * iq2_xs: AVX2 dot product - 19.5 t/s * iq2_xs: faster AVX2 dit product 21.4 t/s for TG-128, 59.2 t/s for PP-512. The latter is 2x compared to the previous version. * iq2_xs: had forgotten to delete iq2-data.h * Add llama enum for IQ2_XS --------- Co-authored-by: Iwan Kawrakow <iwan.kawrakow@gmail.com>
2024-01-11 19:39:39 +00:00
SOTA 2-bit quants (#4773) * iq2_xxs: basics * iq2_xxs: scalar and AVX2 dot products Needed to change Q8_K to have quants in the -127...127 range, else the IQ2_XXS AVX implementation becomes very awkward. The alternative would have been to use Q8_0 instead. Perhaps I'll change later, for now this is what we have. * iq2_xxs: ARM_NEON dot product Somehow strangely slow (112 ms/token). * iq2_xxs: WIP Metal Dequantize works, something is still wrong with the dot product. * iq2_xxs: Metal dot product now works We have PP-512 = 475 t/s TG-128 = 47.3 t/s Not the greatest performance, but not complete garbage either. * iq2_xxs: slighty faster dot product TG-128 is now 48.4 t/s * iq2_xxs: slighty faster dot product TG-128 is now 50.9 t/s * iq2_xxs: even faster Metal dot product TG-128 is now 54.1 t/s. Strangely enough, putting the signs lookup table into shared memory has a bigger impact than the grid values being in shared memory. * iq2_xxs: dequantize CUDA kernel - fix conflict with master * iq2_xxs: quantized CUDA dot product (MMVQ) We get TG-128 = 153.1 t/s * iq2_xxs: slightly faster CUDA dot product TG-128 is now at 155.1 t/s. * iq2_xxs: add to llama ftype enum * iq2_xxs: fix MoE on Metal * Fix missing MMQ ops when on hipBLAS I had put the ggml_supports_mmq call at the wrong place. * Fix bug in qequantize_row_iq2_xxs The 0.25f factor was missing. Great detective work by @ggerganov! * Fixing tests * PR suggestion --------- Co-authored-by: Iwan Kawrakow <iwan.kawrakow@gmail.com>
2024-01-08 15:02:32 +00:00
static const uint8_t kmask_iq2xs[8] = {1, 2, 4, 8, 16, 32, 64, 128};
void dequantize_row_iq2_xxs(const block_iq2_xxs * restrict x, float * restrict y, int k) {
assert(k % QK_K == 0);
const int nb = k / QK_K;
uint32_t aux32[2];
const uint8_t * aux8 = (const uint8_t *)aux32;
for (int i = 0; i < nb; i++) {
const float d = GGML_FP16_TO_FP32(x[i].d);
for (int ib32 = 0; ib32 < QK_K/32; ++ib32) {
memcpy(aux32, x[i].qs + 4*ib32, 2*sizeof(uint32_t));
const float db = d * (0.5f + (aux32[1] >> 28)) * 0.25f;
for (int l = 0; l < 4; ++l) {
const uint8_t * grid = (const uint8_t *)(iq2xxs_grid + aux8[l]);
const uint8_t signs = ksigns_iq2xs[(aux32[1] >> 7*l) & 127];
for (int j = 0; j < 8; ++j) {
y[j] = db * grid[j] * (signs & kmask_iq2xs[j] ? -1.f : 1.f);
}
y += 8;
}
}
}
}
ggml : SOTA 2-bit quants (add IQ2_XS) (#4856) * iq2_xs: basics * iq2_xs: this should have been in the basics * iq2_xs: CUDA and scalar CPU works * iq2_xs: WIP Metal * iq2_xs: Metal now works * iq2_xs: working, but dog slow, ARM_NEON dot product * iq2_xs: better ARM_NEON dot product We are now at 19.5 t/s for TG-128 and 61 t/s for PP-512 when running on the CPU. * iq2_xs: AVX2 dot product - 19.5 t/s * iq2_xs: faster AVX2 dit product 21.4 t/s for TG-128, 59.2 t/s for PP-512. The latter is 2x compared to the previous version. * iq2_xs: had forgotten to delete iq2-data.h * Add llama enum for IQ2_XS --------- Co-authored-by: Iwan Kawrakow <iwan.kawrakow@gmail.com>
2024-01-11 19:39:39 +00:00
// ====================== 2.3125 bpw (de)-quantization
void dequantize_row_iq2_xs(const block_iq2_xs * restrict x, float * restrict y, int k) {
assert(k % QK_K == 0);
const int nb = k / QK_K;
float db[2];
for (int i = 0; i < nb; i++) {
const float d = GGML_FP16_TO_FP32(x[i].d);
for (int ib32 = 0; ib32 < QK_K/32; ++ib32) {
db[0] = d * (0.5f + (x[i].scales[ib32] & 0xf)) * 0.25f;
db[1] = d * (0.5f + (x[i].scales[ib32] >> 4)) * 0.25f;
for (int l = 0; l < 4; ++l) {
const uint8_t * grid = (const uint8_t *)(iq2xs_grid + (x[i].qs[4*ib32 + l] & 511));
const uint8_t signs = ksigns_iq2xs[x[i].qs[4*ib32 + l] >> 9];
for (int j = 0; j < 8; ++j) {
y[j] = db[l/2] * grid[j] * (signs & kmask_iq2xs[j] ? -1.f : 1.f);
}
y += 8;
}
}
}
}
SOTA 3-bit quants (#5196) * iq3_xxs: quantize/dequantize RMSE seems a bit high-ish at about half-way between q2_K and q3_K, so need to check more. * iq3_xxs: CUDA dequantize works * iq2_xxs: tuning quantization * iq3_xxs: starting to look better PPL on wiki.test.raw LLaMA-v1-7B: 6.4218 LLaMA-v2-7B: 6.3560 Mistral-7B : 6.0717 This is better than Q3_K_XS, with a 5% reduction in quantized model size. * iq3_xxs: CUDA dot product We have PP-512: 5891 t/s TG-128: 143.9 t/s * iq3_xxs: scalar and AVX2 dot products * iq3_xxs: ARM_NEON and Metal Metal performance is decent, ARM_NEON is pathetic * iq3_xxs: slightly better grid points * Faster iq3_xxs and iq2_xs dot products on CUDA * iq3_xxs: add some quant mix * iq3_xxs: fix failing quantization test Dot product still fails. Is this real? * iq3_xxs: hopefully fix ROCm * iq3_xxs: failing tests This time the dot product accuracy did find an actual bug in the AVX2 implementation. * Add IQ3_XXS to test-backend-ops --------- Co-authored-by: Iwan Kawrakow <iwan.kawrakow@gmail.com>
2024-01-30 13:14:12 +00:00
// ====================== 3.0625 bpw (de)-quantization
void dequantize_row_iq3_xxs(const block_iq3_xxs * restrict x, float * restrict y, int k) {
assert(k % QK_K == 0);
const int nb = k / QK_K;
uint32_t aux32;
for (int i = 0; i < nb; i++) {
const float d = GGML_FP16_TO_FP32(x[i].d);
const uint8_t * qs = x[i].qs;
const uint8_t * scales_and_signs = qs + QK_K/4;
for (int ib32 = 0; ib32 < QK_K/32; ++ib32) {
memcpy(&aux32, scales_and_signs + 4*ib32, sizeof(uint32_t));
const float db = d * (0.5f + (aux32 >> 28)) * 0.5f;
for (int l = 0; l < 4; ++l) {
const uint8_t signs = ksigns_iq2xs[(aux32 >> 7*l) & 127];
const uint8_t * grid1 = (const uint8_t *)(iq3xxs_grid + qs[2*l+0]);
const uint8_t * grid2 = (const uint8_t *)(iq3xxs_grid + qs[2*l+1]);
for (int j = 0; j < 4; ++j) {
y[j+0] = db * grid1[j] * (signs & kmask_iq2xs[j+0] ? -1.f : 1.f);
y[j+4] = db * grid2[j] * (signs & kmask_iq2xs[j+4] ? -1.f : 1.f);
}
y += 8;
}
qs += 8;
}
}
}
ggml : quantization refactoring (#3833) * ggml : factor all quantization code in ggml-quants ggml-ci * ggml-quants : fix Zig and Swift builds + quantize tool ggml-ci * quantize : --pure option for disabling k-quant mixtures --------- Co-authored-by: cebtenzzre <cebtenzzre@gmail.com>
2023-10-29 16:32:28 +00:00
//===================================== Q8_K ==============================================
void quantize_row_q8_K_reference(const float * restrict x, block_q8_K * restrict y, int k) {
assert(k % QK_K == 0);
const int nb = k / QK_K;
ggml : add SOTA 2,3,4,5,6 bit k-quantizations (#1684) * Starting to add k-quantization to ggml I think it is better to have quantization separate from ggml. For now just adding the k-quants there, but it would be better to also factor out the existing ggml quantizations. * Adding Q3_K and Q8_K (de)-quantization * Q3_K now working on CUDA and AVX2/scalar CUDA is not ideal - ~50% slower than Q4_0 for single token prediction, about the same in batch mode (perplexity). CPU single token is ~55 ms (on Ryzen 7950X). * Some improvement for Q3_K on CUDA It is now ~22.5 ms/token on my GPU, so ~30% slower than Q4_0. * Some more CUDA optimizations for Q3_K Single token is now 20.5 ms/token (~20% slower than Q4_0). Perplexity is on par with Q4_0. * Adding Q4_K - scalar, AVX2, CUDA Performance is the same or perhaps very slightly better than Q4_0 on the CPU. On the GPU, single token prediction is ~10% better than Q4_0, batch mode (perplexity is about the same). * Adding Q6_K - scalar, AVX2, CUDA Performance is ~40% lower compared to Q4_K on the CPU. This is to be expected, considering that we are memory bound on the CPU and the 6-bit model is ~44% larger than the 4-bit. On the GPU, single token prediction is ~6% lower than Q4_0, batch mode (perplexity) is even closer (but still slower). * Adding Q5_K - scalar, AVX2, CUDA Performance is ~20% lower compared to Q4_K on the CPU. This is to be expected, considering that we are memory bound on the CPU and the 5-bit model is ~22% larger than the 4-bit. On the GPU, single token prediction is about the same as Q4_0 for both, single token and batch prediction. * Per convention, all QX_K quantizations use Q5_K for output.weight * Adding quantization mixes * Quantization mixes: didn't quite get what I wanted in the last commit * Q4_K dot product for ARM_NEON * Q6_K dot product for ARM_NEON * Q5_K dot product for ARM_NEON * Adding Q3_K dot for ARM_NEON It is 22% slower than Q4_K, despite the smaller model size. On x86_64, where we are memory bound, the Q3_K model is quite a bit faster than Q4_K. * A very slightly faster ARM_NEON Q3_K dot * Adding Q2_K - just CUDA for now Token prediction is pretty good - about 15.5 ms on a RTX 4080. Perplexity is about the same as Q4_K. * Adding scalar and AVX2 Q2_K dot * Adding ARM_NEON Q2_K dot About the same performance as Q4_K. * A slightly faster ARM_NEON Q2_K dot Single token prediction is now ~36 ms on M2 Max. The code is much simpler too. * Fixed bug in Q2_K CUDA dot product kernel Stranegly enough, for the few prompts I tried with the 7B model the responses looked perfectly reasonable. Only realized something is not quite right when I tried the larger models and started getting nonse back. In any case, Q2_K single token evaluation time on an RTX 4080 in a Ryzen7950X box iusing CUDA and model fully loaded on the GPU are ~15.5 ms for 7B, ~25.4 ms for 13B, and ~55.8 ms for 30B. The max number of layers that fit in VRAM for The 65B is 32. With that, we get ~330 ms per token, which is not that much faster than just running on the CPU (~470 ms per token). * Don't print zeros/NaNs when no count histogram has been collected * A 10% faster CUDA vector dot kernel for Q3_K Q3_K is now running at ~18.5 ms / token on CUDA, so the gap to Q4_0 is only 10%. It seems memory acccess pattern is more important for performance than the amount of computation the kernel does. * A slightly daster Q4_K AVX2 dot product For perplexity, where we are less memory bound, time per pass drops by ~5%. Barely measurable difference for single token prediction. * A slightly faster ARM_NEON A4_K dot product * Minor * Fix quantization error test We cannot possibly be expecting rmse < 0.002 for 2- and 3-bit quantization variants. * Fix docker build I have been sloppy with vector reinterpret casts on ARM_NEON. It seems clang is very forgiving in that regard. * Added forgotten ggml.o dependence on k_quants.h to the Makefile * Had unintentionally committed the Makefile with -Ofast enabled * ggml : rename k_quants -> ggml-quants-k, use lowercase in code --------- Co-authored-by: Iwan Kawrakow <iwan.kawrakow@gmail.com> Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
2023-06-05 19:56:18 +00:00
for (int i = 0; i < nb; i++) {
ggml : quantization refactoring (#3833) * ggml : factor all quantization code in ggml-quants ggml-ci * ggml-quants : fix Zig and Swift builds + quantize tool ggml-ci * quantize : --pure option for disabling k-quant mixtures --------- Co-authored-by: cebtenzzre <cebtenzzre@gmail.com>
2023-10-29 16:32:28 +00:00
float max = 0;
float amax = 0;
for (int j = 0; j < QK_K; ++j) {
float ax = fabsf(x[j]);
if (ax > amax) {
amax = ax; max = x[j];
ggml : add SOTA 2,3,4,5,6 bit k-quantizations (#1684) * Starting to add k-quantization to ggml I think it is better to have quantization separate from ggml. For now just adding the k-quants there, but it would be better to also factor out the existing ggml quantizations. * Adding Q3_K and Q8_K (de)-quantization * Q3_K now working on CUDA and AVX2/scalar CUDA is not ideal - ~50% slower than Q4_0 for single token prediction, about the same in batch mode (perplexity). CPU single token is ~55 ms (on Ryzen 7950X). * Some improvement for Q3_K on CUDA It is now ~22.5 ms/token on my GPU, so ~30% slower than Q4_0. * Some more CUDA optimizations for Q3_K Single token is now 20.5 ms/token (~20% slower than Q4_0). Perplexity is on par with Q4_0. * Adding Q4_K - scalar, AVX2, CUDA Performance is the same or perhaps very slightly better than Q4_0 on the CPU. On the GPU, single token prediction is ~10% better than Q4_0, batch mode (perplexity is about the same). * Adding Q6_K - scalar, AVX2, CUDA Performance is ~40% lower compared to Q4_K on the CPU. This is to be expected, considering that we are memory bound on the CPU and the 6-bit model is ~44% larger than the 4-bit. On the GPU, single token prediction is ~6% lower than Q4_0, batch mode (perplexity) is even closer (but still slower). * Adding Q5_K - scalar, AVX2, CUDA Performance is ~20% lower compared to Q4_K on the CPU. This is to be expected, considering that we are memory bound on the CPU and the 5-bit model is ~22% larger than the 4-bit. On the GPU, single token prediction is about the same as Q4_0 for both, single token and batch prediction. * Per convention, all QX_K quantizations use Q5_K for output.weight * Adding quantization mixes * Quantization mixes: didn't quite get what I wanted in the last commit * Q4_K dot product for ARM_NEON * Q6_K dot product for ARM_NEON * Q5_K dot product for ARM_NEON * Adding Q3_K dot for ARM_NEON It is 22% slower than Q4_K, despite the smaller model size. On x86_64, where we are memory bound, the Q3_K model is quite a bit faster than Q4_K. * A very slightly faster ARM_NEON Q3_K dot * Adding Q2_K - just CUDA for now Token prediction is pretty good - about 15.5 ms on a RTX 4080. Perplexity is about the same as Q4_K. * Adding scalar and AVX2 Q2_K dot * Adding ARM_NEON Q2_K dot About the same performance as Q4_K. * A slightly faster ARM_NEON Q2_K dot Single token prediction is now ~36 ms on M2 Max. The code is much simpler too. * Fixed bug in Q2_K CUDA dot product kernel Stranegly enough, for the few prompts I tried with the 7B model the responses looked perfectly reasonable. Only realized something is not quite right when I tried the larger models and started getting nonse back. In any case, Q2_K single token evaluation time on an RTX 4080 in a Ryzen7950X box iusing CUDA and model fully loaded on the GPU are ~15.5 ms for 7B, ~25.4 ms for 13B, and ~55.8 ms for 30B. The max number of layers that fit in VRAM for The 65B is 32. With that, we get ~330 ms per token, which is not that much faster than just running on the CPU (~470 ms per token). * Don't print zeros/NaNs when no count histogram has been collected * A 10% faster CUDA vector dot kernel for Q3_K Q3_K is now running at ~18.5 ms / token on CUDA, so the gap to Q4_0 is only 10%. It seems memory acccess pattern is more important for performance than the amount of computation the kernel does. * A slightly daster Q4_K AVX2 dot product For perplexity, where we are less memory bound, time per pass drops by ~5%. Barely measurable difference for single token prediction. * A slightly faster ARM_NEON A4_K dot product * Minor * Fix quantization error test We cannot possibly be expecting rmse < 0.002 for 2- and 3-bit quantization variants. * Fix docker build I have been sloppy with vector reinterpret casts on ARM_NEON. It seems clang is very forgiving in that regard. * Added forgotten ggml.o dependence on k_quants.h to the Makefile * Had unintentionally committed the Makefile with -Ofast enabled * ggml : rename k_quants -> ggml-quants-k, use lowercase in code --------- Co-authored-by: Iwan Kawrakow <iwan.kawrakow@gmail.com> Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
2023-06-05 19:56:18 +00:00
}
}
ggml : quantization refactoring (#3833) * ggml : factor all quantization code in ggml-quants ggml-ci * ggml-quants : fix Zig and Swift builds + quantize tool ggml-ci * quantize : --pure option for disabling k-quant mixtures --------- Co-authored-by: cebtenzzre <cebtenzzre@gmail.com>
2023-10-29 16:32:28 +00:00
if (!amax) {
y[i].d = 0;
memset(y[i].qs, 0, QK_K);
x += QK_K;
continue;
}
SOTA 2-bit quants (#4773) * iq2_xxs: basics * iq2_xxs: scalar and AVX2 dot products Needed to change Q8_K to have quants in the -127...127 range, else the IQ2_XXS AVX implementation becomes very awkward. The alternative would have been to use Q8_0 instead. Perhaps I'll change later, for now this is what we have. * iq2_xxs: ARM_NEON dot product Somehow strangely slow (112 ms/token). * iq2_xxs: WIP Metal Dequantize works, something is still wrong with the dot product. * iq2_xxs: Metal dot product now works We have PP-512 = 475 t/s TG-128 = 47.3 t/s Not the greatest performance, but not complete garbage either. * iq2_xxs: slighty faster dot product TG-128 is now 48.4 t/s * iq2_xxs: slighty faster dot product TG-128 is now 50.9 t/s * iq2_xxs: even faster Metal dot product TG-128 is now 54.1 t/s. Strangely enough, putting the signs lookup table into shared memory has a bigger impact than the grid values being in shared memory. * iq2_xxs: dequantize CUDA kernel - fix conflict with master * iq2_xxs: quantized CUDA dot product (MMVQ) We get TG-128 = 153.1 t/s * iq2_xxs: slightly faster CUDA dot product TG-128 is now at 155.1 t/s. * iq2_xxs: add to llama ftype enum * iq2_xxs: fix MoE on Metal * Fix missing MMQ ops when on hipBLAS I had put the ggml_supports_mmq call at the wrong place. * Fix bug in qequantize_row_iq2_xxs The 0.25f factor was missing. Great detective work by @ggerganov! * Fixing tests * PR suggestion --------- Co-authored-by: Iwan Kawrakow <iwan.kawrakow@gmail.com>
2024-01-08 15:02:32 +00:00
//const float iscale = -128.f/max;
// We need this change for IQ2_XXS, else the AVX implementation becomes very awkward
const float iscale = -127.f/max;
ggml : quantization refactoring (#3833) * ggml : factor all quantization code in ggml-quants ggml-ci * ggml-quants : fix Zig and Swift builds + quantize tool ggml-ci * quantize : --pure option for disabling k-quant mixtures --------- Co-authored-by: cebtenzzre <cebtenzzre@gmail.com>
2023-10-29 16:32:28 +00:00
for (int j = 0; j < QK_K; ++j) {
int v = nearest_int(iscale*x[j]);
y[i].qs[j] = MIN(127, v);
ggml : add SOTA 2,3,4,5,6 bit k-quantizations (#1684) * Starting to add k-quantization to ggml I think it is better to have quantization separate from ggml. For now just adding the k-quants there, but it would be better to also factor out the existing ggml quantizations. * Adding Q3_K and Q8_K (de)-quantization * Q3_K now working on CUDA and AVX2/scalar CUDA is not ideal - ~50% slower than Q4_0 for single token prediction, about the same in batch mode (perplexity). CPU single token is ~55 ms (on Ryzen 7950X). * Some improvement for Q3_K on CUDA It is now ~22.5 ms/token on my GPU, so ~30% slower than Q4_0. * Some more CUDA optimizations for Q3_K Single token is now 20.5 ms/token (~20% slower than Q4_0). Perplexity is on par with Q4_0. * Adding Q4_K - scalar, AVX2, CUDA Performance is the same or perhaps very slightly better than Q4_0 on the CPU. On the GPU, single token prediction is ~10% better than Q4_0, batch mode (perplexity is about the same). * Adding Q6_K - scalar, AVX2, CUDA Performance is ~40% lower compared to Q4_K on the CPU. This is to be expected, considering that we are memory bound on the CPU and the 6-bit model is ~44% larger than the 4-bit. On the GPU, single token prediction is ~6% lower than Q4_0, batch mode (perplexity) is even closer (but still slower). * Adding Q5_K - scalar, AVX2, CUDA Performance is ~20% lower compared to Q4_K on the CPU. This is to be expected, considering that we are memory bound on the CPU and the 5-bit model is ~22% larger than the 4-bit. On the GPU, single token prediction is about the same as Q4_0 for both, single token and batch prediction. * Per convention, all QX_K quantizations use Q5_K for output.weight * Adding quantization mixes * Quantization mixes: didn't quite get what I wanted in the last commit * Q4_K dot product for ARM_NEON * Q6_K dot product for ARM_NEON * Q5_K dot product for ARM_NEON * Adding Q3_K dot for ARM_NEON It is 22% slower than Q4_K, despite the smaller model size. On x86_64, where we are memory bound, the Q3_K model is quite a bit faster than Q4_K. * A very slightly faster ARM_NEON Q3_K dot * Adding Q2_K - just CUDA for now Token prediction is pretty good - about 15.5 ms on a RTX 4080. Perplexity is about the same as Q4_K. * Adding scalar and AVX2 Q2_K dot * Adding ARM_NEON Q2_K dot About the same performance as Q4_K. * A slightly faster ARM_NEON Q2_K dot Single token prediction is now ~36 ms on M2 Max. The code is much simpler too. * Fixed bug in Q2_K CUDA dot product kernel Stranegly enough, for the few prompts I tried with the 7B model the responses looked perfectly reasonable. Only realized something is not quite right when I tried the larger models and started getting nonse back. In any case, Q2_K single token evaluation time on an RTX 4080 in a Ryzen7950X box iusing CUDA and model fully loaded on the GPU are ~15.5 ms for 7B, ~25.4 ms for 13B, and ~55.8 ms for 30B. The max number of layers that fit in VRAM for The 65B is 32. With that, we get ~330 ms per token, which is not that much faster than just running on the CPU (~470 ms per token). * Don't print zeros/NaNs when no count histogram has been collected * A 10% faster CUDA vector dot kernel for Q3_K Q3_K is now running at ~18.5 ms / token on CUDA, so the gap to Q4_0 is only 10%. It seems memory acccess pattern is more important for performance than the amount of computation the kernel does. * A slightly daster Q4_K AVX2 dot product For perplexity, where we are less memory bound, time per pass drops by ~5%. Barely measurable difference for single token prediction. * A slightly faster ARM_NEON A4_K dot product * Minor * Fix quantization error test We cannot possibly be expecting rmse < 0.002 for 2- and 3-bit quantization variants. * Fix docker build I have been sloppy with vector reinterpret casts on ARM_NEON. It seems clang is very forgiving in that regard. * Added forgotten ggml.o dependence on k_quants.h to the Makefile * Had unintentionally committed the Makefile with -Ofast enabled * ggml : rename k_quants -> ggml-quants-k, use lowercase in code --------- Co-authored-by: Iwan Kawrakow <iwan.kawrakow@gmail.com> Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
2023-06-05 19:56:18 +00:00
}
for (int j = 0; j < QK_K/16; ++j) {
ggml : quantization refactoring (#3833) * ggml : factor all quantization code in ggml-quants ggml-ci * ggml-quants : fix Zig and Swift builds + quantize tool ggml-ci * quantize : --pure option for disabling k-quant mixtures --------- Co-authored-by: cebtenzzre <cebtenzzre@gmail.com>
2023-10-29 16:32:28 +00:00
int sum = 0;
ggml : add SOTA 2,3,4,5,6 bit k-quantizations (#1684) * Starting to add k-quantization to ggml I think it is better to have quantization separate from ggml. For now just adding the k-quants there, but it would be better to also factor out the existing ggml quantizations. * Adding Q3_K and Q8_K (de)-quantization * Q3_K now working on CUDA and AVX2/scalar CUDA is not ideal - ~50% slower than Q4_0 for single token prediction, about the same in batch mode (perplexity). CPU single token is ~55 ms (on Ryzen 7950X). * Some improvement for Q3_K on CUDA It is now ~22.5 ms/token on my GPU, so ~30% slower than Q4_0. * Some more CUDA optimizations for Q3_K Single token is now 20.5 ms/token (~20% slower than Q4_0). Perplexity is on par with Q4_0. * Adding Q4_K - scalar, AVX2, CUDA Performance is the same or perhaps very slightly better than Q4_0 on the CPU. On the GPU, single token prediction is ~10% better than Q4_0, batch mode (perplexity is about the same). * Adding Q6_K - scalar, AVX2, CUDA Performance is ~40% lower compared to Q4_K on the CPU. This is to be expected, considering that we are memory bound on the CPU and the 6-bit model is ~44% larger than the 4-bit. On the GPU, single token prediction is ~6% lower than Q4_0, batch mode (perplexity) is even closer (but still slower). * Adding Q5_K - scalar, AVX2, CUDA Performance is ~20% lower compared to Q4_K on the CPU. This is to be expected, considering that we are memory bound on the CPU and the 5-bit model is ~22% larger than the 4-bit. On the GPU, single token prediction is about the same as Q4_0 for both, single token and batch prediction. * Per convention, all QX_K quantizations use Q5_K for output.weight * Adding quantization mixes * Quantization mixes: didn't quite get what I wanted in the last commit * Q4_K dot product for ARM_NEON * Q6_K dot product for ARM_NEON * Q5_K dot product for ARM_NEON * Adding Q3_K dot for ARM_NEON It is 22% slower than Q4_K, despite the smaller model size. On x86_64, where we are memory bound, the Q3_K model is quite a bit faster than Q4_K. * A very slightly faster ARM_NEON Q3_K dot * Adding Q2_K - just CUDA for now Token prediction is pretty good - about 15.5 ms on a RTX 4080. Perplexity is about the same as Q4_K. * Adding scalar and AVX2 Q2_K dot * Adding ARM_NEON Q2_K dot About the same performance as Q4_K. * A slightly faster ARM_NEON Q2_K dot Single token prediction is now ~36 ms on M2 Max. The code is much simpler too. * Fixed bug in Q2_K CUDA dot product kernel Stranegly enough, for the few prompts I tried with the 7B model the responses looked perfectly reasonable. Only realized something is not quite right when I tried the larger models and started getting nonse back. In any case, Q2_K single token evaluation time on an RTX 4080 in a Ryzen7950X box iusing CUDA and model fully loaded on the GPU are ~15.5 ms for 7B, ~25.4 ms for 13B, and ~55.8 ms for 30B. The max number of layers that fit in VRAM for The 65B is 32. With that, we get ~330 ms per token, which is not that much faster than just running on the CPU (~470 ms per token). * Don't print zeros/NaNs when no count histogram has been collected * A 10% faster CUDA vector dot kernel for Q3_K Q3_K is now running at ~18.5 ms / token on CUDA, so the gap to Q4_0 is only 10%. It seems memory acccess pattern is more important for performance than the amount of computation the kernel does. * A slightly daster Q4_K AVX2 dot product For perplexity, where we are less memory bound, time per pass drops by ~5%. Barely measurable difference for single token prediction. * A slightly faster ARM_NEON A4_K dot product * Minor * Fix quantization error test We cannot possibly be expecting rmse < 0.002 for 2- and 3-bit quantization variants. * Fix docker build I have been sloppy with vector reinterpret casts on ARM_NEON. It seems clang is very forgiving in that regard. * Added forgotten ggml.o dependence on k_quants.h to the Makefile * Had unintentionally committed the Makefile with -Ofast enabled * ggml : rename k_quants -> ggml-quants-k, use lowercase in code --------- Co-authored-by: Iwan Kawrakow <iwan.kawrakow@gmail.com> Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
2023-06-05 19:56:18 +00:00
for (int ii = 0; ii < 16; ++ii) {
ggml : quantization refactoring (#3833) * ggml : factor all quantization code in ggml-quants ggml-ci * ggml-quants : fix Zig and Swift builds + quantize tool ggml-ci * quantize : --pure option for disabling k-quant mixtures --------- Co-authored-by: cebtenzzre <cebtenzzre@gmail.com>
2023-10-29 16:32:28 +00:00
sum += y[i].qs[j*16 + ii];
ggml : add SOTA 2,3,4,5,6 bit k-quantizations (#1684) * Starting to add k-quantization to ggml I think it is better to have quantization separate from ggml. For now just adding the k-quants there, but it would be better to also factor out the existing ggml quantizations. * Adding Q3_K and Q8_K (de)-quantization * Q3_K now working on CUDA and AVX2/scalar CUDA is not ideal - ~50% slower than Q4_0 for single token prediction, about the same in batch mode (perplexity). CPU single token is ~55 ms (on Ryzen 7950X). * Some improvement for Q3_K on CUDA It is now ~22.5 ms/token on my GPU, so ~30% slower than Q4_0. * Some more CUDA optimizations for Q3_K Single token is now 20.5 ms/token (~20% slower than Q4_0). Perplexity is on par with Q4_0. * Adding Q4_K - scalar, AVX2, CUDA Performance is the same or perhaps very slightly better than Q4_0 on the CPU. On the GPU, single token prediction is ~10% better than Q4_0, batch mode (perplexity is about the same). * Adding Q6_K - scalar, AVX2, CUDA Performance is ~40% lower compared to Q4_K on the CPU. This is to be expected, considering that we are memory bound on the CPU and the 6-bit model is ~44% larger than the 4-bit. On the GPU, single token prediction is ~6% lower than Q4_0, batch mode (perplexity) is even closer (but still slower). * Adding Q5_K - scalar, AVX2, CUDA Performance is ~20% lower compared to Q4_K on the CPU. This is to be expected, considering that we are memory bound on the CPU and the 5-bit model is ~22% larger than the 4-bit. On the GPU, single token prediction is about the same as Q4_0 for both, single token and batch prediction. * Per convention, all QX_K quantizations use Q5_K for output.weight * Adding quantization mixes * Quantization mixes: didn't quite get what I wanted in the last commit * Q4_K dot product for ARM_NEON * Q6_K dot product for ARM_NEON * Q5_K dot product for ARM_NEON * Adding Q3_K dot for ARM_NEON It is 22% slower than Q4_K, despite the smaller model size. On x86_64, where we are memory bound, the Q3_K model is quite a bit faster than Q4_K. * A very slightly faster ARM_NEON Q3_K dot * Adding Q2_K - just CUDA for now Token prediction is pretty good - about 15.5 ms on a RTX 4080. Perplexity is about the same as Q4_K. * Adding scalar and AVX2 Q2_K dot * Adding ARM_NEON Q2_K dot About the same performance as Q4_K. * A slightly faster ARM_NEON Q2_K dot Single token prediction is now ~36 ms on M2 Max. The code is much simpler too. * Fixed bug in Q2_K CUDA dot product kernel Stranegly enough, for the few prompts I tried with the 7B model the responses looked perfectly reasonable. Only realized something is not quite right when I tried the larger models and started getting nonse back. In any case, Q2_K single token evaluation time on an RTX 4080 in a Ryzen7950X box iusing CUDA and model fully loaded on the GPU are ~15.5 ms for 7B, ~25.4 ms for 13B, and ~55.8 ms for 30B. The max number of layers that fit in VRAM for The 65B is 32. With that, we get ~330 ms per token, which is not that much faster than just running on the CPU (~470 ms per token). * Don't print zeros/NaNs when no count histogram has been collected * A 10% faster CUDA vector dot kernel for Q3_K Q3_K is now running at ~18.5 ms / token on CUDA, so the gap to Q4_0 is only 10%. It seems memory acccess pattern is more important for performance than the amount of computation the kernel does. * A slightly daster Q4_K AVX2 dot product For perplexity, where we are less memory bound, time per pass drops by ~5%. Barely measurable difference for single token prediction. * A slightly faster ARM_NEON A4_K dot product * Minor * Fix quantization error test We cannot possibly be expecting rmse < 0.002 for 2- and 3-bit quantization variants. * Fix docker build I have been sloppy with vector reinterpret casts on ARM_NEON. It seems clang is very forgiving in that regard. * Added forgotten ggml.o dependence on k_quants.h to the Makefile * Had unintentionally committed the Makefile with -Ofast enabled * ggml : rename k_quants -> ggml-quants-k, use lowercase in code --------- Co-authored-by: Iwan Kawrakow <iwan.kawrakow@gmail.com> Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
2023-06-05 19:56:18 +00:00
}
ggml : quantization refactoring (#3833) * ggml : factor all quantization code in ggml-quants ggml-ci * ggml-quants : fix Zig and Swift builds + quantize tool ggml-ci * quantize : --pure option for disabling k-quant mixtures --------- Co-authored-by: cebtenzzre <cebtenzzre@gmail.com>
2023-10-29 16:32:28 +00:00
y[i].bsums[j] = sum;
ggml : add SOTA 2,3,4,5,6 bit k-quantizations (#1684) * Starting to add k-quantization to ggml I think it is better to have quantization separate from ggml. For now just adding the k-quants there, but it would be better to also factor out the existing ggml quantizations. * Adding Q3_K and Q8_K (de)-quantization * Q3_K now working on CUDA and AVX2/scalar CUDA is not ideal - ~50% slower than Q4_0 for single token prediction, about the same in batch mode (perplexity). CPU single token is ~55 ms (on Ryzen 7950X). * Some improvement for Q3_K on CUDA It is now ~22.5 ms/token on my GPU, so ~30% slower than Q4_0. * Some more CUDA optimizations for Q3_K Single token is now 20.5 ms/token (~20% slower than Q4_0). Perplexity is on par with Q4_0. * Adding Q4_K - scalar, AVX2, CUDA Performance is the same or perhaps very slightly better than Q4_0 on the CPU. On the GPU, single token prediction is ~10% better than Q4_0, batch mode (perplexity is about the same). * Adding Q6_K - scalar, AVX2, CUDA Performance is ~40% lower compared to Q4_K on the CPU. This is to be expected, considering that we are memory bound on the CPU and the 6-bit model is ~44% larger than the 4-bit. On the GPU, single token prediction is ~6% lower than Q4_0, batch mode (perplexity) is even closer (but still slower). * Adding Q5_K - scalar, AVX2, CUDA Performance is ~20% lower compared to Q4_K on the CPU. This is to be expected, considering that we are memory bound on the CPU and the 5-bit model is ~22% larger than the 4-bit. On the GPU, single token prediction is about the same as Q4_0 for both, single token and batch prediction. * Per convention, all QX_K quantizations use Q5_K for output.weight * Adding quantization mixes * Quantization mixes: didn't quite get what I wanted in the last commit * Q4_K dot product for ARM_NEON * Q6_K dot product for ARM_NEON * Q5_K dot product for ARM_NEON * Adding Q3_K dot for ARM_NEON It is 22% slower than Q4_K, despite the smaller model size. On x86_64, where we are memory bound, the Q3_K model is quite a bit faster than Q4_K. * A very slightly faster ARM_NEON Q3_K dot * Adding Q2_K - just CUDA for now Token prediction is pretty good - about 15.5 ms on a RTX 4080. Perplexity is about the same as Q4_K. * Adding scalar and AVX2 Q2_K dot * Adding ARM_NEON Q2_K dot About the same performance as Q4_K. * A slightly faster ARM_NEON Q2_K dot Single token prediction is now ~36 ms on M2 Max. The code is much simpler too. * Fixed bug in Q2_K CUDA dot product kernel Stranegly enough, for the few prompts I tried with the 7B model the responses looked perfectly reasonable. Only realized something is not quite right when I tried the larger models and started getting nonse back. In any case, Q2_K single token evaluation time on an RTX 4080 in a Ryzen7950X box iusing CUDA and model fully loaded on the GPU are ~15.5 ms for 7B, ~25.4 ms for 13B, and ~55.8 ms for 30B. The max number of layers that fit in VRAM for The 65B is 32. With that, we get ~330 ms per token, which is not that much faster than just running on the CPU (~470 ms per token). * Don't print zeros/NaNs when no count histogram has been collected * A 10% faster CUDA vector dot kernel for Q3_K Q3_K is now running at ~18.5 ms / token on CUDA, so the gap to Q4_0 is only 10%. It seems memory acccess pattern is more important for performance than the amount of computation the kernel does. * A slightly daster Q4_K AVX2 dot product For perplexity, where we are less memory bound, time per pass drops by ~5%. Barely measurable difference for single token prediction. * A slightly faster ARM_NEON A4_K dot product * Minor * Fix quantization error test We cannot possibly be expecting rmse < 0.002 for 2- and 3-bit quantization variants. * Fix docker build I have been sloppy with vector reinterpret casts on ARM_NEON. It seems clang is very forgiving in that regard. * Added forgotten ggml.o dependence on k_quants.h to the Makefile * Had unintentionally committed the Makefile with -Ofast enabled * ggml : rename k_quants -> ggml-quants-k, use lowercase in code --------- Co-authored-by: Iwan Kawrakow <iwan.kawrakow@gmail.com> Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
2023-06-05 19:56:18 +00:00
}
ggml : quantization refactoring (#3833) * ggml : factor all quantization code in ggml-quants ggml-ci * ggml-quants : fix Zig and Swift builds + quantize tool ggml-ci * quantize : --pure option for disabling k-quant mixtures --------- Co-authored-by: cebtenzzre <cebtenzzre@gmail.com>
2023-10-29 16:32:28 +00:00
y[i].d = 1/iscale;
ggml : add SOTA 2,3,4,5,6 bit k-quantizations (#1684) * Starting to add k-quantization to ggml I think it is better to have quantization separate from ggml. For now just adding the k-quants there, but it would be better to also factor out the existing ggml quantizations. * Adding Q3_K and Q8_K (de)-quantization * Q3_K now working on CUDA and AVX2/scalar CUDA is not ideal - ~50% slower than Q4_0 for single token prediction, about the same in batch mode (perplexity). CPU single token is ~55 ms (on Ryzen 7950X). * Some improvement for Q3_K on CUDA It is now ~22.5 ms/token on my GPU, so ~30% slower than Q4_0. * Some more CUDA optimizations for Q3_K Single token is now 20.5 ms/token (~20% slower than Q4_0). Perplexity is on par with Q4_0. * Adding Q4_K - scalar, AVX2, CUDA Performance is the same or perhaps very slightly better than Q4_0 on the CPU. On the GPU, single token prediction is ~10% better than Q4_0, batch mode (perplexity is about the same). * Adding Q6_K - scalar, AVX2, CUDA Performance is ~40% lower compared to Q4_K on the CPU. This is to be expected, considering that we are memory bound on the CPU and the 6-bit model is ~44% larger than the 4-bit. On the GPU, single token prediction is ~6% lower than Q4_0, batch mode (perplexity) is even closer (but still slower). * Adding Q5_K - scalar, AVX2, CUDA Performance is ~20% lower compared to Q4_K on the CPU. This is to be expected, considering that we are memory bound on the CPU and the 5-bit model is ~22% larger than the 4-bit. On the GPU, single token prediction is about the same as Q4_0 for both, single token and batch prediction. * Per convention, all QX_K quantizations use Q5_K for output.weight * Adding quantization mixes * Quantization mixes: didn't quite get what I wanted in the last commit * Q4_K dot product for ARM_NEON * Q6_K dot product for ARM_NEON * Q5_K dot product for ARM_NEON * Adding Q3_K dot for ARM_NEON It is 22% slower than Q4_K, despite the smaller model size. On x86_64, where we are memory bound, the Q3_K model is quite a bit faster than Q4_K. * A very slightly faster ARM_NEON Q3_K dot * Adding Q2_K - just CUDA for now Token prediction is pretty good - about 15.5 ms on a RTX 4080. Perplexity is about the same as Q4_K. * Adding scalar and AVX2 Q2_K dot * Adding ARM_NEON Q2_K dot About the same performance as Q4_K. * A slightly faster ARM_NEON Q2_K dot Single token prediction is now ~36 ms on M2 Max. The code is much simpler too. * Fixed bug in Q2_K CUDA dot product kernel Stranegly enough, for the few prompts I tried with the 7B model the responses looked perfectly reasonable. Only realized something is not quite right when I tried the larger models and started getting nonse back. In any case, Q2_K single token evaluation time on an RTX 4080 in a Ryzen7950X box iusing CUDA and model fully loaded on the GPU are ~15.5 ms for 7B, ~25.4 ms for 13B, and ~55.8 ms for 30B. The max number of layers that fit in VRAM for The 65B is 32. With that, we get ~330 ms per token, which is not that much faster than just running on the CPU (~470 ms per token). * Don't print zeros/NaNs when no count histogram has been collected * A 10% faster CUDA vector dot kernel for Q3_K Q3_K is now running at ~18.5 ms / token on CUDA, so the gap to Q4_0 is only 10%. It seems memory acccess pattern is more important for performance than the amount of computation the kernel does. * A slightly daster Q4_K AVX2 dot product For perplexity, where we are less memory bound, time per pass drops by ~5%. Barely measurable difference for single token prediction. * A slightly faster ARM_NEON A4_K dot product * Minor * Fix quantization error test We cannot possibly be expecting rmse < 0.002 for 2- and 3-bit quantization variants. * Fix docker build I have been sloppy with vector reinterpret casts on ARM_NEON. It seems clang is very forgiving in that regard. * Added forgotten ggml.o dependence on k_quants.h to the Makefile * Had unintentionally committed the Makefile with -Ofast enabled * ggml : rename k_quants -> ggml-quants-k, use lowercase in code --------- Co-authored-by: Iwan Kawrakow <iwan.kawrakow@gmail.com> Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
2023-06-05 19:56:18 +00:00
x += QK_K;
}
}
ggml : quantization refactoring (#3833) * ggml : factor all quantization code in ggml-quants ggml-ci * ggml-quants : fix Zig and Swift builds + quantize tool ggml-ci * quantize : --pure option for disabling k-quant mixtures --------- Co-authored-by: cebtenzzre <cebtenzzre@gmail.com>
2023-10-29 16:32:28 +00:00
void dequantize_row_q8_K(const block_q8_K * restrict x, float * restrict y, int k) {
ggml : add SOTA 2,3,4,5,6 bit k-quantizations (#1684) * Starting to add k-quantization to ggml I think it is better to have quantization separate from ggml. For now just adding the k-quants there, but it would be better to also factor out the existing ggml quantizations. * Adding Q3_K and Q8_K (de)-quantization * Q3_K now working on CUDA and AVX2/scalar CUDA is not ideal - ~50% slower than Q4_0 for single token prediction, about the same in batch mode (perplexity). CPU single token is ~55 ms (on Ryzen 7950X). * Some improvement for Q3_K on CUDA It is now ~22.5 ms/token on my GPU, so ~30% slower than Q4_0. * Some more CUDA optimizations for Q3_K Single token is now 20.5 ms/token (~20% slower than Q4_0). Perplexity is on par with Q4_0. * Adding Q4_K - scalar, AVX2, CUDA Performance is the same or perhaps very slightly better than Q4_0 on the CPU. On the GPU, single token prediction is ~10% better than Q4_0, batch mode (perplexity is about the same). * Adding Q6_K - scalar, AVX2, CUDA Performance is ~40% lower compared to Q4_K on the CPU. This is to be expected, considering that we are memory bound on the CPU and the 6-bit model is ~44% larger than the 4-bit. On the GPU, single token prediction is ~6% lower than Q4_0, batch mode (perplexity) is even closer (but still slower). * Adding Q5_K - scalar, AVX2, CUDA Performance is ~20% lower compared to Q4_K on the CPU. This is to be expected, considering that we are memory bound on the CPU and the 5-bit model is ~22% larger than the 4-bit. On the GPU, single token prediction is about the same as Q4_0 for both, single token and batch prediction. * Per convention, all QX_K quantizations use Q5_K for output.weight * Adding quantization mixes * Quantization mixes: didn't quite get what I wanted in the last commit * Q4_K dot product for ARM_NEON * Q6_K dot product for ARM_NEON * Q5_K dot product for ARM_NEON * Adding Q3_K dot for ARM_NEON It is 22% slower than Q4_K, despite the smaller model size. On x86_64, where we are memory bound, the Q3_K model is quite a bit faster than Q4_K. * A very slightly faster ARM_NEON Q3_K dot * Adding Q2_K - just CUDA for now Token prediction is pretty good - about 15.5 ms on a RTX 4080. Perplexity is about the same as Q4_K. * Adding scalar and AVX2 Q2_K dot * Adding ARM_NEON Q2_K dot About the same performance as Q4_K. * A slightly faster ARM_NEON Q2_K dot Single token prediction is now ~36 ms on M2 Max. The code is much simpler too. * Fixed bug in Q2_K CUDA dot product kernel Stranegly enough, for the few prompts I tried with the 7B model the responses looked perfectly reasonable. Only realized something is not quite right when I tried the larger models and started getting nonse back. In any case, Q2_K single token evaluation time on an RTX 4080 in a Ryzen7950X box iusing CUDA and model fully loaded on the GPU are ~15.5 ms for 7B, ~25.4 ms for 13B, and ~55.8 ms for 30B. The max number of layers that fit in VRAM for The 65B is 32. With that, we get ~330 ms per token, which is not that much faster than just running on the CPU (~470 ms per token). * Don't print zeros/NaNs when no count histogram has been collected * A 10% faster CUDA vector dot kernel for Q3_K Q3_K is now running at ~18.5 ms / token on CUDA, so the gap to Q4_0 is only 10%. It seems memory acccess pattern is more important for performance than the amount of computation the kernel does. * A slightly daster Q4_K AVX2 dot product For perplexity, where we are less memory bound, time per pass drops by ~5%. Barely measurable difference for single token prediction. * A slightly faster ARM_NEON A4_K dot product * Minor * Fix quantization error test We cannot possibly be expecting rmse < 0.002 for 2- and 3-bit quantization variants. * Fix docker build I have been sloppy with vector reinterpret casts on ARM_NEON. It seems clang is very forgiving in that regard. * Added forgotten ggml.o dependence on k_quants.h to the Makefile * Had unintentionally committed the Makefile with -Ofast enabled * ggml : rename k_quants -> ggml-quants-k, use lowercase in code --------- Co-authored-by: Iwan Kawrakow <iwan.kawrakow@gmail.com> Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
2023-06-05 19:56:18 +00:00
assert(k % QK_K == 0);
const int nb = k / QK_K;
for (int i = 0; i < nb; i++) {
ggml : quantization refactoring (#3833) * ggml : factor all quantization code in ggml-quants ggml-ci * ggml-quants : fix Zig and Swift builds + quantize tool ggml-ci * quantize : --pure option for disabling k-quant mixtures --------- Co-authored-by: cebtenzzre <cebtenzzre@gmail.com>
2023-10-29 16:32:28 +00:00
for (int j = 0; j < QK_K; ++j) {
*y++ = x[i].d * x[i].qs[j];
}
}
}
ggml : add SOTA 2,3,4,5,6 bit k-quantizations (#1684) * Starting to add k-quantization to ggml I think it is better to have quantization separate from ggml. For now just adding the k-quants there, but it would be better to also factor out the existing ggml quantizations. * Adding Q3_K and Q8_K (de)-quantization * Q3_K now working on CUDA and AVX2/scalar CUDA is not ideal - ~50% slower than Q4_0 for single token prediction, about the same in batch mode (perplexity). CPU single token is ~55 ms (on Ryzen 7950X). * Some improvement for Q3_K on CUDA It is now ~22.5 ms/token on my GPU, so ~30% slower than Q4_0. * Some more CUDA optimizations for Q3_K Single token is now 20.5 ms/token (~20% slower than Q4_0). Perplexity is on par with Q4_0. * Adding Q4_K - scalar, AVX2, CUDA Performance is the same or perhaps very slightly better than Q4_0 on the CPU. On the GPU, single token prediction is ~10% better than Q4_0, batch mode (perplexity is about the same). * Adding Q6_K - scalar, AVX2, CUDA Performance is ~40% lower compared to Q4_K on the CPU. This is to be expected, considering that we are memory bound on the CPU and the 6-bit model is ~44% larger than the 4-bit. On the GPU, single token prediction is ~6% lower than Q4_0, batch mode (perplexity) is even closer (but still slower). * Adding Q5_K - scalar, AVX2, CUDA Performance is ~20% lower compared to Q4_K on the CPU. This is to be expected, considering that we are memory bound on the CPU and the 5-bit model is ~22% larger than the 4-bit. On the GPU, single token prediction is about the same as Q4_0 for both, single token and batch prediction. * Per convention, all QX_K quantizations use Q5_K for output.weight * Adding quantization mixes * Quantization mixes: didn't quite get what I wanted in the last commit * Q4_K dot product for ARM_NEON * Q6_K dot product for ARM_NEON * Q5_K dot product for ARM_NEON * Adding Q3_K dot for ARM_NEON It is 22% slower than Q4_K, despite the smaller model size. On x86_64, where we are memory bound, the Q3_K model is quite a bit faster than Q4_K. * A very slightly faster ARM_NEON Q3_K dot * Adding Q2_K - just CUDA for now Token prediction is pretty good - about 15.5 ms on a RTX 4080. Perplexity is about the same as Q4_K. * Adding scalar and AVX2 Q2_K dot * Adding ARM_NEON Q2_K dot About the same performance as Q4_K. * A slightly faster ARM_NEON Q2_K dot Single token prediction is now ~36 ms on M2 Max. The code is much simpler too. * Fixed bug in Q2_K CUDA dot product kernel Stranegly enough, for the few prompts I tried with the 7B model the responses looked perfectly reasonable. Only realized something is not quite right when I tried the larger models and started getting nonse back. In any case, Q2_K single token evaluation time on an RTX 4080 in a Ryzen7950X box iusing CUDA and model fully loaded on the GPU are ~15.5 ms for 7B, ~25.4 ms for 13B, and ~55.8 ms for 30B. The max number of layers that fit in VRAM for The 65B is 32. With that, we get ~330 ms per token, which is not that much faster than just running on the CPU (~470 ms per token). * Don't print zeros/NaNs when no count histogram has been collected * A 10% faster CUDA vector dot kernel for Q3_K Q3_K is now running at ~18.5 ms / token on CUDA, so the gap to Q4_0 is only 10%. It seems memory acccess pattern is more important for performance than the amount of computation the kernel does. * A slightly daster Q4_K AVX2 dot product For perplexity, where we are less memory bound, time per pass drops by ~5%. Barely measurable difference for single token prediction. * A slightly faster ARM_NEON A4_K dot product * Minor * Fix quantization error test We cannot possibly be expecting rmse < 0.002 for 2- and 3-bit quantization variants. * Fix docker build I have been sloppy with vector reinterpret casts on ARM_NEON. It seems clang is very forgiving in that regard. * Added forgotten ggml.o dependence on k_quants.h to the Makefile * Had unintentionally committed the Makefile with -Ofast enabled * ggml : rename k_quants -> ggml-quants-k, use lowercase in code --------- Co-authored-by: Iwan Kawrakow <iwan.kawrakow@gmail.com> Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
2023-06-05 19:56:18 +00:00
ggml : quantization refactoring (#3833) * ggml : factor all quantization code in ggml-quants ggml-ci * ggml-quants : fix Zig and Swift builds + quantize tool ggml-ci * quantize : --pure option for disabling k-quant mixtures --------- Co-authored-by: cebtenzzre <cebtenzzre@gmail.com>
2023-10-29 16:32:28 +00:00
void quantize_row_q8_K(const float * restrict x, void * restrict y, int k) {
quantize_row_q8_K_reference(x, y, k);
}
ggml : add SOTA 2,3,4,5,6 bit k-quantizations (#1684) * Starting to add k-quantization to ggml I think it is better to have quantization separate from ggml. For now just adding the k-quants there, but it would be better to also factor out the existing ggml quantizations. * Adding Q3_K and Q8_K (de)-quantization * Q3_K now working on CUDA and AVX2/scalar CUDA is not ideal - ~50% slower than Q4_0 for single token prediction, about the same in batch mode (perplexity). CPU single token is ~55 ms (on Ryzen 7950X). * Some improvement for Q3_K on CUDA It is now ~22.5 ms/token on my GPU, so ~30% slower than Q4_0. * Some more CUDA optimizations for Q3_K Single token is now 20.5 ms/token (~20% slower than Q4_0). Perplexity is on par with Q4_0. * Adding Q4_K - scalar, AVX2, CUDA Performance is the same or perhaps very slightly better than Q4_0 on the CPU. On the GPU, single token prediction is ~10% better than Q4_0, batch mode (perplexity is about the same). * Adding Q6_K - scalar, AVX2, CUDA Performance is ~40% lower compared to Q4_K on the CPU. This is to be expected, considering that we are memory bound on the CPU and the 6-bit model is ~44% larger than the 4-bit. On the GPU, single token prediction is ~6% lower than Q4_0, batch mode (perplexity) is even closer (but still slower). * Adding Q5_K - scalar, AVX2, CUDA Performance is ~20% lower compared to Q4_K on the CPU. This is to be expected, considering that we are memory bound on the CPU and the 5-bit model is ~22% larger than the 4-bit. On the GPU, single token prediction is about the same as Q4_0 for both, single token and batch prediction. * Per convention, all QX_K quantizations use Q5_K for output.weight * Adding quantization mixes * Quantization mixes: didn't quite get what I wanted in the last commit * Q4_K dot product for ARM_NEON * Q6_K dot product for ARM_NEON * Q5_K dot product for ARM_NEON * Adding Q3_K dot for ARM_NEON It is 22% slower than Q4_K, despite the smaller model size. On x86_64, where we are memory bound, the Q3_K model is quite a bit faster than Q4_K. * A very slightly faster ARM_NEON Q3_K dot * Adding Q2_K - just CUDA for now Token prediction is pretty good - about 15.5 ms on a RTX 4080. Perplexity is about the same as Q4_K. * Adding scalar and AVX2 Q2_K dot * Adding ARM_NEON Q2_K dot About the same performance as Q4_K. * A slightly faster ARM_NEON Q2_K dot Single token prediction is now ~36 ms on M2 Max. The code is much simpler too. * Fixed bug in Q2_K CUDA dot product kernel Stranegly enough, for the few prompts I tried with the 7B model the responses looked perfectly reasonable. Only realized something is not quite right when I tried the larger models and started getting nonse back. In any case, Q2_K single token evaluation time on an RTX 4080 in a Ryzen7950X box iusing CUDA and model fully loaded on the GPU are ~15.5 ms for 7B, ~25.4 ms for 13B, and ~55.8 ms for 30B. The max number of layers that fit in VRAM for The 65B is 32. With that, we get ~330 ms per token, which is not that much faster than just running on the CPU (~470 ms per token). * Don't print zeros/NaNs when no count histogram has been collected * A 10% faster CUDA vector dot kernel for Q3_K Q3_K is now running at ~18.5 ms / token on CUDA, so the gap to Q4_0 is only 10%. It seems memory acccess pattern is more important for performance than the amount of computation the kernel does. * A slightly daster Q4_K AVX2 dot product For perplexity, where we are less memory bound, time per pass drops by ~5%. Barely measurable difference for single token prediction. * A slightly faster ARM_NEON A4_K dot product * Minor * Fix quantization error test We cannot possibly be expecting rmse < 0.002 for 2- and 3-bit quantization variants. * Fix docker build I have been sloppy with vector reinterpret casts on ARM_NEON. It seems clang is very forgiving in that regard. * Added forgotten ggml.o dependence on k_quants.h to the Makefile * Had unintentionally committed the Makefile with -Ofast enabled * ggml : rename k_quants -> ggml-quants-k, use lowercase in code --------- Co-authored-by: Iwan Kawrakow <iwan.kawrakow@gmail.com> Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
2023-06-05 19:56:18 +00:00
ggml : quantization refactoring (#3833) * ggml : factor all quantization code in ggml-quants ggml-ci * ggml-quants : fix Zig and Swift builds + quantize tool ggml-ci * quantize : --pure option for disabling k-quant mixtures --------- Co-authored-by: cebtenzzre <cebtenzzre@gmail.com>
2023-10-29 16:32:28 +00:00
//===================================== Dot ptoducts =================================
ggml : add SOTA 2,3,4,5,6 bit k-quantizations (#1684) * Starting to add k-quantization to ggml I think it is better to have quantization separate from ggml. For now just adding the k-quants there, but it would be better to also factor out the existing ggml quantizations. * Adding Q3_K and Q8_K (de)-quantization * Q3_K now working on CUDA and AVX2/scalar CUDA is not ideal - ~50% slower than Q4_0 for single token prediction, about the same in batch mode (perplexity). CPU single token is ~55 ms (on Ryzen 7950X). * Some improvement for Q3_K on CUDA It is now ~22.5 ms/token on my GPU, so ~30% slower than Q4_0. * Some more CUDA optimizations for Q3_K Single token is now 20.5 ms/token (~20% slower than Q4_0). Perplexity is on par with Q4_0. * Adding Q4_K - scalar, AVX2, CUDA Performance is the same or perhaps very slightly better than Q4_0 on the CPU. On the GPU, single token prediction is ~10% better than Q4_0, batch mode (perplexity is about the same). * Adding Q6_K - scalar, AVX2, CUDA Performance is ~40% lower compared to Q4_K on the CPU. This is to be expected, considering that we are memory bound on the CPU and the 6-bit model is ~44% larger than the 4-bit. On the GPU, single token prediction is ~6% lower than Q4_0, batch mode (perplexity) is even closer (but still slower). * Adding Q5_K - scalar, AVX2, CUDA Performance is ~20% lower compared to Q4_K on the CPU. This is to be expected, considering that we are memory bound on the CPU and the 5-bit model is ~22% larger than the 4-bit. On the GPU, single token prediction is about the same as Q4_0 for both, single token and batch prediction. * Per convention, all QX_K quantizations use Q5_K for output.weight * Adding quantization mixes * Quantization mixes: didn't quite get what I wanted in the last commit * Q4_K dot product for ARM_NEON * Q6_K dot product for ARM_NEON * Q5_K dot product for ARM_NEON * Adding Q3_K dot for ARM_NEON It is 22% slower than Q4_K, despite the smaller model size. On x86_64, where we are memory bound, the Q3_K model is quite a bit faster than Q4_K. * A very slightly faster ARM_NEON Q3_K dot * Adding Q2_K - just CUDA for now Token prediction is pretty good - about 15.5 ms on a RTX 4080. Perplexity is about the same as Q4_K. * Adding scalar and AVX2 Q2_K dot * Adding ARM_NEON Q2_K dot About the same performance as Q4_K. * A slightly faster ARM_NEON Q2_K dot Single token prediction is now ~36 ms on M2 Max. The code is much simpler too. * Fixed bug in Q2_K CUDA dot product kernel Stranegly enough, for the few prompts I tried with the 7B model the responses looked perfectly reasonable. Only realized something is not quite right when I tried the larger models and started getting nonse back. In any case, Q2_K single token evaluation time on an RTX 4080 in a Ryzen7950X box iusing CUDA and model fully loaded on the GPU are ~15.5 ms for 7B, ~25.4 ms for 13B, and ~55.8 ms for 30B. The max number of layers that fit in VRAM for The 65B is 32. With that, we get ~330 ms per token, which is not that much faster than just running on the CPU (~470 ms per token). * Don't print zeros/NaNs when no count histogram has been collected * A 10% faster CUDA vector dot kernel for Q3_K Q3_K is now running at ~18.5 ms / token on CUDA, so the gap to Q4_0 is only 10%. It seems memory acccess pattern is more important for performance than the amount of computation the kernel does. * A slightly daster Q4_K AVX2 dot product For perplexity, where we are less memory bound, time per pass drops by ~5%. Barely measurable difference for single token prediction. * A slightly faster ARM_NEON A4_K dot product * Minor * Fix quantization error test We cannot possibly be expecting rmse < 0.002 for 2- and 3-bit quantization variants. * Fix docker build I have been sloppy with vector reinterpret casts on ARM_NEON. It seems clang is very forgiving in that regard. * Added forgotten ggml.o dependence on k_quants.h to the Makefile * Had unintentionally committed the Makefile with -Ofast enabled * ggml : rename k_quants -> ggml-quants-k, use lowercase in code --------- Co-authored-by: Iwan Kawrakow <iwan.kawrakow@gmail.com> Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
2023-06-05 19:56:18 +00:00
ggml : quantization refactoring (#3833) * ggml : factor all quantization code in ggml-quants ggml-ci * ggml-quants : fix Zig and Swift builds + quantize tool ggml-ci * quantize : --pure option for disabling k-quant mixtures --------- Co-authored-by: cebtenzzre <cebtenzzre@gmail.com>
2023-10-29 16:32:28 +00:00
//
// Helper functions
//
#if __AVX__ || __AVX2__ || __AVX512F__
ggml : add SOTA 2,3,4,5,6 bit k-quantizations (#1684) * Starting to add k-quantization to ggml I think it is better to have quantization separate from ggml. For now just adding the k-quants there, but it would be better to also factor out the existing ggml quantizations. * Adding Q3_K and Q8_K (de)-quantization * Q3_K now working on CUDA and AVX2/scalar CUDA is not ideal - ~50% slower than Q4_0 for single token prediction, about the same in batch mode (perplexity). CPU single token is ~55 ms (on Ryzen 7950X). * Some improvement for Q3_K on CUDA It is now ~22.5 ms/token on my GPU, so ~30% slower than Q4_0. * Some more CUDA optimizations for Q3_K Single token is now 20.5 ms/token (~20% slower than Q4_0). Perplexity is on par with Q4_0. * Adding Q4_K - scalar, AVX2, CUDA Performance is the same or perhaps very slightly better than Q4_0 on the CPU. On the GPU, single token prediction is ~10% better than Q4_0, batch mode (perplexity is about the same). * Adding Q6_K - scalar, AVX2, CUDA Performance is ~40% lower compared to Q4_K on the CPU. This is to be expected, considering that we are memory bound on the CPU and the 6-bit model is ~44% larger than the 4-bit. On the GPU, single token prediction is ~6% lower than Q4_0, batch mode (perplexity) is even closer (but still slower). * Adding Q5_K - scalar, AVX2, CUDA Performance is ~20% lower compared to Q4_K on the CPU. This is to be expected, considering that we are memory bound on the CPU and the 5-bit model is ~22% larger than the 4-bit. On the GPU, single token prediction is about the same as Q4_0 for both, single token and batch prediction. * Per convention, all QX_K quantizations use Q5_K for output.weight * Adding quantization mixes * Quantization mixes: didn't quite get what I wanted in the last commit * Q4_K dot product for ARM_NEON * Q6_K dot product for ARM_NEON * Q5_K dot product for ARM_NEON * Adding Q3_K dot for ARM_NEON It is 22% slower than Q4_K, despite the smaller model size. On x86_64, where we are memory bound, the Q3_K model is quite a bit faster than Q4_K. * A very slightly faster ARM_NEON Q3_K dot * Adding Q2_K - just CUDA for now Token prediction is pretty good - about 15.5 ms on a RTX 4080. Perplexity is about the same as Q4_K. * Adding scalar and AVX2 Q2_K dot * Adding ARM_NEON Q2_K dot About the same performance as Q4_K. * A slightly faster ARM_NEON Q2_K dot Single token prediction is now ~36 ms on M2 Max. The code is much simpler too. * Fixed bug in Q2_K CUDA dot product kernel Stranegly enough, for the few prompts I tried with the 7B model the responses looked perfectly reasonable. Only realized something is not quite right when I tried the larger models and started getting nonse back. In any case, Q2_K single token evaluation time on an RTX 4080 in a Ryzen7950X box iusing CUDA and model fully loaded on the GPU are ~15.5 ms for 7B, ~25.4 ms for 13B, and ~55.8 ms for 30B. The max number of layers that fit in VRAM for The 65B is 32. With that, we get ~330 ms per token, which is not that much faster than just running on the CPU (~470 ms per token). * Don't print zeros/NaNs when no count histogram has been collected * A 10% faster CUDA vector dot kernel for Q3_K Q3_K is now running at ~18.5 ms / token on CUDA, so the gap to Q4_0 is only 10%. It seems memory acccess pattern is more important for performance than the amount of computation the kernel does. * A slightly daster Q4_K AVX2 dot product For perplexity, where we are less memory bound, time per pass drops by ~5%. Barely measurable difference for single token prediction. * A slightly faster ARM_NEON A4_K dot product * Minor * Fix quantization error test We cannot possibly be expecting rmse < 0.002 for 2- and 3-bit quantization variants. * Fix docker build I have been sloppy with vector reinterpret casts on ARM_NEON. It seems clang is very forgiving in that regard. * Added forgotten ggml.o dependence on k_quants.h to the Makefile * Had unintentionally committed the Makefile with -Ofast enabled * ggml : rename k_quants -> ggml-quants-k, use lowercase in code --------- Co-authored-by: Iwan Kawrakow <iwan.kawrakow@gmail.com> Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
2023-06-05 19:56:18 +00:00
ggml : quantization refactoring (#3833) * ggml : factor all quantization code in ggml-quants ggml-ci * ggml-quants : fix Zig and Swift builds + quantize tool ggml-ci * quantize : --pure option for disabling k-quant mixtures --------- Co-authored-by: cebtenzzre <cebtenzzre@gmail.com>
2023-10-29 16:32:28 +00:00
// shuffles to pick the required scales in dot products
static inline __m256i get_scale_shuffle_q3k(int i) {
static const uint8_t k_shuffle[128] = {
0, 1, 0, 1, 0, 1, 0, 1, 0, 1, 0, 1, 0, 1, 0, 1, 2, 3, 2, 3, 2, 3, 2, 3, 2, 3, 2, 3, 2, 3, 2, 3,
4, 5, 4, 5, 4, 5, 4, 5, 4, 5, 4, 5, 4, 5, 4, 5, 6, 7, 6, 7, 6, 7, 6, 7, 6, 7, 6, 7, 6, 7, 6, 7,
8, 9, 8, 9, 8, 9, 8, 9, 8, 9, 8, 9, 8, 9, 8, 9, 10,11,10,11,10,11,10,11,10,11,10,11,10,11,10,11,
12,13,12,13,12,13,12,13,12,13,12,13,12,13,12,13, 14,15,14,15,14,15,14,15,14,15,14,15,14,15,14,15,
};
return _mm256_loadu_si256((const __m256i*)k_shuffle + i);
}
static inline __m256i get_scale_shuffle_k4(int i) {
static const uint8_t k_shuffle[256] = {
0, 1, 0, 1, 0, 1, 0, 1, 0, 1, 0, 1, 0, 1, 0, 1, 0, 1, 0, 1, 0, 1, 0, 1, 0, 1, 0, 1, 0, 1, 0, 1,
2, 3, 2, 3, 2, 3, 2, 3, 2, 3, 2, 3, 2, 3, 2, 3, 2, 3, 2, 3, 2, 3, 2, 3, 2, 3, 2, 3, 2, 3, 2, 3,
4, 5, 4, 5, 4, 5, 4, 5, 4, 5, 4, 5, 4, 5, 4, 5, 4, 5, 4, 5, 4, 5, 4, 5, 4, 5, 4, 5, 4, 5, 4, 5,
6, 7, 6, 7, 6, 7, 6, 7, 6, 7, 6, 7, 6, 7, 6, 7, 6, 7, 6, 7, 6, 7, 6, 7, 6, 7, 6, 7, 6, 7, 6, 7,
8, 9, 8, 9, 8, 9, 8, 9, 8, 9, 8, 9, 8, 9, 8, 9, 8, 9, 8, 9, 8, 9, 8, 9, 8, 9, 8, 9, 8, 9, 8, 9,
10,11,10,11,10,11,10,11,10,11,10,11,10,11,10,11,10,11,10,11,10,11,10,11,10,11,10,11,10,11,10,11,
12,13,12,13,12,13,12,13,12,13,12,13,12,13,12,13,12,13,12,13,12,13,12,13,12,13,12,13,12,13,12,13,
14,15,14,15,14,15,14,15,14,15,14,15,14,15,14,15,14,15,14,15,14,15,14,15,14,15,14,15,14,15,14,15
};
return _mm256_loadu_si256((const __m256i*)k_shuffle + i);
}
static inline __m128i get_scale_shuffle(int i) {
static const uint8_t k_shuffle[128] = {
0, 0, 0, 0, 0, 0, 0, 0, 1, 1, 1, 1, 1, 1, 1, 1,
2, 2, 2, 2, 2, 2, 2, 2, 3, 3, 3, 3, 3, 3, 3, 3,
4, 4, 4, 4, 4, 4, 4, 4, 5, 5, 5, 5, 5, 5, 5, 5,
6, 6, 6, 6, 6, 6, 6, 6, 7, 7, 7, 7, 7, 7, 7, 7,
8, 8, 8, 8, 8, 8, 8, 8, 9, 9, 9, 9, 9, 9, 9, 9,
10,10,10,10,10,10,10,10, 11,11,11,11,11,11,11,11,
12,12,12,12,12,12,12,12, 13,13,13,13,13,13,13,13,
14,14,14,14,14,14,14,14, 15,15,15,15,15,15,15,15
};
return _mm_loadu_si128((const __m128i*)k_shuffle + i);
}
#endif
ggml : add SOTA 2,3,4,5,6 bit k-quantizations (#1684) * Starting to add k-quantization to ggml I think it is better to have quantization separate from ggml. For now just adding the k-quants there, but it would be better to also factor out the existing ggml quantizations. * Adding Q3_K and Q8_K (de)-quantization * Q3_K now working on CUDA and AVX2/scalar CUDA is not ideal - ~50% slower than Q4_0 for single token prediction, about the same in batch mode (perplexity). CPU single token is ~55 ms (on Ryzen 7950X). * Some improvement for Q3_K on CUDA It is now ~22.5 ms/token on my GPU, so ~30% slower than Q4_0. * Some more CUDA optimizations for Q3_K Single token is now 20.5 ms/token (~20% slower than Q4_0). Perplexity is on par with Q4_0. * Adding Q4_K - scalar, AVX2, CUDA Performance is the same or perhaps very slightly better than Q4_0 on the CPU. On the GPU, single token prediction is ~10% better than Q4_0, batch mode (perplexity is about the same). * Adding Q6_K - scalar, AVX2, CUDA Performance is ~40% lower compared to Q4_K on the CPU. This is to be expected, considering that we are memory bound on the CPU and the 6-bit model is ~44% larger than the 4-bit. On the GPU, single token prediction is ~6% lower than Q4_0, batch mode (perplexity) is even closer (but still slower). * Adding Q5_K - scalar, AVX2, CUDA Performance is ~20% lower compared to Q4_K on the CPU. This is to be expected, considering that we are memory bound on the CPU and the 5-bit model is ~22% larger than the 4-bit. On the GPU, single token prediction is about the same as Q4_0 for both, single token and batch prediction. * Per convention, all QX_K quantizations use Q5_K for output.weight * Adding quantization mixes * Quantization mixes: didn't quite get what I wanted in the last commit * Q4_K dot product for ARM_NEON * Q6_K dot product for ARM_NEON * Q5_K dot product for ARM_NEON * Adding Q3_K dot for ARM_NEON It is 22% slower than Q4_K, despite the smaller model size. On x86_64, where we are memory bound, the Q3_K model is quite a bit faster than Q4_K. * A very slightly faster ARM_NEON Q3_K dot * Adding Q2_K - just CUDA for now Token prediction is pretty good - about 15.5 ms on a RTX 4080. Perplexity is about the same as Q4_K. * Adding scalar and AVX2 Q2_K dot * Adding ARM_NEON Q2_K dot About the same performance as Q4_K. * A slightly faster ARM_NEON Q2_K dot Single token prediction is now ~36 ms on M2 Max. The code is much simpler too. * Fixed bug in Q2_K CUDA dot product kernel Stranegly enough, for the few prompts I tried with the 7B model the responses looked perfectly reasonable. Only realized something is not quite right when I tried the larger models and started getting nonse back. In any case, Q2_K single token evaluation time on an RTX 4080 in a Ryzen7950X box iusing CUDA and model fully loaded on the GPU are ~15.5 ms for 7B, ~25.4 ms for 13B, and ~55.8 ms for 30B. The max number of layers that fit in VRAM for The 65B is 32. With that, we get ~330 ms per token, which is not that much faster than just running on the CPU (~470 ms per token). * Don't print zeros/NaNs when no count histogram has been collected * A 10% faster CUDA vector dot kernel for Q3_K Q3_K is now running at ~18.5 ms / token on CUDA, so the gap to Q4_0 is only 10%. It seems memory acccess pattern is more important for performance than the amount of computation the kernel does. * A slightly daster Q4_K AVX2 dot product For perplexity, where we are less memory bound, time per pass drops by ~5%. Barely measurable difference for single token prediction. * A slightly faster ARM_NEON A4_K dot product * Minor * Fix quantization error test We cannot possibly be expecting rmse < 0.002 for 2- and 3-bit quantization variants. * Fix docker build I have been sloppy with vector reinterpret casts on ARM_NEON. It seems clang is very forgiving in that regard. * Added forgotten ggml.o dependence on k_quants.h to the Makefile * Had unintentionally committed the Makefile with -Ofast enabled * ggml : rename k_quants -> ggml-quants-k, use lowercase in code --------- Co-authored-by: Iwan Kawrakow <iwan.kawrakow@gmail.com> Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
2023-06-05 19:56:18 +00:00
ggml : quantization refactoring (#3833) * ggml : factor all quantization code in ggml-quants ggml-ci * ggml-quants : fix Zig and Swift builds + quantize tool ggml-ci * quantize : --pure option for disabling k-quant mixtures --------- Co-authored-by: cebtenzzre <cebtenzzre@gmail.com>
2023-10-29 16:32:28 +00:00
void ggml_vec_dot_q4_0_q8_0(int n, float * restrict s, const void * restrict vx, const void * restrict vy) {
const int qk = QK8_0;
const int nb = n / qk;
ggml : add SOTA 2,3,4,5,6 bit k-quantizations (#1684) * Starting to add k-quantization to ggml I think it is better to have quantization separate from ggml. For now just adding the k-quants there, but it would be better to also factor out the existing ggml quantizations. * Adding Q3_K and Q8_K (de)-quantization * Q3_K now working on CUDA and AVX2/scalar CUDA is not ideal - ~50% slower than Q4_0 for single token prediction, about the same in batch mode (perplexity). CPU single token is ~55 ms (on Ryzen 7950X). * Some improvement for Q3_K on CUDA It is now ~22.5 ms/token on my GPU, so ~30% slower than Q4_0. * Some more CUDA optimizations for Q3_K Single token is now 20.5 ms/token (~20% slower than Q4_0). Perplexity is on par with Q4_0. * Adding Q4_K - scalar, AVX2, CUDA Performance is the same or perhaps very slightly better than Q4_0 on the CPU. On the GPU, single token prediction is ~10% better than Q4_0, batch mode (perplexity is about the same). * Adding Q6_K - scalar, AVX2, CUDA Performance is ~40% lower compared to Q4_K on the CPU. This is to be expected, considering that we are memory bound on the CPU and the 6-bit model is ~44% larger than the 4-bit. On the GPU, single token prediction is ~6% lower than Q4_0, batch mode (perplexity) is even closer (but still slower). * Adding Q5_K - scalar, AVX2, CUDA Performance is ~20% lower compared to Q4_K on the CPU. This is to be expected, considering that we are memory bound on the CPU and the 5-bit model is ~22% larger than the 4-bit. On the GPU, single token prediction is about the same as Q4_0 for both, single token and batch prediction. * Per convention, all QX_K quantizations use Q5_K for output.weight * Adding quantization mixes * Quantization mixes: didn't quite get what I wanted in the last commit * Q4_K dot product for ARM_NEON * Q6_K dot product for ARM_NEON * Q5_K dot product for ARM_NEON * Adding Q3_K dot for ARM_NEON It is 22% slower than Q4_K, despite the smaller model size. On x86_64, where we are memory bound, the Q3_K model is quite a bit faster than Q4_K. * A very slightly faster ARM_NEON Q3_K dot * Adding Q2_K - just CUDA for now Token prediction is pretty good - about 15.5 ms on a RTX 4080. Perplexity is about the same as Q4_K. * Adding scalar and AVX2 Q2_K dot * Adding ARM_NEON Q2_K dot About the same performance as Q4_K. * A slightly faster ARM_NEON Q2_K dot Single token prediction is now ~36 ms on M2 Max. The code is much simpler too. * Fixed bug in Q2_K CUDA dot product kernel Stranegly enough, for the few prompts I tried with the 7B model the responses looked perfectly reasonable. Only realized something is not quite right when I tried the larger models and started getting nonse back. In any case, Q2_K single token evaluation time on an RTX 4080 in a Ryzen7950X box iusing CUDA and model fully loaded on the GPU are ~15.5 ms for 7B, ~25.4 ms for 13B, and ~55.8 ms for 30B. The max number of layers that fit in VRAM for The 65B is 32. With that, we get ~330 ms per token, which is not that much faster than just running on the CPU (~470 ms per token). * Don't print zeros/NaNs when no count histogram has been collected * A 10% faster CUDA vector dot kernel for Q3_K Q3_K is now running at ~18.5 ms / token on CUDA, so the gap to Q4_0 is only 10%. It seems memory acccess pattern is more important for performance than the amount of computation the kernel does. * A slightly daster Q4_K AVX2 dot product For perplexity, where we are less memory bound, time per pass drops by ~5%. Barely measurable difference for single token prediction. * A slightly faster ARM_NEON A4_K dot product * Minor * Fix quantization error test We cannot possibly be expecting rmse < 0.002 for 2- and 3-bit quantization variants. * Fix docker build I have been sloppy with vector reinterpret casts on ARM_NEON. It seems clang is very forgiving in that regard. * Added forgotten ggml.o dependence on k_quants.h to the Makefile * Had unintentionally committed the Makefile with -Ofast enabled * ggml : rename k_quants -> ggml-quants-k, use lowercase in code --------- Co-authored-by: Iwan Kawrakow <iwan.kawrakow@gmail.com> Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
2023-06-05 19:56:18 +00:00
ggml : quantization refactoring (#3833) * ggml : factor all quantization code in ggml-quants ggml-ci * ggml-quants : fix Zig and Swift builds + quantize tool ggml-ci * quantize : --pure option for disabling k-quant mixtures --------- Co-authored-by: cebtenzzre <cebtenzzre@gmail.com>
2023-10-29 16:32:28 +00:00
assert(n % qk == 0);
const block_q4_0 * restrict x = vx;
const block_q8_0 * restrict y = vy;
#if defined(__ARM_NEON)
float32x4_t sumv0 = vdupq_n_f32(0.0f);
float32x4_t sumv1 = vdupq_n_f32(0.0f);
assert(nb % 2 == 0); // TODO: handle odd nb
for (int i = 0; i < nb; i += 2) {
const block_q4_0 * restrict x0 = &x[i + 0];
const block_q4_0 * restrict x1 = &x[i + 1];
const block_q8_0 * restrict y0 = &y[i + 0];
const block_q8_0 * restrict y1 = &y[i + 1];
const uint8x16_t m4b = vdupq_n_u8(0x0F);
const int8x16_t s8b = vdupq_n_s8(0x8);
const uint8x16_t v0_0 = vld1q_u8(x0->qs);
const uint8x16_t v0_1 = vld1q_u8(x1->qs);
// 4-bit -> 8-bit
const int8x16_t v0_0l = vreinterpretq_s8_u8(vandq_u8 (v0_0, m4b));
const int8x16_t v0_0h = vreinterpretq_s8_u8(vshrq_n_u8(v0_0, 4));
const int8x16_t v0_1l = vreinterpretq_s8_u8(vandq_u8 (v0_1, m4b));
const int8x16_t v0_1h = vreinterpretq_s8_u8(vshrq_n_u8(v0_1, 4));
// sub 8
const int8x16_t v0_0ls = vsubq_s8(v0_0l, s8b);
const int8x16_t v0_0hs = vsubq_s8(v0_0h, s8b);
const int8x16_t v0_1ls = vsubq_s8(v0_1l, s8b);
const int8x16_t v0_1hs = vsubq_s8(v0_1h, s8b);
// load y
const int8x16_t v1_0l = vld1q_s8(y0->qs);
const int8x16_t v1_0h = vld1q_s8(y0->qs + 16);
const int8x16_t v1_1l = vld1q_s8(y1->qs);
const int8x16_t v1_1h = vld1q_s8(y1->qs + 16);
// dot product into int32x4_t
ggml : add ggml_vdotq_s32 alias (#4715) ggml-ci
2023-12-31 09:43:31 +00:00
const int32x4_t p_0 = ggml_vdotq_s32(ggml_vdotq_s32(vdupq_n_s32(0), v0_0ls, v1_0l), v0_0hs, v1_0h);
const int32x4_t p_1 = ggml_vdotq_s32(ggml_vdotq_s32(vdupq_n_s32(0), v0_1ls, v1_1l), v0_1hs, v1_1h);
ggml : quantization refactoring (#3833) * ggml : factor all quantization code in ggml-quants ggml-ci * ggml-quants : fix Zig and Swift builds + quantize tool ggml-ci * quantize : --pure option for disabling k-quant mixtures --------- Co-authored-by: cebtenzzre <cebtenzzre@gmail.com>
2023-10-29 16:32:28 +00:00
ggml : move FP16 <-> FP32 code to ggml-impl.h (#3861) * ggml : move FP16 <-> FP32 stuff to ggml-impl.h ggml-ci * tests : fix ARM build * ggml : explicitly initialize deprecated type traits * ggml : add math.h to ggml-impl.h * ggml : remove duplicate static assert macros * ggml : prefix lookup tables with ggml_ ggml-ci * ggml-impl : move extern "C" to start of file
2023-10-30 17:19:15 +00:00
sumv0 = vmlaq_n_f32(sumv0, vcvtq_f32_s32(p_0), GGML_FP16_TO_FP32(x0->d)*GGML_FP16_TO_FP32(y0->d));
sumv1 = vmlaq_n_f32(sumv1, vcvtq_f32_s32(p_1), GGML_FP16_TO_FP32(x1->d)*GGML_FP16_TO_FP32(y1->d));
ggml : add SOTA 2,3,4,5,6 bit k-quantizations (#1684) * Starting to add k-quantization to ggml I think it is better to have quantization separate from ggml. For now just adding the k-quants there, but it would be better to also factor out the existing ggml quantizations. * Adding Q3_K and Q8_K (de)-quantization * Q3_K now working on CUDA and AVX2/scalar CUDA is not ideal - ~50% slower than Q4_0 for single token prediction, about the same in batch mode (perplexity). CPU single token is ~55 ms (on Ryzen 7950X). * Some improvement for Q3_K on CUDA It is now ~22.5 ms/token on my GPU, so ~30% slower than Q4_0. * Some more CUDA optimizations for Q3_K Single token is now 20.5 ms/token (~20% slower than Q4_0). Perplexity is on par with Q4_0. * Adding Q4_K - scalar, AVX2, CUDA Performance is the same or perhaps very slightly better than Q4_0 on the CPU. On the GPU, single token prediction is ~10% better than Q4_0, batch mode (perplexity is about the same). * Adding Q6_K - scalar, AVX2, CUDA Performance is ~40% lower compared to Q4_K on the CPU. This is to be expected, considering that we are memory bound on the CPU and the 6-bit model is ~44% larger than the 4-bit. On the GPU, single token prediction is ~6% lower than Q4_0, batch mode (perplexity) is even closer (but still slower). * Adding Q5_K - scalar, AVX2, CUDA Performance is ~20% lower compared to Q4_K on the CPU. This is to be expected, considering that we are memory bound on the CPU and the 5-bit model is ~22% larger than the 4-bit. On the GPU, single token prediction is about the same as Q4_0 for both, single token and batch prediction. * Per convention, all QX_K quantizations use Q5_K for output.weight * Adding quantization mixes * Quantization mixes: didn't quite get what I wanted in the last commit * Q4_K dot product for ARM_NEON * Q6_K dot product for ARM_NEON * Q5_K dot product for ARM_NEON * Adding Q3_K dot for ARM_NEON It is 22% slower than Q4_K, despite the smaller model size. On x86_64, where we are memory bound, the Q3_K model is quite a bit faster than Q4_K. * A very slightly faster ARM_NEON Q3_K dot * Adding Q2_K - just CUDA for now Token prediction is pretty good - about 15.5 ms on a RTX 4080. Perplexity is about the same as Q4_K. * Adding scalar and AVX2 Q2_K dot * Adding ARM_NEON Q2_K dot About the same performance as Q4_K. * A slightly faster ARM_NEON Q2_K dot Single token prediction is now ~36 ms on M2 Max. The code is much simpler too. * Fixed bug in Q2_K CUDA dot product kernel Stranegly enough, for the few prompts I tried with the 7B model the responses looked perfectly reasonable. Only realized something is not quite right when I tried the larger models and started getting nonse back. In any case, Q2_K single token evaluation time on an RTX 4080 in a Ryzen7950X box iusing CUDA and model fully loaded on the GPU are ~15.5 ms for 7B, ~25.4 ms for 13B, and ~55.8 ms for 30B. The max number of layers that fit in VRAM for The 65B is 32. With that, we get ~330 ms per token, which is not that much faster than just running on the CPU (~470 ms per token). * Don't print zeros/NaNs when no count histogram has been collected * A 10% faster CUDA vector dot kernel for Q3_K Q3_K is now running at ~18.5 ms / token on CUDA, so the gap to Q4_0 is only 10%. It seems memory acccess pattern is more important for performance than the amount of computation the kernel does. * A slightly daster Q4_K AVX2 dot product For perplexity, where we are less memory bound, time per pass drops by ~5%. Barely measurable difference for single token prediction. * A slightly faster ARM_NEON A4_K dot product * Minor * Fix quantization error test We cannot possibly be expecting rmse < 0.002 for 2- and 3-bit quantization variants. * Fix docker build I have been sloppy with vector reinterpret casts on ARM_NEON. It seems clang is very forgiving in that regard. * Added forgotten ggml.o dependence on k_quants.h to the Makefile * Had unintentionally committed the Makefile with -Ofast enabled * ggml : rename k_quants -> ggml-quants-k, use lowercase in code --------- Co-authored-by: Iwan Kawrakow <iwan.kawrakow@gmail.com> Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
2023-06-05 19:56:18 +00:00
}
ggml : quantization refactoring (#3833) * ggml : factor all quantization code in ggml-quants ggml-ci * ggml-quants : fix Zig and Swift builds + quantize tool ggml-ci * quantize : --pure option for disabling k-quant mixtures --------- Co-authored-by: cebtenzzre <cebtenzzre@gmail.com>
2023-10-29 16:32:28 +00:00
*s = vaddvq_f32(sumv0) + vaddvq_f32(sumv1);
#elif defined(__AVX2__)
// Initialize accumulator with zeros
__m256 acc = _mm256_setzero_ps();
ggml : add SOTA 2,3,4,5,6 bit k-quantizations (#1684) * Starting to add k-quantization to ggml I think it is better to have quantization separate from ggml. For now just adding the k-quants there, but it would be better to also factor out the existing ggml quantizations. * Adding Q3_K and Q8_K (de)-quantization * Q3_K now working on CUDA and AVX2/scalar CUDA is not ideal - ~50% slower than Q4_0 for single token prediction, about the same in batch mode (perplexity). CPU single token is ~55 ms (on Ryzen 7950X). * Some improvement for Q3_K on CUDA It is now ~22.5 ms/token on my GPU, so ~30% slower than Q4_0. * Some more CUDA optimizations for Q3_K Single token is now 20.5 ms/token (~20% slower than Q4_0). Perplexity is on par with Q4_0. * Adding Q4_K - scalar, AVX2, CUDA Performance is the same or perhaps very slightly better than Q4_0 on the CPU. On the GPU, single token prediction is ~10% better than Q4_0, batch mode (perplexity is about the same). * Adding Q6_K - scalar, AVX2, CUDA Performance is ~40% lower compared to Q4_K on the CPU. This is to be expected, considering that we are memory bound on the CPU and the 6-bit model is ~44% larger than the 4-bit. On the GPU, single token prediction is ~6% lower than Q4_0, batch mode (perplexity) is even closer (but still slower). * Adding Q5_K - scalar, AVX2, CUDA Performance is ~20% lower compared to Q4_K on the CPU. This is to be expected, considering that we are memory bound on the CPU and the 5-bit model is ~22% larger than the 4-bit. On the GPU, single token prediction is about the same as Q4_0 for both, single token and batch prediction. * Per convention, all QX_K quantizations use Q5_K for output.weight * Adding quantization mixes * Quantization mixes: didn't quite get what I wanted in the last commit * Q4_K dot product for ARM_NEON * Q6_K dot product for ARM_NEON * Q5_K dot product for ARM_NEON * Adding Q3_K dot for ARM_NEON It is 22% slower than Q4_K, despite the smaller model size. On x86_64, where we are memory bound, the Q3_K model is quite a bit faster than Q4_K. * A very slightly faster ARM_NEON Q3_K dot * Adding Q2_K - just CUDA for now Token prediction is pretty good - about 15.5 ms on a RTX 4080. Perplexity is about the same as Q4_K. * Adding scalar and AVX2 Q2_K dot * Adding ARM_NEON Q2_K dot About the same performance as Q4_K. * A slightly faster ARM_NEON Q2_K dot Single token prediction is now ~36 ms on M2 Max. The code is much simpler too. * Fixed bug in Q2_K CUDA dot product kernel Stranegly enough, for the few prompts I tried with the 7B model the responses looked perfectly reasonable. Only realized something is not quite right when I tried the larger models and started getting nonse back. In any case, Q2_K single token evaluation time on an RTX 4080 in a Ryzen7950X box iusing CUDA and model fully loaded on the GPU are ~15.5 ms for 7B, ~25.4 ms for 13B, and ~55.8 ms for 30B. The max number of layers that fit in VRAM for The 65B is 32. With that, we get ~330 ms per token, which is not that much faster than just running on the CPU (~470 ms per token). * Don't print zeros/NaNs when no count histogram has been collected * A 10% faster CUDA vector dot kernel for Q3_K Q3_K is now running at ~18.5 ms / token on CUDA, so the gap to Q4_0 is only 10%. It seems memory acccess pattern is more important for performance than the amount of computation the kernel does. * A slightly daster Q4_K AVX2 dot product For perplexity, where we are less memory bound, time per pass drops by ~5%. Barely measurable difference for single token prediction. * A slightly faster ARM_NEON A4_K dot product * Minor * Fix quantization error test We cannot possibly be expecting rmse < 0.002 for 2- and 3-bit quantization variants. * Fix docker build I have been sloppy with vector reinterpret casts on ARM_NEON. It seems clang is very forgiving in that regard. * Added forgotten ggml.o dependence on k_quants.h to the Makefile * Had unintentionally committed the Makefile with -Ofast enabled * ggml : rename k_quants -> ggml-quants-k, use lowercase in code --------- Co-authored-by: Iwan Kawrakow <iwan.kawrakow@gmail.com> Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
2023-06-05 19:56:18 +00:00
ggml : quantization refactoring (#3833) * ggml : factor all quantization code in ggml-quants ggml-ci * ggml-quants : fix Zig and Swift builds + quantize tool ggml-ci * quantize : --pure option for disabling k-quant mixtures --------- Co-authored-by: cebtenzzre <cebtenzzre@gmail.com>
2023-10-29 16:32:28 +00:00
// Main loop
for (int i = 0; i < nb; ++i) {
/* Compute combined scale for the block */
ggml : move FP16 <-> FP32 code to ggml-impl.h (#3861) * ggml : move FP16 <-> FP32 stuff to ggml-impl.h ggml-ci * tests : fix ARM build * ggml : explicitly initialize deprecated type traits * ggml : add math.h to ggml-impl.h * ggml : remove duplicate static assert macros * ggml : prefix lookup tables with ggml_ ggml-ci * ggml-impl : move extern "C" to start of file
2023-10-30 17:19:15 +00:00
const __m256 d = _mm256_set1_ps( GGML_FP16_TO_FP32(x[i].d) * GGML_FP16_TO_FP32(y[i].d) );
ggml : add SOTA 2,3,4,5,6 bit k-quantizations (#1684) * Starting to add k-quantization to ggml I think it is better to have quantization separate from ggml. For now just adding the k-quants there, but it would be better to also factor out the existing ggml quantizations. * Adding Q3_K and Q8_K (de)-quantization * Q3_K now working on CUDA and AVX2/scalar CUDA is not ideal - ~50% slower than Q4_0 for single token prediction, about the same in batch mode (perplexity). CPU single token is ~55 ms (on Ryzen 7950X). * Some improvement for Q3_K on CUDA It is now ~22.5 ms/token on my GPU, so ~30% slower than Q4_0. * Some more CUDA optimizations for Q3_K Single token is now 20.5 ms/token (~20% slower than Q4_0). Perplexity is on par with Q4_0. * Adding Q4_K - scalar, AVX2, CUDA Performance is the same or perhaps very slightly better than Q4_0 on the CPU. On the GPU, single token prediction is ~10% better than Q4_0, batch mode (perplexity is about the same). * Adding Q6_K - scalar, AVX2, CUDA Performance is ~40% lower compared to Q4_K on the CPU. This is to be expected, considering that we are memory bound on the CPU and the 6-bit model is ~44% larger than the 4-bit. On the GPU, single token prediction is ~6% lower than Q4_0, batch mode (perplexity) is even closer (but still slower). * Adding Q5_K - scalar, AVX2, CUDA Performance is ~20% lower compared to Q4_K on the CPU. This is to be expected, considering that we are memory bound on the CPU and the 5-bit model is ~22% larger than the 4-bit. On the GPU, single token prediction is about the same as Q4_0 for both, single token and batch prediction. * Per convention, all QX_K quantizations use Q5_K for output.weight * Adding quantization mixes * Quantization mixes: didn't quite get what I wanted in the last commit * Q4_K dot product for ARM_NEON * Q6_K dot product for ARM_NEON * Q5_K dot product for ARM_NEON * Adding Q3_K dot for ARM_NEON It is 22% slower than Q4_K, despite the smaller model size. On x86_64, where we are memory bound, the Q3_K model is quite a bit faster than Q4_K. * A very slightly faster ARM_NEON Q3_K dot * Adding Q2_K - just CUDA for now Token prediction is pretty good - about 15.5 ms on a RTX 4080. Perplexity is about the same as Q4_K. * Adding scalar and AVX2 Q2_K dot * Adding ARM_NEON Q2_K dot About the same performance as Q4_K. * A slightly faster ARM_NEON Q2_K dot Single token prediction is now ~36 ms on M2 Max. The code is much simpler too. * Fixed bug in Q2_K CUDA dot product kernel Stranegly enough, for the few prompts I tried with the 7B model the responses looked perfectly reasonable. Only realized something is not quite right when I tried the larger models and started getting nonse back. In any case, Q2_K single token evaluation time on an RTX 4080 in a Ryzen7950X box iusing CUDA and model fully loaded on the GPU are ~15.5 ms for 7B, ~25.4 ms for 13B, and ~55.8 ms for 30B. The max number of layers that fit in VRAM for The 65B is 32. With that, we get ~330 ms per token, which is not that much faster than just running on the CPU (~470 ms per token). * Don't print zeros/NaNs when no count histogram has been collected * A 10% faster CUDA vector dot kernel for Q3_K Q3_K is now running at ~18.5 ms / token on CUDA, so the gap to Q4_0 is only 10%. It seems memory acccess pattern is more important for performance than the amount of computation the kernel does. * A slightly daster Q4_K AVX2 dot product For perplexity, where we are less memory bound, time per pass drops by ~5%. Barely measurable difference for single token prediction. * A slightly faster ARM_NEON A4_K dot product * Minor * Fix quantization error test We cannot possibly be expecting rmse < 0.002 for 2- and 3-bit quantization variants. * Fix docker build I have been sloppy with vector reinterpret casts on ARM_NEON. It seems clang is very forgiving in that regard. * Added forgotten ggml.o dependence on k_quants.h to the Makefile * Had unintentionally committed the Makefile with -Ofast enabled * ggml : rename k_quants -> ggml-quants-k, use lowercase in code --------- Co-authored-by: Iwan Kawrakow <iwan.kawrakow@gmail.com> Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
2023-06-05 19:56:18 +00:00
ggml : quantization refactoring (#3833) * ggml : factor all quantization code in ggml-quants ggml-ci * ggml-quants : fix Zig and Swift builds + quantize tool ggml-ci * quantize : --pure option for disabling k-quant mixtures --------- Co-authored-by: cebtenzzre <cebtenzzre@gmail.com>
2023-10-29 16:32:28 +00:00
__m256i bx = bytes_from_nibbles_32(x[i].qs);
// Now we have a vector with bytes in [ 0 .. 15 ] interval. Offset them into [ -8 .. +7 ] interval.
const __m256i off = _mm256_set1_epi8( 8 );
bx = _mm256_sub_epi8( bx, off );
__m256i by = _mm256_loadu_si256((const __m256i *)y[i].qs);
const __m256 q = mul_sum_i8_pairs_float(bx, by);
/* Multiply q with scale and accumulate */
acc = _mm256_fmadd_ps( d, q, acc );
ggml : add SOTA 2,3,4,5,6 bit k-quantizations (#1684) * Starting to add k-quantization to ggml I think it is better to have quantization separate from ggml. For now just adding the k-quants there, but it would be better to also factor out the existing ggml quantizations. * Adding Q3_K and Q8_K (de)-quantization * Q3_K now working on CUDA and AVX2/scalar CUDA is not ideal - ~50% slower than Q4_0 for single token prediction, about the same in batch mode (perplexity). CPU single token is ~55 ms (on Ryzen 7950X). * Some improvement for Q3_K on CUDA It is now ~22.5 ms/token on my GPU, so ~30% slower than Q4_0. * Some more CUDA optimizations for Q3_K Single token is now 20.5 ms/token (~20% slower than Q4_0). Perplexity is on par with Q4_0. * Adding Q4_K - scalar, AVX2, CUDA Performance is the same or perhaps very slightly better than Q4_0 on the CPU. On the GPU, single token prediction is ~10% better than Q4_0, batch mode (perplexity is about the same). * Adding Q6_K - scalar, AVX2, CUDA Performance is ~40% lower compared to Q4_K on the CPU. This is to be expected, considering that we are memory bound on the CPU and the 6-bit model is ~44% larger than the 4-bit. On the GPU, single token prediction is ~6% lower than Q4_0, batch mode (perplexity) is even closer (but still slower). * Adding Q5_K - scalar, AVX2, CUDA Performance is ~20% lower compared to Q4_K on the CPU. This is to be expected, considering that we are memory bound on the CPU and the 5-bit model is ~22% larger than the 4-bit. On the GPU, single token prediction is about the same as Q4_0 for both, single token and batch prediction. * Per convention, all QX_K quantizations use Q5_K for output.weight * Adding quantization mixes * Quantization mixes: didn't quite get what I wanted in the last commit * Q4_K dot product for ARM_NEON * Q6_K dot product for ARM_NEON * Q5_K dot product for ARM_NEON * Adding Q3_K dot for ARM_NEON It is 22% slower than Q4_K, despite the smaller model size. On x86_64, where we are memory bound, the Q3_K model is quite a bit faster than Q4_K. * A very slightly faster ARM_NEON Q3_K dot * Adding Q2_K - just CUDA for now Token prediction is pretty good - about 15.5 ms on a RTX 4080. Perplexity is about the same as Q4_K. * Adding scalar and AVX2 Q2_K dot * Adding ARM_NEON Q2_K dot About the same performance as Q4_K. * A slightly faster ARM_NEON Q2_K dot Single token prediction is now ~36 ms on M2 Max. The code is much simpler too. * Fixed bug in Q2_K CUDA dot product kernel Stranegly enough, for the few prompts I tried with the 7B model the responses looked perfectly reasonable. Only realized something is not quite right when I tried the larger models and started getting nonse back. In any case, Q2_K single token evaluation time on an RTX 4080 in a Ryzen7950X box iusing CUDA and model fully loaded on the GPU are ~15.5 ms for 7B, ~25.4 ms for 13B, and ~55.8 ms for 30B. The max number of layers that fit in VRAM for The 65B is 32. With that, we get ~330 ms per token, which is not that much faster than just running on the CPU (~470 ms per token). * Don't print zeros/NaNs when no count histogram has been collected * A 10% faster CUDA vector dot kernel for Q3_K Q3_K is now running at ~18.5 ms / token on CUDA, so the gap to Q4_0 is only 10%. It seems memory acccess pattern is more important for performance than the amount of computation the kernel does. * A slightly daster Q4_K AVX2 dot product For perplexity, where we are less memory bound, time per pass drops by ~5%. Barely measurable difference for single token prediction. * A slightly faster ARM_NEON A4_K dot product * Minor * Fix quantization error test We cannot possibly be expecting rmse < 0.002 for 2- and 3-bit quantization variants. * Fix docker build I have been sloppy with vector reinterpret casts on ARM_NEON. It seems clang is very forgiving in that regard. * Added forgotten ggml.o dependence on k_quants.h to the Makefile * Had unintentionally committed the Makefile with -Ofast enabled * ggml : rename k_quants -> ggml-quants-k, use lowercase in code --------- Co-authored-by: Iwan Kawrakow <iwan.kawrakow@gmail.com> Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
2023-06-05 19:56:18 +00:00
}
ggml : quantization refactoring (#3833) * ggml : factor all quantization code in ggml-quants ggml-ci * ggml-quants : fix Zig and Swift builds + quantize tool ggml-ci * quantize : --pure option for disabling k-quant mixtures --------- Co-authored-by: cebtenzzre <cebtenzzre@gmail.com>
2023-10-29 16:32:28 +00:00
*s = hsum_float_8(acc);
#elif defined(__AVX__)
// Initialize accumulator with zeros
__m256 acc = _mm256_setzero_ps();
ggml : add SOTA 2,3,4,5,6 bit k-quantizations (#1684) * Starting to add k-quantization to ggml I think it is better to have quantization separate from ggml. For now just adding the k-quants there, but it would be better to also factor out the existing ggml quantizations. * Adding Q3_K and Q8_K (de)-quantization * Q3_K now working on CUDA and AVX2/scalar CUDA is not ideal - ~50% slower than Q4_0 for single token prediction, about the same in batch mode (perplexity). CPU single token is ~55 ms (on Ryzen 7950X). * Some improvement for Q3_K on CUDA It is now ~22.5 ms/token on my GPU, so ~30% slower than Q4_0. * Some more CUDA optimizations for Q3_K Single token is now 20.5 ms/token (~20% slower than Q4_0). Perplexity is on par with Q4_0. * Adding Q4_K - scalar, AVX2, CUDA Performance is the same or perhaps very slightly better than Q4_0 on the CPU. On the GPU, single token prediction is ~10% better than Q4_0, batch mode (perplexity is about the same). * Adding Q6_K - scalar, AVX2, CUDA Performance is ~40% lower compared to Q4_K on the CPU. This is to be expected, considering that we are memory bound on the CPU and the 6-bit model is ~44% larger than the 4-bit. On the GPU, single token prediction is ~6% lower than Q4_0, batch mode (perplexity) is even closer (but still slower). * Adding Q5_K - scalar, AVX2, CUDA Performance is ~20% lower compared to Q4_K on the CPU. This is to be expected, considering that we are memory bound on the CPU and the 5-bit model is ~22% larger than the 4-bit. On the GPU, single token prediction is about the same as Q4_0 for both, single token and batch prediction. * Per convention, all QX_K quantizations use Q5_K for output.weight * Adding quantization mixes * Quantization mixes: didn't quite get what I wanted in the last commit * Q4_K dot product for ARM_NEON * Q6_K dot product for ARM_NEON * Q5_K dot product for ARM_NEON * Adding Q3_K dot for ARM_NEON It is 22% slower than Q4_K, despite the smaller model size. On x86_64, where we are memory bound, the Q3_K model is quite a bit faster than Q4_K. * A very slightly faster ARM_NEON Q3_K dot * Adding Q2_K - just CUDA for now Token prediction is pretty good - about 15.5 ms on a RTX 4080. Perplexity is about the same as Q4_K. * Adding scalar and AVX2 Q2_K dot * Adding ARM_NEON Q2_K dot About the same performance as Q4_K. * A slightly faster ARM_NEON Q2_K dot Single token prediction is now ~36 ms on M2 Max. The code is much simpler too. * Fixed bug in Q2_K CUDA dot product kernel Stranegly enough, for the few prompts I tried with the 7B model the responses looked perfectly reasonable. Only realized something is not quite right when I tried the larger models and started getting nonse back. In any case, Q2_K single token evaluation time on an RTX 4080 in a Ryzen7950X box iusing CUDA and model fully loaded on the GPU are ~15.5 ms for 7B, ~25.4 ms for 13B, and ~55.8 ms for 30B. The max number of layers that fit in VRAM for The 65B is 32. With that, we get ~330 ms per token, which is not that much faster than just running on the CPU (~470 ms per token). * Don't print zeros/NaNs when no count histogram has been collected * A 10% faster CUDA vector dot kernel for Q3_K Q3_K is now running at ~18.5 ms / token on CUDA, so the gap to Q4_0 is only 10%. It seems memory acccess pattern is more important for performance than the amount of computation the kernel does. * A slightly daster Q4_K AVX2 dot product For perplexity, where we are less memory bound, time per pass drops by ~5%. Barely measurable difference for single token prediction. * A slightly faster ARM_NEON A4_K dot product * Minor * Fix quantization error test We cannot possibly be expecting rmse < 0.002 for 2- and 3-bit quantization variants. * Fix docker build I have been sloppy with vector reinterpret casts on ARM_NEON. It seems clang is very forgiving in that regard. * Added forgotten ggml.o dependence on k_quants.h to the Makefile * Had unintentionally committed the Makefile with -Ofast enabled * ggml : rename k_quants -> ggml-quants-k, use lowercase in code --------- Co-authored-by: Iwan Kawrakow <iwan.kawrakow@gmail.com> Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
2023-06-05 19:56:18 +00:00
ggml : quantization refactoring (#3833) * ggml : factor all quantization code in ggml-quants ggml-ci * ggml-quants : fix Zig and Swift builds + quantize tool ggml-ci * quantize : --pure option for disabling k-quant mixtures --------- Co-authored-by: cebtenzzre <cebtenzzre@gmail.com>
2023-10-29 16:32:28 +00:00
// Main loop
for (int i = 0; i < nb; ++i) {
// Compute combined scale for the block
ggml : move FP16 <-> FP32 code to ggml-impl.h (#3861) * ggml : move FP16 <-> FP32 stuff to ggml-impl.h ggml-ci * tests : fix ARM build * ggml : explicitly initialize deprecated type traits * ggml : add math.h to ggml-impl.h * ggml : remove duplicate static assert macros * ggml : prefix lookup tables with ggml_ ggml-ci * ggml-impl : move extern "C" to start of file
2023-10-30 17:19:15 +00:00
const __m256 d = _mm256_set1_ps( GGML_FP16_TO_FP32(x[i].d) * GGML_FP16_TO_FP32(y[i].d) );
ggml : add SOTA 2,3,4,5,6 bit k-quantizations (#1684) * Starting to add k-quantization to ggml I think it is better to have quantization separate from ggml. For now just adding the k-quants there, but it would be better to also factor out the existing ggml quantizations. * Adding Q3_K and Q8_K (de)-quantization * Q3_K now working on CUDA and AVX2/scalar CUDA is not ideal - ~50% slower than Q4_0 for single token prediction, about the same in batch mode (perplexity). CPU single token is ~55 ms (on Ryzen 7950X). * Some improvement for Q3_K on CUDA It is now ~22.5 ms/token on my GPU, so ~30% slower than Q4_0. * Some more CUDA optimizations for Q3_K Single token is now 20.5 ms/token (~20% slower than Q4_0). Perplexity is on par with Q4_0. * Adding Q4_K - scalar, AVX2, CUDA Performance is the same or perhaps very slightly better than Q4_0 on the CPU. On the GPU, single token prediction is ~10% better than Q4_0, batch mode (perplexity is about the same). * Adding Q6_K - scalar, AVX2, CUDA Performance is ~40% lower compared to Q4_K on the CPU. This is to be expected, considering that we are memory bound on the CPU and the 6-bit model is ~44% larger than the 4-bit. On the GPU, single token prediction is ~6% lower than Q4_0, batch mode (perplexity) is even closer (but still slower). * Adding Q5_K - scalar, AVX2, CUDA Performance is ~20% lower compared to Q4_K on the CPU. This is to be expected, considering that we are memory bound on the CPU and the 5-bit model is ~22% larger than the 4-bit. On the GPU, single token prediction is about the same as Q4_0 for both, single token and batch prediction. * Per convention, all QX_K quantizations use Q5_K for output.weight * Adding quantization mixes * Quantization mixes: didn't quite get what I wanted in the last commit * Q4_K dot product for ARM_NEON * Q6_K dot product for ARM_NEON * Q5_K dot product for ARM_NEON * Adding Q3_K dot for ARM_NEON It is 22% slower than Q4_K, despite the smaller model size. On x86_64, where we are memory bound, the Q3_K model is quite a bit faster than Q4_K. * A very slightly faster ARM_NEON Q3_K dot * Adding Q2_K - just CUDA for now Token prediction is pretty good - about 15.5 ms on a RTX 4080. Perplexity is about the same as Q4_K. * Adding scalar and AVX2 Q2_K dot * Adding ARM_NEON Q2_K dot About the same performance as Q4_K. * A slightly faster ARM_NEON Q2_K dot Single token prediction is now ~36 ms on M2 Max. The code is much simpler too. * Fixed bug in Q2_K CUDA dot product kernel Stranegly enough, for the few prompts I tried with the 7B model the responses looked perfectly reasonable. Only realized something is not quite right when I tried the larger models and started getting nonse back. In any case, Q2_K single token evaluation time on an RTX 4080 in a Ryzen7950X box iusing CUDA and model fully loaded on the GPU are ~15.5 ms for 7B, ~25.4 ms for 13B, and ~55.8 ms for 30B. The max number of layers that fit in VRAM for The 65B is 32. With that, we get ~330 ms per token, which is not that much faster than just running on the CPU (~470 ms per token). * Don't print zeros/NaNs when no count histogram has been collected * A 10% faster CUDA vector dot kernel for Q3_K Q3_K is now running at ~18.5 ms / token on CUDA, so the gap to Q4_0 is only 10%. It seems memory acccess pattern is more important for performance than the amount of computation the kernel does. * A slightly daster Q4_K AVX2 dot product For perplexity, where we are less memory bound, time per pass drops by ~5%. Barely measurable difference for single token prediction. * A slightly faster ARM_NEON A4_K dot product * Minor * Fix quantization error test We cannot possibly be expecting rmse < 0.002 for 2- and 3-bit quantization variants. * Fix docker build I have been sloppy with vector reinterpret casts on ARM_NEON. It seems clang is very forgiving in that regard. * Added forgotten ggml.o dependence on k_quants.h to the Makefile * Had unintentionally committed the Makefile with -Ofast enabled * ggml : rename k_quants -> ggml-quants-k, use lowercase in code --------- Co-authored-by: Iwan Kawrakow <iwan.kawrakow@gmail.com> Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
2023-06-05 19:56:18 +00:00
ggml : quantization refactoring (#3833) * ggml : factor all quantization code in ggml-quants ggml-ci * ggml-quants : fix Zig and Swift builds + quantize tool ggml-ci * quantize : --pure option for disabling k-quant mixtures --------- Co-authored-by: cebtenzzre <cebtenzzre@gmail.com>
2023-10-29 16:32:28 +00:00
const __m128i lowMask = _mm_set1_epi8(0xF);
const __m128i off = _mm_set1_epi8(8);
const __m128i tmp = _mm_loadu_si128((const __m128i *)x[i].qs);
__m128i bx = _mm_and_si128(lowMask, tmp);
__m128i by = _mm_loadu_si128((const __m128i *)y[i].qs);
bx = _mm_sub_epi8(bx, off);
const __m128i i32_0 = mul_sum_i8_pairs(bx, by);
bx = _mm_and_si128(lowMask, _mm_srli_epi64(tmp, 4));
by = _mm_loadu_si128((const __m128i *)(y[i].qs + 16));
bx = _mm_sub_epi8(bx, off);
const __m128i i32_1 = mul_sum_i8_pairs(bx, by);
// Convert int32_t to float
__m256 p = _mm256_cvtepi32_ps(MM256_SET_M128I(i32_0, i32_1));
// Apply the scale, and accumulate
acc = _mm256_add_ps(_mm256_mul_ps( d, p ), acc);
}
*s = hsum_float_8(acc);
#elif defined(__SSSE3__)
// set constants
const __m128i lowMask = _mm_set1_epi8(0xF);
const __m128i off = _mm_set1_epi8(8);
// Initialize accumulator with zeros
__m128 acc_0 = _mm_setzero_ps();
__m128 acc_1 = _mm_setzero_ps();
__m128 acc_2 = _mm_setzero_ps();
__m128 acc_3 = _mm_setzero_ps();
// First round without accumulation
{
_mm_prefetch(&x[0] + sizeof(block_q4_0), _MM_HINT_T0);
_mm_prefetch(&y[0] + sizeof(block_q8_0), _MM_HINT_T0);
// Compute combined scale for the block 0 and 1
ggml : move FP16 <-> FP32 code to ggml-impl.h (#3861) * ggml : move FP16 <-> FP32 stuff to ggml-impl.h ggml-ci * tests : fix ARM build * ggml : explicitly initialize deprecated type traits * ggml : add math.h to ggml-impl.h * ggml : remove duplicate static assert macros * ggml : prefix lookup tables with ggml_ ggml-ci * ggml-impl : move extern "C" to start of file
2023-10-30 17:19:15 +00:00
const __m128 d_0_1 = _mm_set1_ps( GGML_FP16_TO_FP32(x[0].d) * GGML_FP16_TO_FP32(y[0].d) );
ggml : quantization refactoring (#3833) * ggml : factor all quantization code in ggml-quants ggml-ci * ggml-quants : fix Zig and Swift builds + quantize tool ggml-ci * quantize : --pure option for disabling k-quant mixtures --------- Co-authored-by: cebtenzzre <cebtenzzre@gmail.com>
2023-10-29 16:32:28 +00:00
const __m128i tmp_0_1 = _mm_loadu_si128((const __m128i *)x[0].qs);
__m128i bx_0 = _mm_and_si128(lowMask, tmp_0_1);
__m128i by_0 = _mm_loadu_si128((const __m128i *)y[0].qs);
bx_0 = _mm_sub_epi8(bx_0, off);
const __m128i i32_0 = mul_sum_i8_pairs(bx_0, by_0);
__m128i bx_1 = _mm_and_si128(lowMask, _mm_srli_epi64(tmp_0_1, 4));
__m128i by_1 = _mm_loadu_si128((const __m128i *)(y[0].qs + 16));
bx_1 = _mm_sub_epi8(bx_1, off);
const __m128i i32_1 = mul_sum_i8_pairs(bx_1, by_1);
_mm_prefetch(&x[1] + sizeof(block_q4_0), _MM_HINT_T0);
_mm_prefetch(&y[1] + sizeof(block_q8_0), _MM_HINT_T0);
// Compute combined scale for the block 2 and 3
ggml : move FP16 <-> FP32 code to ggml-impl.h (#3861) * ggml : move FP16 <-> FP32 stuff to ggml-impl.h ggml-ci * tests : fix ARM build * ggml : explicitly initialize deprecated type traits * ggml : add math.h to ggml-impl.h * ggml : remove duplicate static assert macros * ggml : prefix lookup tables with ggml_ ggml-ci * ggml-impl : move extern "C" to start of file
2023-10-30 17:19:15 +00:00
const __m128 d_2_3 = _mm_set1_ps( GGML_FP16_TO_FP32(x[1].d) * GGML_FP16_TO_FP32(y[1].d) );
ggml : quantization refactoring (#3833) * ggml : factor all quantization code in ggml-quants ggml-ci * ggml-quants : fix Zig and Swift builds + quantize tool ggml-ci * quantize : --pure option for disabling k-quant mixtures --------- Co-authored-by: cebtenzzre <cebtenzzre@gmail.com>
2023-10-29 16:32:28 +00:00
const __m128i tmp_2_3 = _mm_loadu_si128((const __m128i *)x[1].qs);
__m128i bx_2 = _mm_and_si128(lowMask, tmp_2_3);
__m128i by_2 = _mm_loadu_si128((const __m128i *)y[1].qs);
bx_2 = _mm_sub_epi8(bx_2, off);
const __m128i i32_2 = mul_sum_i8_pairs(bx_2, by_2);
__m128i bx_3 = _mm_and_si128(lowMask, _mm_srli_epi64(tmp_2_3, 4));
__m128i by_3 = _mm_loadu_si128((const __m128i *)(y[1].qs + 16));
bx_3 = _mm_sub_epi8(bx_3, off);
const __m128i i32_3 = mul_sum_i8_pairs(bx_3, by_3);
// Convert int32_t to float
__m128 p0 = _mm_cvtepi32_ps(i32_0);
__m128 p1 = _mm_cvtepi32_ps(i32_1);
__m128 p2 = _mm_cvtepi32_ps(i32_2);
__m128 p3 = _mm_cvtepi32_ps(i32_3);
// Apply the scale
acc_0 = _mm_mul_ps( d_0_1, p0 );
acc_1 = _mm_mul_ps( d_0_1, p1 );
acc_2 = _mm_mul_ps( d_2_3, p2 );
acc_3 = _mm_mul_ps( d_2_3, p3 );
}
assert(nb % 2 == 0); // TODO: handle odd nb
// Main loop
for (int i = 2; i < nb; i+=2) {
_mm_prefetch(&x[i] + sizeof(block_q4_0), _MM_HINT_T0);
_mm_prefetch(&y[i] + sizeof(block_q8_0), _MM_HINT_T0);
// Compute combined scale for the block 0 and 1
ggml : move FP16 <-> FP32 code to ggml-impl.h (#3861) * ggml : move FP16 <-> FP32 stuff to ggml-impl.h ggml-ci * tests : fix ARM build * ggml : explicitly initialize deprecated type traits * ggml : add math.h to ggml-impl.h * ggml : remove duplicate static assert macros * ggml : prefix lookup tables with ggml_ ggml-ci * ggml-impl : move extern "C" to start of file
2023-10-30 17:19:15 +00:00
const __m128 d_0_1 = _mm_set1_ps( GGML_FP16_TO_FP32(x[i].d) * GGML_FP16_TO_FP32(y[i].d) );
ggml : quantization refactoring (#3833) * ggml : factor all quantization code in ggml-quants ggml-ci * ggml-quants : fix Zig and Swift builds + quantize tool ggml-ci * quantize : --pure option for disabling k-quant mixtures --------- Co-authored-by: cebtenzzre <cebtenzzre@gmail.com>
2023-10-29 16:32:28 +00:00
const __m128i tmp_0_1 = _mm_loadu_si128((const __m128i *)x[i].qs);
__m128i bx_0 = _mm_and_si128(lowMask, tmp_0_1);
__m128i by_0 = _mm_loadu_si128((const __m128i *)y[i].qs);
bx_0 = _mm_sub_epi8(bx_0, off);
const __m128i i32_0 = mul_sum_i8_pairs(bx_0, by_0);
__m128i bx_1 = _mm_and_si128(lowMask, _mm_srli_epi64(tmp_0_1, 4));
__m128i by_1 = _mm_loadu_si128((const __m128i *)(y[i].qs + 16));
bx_1 = _mm_sub_epi8(bx_1, off);
const __m128i i32_1 = mul_sum_i8_pairs(bx_1, by_1);
_mm_prefetch(&x[i] + 2 * sizeof(block_q4_0), _MM_HINT_T0);
_mm_prefetch(&y[i] + 2 * sizeof(block_q8_0), _MM_HINT_T0);
// Compute combined scale for the block 2 and 3
ggml : move FP16 <-> FP32 code to ggml-impl.h (#3861) * ggml : move FP16 <-> FP32 stuff to ggml-impl.h ggml-ci * tests : fix ARM build * ggml : explicitly initialize deprecated type traits * ggml : add math.h to ggml-impl.h * ggml : remove duplicate static assert macros * ggml : prefix lookup tables with ggml_ ggml-ci * ggml-impl : move extern "C" to start of file
2023-10-30 17:19:15 +00:00
const __m128 d_2_3 = _mm_set1_ps( GGML_FP16_TO_FP32(x[i + 1].d) * GGML_FP16_TO_FP32(y[i + 1].d) );
ggml : quantization refactoring (#3833) * ggml : factor all quantization code in ggml-quants ggml-ci * ggml-quants : fix Zig and Swift builds + quantize tool ggml-ci * quantize : --pure option for disabling k-quant mixtures --------- Co-authored-by: cebtenzzre <cebtenzzre@gmail.com>
2023-10-29 16:32:28 +00:00
const __m128i tmp_2_3 = _mm_loadu_si128((const __m128i *)x[i + 1].qs);
__m128i bx_2 = _mm_and_si128(lowMask, tmp_2_3);
__m128i by_2 = _mm_loadu_si128((const __m128i *)y[i + 1].qs);
bx_2 = _mm_sub_epi8(bx_2, off);
const __m128i i32_2 = mul_sum_i8_pairs(bx_2, by_2);
__m128i bx_3 = _mm_and_si128(lowMask, _mm_srli_epi64(tmp_2_3, 4));
__m128i by_3 = _mm_loadu_si128((const __m128i *)(y[i + 1].qs + 16));
bx_3 = _mm_sub_epi8(bx_3, off);
const __m128i i32_3 = mul_sum_i8_pairs(bx_3, by_3);
// Convert int32_t to float
__m128 p0 = _mm_cvtepi32_ps(i32_0);
__m128 p1 = _mm_cvtepi32_ps(i32_1);
__m128 p2 = _mm_cvtepi32_ps(i32_2);
__m128 p3 = _mm_cvtepi32_ps(i32_3);
// Apply the scale
__m128 p0_d = _mm_mul_ps( d_0_1, p0 );
__m128 p1_d = _mm_mul_ps( d_0_1, p1 );
__m128 p2_d = _mm_mul_ps( d_2_3, p2 );
__m128 p3_d = _mm_mul_ps( d_2_3, p3 );
// Acummulate
acc_0 = _mm_add_ps(p0_d, acc_0);
acc_1 = _mm_add_ps(p1_d, acc_1);
acc_2 = _mm_add_ps(p2_d, acc_2);
acc_3 = _mm_add_ps(p3_d, acc_3);
}
*s = hsum_float_4x4(acc_0, acc_1, acc_2, acc_3);
#elif defined(__riscv_v_intrinsic)
float sumf = 0.0;
size_t vl = __riscv_vsetvl_e8m1(qk/2);
ggml : add SOTA 2,3,4,5,6 bit k-quantizations (#1684) * Starting to add k-quantization to ggml I think it is better to have quantization separate from ggml. For now just adding the k-quants there, but it would be better to also factor out the existing ggml quantizations. * Adding Q3_K and Q8_K (de)-quantization * Q3_K now working on CUDA and AVX2/scalar CUDA is not ideal - ~50% slower than Q4_0 for single token prediction, about the same in batch mode (perplexity). CPU single token is ~55 ms (on Ryzen 7950X). * Some improvement for Q3_K on CUDA It is now ~22.5 ms/token on my GPU, so ~30% slower than Q4_0. * Some more CUDA optimizations for Q3_K Single token is now 20.5 ms/token (~20% slower than Q4_0). Perplexity is on par with Q4_0. * Adding Q4_K - scalar, AVX2, CUDA Performance is the same or perhaps very slightly better than Q4_0 on the CPU. On the GPU, single token prediction is ~10% better than Q4_0, batch mode (perplexity is about the same). * Adding Q6_K - scalar, AVX2, CUDA Performance is ~40% lower compared to Q4_K on the CPU. This is to be expected, considering that we are memory bound on the CPU and the 6-bit model is ~44% larger than the 4-bit. On the GPU, single token prediction is ~6% lower than Q4_0, batch mode (perplexity) is even closer (but still slower). * Adding Q5_K - scalar, AVX2, CUDA Performance is ~20% lower compared to Q4_K on the CPU. This is to be expected, considering that we are memory bound on the CPU and the 5-bit model is ~22% larger than the 4-bit. On the GPU, single token prediction is about the same as Q4_0 for both, single token and batch prediction. * Per convention, all QX_K quantizations use Q5_K for output.weight * Adding quantization mixes * Quantization mixes: didn't quite get what I wanted in the last commit * Q4_K dot product for ARM_NEON * Q6_K dot product for ARM_NEON * Q5_K dot product for ARM_NEON * Adding Q3_K dot for ARM_NEON It is 22% slower than Q4_K, despite the smaller model size. On x86_64, where we are memory bound, the Q3_K model is quite a bit faster than Q4_K. * A very slightly faster ARM_NEON Q3_K dot * Adding Q2_K - just CUDA for now Token prediction is pretty good - about 15.5 ms on a RTX 4080. Perplexity is about the same as Q4_K. * Adding scalar and AVX2 Q2_K dot * Adding ARM_NEON Q2_K dot About the same performance as Q4_K. * A slightly faster ARM_NEON Q2_K dot Single token prediction is now ~36 ms on M2 Max. The code is much simpler too. * Fixed bug in Q2_K CUDA dot product kernel Stranegly enough, for the few prompts I tried with the 7B model the responses looked perfectly reasonable. Only realized something is not quite right when I tried the larger models and started getting nonse back. In any case, Q2_K single token evaluation time on an RTX 4080 in a Ryzen7950X box iusing CUDA and model fully loaded on the GPU are ~15.5 ms for 7B, ~25.4 ms for 13B, and ~55.8 ms for 30B. The max number of layers that fit in VRAM for The 65B is 32. With that, we get ~330 ms per token, which is not that much faster than just running on the CPU (~470 ms per token). * Don't print zeros/NaNs when no count histogram has been collected * A 10% faster CUDA vector dot kernel for Q3_K Q3_K is now running at ~18.5 ms / token on CUDA, so the gap to Q4_0 is only 10%. It seems memory acccess pattern is more important for performance than the amount of computation the kernel does. * A slightly daster Q4_K AVX2 dot product For perplexity, where we are less memory bound, time per pass drops by ~5%. Barely measurable difference for single token prediction. * A slightly faster ARM_NEON A4_K dot product * Minor * Fix quantization error test We cannot possibly be expecting rmse < 0.002 for 2- and 3-bit quantization variants. * Fix docker build I have been sloppy with vector reinterpret casts on ARM_NEON. It seems clang is very forgiving in that regard. * Added forgotten ggml.o dependence on k_quants.h to the Makefile * Had unintentionally committed the Makefile with -Ofast enabled * ggml : rename k_quants -> ggml-quants-k, use lowercase in code --------- Co-authored-by: Iwan Kawrakow <iwan.kawrakow@gmail.com> Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
2023-06-05 19:56:18 +00:00
for (int i = 0; i < nb; i++) {
ggml : quantization refactoring (#3833) * ggml : factor all quantization code in ggml-quants ggml-ci * ggml-quants : fix Zig and Swift builds + quantize tool ggml-ci * quantize : --pure option for disabling k-quant mixtures --------- Co-authored-by: cebtenzzre <cebtenzzre@gmail.com>
2023-10-29 16:32:28 +00:00
// load elements
vuint8mf2_t tx = __riscv_vle8_v_u8mf2(x[i].qs, vl);
ggml : add SOTA 2,3,4,5,6 bit k-quantizations (#1684) * Starting to add k-quantization to ggml I think it is better to have quantization separate from ggml. For now just adding the k-quants there, but it would be better to also factor out the existing ggml quantizations. * Adding Q3_K and Q8_K (de)-quantization * Q3_K now working on CUDA and AVX2/scalar CUDA is not ideal - ~50% slower than Q4_0 for single token prediction, about the same in batch mode (perplexity). CPU single token is ~55 ms (on Ryzen 7950X). * Some improvement for Q3_K on CUDA It is now ~22.5 ms/token on my GPU, so ~30% slower than Q4_0. * Some more CUDA optimizations for Q3_K Single token is now 20.5 ms/token (~20% slower than Q4_0). Perplexity is on par with Q4_0. * Adding Q4_K - scalar, AVX2, CUDA Performance is the same or perhaps very slightly better than Q4_0 on the CPU. On the GPU, single token prediction is ~10% better than Q4_0, batch mode (perplexity is about the same). * Adding Q6_K - scalar, AVX2, CUDA Performance is ~40% lower compared to Q4_K on the CPU. This is to be expected, considering that we are memory bound on the CPU and the 6-bit model is ~44% larger than the 4-bit. On the GPU, single token prediction is ~6% lower than Q4_0, batch mode (perplexity) is even closer (but still slower). * Adding Q5_K - scalar, AVX2, CUDA Performance is ~20% lower compared to Q4_K on the CPU. This is to be expected, considering that we are memory bound on the CPU and the 5-bit model is ~22% larger than the 4-bit. On the GPU, single token prediction is about the same as Q4_0 for both, single token and batch prediction. * Per convention, all QX_K quantizations use Q5_K for output.weight * Adding quantization mixes * Quantization mixes: didn't quite get what I wanted in the last commit * Q4_K dot product for ARM_NEON * Q6_K dot product for ARM_NEON * Q5_K dot product for ARM_NEON * Adding Q3_K dot for ARM_NEON It is 22% slower than Q4_K, despite the smaller model size. On x86_64, where we are memory bound, the Q3_K model is quite a bit faster than Q4_K. * A very slightly faster ARM_NEON Q3_K dot * Adding Q2_K - just CUDA for now Token prediction is pretty good - about 15.5 ms on a RTX 4080. Perplexity is about the same as Q4_K. * Adding scalar and AVX2 Q2_K dot * Adding ARM_NEON Q2_K dot About the same performance as Q4_K. * A slightly faster ARM_NEON Q2_K dot Single token prediction is now ~36 ms on M2 Max. The code is much simpler too. * Fixed bug in Q2_K CUDA dot product kernel Stranegly enough, for the few prompts I tried with the 7B model the responses looked perfectly reasonable. Only realized something is not quite right when I tried the larger models and started getting nonse back. In any case, Q2_K single token evaluation time on an RTX 4080 in a Ryzen7950X box iusing CUDA and model fully loaded on the GPU are ~15.5 ms for 7B, ~25.4 ms for 13B, and ~55.8 ms for 30B. The max number of layers that fit in VRAM for The 65B is 32. With that, we get ~330 ms per token, which is not that much faster than just running on the CPU (~470 ms per token). * Don't print zeros/NaNs when no count histogram has been collected * A 10% faster CUDA vector dot kernel for Q3_K Q3_K is now running at ~18.5 ms / token on CUDA, so the gap to Q4_0 is only 10%. It seems memory acccess pattern is more important for performance than the amount of computation the kernel does. * A slightly daster Q4_K AVX2 dot product For perplexity, where we are less memory bound, time per pass drops by ~5%. Barely measurable difference for single token prediction. * A slightly faster ARM_NEON A4_K dot product * Minor * Fix quantization error test We cannot possibly be expecting rmse < 0.002 for 2- and 3-bit quantization variants. * Fix docker build I have been sloppy with vector reinterpret casts on ARM_NEON. It seems clang is very forgiving in that regard. * Added forgotten ggml.o dependence on k_quants.h to the Makefile * Had unintentionally committed the Makefile with -Ofast enabled * ggml : rename k_quants -> ggml-quants-k, use lowercase in code --------- Co-authored-by: Iwan Kawrakow <iwan.kawrakow@gmail.com> Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
2023-06-05 19:56:18 +00:00
ggml : quantization refactoring (#3833) * ggml : factor all quantization code in ggml-quants ggml-ci * ggml-quants : fix Zig and Swift builds + quantize tool ggml-ci * quantize : --pure option for disabling k-quant mixtures --------- Co-authored-by: cebtenzzre <cebtenzzre@gmail.com>
2023-10-29 16:32:28 +00:00
vint8mf2_t y0 = __riscv_vle8_v_i8mf2(y[i].qs, vl);
vint8mf2_t y1 = __riscv_vle8_v_i8mf2(y[i].qs+16, vl);
ggml : add SOTA 2,3,4,5,6 bit k-quantizations (#1684) * Starting to add k-quantization to ggml I think it is better to have quantization separate from ggml. For now just adding the k-quants there, but it would be better to also factor out the existing ggml quantizations. * Adding Q3_K and Q8_K (de)-quantization * Q3_K now working on CUDA and AVX2/scalar CUDA is not ideal - ~50% slower than Q4_0 for single token prediction, about the same in batch mode (perplexity). CPU single token is ~55 ms (on Ryzen 7950X). * Some improvement for Q3_K on CUDA It is now ~22.5 ms/token on my GPU, so ~30% slower than Q4_0. * Some more CUDA optimizations for Q3_K Single token is now 20.5 ms/token (~20% slower than Q4_0). Perplexity is on par with Q4_0. * Adding Q4_K - scalar, AVX2, CUDA Performance is the same or perhaps very slightly better than Q4_0 on the CPU. On the GPU, single token prediction is ~10% better than Q4_0, batch mode (perplexity is about the same). * Adding Q6_K - scalar, AVX2, CUDA Performance is ~40% lower compared to Q4_K on the CPU. This is to be expected, considering that we are memory bound on the CPU and the 6-bit model is ~44% larger than the 4-bit. On the GPU, single token prediction is ~6% lower than Q4_0, batch mode (perplexity) is even closer (but still slower). * Adding Q5_K - scalar, AVX2, CUDA Performance is ~20% lower compared to Q4_K on the CPU. This is to be expected, considering that we are memory bound on the CPU and the 5-bit model is ~22% larger than the 4-bit. On the GPU, single token prediction is about the same as Q4_0 for both, single token and batch prediction. * Per convention, all QX_K quantizations use Q5_K for output.weight * Adding quantization mixes * Quantization mixes: didn't quite get what I wanted in the last commit * Q4_K dot product for ARM_NEON * Q6_K dot product for ARM_NEON * Q5_K dot product for ARM_NEON * Adding Q3_K dot for ARM_NEON It is 22% slower than Q4_K, despite the smaller model size. On x86_64, where we are memory bound, the Q3_K model is quite a bit faster than Q4_K. * A very slightly faster ARM_NEON Q3_K dot * Adding Q2_K - just CUDA for now Token prediction is pretty good - about 15.5 ms on a RTX 4080. Perplexity is about the same as Q4_K. * Adding scalar and AVX2 Q2_K dot * Adding ARM_NEON Q2_K dot About the same performance as Q4_K. * A slightly faster ARM_NEON Q2_K dot Single token prediction is now ~36 ms on M2 Max. The code is much simpler too. * Fixed bug in Q2_K CUDA dot product kernel Stranegly enough, for the few prompts I tried with the 7B model the responses looked perfectly reasonable. Only realized something is not quite right when I tried the larger models and started getting nonse back. In any case, Q2_K single token evaluation time on an RTX 4080 in a Ryzen7950X box iusing CUDA and model fully loaded on the GPU are ~15.5 ms for 7B, ~25.4 ms for 13B, and ~55.8 ms for 30B. The max number of layers that fit in VRAM for The 65B is 32. With that, we get ~330 ms per token, which is not that much faster than just running on the CPU (~470 ms per token). * Don't print zeros/NaNs when no count histogram has been collected * A 10% faster CUDA vector dot kernel for Q3_K Q3_K is now running at ~18.5 ms / token on CUDA, so the gap to Q4_0 is only 10%. It seems memory acccess pattern is more important for performance than the amount of computation the kernel does. * A slightly daster Q4_K AVX2 dot product For perplexity, where we are less memory bound, time per pass drops by ~5%. Barely measurable difference for single token prediction. * A slightly faster ARM_NEON A4_K dot product * Minor * Fix quantization error test We cannot possibly be expecting rmse < 0.002 for 2- and 3-bit quantization variants. * Fix docker build I have been sloppy with vector reinterpret casts on ARM_NEON. It seems clang is very forgiving in that regard. * Added forgotten ggml.o dependence on k_quants.h to the Makefile * Had unintentionally committed the Makefile with -Ofast enabled * ggml : rename k_quants -> ggml-quants-k, use lowercase in code --------- Co-authored-by: Iwan Kawrakow <iwan.kawrakow@gmail.com> Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
2023-06-05 19:56:18 +00:00
ggml : quantization refactoring (#3833) * ggml : factor all quantization code in ggml-quants ggml-ci * ggml-quants : fix Zig and Swift builds + quantize tool ggml-ci * quantize : --pure option for disabling k-quant mixtures --------- Co-authored-by: cebtenzzre <cebtenzzre@gmail.com>
2023-10-29 16:32:28 +00:00
// mask and store lower part of x, and then upper part
vuint8mf2_t x_a = __riscv_vand_vx_u8mf2(tx, 0x0F, vl);
vuint8mf2_t x_l = __riscv_vsrl_vx_u8mf2(tx, 0x04, vl);
ggml : add SOTA 2,3,4,5,6 bit k-quantizations (#1684) * Starting to add k-quantization to ggml I think it is better to have quantization separate from ggml. For now just adding the k-quants there, but it would be better to also factor out the existing ggml quantizations. * Adding Q3_K and Q8_K (de)-quantization * Q3_K now working on CUDA and AVX2/scalar CUDA is not ideal - ~50% slower than Q4_0 for single token prediction, about the same in batch mode (perplexity). CPU single token is ~55 ms (on Ryzen 7950X). * Some improvement for Q3_K on CUDA It is now ~22.5 ms/token on my GPU, so ~30% slower than Q4_0. * Some more CUDA optimizations for Q3_K Single token is now 20.5 ms/token (~20% slower than Q4_0). Perplexity is on par with Q4_0. * Adding Q4_K - scalar, AVX2, CUDA Performance is the same or perhaps very slightly better than Q4_0 on the CPU. On the GPU, single token prediction is ~10% better than Q4_0, batch mode (perplexity is about the same). * Adding Q6_K - scalar, AVX2, CUDA Performance is ~40% lower compared to Q4_K on the CPU. This is to be expected, considering that we are memory bound on the CPU and the 6-bit model is ~44% larger than the 4-bit. On the GPU, single token prediction is ~6% lower than Q4_0, batch mode (perplexity) is even closer (but still slower). * Adding Q5_K - scalar, AVX2, CUDA Performance is ~20% lower compared to Q4_K on the CPU. This is to be expected, considering that we are memory bound on the CPU and the 5-bit model is ~22% larger than the 4-bit. On the GPU, single token prediction is about the same as Q4_0 for both, single token and batch prediction. * Per convention, all QX_K quantizations use Q5_K for output.weight * Adding quantization mixes * Quantization mixes: didn't quite get what I wanted in the last commit * Q4_K dot product for ARM_NEON * Q6_K dot product for ARM_NEON * Q5_K dot product for ARM_NEON * Adding Q3_K dot for ARM_NEON It is 22% slower than Q4_K, despite the smaller model size. On x86_64, where we are memory bound, the Q3_K model is quite a bit faster than Q4_K. * A very slightly faster ARM_NEON Q3_K dot * Adding Q2_K - just CUDA for now Token prediction is pretty good - about 15.5 ms on a RTX 4080. Perplexity is about the same as Q4_K. * Adding scalar and AVX2 Q2_K dot * Adding ARM_NEON Q2_K dot About the same performance as Q4_K. * A slightly faster ARM_NEON Q2_K dot Single token prediction is now ~36 ms on M2 Max. The code is much simpler too. * Fixed bug in Q2_K CUDA dot product kernel Stranegly enough, for the few prompts I tried with the 7B model the responses looked perfectly reasonable. Only realized something is not quite right when I tried the larger models and started getting nonse back. In any case, Q2_K single token evaluation time on an RTX 4080 in a Ryzen7950X box iusing CUDA and model fully loaded on the GPU are ~15.5 ms for 7B, ~25.4 ms for 13B, and ~55.8 ms for 30B. The max number of layers that fit in VRAM for The 65B is 32. With that, we get ~330 ms per token, which is not that much faster than just running on the CPU (~470 ms per token). * Don't print zeros/NaNs when no count histogram has been collected * A 10% faster CUDA vector dot kernel for Q3_K Q3_K is now running at ~18.5 ms / token on CUDA, so the gap to Q4_0 is only 10%. It seems memory acccess pattern is more important for performance than the amount of computation the kernel does. * A slightly daster Q4_K AVX2 dot product For perplexity, where we are less memory bound, time per pass drops by ~5%. Barely measurable difference for single token prediction. * A slightly faster ARM_NEON A4_K dot product * Minor * Fix quantization error test We cannot possibly be expecting rmse < 0.002 for 2- and 3-bit quantization variants. * Fix docker build I have been sloppy with vector reinterpret casts on ARM_NEON. It seems clang is very forgiving in that regard. * Added forgotten ggml.o dependence on k_quants.h to the Makefile * Had unintentionally committed the Makefile with -Ofast enabled * ggml : rename k_quants -> ggml-quants-k, use lowercase in code --------- Co-authored-by: Iwan Kawrakow <iwan.kawrakow@gmail.com> Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
2023-06-05 19:56:18 +00:00
ggml : quantization refactoring (#3833) * ggml : factor all quantization code in ggml-quants ggml-ci * ggml-quants : fix Zig and Swift builds + quantize tool ggml-ci * quantize : --pure option for disabling k-quant mixtures --------- Co-authored-by: cebtenzzre <cebtenzzre@gmail.com>
2023-10-29 16:32:28 +00:00
vint8mf2_t x_ai = __riscv_vreinterpret_v_u8mf2_i8mf2(x_a);
vint8mf2_t x_li = __riscv_vreinterpret_v_u8mf2_i8mf2(x_l);
// subtract offset
vint8mf2_t v0 = __riscv_vsub_vx_i8mf2(x_ai, 8, vl);
vint8mf2_t v1 = __riscv_vsub_vx_i8mf2(x_li, 8, vl);
vint16m1_t vec_mul1 = __riscv_vwmul_vv_i16m1(v0, y0, vl);
vint16m1_t vec_mul2 = __riscv_vwmul_vv_i16m1(v1, y1, vl);
vint32m1_t vec_zero = __riscv_vmv_v_x_i32m1(0, vl);
vint32m1_t vs1 = __riscv_vwredsum_vs_i16m1_i32m1(vec_mul1, vec_zero, vl);
vint32m1_t vs2 = __riscv_vwredsum_vs_i16m1_i32m1(vec_mul2, vs1, vl);
int sumi = __riscv_vmv_x_s_i32m1_i32(vs2);
ggml : add SOTA 2,3,4,5,6 bit k-quantizations (#1684) * Starting to add k-quantization to ggml I think it is better to have quantization separate from ggml. For now just adding the k-quants there, but it would be better to also factor out the existing ggml quantizations. * Adding Q3_K and Q8_K (de)-quantization * Q3_K now working on CUDA and AVX2/scalar CUDA is not ideal - ~50% slower than Q4_0 for single token prediction, about the same in batch mode (perplexity). CPU single token is ~55 ms (on Ryzen 7950X). * Some improvement for Q3_K on CUDA It is now ~22.5 ms/token on my GPU, so ~30% slower than Q4_0. * Some more CUDA optimizations for Q3_K Single token is now 20.5 ms/token (~20% slower than Q4_0). Perplexity is on par with Q4_0. * Adding Q4_K - scalar, AVX2, CUDA Performance is the same or perhaps very slightly better than Q4_0 on the CPU. On the GPU, single token prediction is ~10% better than Q4_0, batch mode (perplexity is about the same). * Adding Q6_K - scalar, AVX2, CUDA Performance is ~40% lower compared to Q4_K on the CPU. This is to be expected, considering that we are memory bound on the CPU and the 6-bit model is ~44% larger than the 4-bit. On the GPU, single token prediction is ~6% lower than Q4_0, batch mode (perplexity) is even closer (but still slower). * Adding Q5_K - scalar, AVX2, CUDA Performance is ~20% lower compared to Q4_K on the CPU. This is to be expected, considering that we are memory bound on the CPU and the 5-bit model is ~22% larger than the 4-bit. On the GPU, single token prediction is about the same as Q4_0 for both, single token and batch prediction. * Per convention, all QX_K quantizations use Q5_K for output.weight * Adding quantization mixes * Quantization mixes: didn't quite get what I wanted in the last commit * Q4_K dot product for ARM_NEON * Q6_K dot product for ARM_NEON * Q5_K dot product for ARM_NEON * Adding Q3_K dot for ARM_NEON It is 22% slower than Q4_K, despite the smaller model size. On x86_64, where we are memory bound, the Q3_K model is quite a bit faster than Q4_K. * A very slightly faster ARM_NEON Q3_K dot * Adding Q2_K - just CUDA for now Token prediction is pretty good - about 15.5 ms on a RTX 4080. Perplexity is about the same as Q4_K. * Adding scalar and AVX2 Q2_K dot * Adding ARM_NEON Q2_K dot About the same performance as Q4_K. * A slightly faster ARM_NEON Q2_K dot Single token prediction is now ~36 ms on M2 Max. The code is much simpler too. * Fixed bug in Q2_K CUDA dot product kernel Stranegly enough, for the few prompts I tried with the 7B model the responses looked perfectly reasonable. Only realized something is not quite right when I tried the larger models and started getting nonse back. In any case, Q2_K single token evaluation time on an RTX 4080 in a Ryzen7950X box iusing CUDA and model fully loaded on the GPU are ~15.5 ms for 7B, ~25.4 ms for 13B, and ~55.8 ms for 30B. The max number of layers that fit in VRAM for The 65B is 32. With that, we get ~330 ms per token, which is not that much faster than just running on the CPU (~470 ms per token). * Don't print zeros/NaNs when no count histogram has been collected * A 10% faster CUDA vector dot kernel for Q3_K Q3_K is now running at ~18.5 ms / token on CUDA, so the gap to Q4_0 is only 10%. It seems memory acccess pattern is more important for performance than the amount of computation the kernel does. * A slightly daster Q4_K AVX2 dot product For perplexity, where we are less memory bound, time per pass drops by ~5%. Barely measurable difference for single token prediction. * A slightly faster ARM_NEON A4_K dot product * Minor * Fix quantization error test We cannot possibly be expecting rmse < 0.002 for 2- and 3-bit quantization variants. * Fix docker build I have been sloppy with vector reinterpret casts on ARM_NEON. It seems clang is very forgiving in that regard. * Added forgotten ggml.o dependence on k_quants.h to the Makefile * Had unintentionally committed the Makefile with -Ofast enabled * ggml : rename k_quants -> ggml-quants-k, use lowercase in code --------- Co-authored-by: Iwan Kawrakow <iwan.kawrakow@gmail.com> Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
2023-06-05 19:56:18 +00:00
ggml : move FP16 <-> FP32 code to ggml-impl.h (#3861) * ggml : move FP16 <-> FP32 stuff to ggml-impl.h ggml-ci * tests : fix ARM build * ggml : explicitly initialize deprecated type traits * ggml : add math.h to ggml-impl.h * ggml : remove duplicate static assert macros * ggml : prefix lookup tables with ggml_ ggml-ci * ggml-impl : move extern "C" to start of file
2023-10-30 17:19:15 +00:00
sumf += sumi*GGML_FP16_TO_FP32(x[i].d)*GGML_FP16_TO_FP32(y[i].d);
ggml : quantization refactoring (#3833) * ggml : factor all quantization code in ggml-quants ggml-ci * ggml-quants : fix Zig and Swift builds + quantize tool ggml-ci * quantize : --pure option for disabling k-quant mixtures --------- Co-authored-by: cebtenzzre <cebtenzzre@gmail.com>
2023-10-29 16:32:28 +00:00
}
*s = sumf;
k-quants : support for super-block size of 64 (#2001) * k_quants: WIP super-blocks with 64 weights * k_quants: WIP super-blocks with 64 weights Q6_K scalar and AVX2 works * k_quants: WIP super-blocks with 64 weights Q4_K scalar and AVX2 works * k_quants: WIP super-blocks with 64 weights Q2_K scalar and AVX2 works. Q2_K is way too slow (it is actually slower than the scalar implementation) * k_quants: WIP super-blocks with 64 weights Q3_K scalar and AVX2 works. * k_quants: WIP super-blocks with 64 weights Q5_K scalar and AVX2 works, and with that all k_quants are done on AVX2 and scalar * k_quants: WIP super-blocks with 64 weights Q6_K working on CUDA. Cannot make it run quite as gast as with super-blocks with 256 weigths: 8% slower on 4080, 20% slower on the 1660 (but there we fit 1 less layer on the GPU because pf the larger model size), so some fraction of these 20% is due to that, * k_quants: WIP super-blocks with 64 weights Q4_K working on CUDA. ~10% slower on GTX-1660, 16% slower on 4080. * k_quants: WIP super-blocks with 64 weights Q2_K working on CUDA. ~3% slower on GTX-1660, 10% slower on 4080. * k_quants: WIP super-blocks with 64 weights Q3_K working on CUDA. * k_quants: WIP super-blocks with 64 weights Q5_K working on CUDA, and with this CUDA is done. * k_quants: WIP super-blocks with 64 weights Q6_K working on ARM_NEON * k_quants: WIP super-blocks with 64 weights Q4_K working on ARM_NEON, but quite a bit slower than 256 weights * k_quants: WIP super-blocks with 64 weights Q2_K working on ARM_NEON, but quite a bit slower than 256 weights * k_quants: WIP super-blocks with 64 weights Q3_K working on ARM_NEON, but quite a bit slower than 256 weights. * k_quants: WIP super-blocks with 64 weights Q5_K working on ARM_NEON, but quite a bit slower than 256 weights. With that, we have full support for ARM_NEON, although performance is not quite there. * k_quants: WIP super-blocks with 64 weights Slightly more efficient Q3_K and Q5_K * k_quants: WIP super-blocks with 64 weights Another small improvement for Q3_K and Q5_K on ARM_NEON * k_quants: WIP super-blocks with 64 weights Yet another speedup for Q5_K on ARM_NEON. We are now within 10% of the QK_K = 256 version. * k_quants: WIP super-blocks with 64 weights * We are able to pass preprocessor macros to the Metal compiler * Q6_K works and is actually slightly more efficient than the QK_K = 256 version (25.2 ms vs 25.8 ms) * k_quants: WIP super-blocks with 64 weights Q4_K works on Metal and is actually slightly faster than QK_K = 256 (21.95 ms vs 24.0 ms). * k_quants: WIP super-blocks with 64 weights Q2_K works on Metal and is very slightly faster than QK_K = 256 (23.8 ms vs 24.2 ms). * k_quants: WIP super-blocks with 64 weights Q3_K works on Metal and is slightly faster than QK_K = 256 (26.6 ms vs 28.3 ms). * k_quants: WIP super-blocks with 64 weights Q5_K works on Metal and is slightly faster than QK_K = 256 (23.7 ms vs 26.3 ms). * k_quants: call them _K, not _k, also on Metal * k_quants: correctly define QK_K in llama.cpp * Fixed bug in q4_K quantization added with the 64-block addition * Simplify via lambda * k_quants: swicth Q3_K to 4-bit scales when QK_K = 64 Otherwise there isn't much benefit from this quantization type. There is some very slight loss in accuracy, but we reduce size by ~7%. E.g., for OpenLLaMA-3B, Q3_K_S perplexity is 8.6131 with 8-bit scales and 8.6352 with 4-bit, while file size decreases from 1.53G to 1.44G. * k_quants: switch Q4_K to 4-bit scales when QK_K = 64 Here the loss in accuracy is greater than for Q3_K, but the Q4_K points still move further to the left on the perplexity vs size curve. * k_quants: forgot to add the Metal changes in last commit * k_quants: change Q5_K to be type 0 when QK_K = 64 Still needs AVX2 implementation * k_quants: AVX2 implementation for new 64-weight Q5_K * k_quants: 10% faster ARM_NEON Q5_K dot product * k_quants: fixed issue caused by merging with master --------- Co-authored-by: Iwan Kawrakow <iwan.kawrakow@gmail.com>
2023-06-26 16:43:07 +00:00
#else
ggml : quantization refactoring (#3833) * ggml : factor all quantization code in ggml-quants ggml-ci * ggml-quants : fix Zig and Swift builds + quantize tool ggml-ci * quantize : --pure option for disabling k-quant mixtures --------- Co-authored-by: cebtenzzre <cebtenzzre@gmail.com>
2023-10-29 16:32:28 +00:00
// scalar
float sumf = 0.0;
for (int i = 0; i < nb; i++) {
int sumi = 0;
for (int j = 0; j < qk/2; ++j) {
const int v0 = (x[i].qs[j] & 0x0F) - 8;
const int v1 = (x[i].qs[j] >> 4) - 8;
sumi += (v0 * y[i].qs[j]) + (v1 * y[i].qs[j + qk/2]);
k-quants : support for super-block size of 64 (#2001) * k_quants: WIP super-blocks with 64 weights * k_quants: WIP super-blocks with 64 weights Q6_K scalar and AVX2 works * k_quants: WIP super-blocks with 64 weights Q4_K scalar and AVX2 works * k_quants: WIP super-blocks with 64 weights Q2_K scalar and AVX2 works. Q2_K is way too slow (it is actually slower than the scalar implementation) * k_quants: WIP super-blocks with 64 weights Q3_K scalar and AVX2 works. * k_quants: WIP super-blocks with 64 weights Q5_K scalar and AVX2 works, and with that all k_quants are done on AVX2 and scalar * k_quants: WIP super-blocks with 64 weights Q6_K working on CUDA. Cannot make it run quite as gast as with super-blocks with 256 weigths: 8% slower on 4080, 20% slower on the 1660 (but there we fit 1 less layer on the GPU because pf the larger model size), so some fraction of these 20% is due to that, * k_quants: WIP super-blocks with 64 weights Q4_K working on CUDA. ~10% slower on GTX-1660, 16% slower on 4080. * k_quants: WIP super-blocks with 64 weights Q2_K working on CUDA. ~3% slower on GTX-1660, 10% slower on 4080. * k_quants: WIP super-blocks with 64 weights Q3_K working on CUDA. * k_quants: WIP super-blocks with 64 weights Q5_K working on CUDA, and with this CUDA is done. * k_quants: WIP super-blocks with 64 weights Q6_K working on ARM_NEON * k_quants: WIP super-blocks with 64 weights Q4_K working on ARM_NEON, but quite a bit slower than 256 weights * k_quants: WIP super-blocks with 64 weights Q2_K working on ARM_NEON, but quite a bit slower than 256 weights * k_quants: WIP super-blocks with 64 weights Q3_K working on ARM_NEON, but quite a bit slower than 256 weights. * k_quants: WIP super-blocks with 64 weights Q5_K working on ARM_NEON, but quite a bit slower than 256 weights. With that, we have full support for ARM_NEON, although performance is not quite there. * k_quants: WIP super-blocks with 64 weights Slightly more efficient Q3_K and Q5_K * k_quants: WIP super-blocks with 64 weights Another small improvement for Q3_K and Q5_K on ARM_NEON * k_quants: WIP super-blocks with 64 weights Yet another speedup for Q5_K on ARM_NEON. We are now within 10% of the QK_K = 256 version. * k_quants: WIP super-blocks with 64 weights * We are able to pass preprocessor macros to the Metal compiler * Q6_K works and is actually slightly more efficient than the QK_K = 256 version (25.2 ms vs 25.8 ms) * k_quants: WIP super-blocks with 64 weights Q4_K works on Metal and is actually slightly faster than QK_K = 256 (21.95 ms vs 24.0 ms). * k_quants: WIP super-blocks with 64 weights Q2_K works on Metal and is very slightly faster than QK_K = 256 (23.8 ms vs 24.2 ms). * k_quants: WIP super-blocks with 64 weights Q3_K works on Metal and is slightly faster than QK_K = 256 (26.6 ms vs 28.3 ms). * k_quants: WIP super-blocks with 64 weights Q5_K works on Metal and is slightly faster than QK_K = 256 (23.7 ms vs 26.3 ms). * k_quants: call them _K, not _k, also on Metal * k_quants: correctly define QK_K in llama.cpp * Fixed bug in q4_K quantization added with the 64-block addition * Simplify via lambda * k_quants: swicth Q3_K to 4-bit scales when QK_K = 64 Otherwise there isn't much benefit from this quantization type. There is some very slight loss in accuracy, but we reduce size by ~7%. E.g., for OpenLLaMA-3B, Q3_K_S perplexity is 8.6131 with 8-bit scales and 8.6352 with 4-bit, while file size decreases from 1.53G to 1.44G. * k_quants: switch Q4_K to 4-bit scales when QK_K = 64 Here the loss in accuracy is greater than for Q3_K, but the Q4_K points still move further to the left on the perplexity vs size curve. * k_quants: forgot to add the Metal changes in last commit * k_quants: change Q5_K to be type 0 when QK_K = 64 Still needs AVX2 implementation * k_quants: AVX2 implementation for new 64-weight Q5_K * k_quants: 10% faster ARM_NEON Q5_K dot product * k_quants: fixed issue caused by merging with master --------- Co-authored-by: Iwan Kawrakow <iwan.kawrakow@gmail.com>
2023-06-26 16:43:07 +00:00
}
ggml : add SOTA 2,3,4,5,6 bit k-quantizations (#1684) * Starting to add k-quantization to ggml I think it is better to have quantization separate from ggml. For now just adding the k-quants there, but it would be better to also factor out the existing ggml quantizations. * Adding Q3_K and Q8_K (de)-quantization * Q3_K now working on CUDA and AVX2/scalar CUDA is not ideal - ~50% slower than Q4_0 for single token prediction, about the same in batch mode (perplexity). CPU single token is ~55 ms (on Ryzen 7950X). * Some improvement for Q3_K on CUDA It is now ~22.5 ms/token on my GPU, so ~30% slower than Q4_0. * Some more CUDA optimizations for Q3_K Single token is now 20.5 ms/token (~20% slower than Q4_0). Perplexity is on par with Q4_0. * Adding Q4_K - scalar, AVX2, CUDA Performance is the same or perhaps very slightly better than Q4_0 on the CPU. On the GPU, single token prediction is ~10% better than Q4_0, batch mode (perplexity is about the same). * Adding Q6_K - scalar, AVX2, CUDA Performance is ~40% lower compared to Q4_K on the CPU. This is to be expected, considering that we are memory bound on the CPU and the 6-bit model is ~44% larger than the 4-bit. On the GPU, single token prediction is ~6% lower than Q4_0, batch mode (perplexity) is even closer (but still slower). * Adding Q5_K - scalar, AVX2, CUDA Performance is ~20% lower compared to Q4_K on the CPU. This is to be expected, considering that we are memory bound on the CPU and the 5-bit model is ~22% larger than the 4-bit. On the GPU, single token prediction is about the same as Q4_0 for both, single token and batch prediction. * Per convention, all QX_K quantizations use Q5_K for output.weight * Adding quantization mixes * Quantization mixes: didn't quite get what I wanted in the last commit * Q4_K dot product for ARM_NEON * Q6_K dot product for ARM_NEON * Q5_K dot product for ARM_NEON * Adding Q3_K dot for ARM_NEON It is 22% slower than Q4_K, despite the smaller model size. On x86_64, where we are memory bound, the Q3_K model is quite a bit faster than Q4_K. * A very slightly faster ARM_NEON Q3_K dot * Adding Q2_K - just CUDA for now Token prediction is pretty good - about 15.5 ms on a RTX 4080. Perplexity is about the same as Q4_K. * Adding scalar and AVX2 Q2_K dot * Adding ARM_NEON Q2_K dot About the same performance as Q4_K. * A slightly faster ARM_NEON Q2_K dot Single token prediction is now ~36 ms on M2 Max. The code is much simpler too. * Fixed bug in Q2_K CUDA dot product kernel Stranegly enough, for the few prompts I tried with the 7B model the responses looked perfectly reasonable. Only realized something is not quite right when I tried the larger models and started getting nonse back. In any case, Q2_K single token evaluation time on an RTX 4080 in a Ryzen7950X box iusing CUDA and model fully loaded on the GPU are ~15.5 ms for 7B, ~25.4 ms for 13B, and ~55.8 ms for 30B. The max number of layers that fit in VRAM for The 65B is 32. With that, we get ~330 ms per token, which is not that much faster than just running on the CPU (~470 ms per token). * Don't print zeros/NaNs when no count histogram has been collected * A 10% faster CUDA vector dot kernel for Q3_K Q3_K is now running at ~18.5 ms / token on CUDA, so the gap to Q4_0 is only 10%. It seems memory acccess pattern is more important for performance than the amount of computation the kernel does. * A slightly daster Q4_K AVX2 dot product For perplexity, where we are less memory bound, time per pass drops by ~5%. Barely measurable difference for single token prediction. * A slightly faster ARM_NEON A4_K dot product * Minor * Fix quantization error test We cannot possibly be expecting rmse < 0.002 for 2- and 3-bit quantization variants. * Fix docker build I have been sloppy with vector reinterpret casts on ARM_NEON. It seems clang is very forgiving in that regard. * Added forgotten ggml.o dependence on k_quants.h to the Makefile * Had unintentionally committed the Makefile with -Ofast enabled * ggml : rename k_quants -> ggml-quants-k, use lowercase in code --------- Co-authored-by: Iwan Kawrakow <iwan.kawrakow@gmail.com> Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
2023-06-05 19:56:18 +00:00
ggml : move FP16 <-> FP32 code to ggml-impl.h (#3861) * ggml : move FP16 <-> FP32 stuff to ggml-impl.h ggml-ci * tests : fix ARM build * ggml : explicitly initialize deprecated type traits * ggml : add math.h to ggml-impl.h * ggml : remove duplicate static assert macros * ggml : prefix lookup tables with ggml_ ggml-ci * ggml-impl : move extern "C" to start of file
2023-10-30 17:19:15 +00:00
sumf += sumi*GGML_FP16_TO_FP32(x[i].d)*GGML_FP16_TO_FP32(y[i].d);
ggml : add SOTA 2,3,4,5,6 bit k-quantizations (#1684) * Starting to add k-quantization to ggml I think it is better to have quantization separate from ggml. For now just adding the k-quants there, but it would be better to also factor out the existing ggml quantizations. * Adding Q3_K and Q8_K (de)-quantization * Q3_K now working on CUDA and AVX2/scalar CUDA is not ideal - ~50% slower than Q4_0 for single token prediction, about the same in batch mode (perplexity). CPU single token is ~55 ms (on Ryzen 7950X). * Some improvement for Q3_K on CUDA It is now ~22.5 ms/token on my GPU, so ~30% slower than Q4_0. * Some more CUDA optimizations for Q3_K Single token is now 20.5 ms/token (~20% slower than Q4_0). Perplexity is on par with Q4_0. * Adding Q4_K - scalar, AVX2, CUDA Performance is the same or perhaps very slightly better than Q4_0 on the CPU. On the GPU, single token prediction is ~10% better than Q4_0, batch mode (perplexity is about the same). * Adding Q6_K - scalar, AVX2, CUDA Performance is ~40% lower compared to Q4_K on the CPU. This is to be expected, considering that we are memory bound on the CPU and the 6-bit model is ~44% larger than the 4-bit. On the GPU, single token prediction is ~6% lower than Q4_0, batch mode (perplexity) is even closer (but still slower). * Adding Q5_K - scalar, AVX2, CUDA Performance is ~20% lower compared to Q4_K on the CPU. This is to be expected, considering that we are memory bound on the CPU and the 5-bit model is ~22% larger than the 4-bit. On the GPU, single token prediction is about the same as Q4_0 for both, single token and batch prediction. * Per convention, all QX_K quantizations use Q5_K for output.weight * Adding quantization mixes * Quantization mixes: didn't quite get what I wanted in the last commit * Q4_K dot product for ARM_NEON * Q6_K dot product for ARM_NEON * Q5_K dot product for ARM_NEON * Adding Q3_K dot for ARM_NEON It is 22% slower than Q4_K, despite the smaller model size. On x86_64, where we are memory bound, the Q3_K model is quite a bit faster than Q4_K. * A very slightly faster ARM_NEON Q3_K dot * Adding Q2_K - just CUDA for now Token prediction is pretty good - about 15.5 ms on a RTX 4080. Perplexity is about the same as Q4_K. * Adding scalar and AVX2 Q2_K dot * Adding ARM_NEON Q2_K dot About the same performance as Q4_K. * A slightly faster ARM_NEON Q2_K dot Single token prediction is now ~36 ms on M2 Max. The code is much simpler too. * Fixed bug in Q2_K CUDA dot product kernel Stranegly enough, for the few prompts I tried with the 7B model the responses looked perfectly reasonable. Only realized something is not quite right when I tried the larger models and started getting nonse back. In any case, Q2_K single token evaluation time on an RTX 4080 in a Ryzen7950X box iusing CUDA and model fully loaded on the GPU are ~15.5 ms for 7B, ~25.4 ms for 13B, and ~55.8 ms for 30B. The max number of layers that fit in VRAM for The 65B is 32. With that, we get ~330 ms per token, which is not that much faster than just running on the CPU (~470 ms per token). * Don't print zeros/NaNs when no count histogram has been collected * A 10% faster CUDA vector dot kernel for Q3_K Q3_K is now running at ~18.5 ms / token on CUDA, so the gap to Q4_0 is only 10%. It seems memory acccess pattern is more important for performance than the amount of computation the kernel does. * A slightly daster Q4_K AVX2 dot product For perplexity, where we are less memory bound, time per pass drops by ~5%. Barely measurable difference for single token prediction. * A slightly faster ARM_NEON A4_K dot product * Minor * Fix quantization error test We cannot possibly be expecting rmse < 0.002 for 2- and 3-bit quantization variants. * Fix docker build I have been sloppy with vector reinterpret casts on ARM_NEON. It seems clang is very forgiving in that regard. * Added forgotten ggml.o dependence on k_quants.h to the Makefile * Had unintentionally committed the Makefile with -Ofast enabled * ggml : rename k_quants -> ggml-quants-k, use lowercase in code --------- Co-authored-by: Iwan Kawrakow <iwan.kawrakow@gmail.com> Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
2023-06-05 19:56:18 +00:00
}
ggml : quantization refactoring (#3833) * ggml : factor all quantization code in ggml-quants ggml-ci * ggml-quants : fix Zig and Swift builds + quantize tool ggml-ci * quantize : --pure option for disabling k-quant mixtures --------- Co-authored-by: cebtenzzre <cebtenzzre@gmail.com>
2023-10-29 16:32:28 +00:00
*s = sumf;
#endif
ggml : add SOTA 2,3,4,5,6 bit k-quantizations (#1684) * Starting to add k-quantization to ggml I think it is better to have quantization separate from ggml. For now just adding the k-quants there, but it would be better to also factor out the existing ggml quantizations. * Adding Q3_K and Q8_K (de)-quantization * Q3_K now working on CUDA and AVX2/scalar CUDA is not ideal - ~50% slower than Q4_0 for single token prediction, about the same in batch mode (perplexity). CPU single token is ~55 ms (on Ryzen 7950X). * Some improvement for Q3_K on CUDA It is now ~22.5 ms/token on my GPU, so ~30% slower than Q4_0. * Some more CUDA optimizations for Q3_K Single token is now 20.5 ms/token (~20% slower than Q4_0). Perplexity is on par with Q4_0. * Adding Q4_K - scalar, AVX2, CUDA Performance is the same or perhaps very slightly better than Q4_0 on the CPU. On the GPU, single token prediction is ~10% better than Q4_0, batch mode (perplexity is about the same). * Adding Q6_K - scalar, AVX2, CUDA Performance is ~40% lower compared to Q4_K on the CPU. This is to be expected, considering that we are memory bound on the CPU and the 6-bit model is ~44% larger than the 4-bit. On the GPU, single token prediction is ~6% lower than Q4_0, batch mode (perplexity) is even closer (but still slower). * Adding Q5_K - scalar, AVX2, CUDA Performance is ~20% lower compared to Q4_K on the CPU. This is to be expected, considering that we are memory bound on the CPU and the 5-bit model is ~22% larger than the 4-bit. On the GPU, single token prediction is about the same as Q4_0 for both, single token and batch prediction. * Per convention, all QX_K quantizations use Q5_K for output.weight * Adding quantization mixes * Quantization mixes: didn't quite get what I wanted in the last commit * Q4_K dot product for ARM_NEON * Q6_K dot product for ARM_NEON * Q5_K dot product for ARM_NEON * Adding Q3_K dot for ARM_NEON It is 22% slower than Q4_K, despite the smaller model size. On x86_64, where we are memory bound, the Q3_K model is quite a bit faster than Q4_K. * A very slightly faster ARM_NEON Q3_K dot * Adding Q2_K - just CUDA for now Token prediction is pretty good - about 15.5 ms on a RTX 4080. Perplexity is about the same as Q4_K. * Adding scalar and AVX2 Q2_K dot * Adding ARM_NEON Q2_K dot About the same performance as Q4_K. * A slightly faster ARM_NEON Q2_K dot Single token prediction is now ~36 ms on M2 Max. The code is much simpler too. * Fixed bug in Q2_K CUDA dot product kernel Stranegly enough, for the few prompts I tried with the 7B model the responses looked perfectly reasonable. Only realized something is not quite right when I tried the larger models and started getting nonse back. In any case, Q2_K single token evaluation time on an RTX 4080 in a Ryzen7950X box iusing CUDA and model fully loaded on the GPU are ~15.5 ms for 7B, ~25.4 ms for 13B, and ~55.8 ms for 30B. The max number of layers that fit in VRAM for The 65B is 32. With that, we get ~330 ms per token, which is not that much faster than just running on the CPU (~470 ms per token). * Don't print zeros/NaNs when no count histogram has been collected * A 10% faster CUDA vector dot kernel for Q3_K Q3_K is now running at ~18.5 ms / token on CUDA, so the gap to Q4_0 is only 10%. It seems memory acccess pattern is more important for performance than the amount of computation the kernel does. * A slightly daster Q4_K AVX2 dot product For perplexity, where we are less memory bound, time per pass drops by ~5%. Barely measurable difference for single token prediction. * A slightly faster ARM_NEON A4_K dot product * Minor * Fix quantization error test We cannot possibly be expecting rmse < 0.002 for 2- and 3-bit quantization variants. * Fix docker build I have been sloppy with vector reinterpret casts on ARM_NEON. It seems clang is very forgiving in that regard. * Added forgotten ggml.o dependence on k_quants.h to the Makefile * Had unintentionally committed the Makefile with -Ofast enabled * ggml : rename k_quants -> ggml-quants-k, use lowercase in code --------- Co-authored-by: Iwan Kawrakow <iwan.kawrakow@gmail.com> Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
2023-06-05 19:56:18 +00:00
}
ggml : quantization refactoring (#3833) * ggml : factor all quantization code in ggml-quants ggml-ci * ggml-quants : fix Zig and Swift builds + quantize tool ggml-ci * quantize : --pure option for disabling k-quant mixtures --------- Co-authored-by: cebtenzzre <cebtenzzre@gmail.com>
2023-10-29 16:32:28 +00:00
void ggml_vec_dot_q4_1_q8_1(const int n, float * restrict s, const void * restrict vx, const void * restrict vy) {
const int qk = QK8_1;
const int nb = n / qk;
ggml : add SOTA 2,3,4,5,6 bit k-quantizations (#1684) * Starting to add k-quantization to ggml I think it is better to have quantization separate from ggml. For now just adding the k-quants there, but it would be better to also factor out the existing ggml quantizations. * Adding Q3_K and Q8_K (de)-quantization * Q3_K now working on CUDA and AVX2/scalar CUDA is not ideal - ~50% slower than Q4_0 for single token prediction, about the same in batch mode (perplexity). CPU single token is ~55 ms (on Ryzen 7950X). * Some improvement for Q3_K on CUDA It is now ~22.5 ms/token on my GPU, so ~30% slower than Q4_0. * Some more CUDA optimizations for Q3_K Single token is now 20.5 ms/token (~20% slower than Q4_0). Perplexity is on par with Q4_0. * Adding Q4_K - scalar, AVX2, CUDA Performance is the same or perhaps very slightly better than Q4_0 on the CPU. On the GPU, single token prediction is ~10% better than Q4_0, batch mode (perplexity is about the same). * Adding Q6_K - scalar, AVX2, CUDA Performance is ~40% lower compared to Q4_K on the CPU. This is to be expected, considering that we are memory bound on the CPU and the 6-bit model is ~44% larger than the 4-bit. On the GPU, single token prediction is ~6% lower than Q4_0, batch mode (perplexity) is even closer (but still slower). * Adding Q5_K - scalar, AVX2, CUDA Performance is ~20% lower compared to Q4_K on the CPU. This is to be expected, considering that we are memory bound on the CPU and the 5-bit model is ~22% larger than the 4-bit. On the GPU, single token prediction is about the same as Q4_0 for both, single token and batch prediction. * Per convention, all QX_K quantizations use Q5_K for output.weight * Adding quantization mixes * Quantization mixes: didn't quite get what I wanted in the last commit * Q4_K dot product for ARM_NEON * Q6_K dot product for ARM_NEON * Q5_K dot product for ARM_NEON * Adding Q3_K dot for ARM_NEON It is 22% slower than Q4_K, despite the smaller model size. On x86_64, where we are memory bound, the Q3_K model is quite a bit faster than Q4_K. * A very slightly faster ARM_NEON Q3_K dot * Adding Q2_K - just CUDA for now Token prediction is pretty good - about 15.5 ms on a RTX 4080. Perplexity is about the same as Q4_K. * Adding scalar and AVX2 Q2_K dot * Adding ARM_NEON Q2_K dot About the same performance as Q4_K. * A slightly faster ARM_NEON Q2_K dot Single token prediction is now ~36 ms on M2 Max. The code is much simpler too. * Fixed bug in Q2_K CUDA dot product kernel Stranegly enough, for the few prompts I tried with the 7B model the responses looked perfectly reasonable. Only realized something is not quite right when I tried the larger models and started getting nonse back. In any case, Q2_K single token evaluation time on an RTX 4080 in a Ryzen7950X box iusing CUDA and model fully loaded on the GPU are ~15.5 ms for 7B, ~25.4 ms for 13B, and ~55.8 ms for 30B. The max number of layers that fit in VRAM for The 65B is 32. With that, we get ~330 ms per token, which is not that much faster than just running on the CPU (~470 ms per token). * Don't print zeros/NaNs when no count histogram has been collected * A 10% faster CUDA vector dot kernel for Q3_K Q3_K is now running at ~18.5 ms / token on CUDA, so the gap to Q4_0 is only 10%. It seems memory acccess pattern is more important for performance than the amount of computation the kernel does. * A slightly daster Q4_K AVX2 dot product For perplexity, where we are less memory bound, time per pass drops by ~5%. Barely measurable difference for single token prediction. * A slightly faster ARM_NEON A4_K dot product * Minor * Fix quantization error test We cannot possibly be expecting rmse < 0.002 for 2- and 3-bit quantization variants. * Fix docker build I have been sloppy with vector reinterpret casts on ARM_NEON. It seems clang is very forgiving in that regard. * Added forgotten ggml.o dependence on k_quants.h to the Makefile * Had unintentionally committed the Makefile with -Ofast enabled * ggml : rename k_quants -> ggml-quants-k, use lowercase in code --------- Co-authored-by: Iwan Kawrakow <iwan.kawrakow@gmail.com> Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
2023-06-05 19:56:18 +00:00
ggml : quantization refactoring (#3833) * ggml : factor all quantization code in ggml-quants ggml-ci * ggml-quants : fix Zig and Swift builds + quantize tool ggml-ci * quantize : --pure option for disabling k-quant mixtures --------- Co-authored-by: cebtenzzre <cebtenzzre@gmail.com>
2023-10-29 16:32:28 +00:00
assert(n % qk == 0);
ggml : add SOTA 2,3,4,5,6 bit k-quantizations (#1684) * Starting to add k-quantization to ggml I think it is better to have quantization separate from ggml. For now just adding the k-quants there, but it would be better to also factor out the existing ggml quantizations. * Adding Q3_K and Q8_K (de)-quantization * Q3_K now working on CUDA and AVX2/scalar CUDA is not ideal - ~50% slower than Q4_0 for single token prediction, about the same in batch mode (perplexity). CPU single token is ~55 ms (on Ryzen 7950X). * Some improvement for Q3_K on CUDA It is now ~22.5 ms/token on my GPU, so ~30% slower than Q4_0. * Some more CUDA optimizations for Q3_K Single token is now 20.5 ms/token (~20% slower than Q4_0). Perplexity is on par with Q4_0. * Adding Q4_K - scalar, AVX2, CUDA Performance is the same or perhaps very slightly better than Q4_0 on the CPU. On the GPU, single token prediction is ~10% better than Q4_0, batch mode (perplexity is about the same). * Adding Q6_K - scalar, AVX2, CUDA Performance is ~40% lower compared to Q4_K on the CPU. This is to be expected, considering that we are memory bound on the CPU and the 6-bit model is ~44% larger than the 4-bit. On the GPU, single token prediction is ~6% lower than Q4_0, batch mode (perplexity) is even closer (but still slower). * Adding Q5_K - scalar, AVX2, CUDA Performance is ~20% lower compared to Q4_K on the CPU. This is to be expected, considering that we are memory bound on the CPU and the 5-bit model is ~22% larger than the 4-bit. On the GPU, single token prediction is about the same as Q4_0 for both, single token and batch prediction. * Per convention, all QX_K quantizations use Q5_K for output.weight * Adding quantization mixes * Quantization mixes: didn't quite get what I wanted in the last commit * Q4_K dot product for ARM_NEON * Q6_K dot product for ARM_NEON * Q5_K dot product for ARM_NEON * Adding Q3_K dot for ARM_NEON It is 22% slower than Q4_K, despite the smaller model size. On x86_64, where we are memory bound, the Q3_K model is quite a bit faster than Q4_K. * A very slightly faster ARM_NEON Q3_K dot * Adding Q2_K - just CUDA for now Token prediction is pretty good - about 15.5 ms on a RTX 4080. Perplexity is about the same as Q4_K. * Adding scalar and AVX2 Q2_K dot * Adding ARM_NEON Q2_K dot About the same performance as Q4_K. * A slightly faster ARM_NEON Q2_K dot Single token prediction is now ~36 ms on M2 Max. The code is much simpler too. * Fixed bug in Q2_K CUDA dot product kernel Stranegly enough, for the few prompts I tried with the 7B model the responses looked perfectly reasonable. Only realized something is not quite right when I tried the larger models and started getting nonse back. In any case, Q2_K single token evaluation time on an RTX 4080 in a Ryzen7950X box iusing CUDA and model fully loaded on the GPU are ~15.5 ms for 7B, ~25.4 ms for 13B, and ~55.8 ms for 30B. The max number of layers that fit in VRAM for The 65B is 32. With that, we get ~330 ms per token, which is not that much faster than just running on the CPU (~470 ms per token). * Don't print zeros/NaNs when no count histogram has been collected * A 10% faster CUDA vector dot kernel for Q3_K Q3_K is now running at ~18.5 ms / token on CUDA, so the gap to Q4_0 is only 10%. It seems memory acccess pattern is more important for performance than the amount of computation the kernel does. * A slightly daster Q4_K AVX2 dot product For perplexity, where we are less memory bound, time per pass drops by ~5%. Barely measurable difference for single token prediction. * A slightly faster ARM_NEON A4_K dot product * Minor * Fix quantization error test We cannot possibly be expecting rmse < 0.002 for 2- and 3-bit quantization variants. * Fix docker build I have been sloppy with vector reinterpret casts on ARM_NEON. It seems clang is very forgiving in that regard. * Added forgotten ggml.o dependence on k_quants.h to the Makefile * Had unintentionally committed the Makefile with -Ofast enabled * ggml : rename k_quants -> ggml-quants-k, use lowercase in code --------- Co-authored-by: Iwan Kawrakow <iwan.kawrakow@gmail.com> Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
2023-06-05 19:56:18 +00:00
ggml : quantization refactoring (#3833) * ggml : factor all quantization code in ggml-quants ggml-ci * ggml-quants : fix Zig and Swift builds + quantize tool ggml-ci * quantize : --pure option for disabling k-quant mixtures --------- Co-authored-by: cebtenzzre <cebtenzzre@gmail.com>
2023-10-29 16:32:28 +00:00
const block_q4_1 * restrict x = vx;
const block_q8_1 * restrict y = vy;
ggml : add SOTA 2,3,4,5,6 bit k-quantizations (#1684) * Starting to add k-quantization to ggml I think it is better to have quantization separate from ggml. For now just adding the k-quants there, but it would be better to also factor out the existing ggml quantizations. * Adding Q3_K and Q8_K (de)-quantization * Q3_K now working on CUDA and AVX2/scalar CUDA is not ideal - ~50% slower than Q4_0 for single token prediction, about the same in batch mode (perplexity). CPU single token is ~55 ms (on Ryzen 7950X). * Some improvement for Q3_K on CUDA It is now ~22.5 ms/token on my GPU, so ~30% slower than Q4_0. * Some more CUDA optimizations for Q3_K Single token is now 20.5 ms/token (~20% slower than Q4_0). Perplexity is on par with Q4_0. * Adding Q4_K - scalar, AVX2, CUDA Performance is the same or perhaps very slightly better than Q4_0 on the CPU. On the GPU, single token prediction is ~10% better than Q4_0, batch mode (perplexity is about the same). * Adding Q6_K - scalar, AVX2, CUDA Performance is ~40% lower compared to Q4_K on the CPU. This is to be expected, considering that we are memory bound on the CPU and the 6-bit model is ~44% larger than the 4-bit. On the GPU, single token prediction is ~6% lower than Q4_0, batch mode (perplexity) is even closer (but still slower). * Adding Q5_K - scalar, AVX2, CUDA Performance is ~20% lower compared to Q4_K on the CPU. This is to be expected, considering that we are memory bound on the CPU and the 5-bit model is ~22% larger than the 4-bit. On the GPU, single token prediction is about the same as Q4_0 for both, single token and batch prediction. * Per convention, all QX_K quantizations use Q5_K for output.weight * Adding quantization mixes * Quantization mixes: didn't quite get what I wanted in the last commit * Q4_K dot product for ARM_NEON * Q6_K dot product for ARM_NEON * Q5_K dot product for ARM_NEON * Adding Q3_K dot for ARM_NEON It is 22% slower than Q4_K, despite the smaller model size. On x86_64, where we are memory bound, the Q3_K model is quite a bit faster than Q4_K. * A very slightly faster ARM_NEON Q3_K dot * Adding Q2_K - just CUDA for now Token prediction is pretty good - about 15.5 ms on a RTX 4080. Perplexity is about the same as Q4_K. * Adding scalar and AVX2 Q2_K dot * Adding ARM_NEON Q2_K dot About the same performance as Q4_K. * A slightly faster ARM_NEON Q2_K dot Single token prediction is now ~36 ms on M2 Max. The code is much simpler too. * Fixed bug in Q2_K CUDA dot product kernel Stranegly enough, for the few prompts I tried with the 7B model the responses looked perfectly reasonable. Only realized something is not quite right when I tried the larger models and started getting nonse back. In any case, Q2_K single token evaluation time on an RTX 4080 in a Ryzen7950X box iusing CUDA and model fully loaded on the GPU are ~15.5 ms for 7B, ~25.4 ms for 13B, and ~55.8 ms for 30B. The max number of layers that fit in VRAM for The 65B is 32. With that, we get ~330 ms per token, which is not that much faster than just running on the CPU (~470 ms per token). * Don't print zeros/NaNs when no count histogram has been collected * A 10% faster CUDA vector dot kernel for Q3_K Q3_K is now running at ~18.5 ms / token on CUDA, so the gap to Q4_0 is only 10%. It seems memory acccess pattern is more important for performance than the amount of computation the kernel does. * A slightly daster Q4_K AVX2 dot product For perplexity, where we are less memory bound, time per pass drops by ~5%. Barely measurable difference for single token prediction. * A slightly faster ARM_NEON A4_K dot product * Minor * Fix quantization error test We cannot possibly be expecting rmse < 0.002 for 2- and 3-bit quantization variants. * Fix docker build I have been sloppy with vector reinterpret casts on ARM_NEON. It seems clang is very forgiving in that regard. * Added forgotten ggml.o dependence on k_quants.h to the Makefile * Had unintentionally committed the Makefile with -Ofast enabled * ggml : rename k_quants -> ggml-quants-k, use lowercase in code --------- Co-authored-by: Iwan Kawrakow <iwan.kawrakow@gmail.com> Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
2023-06-05 19:56:18 +00:00
ggml : quantization refactoring (#3833) * ggml : factor all quantization code in ggml-quants ggml-ci * ggml-quants : fix Zig and Swift builds + quantize tool ggml-ci * quantize : --pure option for disabling k-quant mixtures --------- Co-authored-by: cebtenzzre <cebtenzzre@gmail.com>
2023-10-29 16:32:28 +00:00
// TODO: add WASM SIMD
#if defined(__ARM_NEON)
float32x4_t sumv0 = vdupq_n_f32(0.0f);
float32x4_t sumv1 = vdupq_n_f32(0.0f);
ggml : add SOTA 2,3,4,5,6 bit k-quantizations (#1684) * Starting to add k-quantization to ggml I think it is better to have quantization separate from ggml. For now just adding the k-quants there, but it would be better to also factor out the existing ggml quantizations. * Adding Q3_K and Q8_K (de)-quantization * Q3_K now working on CUDA and AVX2/scalar CUDA is not ideal - ~50% slower than Q4_0 for single token prediction, about the same in batch mode (perplexity). CPU single token is ~55 ms (on Ryzen 7950X). * Some improvement for Q3_K on CUDA It is now ~22.5 ms/token on my GPU, so ~30% slower than Q4_0. * Some more CUDA optimizations for Q3_K Single token is now 20.5 ms/token (~20% slower than Q4_0). Perplexity is on par with Q4_0. * Adding Q4_K - scalar, AVX2, CUDA Performance is the same or perhaps very slightly better than Q4_0 on the CPU. On the GPU, single token prediction is ~10% better than Q4_0, batch mode (perplexity is about the same). * Adding Q6_K - scalar, AVX2, CUDA Performance is ~40% lower compared to Q4_K on the CPU. This is to be expected, considering that we are memory bound on the CPU and the 6-bit model is ~44% larger than the 4-bit. On the GPU, single token prediction is ~6% lower than Q4_0, batch mode (perplexity) is even closer (but still slower). * Adding Q5_K - scalar, AVX2, CUDA Performance is ~20% lower compared to Q4_K on the CPU. This is to be expected, considering that we are memory bound on the CPU and the 5-bit model is ~22% larger than the 4-bit. On the GPU, single token prediction is about the same as Q4_0 for both, single token and batch prediction. * Per convention, all QX_K quantizations use Q5_K for output.weight * Adding quantization mixes * Quantization mixes: didn't quite get what I wanted in the last commit * Q4_K dot product for ARM_NEON * Q6_K dot product for ARM_NEON * Q5_K dot product for ARM_NEON * Adding Q3_K dot for ARM_NEON It is 22% slower than Q4_K, despite the smaller model size. On x86_64, where we are memory bound, the Q3_K model is quite a bit faster than Q4_K. * A very slightly faster ARM_NEON Q3_K dot * Adding Q2_K - just CUDA for now Token prediction is pretty good - about 15.5 ms on a RTX 4080. Perplexity is about the same as Q4_K. * Adding scalar and AVX2 Q2_K dot * Adding ARM_NEON Q2_K dot About the same performance as Q4_K. * A slightly faster ARM_NEON Q2_K dot Single token prediction is now ~36 ms on M2 Max. The code is much simpler too. * Fixed bug in Q2_K CUDA dot product kernel Stranegly enough, for the few prompts I tried with the 7B model the responses looked perfectly reasonable. Only realized something is not quite right when I tried the larger models and started getting nonse back. In any case, Q2_K single token evaluation time on an RTX 4080 in a Ryzen7950X box iusing CUDA and model fully loaded on the GPU are ~15.5 ms for 7B, ~25.4 ms for 13B, and ~55.8 ms for 30B. The max number of layers that fit in VRAM for The 65B is 32. With that, we get ~330 ms per token, which is not that much faster than just running on the CPU (~470 ms per token). * Don't print zeros/NaNs when no count histogram has been collected * A 10% faster CUDA vector dot kernel for Q3_K Q3_K is now running at ~18.5 ms / token on CUDA, so the gap to Q4_0 is only 10%. It seems memory acccess pattern is more important for performance than the amount of computation the kernel does. * A slightly daster Q4_K AVX2 dot product For perplexity, where we are less memory bound, time per pass drops by ~5%. Barely measurable difference for single token prediction. * A slightly faster ARM_NEON A4_K dot product * Minor * Fix quantization error test We cannot possibly be expecting rmse < 0.002 for 2- and 3-bit quantization variants. * Fix docker build I have been sloppy with vector reinterpret casts on ARM_NEON. It seems clang is very forgiving in that regard. * Added forgotten ggml.o dependence on k_quants.h to the Makefile * Had unintentionally committed the Makefile with -Ofast enabled * ggml : rename k_quants -> ggml-quants-k, use lowercase in code --------- Co-authored-by: Iwan Kawrakow <iwan.kawrakow@gmail.com> Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
2023-06-05 19:56:18 +00:00
ggml : quantization refactoring (#3833) * ggml : factor all quantization code in ggml-quants ggml-ci * ggml-quants : fix Zig and Swift builds + quantize tool ggml-ci * quantize : --pure option for disabling k-quant mixtures --------- Co-authored-by: cebtenzzre <cebtenzzre@gmail.com>
2023-10-29 16:32:28 +00:00
float summs = 0;
ggml : add SOTA 2,3,4,5,6 bit k-quantizations (#1684) * Starting to add k-quantization to ggml I think it is better to have quantization separate from ggml. For now just adding the k-quants there, but it would be better to also factor out the existing ggml quantizations. * Adding Q3_K and Q8_K (de)-quantization * Q3_K now working on CUDA and AVX2/scalar CUDA is not ideal - ~50% slower than Q4_0 for single token prediction, about the same in batch mode (perplexity). CPU single token is ~55 ms (on Ryzen 7950X). * Some improvement for Q3_K on CUDA It is now ~22.5 ms/token on my GPU, so ~30% slower than Q4_0. * Some more CUDA optimizations for Q3_K Single token is now 20.5 ms/token (~20% slower than Q4_0). Perplexity is on par with Q4_0. * Adding Q4_K - scalar, AVX2, CUDA Performance is the same or perhaps very slightly better than Q4_0 on the CPU. On the GPU, single token prediction is ~10% better than Q4_0, batch mode (perplexity is about the same). * Adding Q6_K - scalar, AVX2, CUDA Performance is ~40% lower compared to Q4_K on the CPU. This is to be expected, considering that we are memory bound on the CPU and the 6-bit model is ~44% larger than the 4-bit. On the GPU, single token prediction is ~6% lower than Q4_0, batch mode (perplexity) is even closer (but still slower). * Adding Q5_K - scalar, AVX2, CUDA Performance is ~20% lower compared to Q4_K on the CPU. This is to be expected, considering that we are memory bound on the CPU and the 5-bit model is ~22% larger than the 4-bit. On the GPU, single token prediction is about the same as Q4_0 for both, single token and batch prediction. * Per convention, all QX_K quantizations use Q5_K for output.weight * Adding quantization mixes * Quantization mixes: didn't quite get what I wanted in the last commit * Q4_K dot product for ARM_NEON * Q6_K dot product for ARM_NEON * Q5_K dot product for ARM_NEON * Adding Q3_K dot for ARM_NEON It is 22% slower than Q4_K, despite the smaller model size. On x86_64, where we are memory bound, the Q3_K model is quite a bit faster than Q4_K. * A very slightly faster ARM_NEON Q3_K dot * Adding Q2_K - just CUDA for now Token prediction is pretty good - about 15.5 ms on a RTX 4080. Perplexity is about the same as Q4_K. * Adding scalar and AVX2 Q2_K dot * Adding ARM_NEON Q2_K dot About the same performance as Q4_K. * A slightly faster ARM_NEON Q2_K dot Single token prediction is now ~36 ms on M2 Max. The code is much simpler too. * Fixed bug in Q2_K CUDA dot product kernel Stranegly enough, for the few prompts I tried with the 7B model the responses looked perfectly reasonable. Only realized something is not quite right when I tried the larger models and started getting nonse back. In any case, Q2_K single token evaluation time on an RTX 4080 in a Ryzen7950X box iusing CUDA and model fully loaded on the GPU are ~15.5 ms for 7B, ~25.4 ms for 13B, and ~55.8 ms for 30B. The max number of layers that fit in VRAM for The 65B is 32. With that, we get ~330 ms per token, which is not that much faster than just running on the CPU (~470 ms per token). * Don't print zeros/NaNs when no count histogram has been collected * A 10% faster CUDA vector dot kernel for Q3_K Q3_K is now running at ~18.5 ms / token on CUDA, so the gap to Q4_0 is only 10%. It seems memory acccess pattern is more important for performance than the amount of computation the kernel does. * A slightly daster Q4_K AVX2 dot product For perplexity, where we are less memory bound, time per pass drops by ~5%. Barely measurable difference for single token prediction. * A slightly faster ARM_NEON A4_K dot product * Minor * Fix quantization error test We cannot possibly be expecting rmse < 0.002 for 2- and 3-bit quantization variants. * Fix docker build I have been sloppy with vector reinterpret casts on ARM_NEON. It seems clang is very forgiving in that regard. * Added forgotten ggml.o dependence on k_quants.h to the Makefile * Had unintentionally committed the Makefile with -Ofast enabled * ggml : rename k_quants -> ggml-quants-k, use lowercase in code --------- Co-authored-by: Iwan Kawrakow <iwan.kawrakow@gmail.com> Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
2023-06-05 19:56:18 +00:00
ggml : quantization refactoring (#3833) * ggml : factor all quantization code in ggml-quants ggml-ci * ggml-quants : fix Zig and Swift builds + quantize tool ggml-ci * quantize : --pure option for disabling k-quant mixtures --------- Co-authored-by: cebtenzzre <cebtenzzre@gmail.com>
2023-10-29 16:32:28 +00:00
assert(nb % 2 == 0); // TODO: handle odd nb
ggml : add SOTA 2,3,4,5,6 bit k-quantizations (#1684) * Starting to add k-quantization to ggml I think it is better to have quantization separate from ggml. For now just adding the k-quants there, but it would be better to also factor out the existing ggml quantizations. * Adding Q3_K and Q8_K (de)-quantization * Q3_K now working on CUDA and AVX2/scalar CUDA is not ideal - ~50% slower than Q4_0 for single token prediction, about the same in batch mode (perplexity). CPU single token is ~55 ms (on Ryzen 7950X). * Some improvement for Q3_K on CUDA It is now ~22.5 ms/token on my GPU, so ~30% slower than Q4_0. * Some more CUDA optimizations for Q3_K Single token is now 20.5 ms/token (~20% slower than Q4_0). Perplexity is on par with Q4_0. * Adding Q4_K - scalar, AVX2, CUDA Performance is the same or perhaps very slightly better than Q4_0 on the CPU. On the GPU, single token prediction is ~10% better than Q4_0, batch mode (perplexity is about the same). * Adding Q6_K - scalar, AVX2, CUDA Performance is ~40% lower compared to Q4_K on the CPU. This is to be expected, considering that we are memory bound on the CPU and the 6-bit model is ~44% larger than the 4-bit. On the GPU, single token prediction is ~6% lower than Q4_0, batch mode (perplexity) is even closer (but still slower). * Adding Q5_K - scalar, AVX2, CUDA Performance is ~20% lower compared to Q4_K on the CPU. This is to be expected, considering that we are memory bound on the CPU and the 5-bit model is ~22% larger than the 4-bit. On the GPU, single token prediction is about the same as Q4_0 for both, single token and batch prediction. * Per convention, all QX_K quantizations use Q5_K for output.weight * Adding quantization mixes * Quantization mixes: didn't quite get what I wanted in the last commit * Q4_K dot product for ARM_NEON * Q6_K dot product for ARM_NEON * Q5_K dot product for ARM_NEON * Adding Q3_K dot for ARM_NEON It is 22% slower than Q4_K, despite the smaller model size. On x86_64, where we are memory bound, the Q3_K model is quite a bit faster than Q4_K. * A very slightly faster ARM_NEON Q3_K dot * Adding Q2_K - just CUDA for now Token prediction is pretty good - about 15.5 ms on a RTX 4080. Perplexity is about the same as Q4_K. * Adding scalar and AVX2 Q2_K dot * Adding ARM_NEON Q2_K dot About the same performance as Q4_K. * A slightly faster ARM_NEON Q2_K dot Single token prediction is now ~36 ms on M2 Max. The code is much simpler too. * Fixed bug in Q2_K CUDA dot product kernel Stranegly enough, for the few prompts I tried with the 7B model the responses looked perfectly reasonable. Only realized something is not quite right when I tried the larger models and started getting nonse back. In any case, Q2_K single token evaluation time on an RTX 4080 in a Ryzen7950X box iusing CUDA and model fully loaded on the GPU are ~15.5 ms for 7B, ~25.4 ms for 13B, and ~55.8 ms for 30B. The max number of layers that fit in VRAM for The 65B is 32. With that, we get ~330 ms per token, which is not that much faster than just running on the CPU (~470 ms per token). * Don't print zeros/NaNs when no count histogram has been collected * A 10% faster CUDA vector dot kernel for Q3_K Q3_K is now running at ~18.5 ms / token on CUDA, so the gap to Q4_0 is only 10%. It seems memory acccess pattern is more important for performance than the amount of computation the kernel does. * A slightly daster Q4_K AVX2 dot product For perplexity, where we are less memory bound, time per pass drops by ~5%. Barely measurable difference for single token prediction. * A slightly faster ARM_NEON A4_K dot product * Minor * Fix quantization error test We cannot possibly be expecting rmse < 0.002 for 2- and 3-bit quantization variants. * Fix docker build I have been sloppy with vector reinterpret casts on ARM_NEON. It seems clang is very forgiving in that regard. * Added forgotten ggml.o dependence on k_quants.h to the Makefile * Had unintentionally committed the Makefile with -Ofast enabled * ggml : rename k_quants -> ggml-quants-k, use lowercase in code --------- Co-authored-by: Iwan Kawrakow <iwan.kawrakow@gmail.com> Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
2023-06-05 19:56:18 +00:00
ggml : quantization refactoring (#3833) * ggml : factor all quantization code in ggml-quants ggml-ci * ggml-quants : fix Zig and Swift builds + quantize tool ggml-ci * quantize : --pure option for disabling k-quant mixtures --------- Co-authored-by: cebtenzzre <cebtenzzre@gmail.com>
2023-10-29 16:32:28 +00:00
for (int i = 0; i < nb; i += 2) {
const block_q4_1 * restrict x0 = &x[i + 0];
const block_q4_1 * restrict x1 = &x[i + 1];
const block_q8_1 * restrict y0 = &y[i + 0];
const block_q8_1 * restrict y1 = &y[i + 1];
ggml : add SOTA 2,3,4,5,6 bit k-quantizations (#1684) * Starting to add k-quantization to ggml I think it is better to have quantization separate from ggml. For now just adding the k-quants there, but it would be better to also factor out the existing ggml quantizations. * Adding Q3_K and Q8_K (de)-quantization * Q3_K now working on CUDA and AVX2/scalar CUDA is not ideal - ~50% slower than Q4_0 for single token prediction, about the same in batch mode (perplexity). CPU single token is ~55 ms (on Ryzen 7950X). * Some improvement for Q3_K on CUDA It is now ~22.5 ms/token on my GPU, so ~30% slower than Q4_0. * Some more CUDA optimizations for Q3_K Single token is now 20.5 ms/token (~20% slower than Q4_0). Perplexity is on par with Q4_0. * Adding Q4_K - scalar, AVX2, CUDA Performance is the same or perhaps very slightly better than Q4_0 on the CPU. On the GPU, single token prediction is ~10% better than Q4_0, batch mode (perplexity is about the same). * Adding Q6_K - scalar, AVX2, CUDA Performance is ~40% lower compared to Q4_K on the CPU. This is to be expected, considering that we are memory bound on the CPU and the 6-bit model is ~44% larger than the 4-bit. On the GPU, single token prediction is ~6% lower than Q4_0, batch mode (perplexity) is even closer (but still slower). * Adding Q5_K - scalar, AVX2, CUDA Performance is ~20% lower compared to Q4_K on the CPU. This is to be expected, considering that we are memory bound on the CPU and the 5-bit model is ~22% larger than the 4-bit. On the GPU, single token prediction is about the same as Q4_0 for both, single token and batch prediction. * Per convention, all QX_K quantizations use Q5_K for output.weight * Adding quantization mixes * Quantization mixes: didn't quite get what I wanted in the last commit * Q4_K dot product for ARM_NEON * Q6_K dot product for ARM_NEON * Q5_K dot product for ARM_NEON * Adding Q3_K dot for ARM_NEON It is 22% slower than Q4_K, despite the smaller model size. On x86_64, where we are memory bound, the Q3_K model is quite a bit faster than Q4_K. * A very slightly faster ARM_NEON Q3_K dot * Adding Q2_K - just CUDA for now Token prediction is pretty good - about 15.5 ms on a RTX 4080. Perplexity is about the same as Q4_K. * Adding scalar and AVX2 Q2_K dot * Adding ARM_NEON Q2_K dot About the same performance as Q4_K. * A slightly faster ARM_NEON Q2_K dot Single token prediction is now ~36 ms on M2 Max. The code is much simpler too. * Fixed bug in Q2_K CUDA dot product kernel Stranegly enough, for the few prompts I tried with the 7B model the responses looked perfectly reasonable. Only realized something is not quite right when I tried the larger models and started getting nonse back. In any case, Q2_K single token evaluation time on an RTX 4080 in a Ryzen7950X box iusing CUDA and model fully loaded on the GPU are ~15.5 ms for 7B, ~25.4 ms for 13B, and ~55.8 ms for 30B. The max number of layers that fit in VRAM for The 65B is 32. With that, we get ~330 ms per token, which is not that much faster than just running on the CPU (~470 ms per token). * Don't print zeros/NaNs when no count histogram has been collected * A 10% faster CUDA vector dot kernel for Q3_K Q3_K is now running at ~18.5 ms / token on CUDA, so the gap to Q4_0 is only 10%. It seems memory acccess pattern is more important for performance than the amount of computation the kernel does. * A slightly daster Q4_K AVX2 dot product For perplexity, where we are less memory bound, time per pass drops by ~5%. Barely measurable difference for single token prediction. * A slightly faster ARM_NEON A4_K dot product * Minor * Fix quantization error test We cannot possibly be expecting rmse < 0.002 for 2- and 3-bit quantization variants. * Fix docker build I have been sloppy with vector reinterpret casts on ARM_NEON. It seems clang is very forgiving in that regard. * Added forgotten ggml.o dependence on k_quants.h to the Makefile * Had unintentionally committed the Makefile with -Ofast enabled * ggml : rename k_quants -> ggml-quants-k, use lowercase in code --------- Co-authored-by: Iwan Kawrakow <iwan.kawrakow@gmail.com> Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
2023-06-05 19:56:18 +00:00
ggml : move FP16 <-> FP32 code to ggml-impl.h (#3861) * ggml : move FP16 <-> FP32 stuff to ggml-impl.h ggml-ci * tests : fix ARM build * ggml : explicitly initialize deprecated type traits * ggml : add math.h to ggml-impl.h * ggml : remove duplicate static assert macros * ggml : prefix lookup tables with ggml_ ggml-ci * ggml-impl : move extern "C" to start of file
2023-10-30 17:19:15 +00:00
summs += GGML_FP16_TO_FP32(x0->m) * y0->s + GGML_FP16_TO_FP32(x1->m) * y1->s;
ggml : add SOTA 2,3,4,5,6 bit k-quantizations (#1684) * Starting to add k-quantization to ggml I think it is better to have quantization separate from ggml. For now just adding the k-quants there, but it would be better to also factor out the existing ggml quantizations. * Adding Q3_K and Q8_K (de)-quantization * Q3_K now working on CUDA and AVX2/scalar CUDA is not ideal - ~50% slower than Q4_0 for single token prediction, about the same in batch mode (perplexity). CPU single token is ~55 ms (on Ryzen 7950X). * Some improvement for Q3_K on CUDA It is now ~22.5 ms/token on my GPU, so ~30% slower than Q4_0. * Some more CUDA optimizations for Q3_K Single token is now 20.5 ms/token (~20% slower than Q4_0). Perplexity is on par with Q4_0. * Adding Q4_K - scalar, AVX2, CUDA Performance is the same or perhaps very slightly better than Q4_0 on the CPU. On the GPU, single token prediction is ~10% better than Q4_0, batch mode (perplexity is about the same). * Adding Q6_K - scalar, AVX2, CUDA Performance is ~40% lower compared to Q4_K on the CPU. This is to be expected, considering that we are memory bound on the CPU and the 6-bit model is ~44% larger than the 4-bit. On the GPU, single token prediction is ~6% lower than Q4_0, batch mode (perplexity) is even closer (but still slower). * Adding Q5_K - scalar, AVX2, CUDA Performance is ~20% lower compared to Q4_K on the CPU. This is to be expected, considering that we are memory bound on the CPU and the 5-bit model is ~22% larger than the 4-bit. On the GPU, single token prediction is about the same as Q4_0 for both, single token and batch prediction. * Per convention, all QX_K quantizations use Q5_K for output.weight * Adding quantization mixes * Quantization mixes: didn't quite get what I wanted in the last commit * Q4_K dot product for ARM_NEON * Q6_K dot product for ARM_NEON * Q5_K dot product for ARM_NEON * Adding Q3_K dot for ARM_NEON It is 22% slower than Q4_K, despite the smaller model size. On x86_64, where we are memory bound, the Q3_K model is quite a bit faster than Q4_K. * A very slightly faster ARM_NEON Q3_K dot * Adding Q2_K - just CUDA for now Token prediction is pretty good - about 15.5 ms on a RTX 4080. Perplexity is about the same as Q4_K. * Adding scalar and AVX2 Q2_K dot * Adding ARM_NEON Q2_K dot About the same performance as Q4_K. * A slightly faster ARM_NEON Q2_K dot Single token prediction is now ~36 ms on M2 Max. The code is much simpler too. * Fixed bug in Q2_K CUDA dot product kernel Stranegly enough, for the few prompts I tried with the 7B model the responses looked perfectly reasonable. Only realized something is not quite right when I tried the larger models and started getting nonse back. In any case, Q2_K single token evaluation time on an RTX 4080 in a Ryzen7950X box iusing CUDA and model fully loaded on the GPU are ~15.5 ms for 7B, ~25.4 ms for 13B, and ~55.8 ms for 30B. The max number of layers that fit in VRAM for The 65B is 32. With that, we get ~330 ms per token, which is not that much faster than just running on the CPU (~470 ms per token). * Don't print zeros/NaNs when no count histogram has been collected * A 10% faster CUDA vector dot kernel for Q3_K Q3_K is now running at ~18.5 ms / token on CUDA, so the gap to Q4_0 is only 10%. It seems memory acccess pattern is more important for performance than the amount of computation the kernel does. * A slightly daster Q4_K AVX2 dot product For perplexity, where we are less memory bound, time per pass drops by ~5%. Barely measurable difference for single token prediction. * A slightly faster ARM_NEON A4_K dot product * Minor * Fix quantization error test We cannot possibly be expecting rmse < 0.002 for 2- and 3-bit quantization variants. * Fix docker build I have been sloppy with vector reinterpret casts on ARM_NEON. It seems clang is very forgiving in that regard. * Added forgotten ggml.o dependence on k_quants.h to the Makefile * Had unintentionally committed the Makefile with -Ofast enabled * ggml : rename k_quants -> ggml-quants-k, use lowercase in code --------- Co-authored-by: Iwan Kawrakow <iwan.kawrakow@gmail.com> Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
2023-06-05 19:56:18 +00:00
ggml : quantization refactoring (#3833) * ggml : factor all quantization code in ggml-quants ggml-ci * ggml-quants : fix Zig and Swift builds + quantize tool ggml-ci * quantize : --pure option for disabling k-quant mixtures --------- Co-authored-by: cebtenzzre <cebtenzzre@gmail.com>
2023-10-29 16:32:28 +00:00
const uint8x16_t m4b = vdupq_n_u8(0x0F);
ggml : add SOTA 2,3,4,5,6 bit k-quantizations (#1684) * Starting to add k-quantization to ggml I think it is better to have quantization separate from ggml. For now just adding the k-quants there, but it would be better to also factor out the existing ggml quantizations. * Adding Q3_K and Q8_K (de)-quantization * Q3_K now working on CUDA and AVX2/scalar CUDA is not ideal - ~50% slower than Q4_0 for single token prediction, about the same in batch mode (perplexity). CPU single token is ~55 ms (on Ryzen 7950X). * Some improvement for Q3_K on CUDA It is now ~22.5 ms/token on my GPU, so ~30% slower than Q4_0. * Some more CUDA optimizations for Q3_K Single token is now 20.5 ms/token (~20% slower than Q4_0). Perplexity is on par with Q4_0. * Adding Q4_K - scalar, AVX2, CUDA Performance is the same or perhaps very slightly better than Q4_0 on the CPU. On the GPU, single token prediction is ~10% better than Q4_0, batch mode (perplexity is about the same). * Adding Q6_K - scalar, AVX2, CUDA Performance is ~40% lower compared to Q4_K on the CPU. This is to be expected, considering that we are memory bound on the CPU and the 6-bit model is ~44% larger than the 4-bit. On the GPU, single token prediction is ~6% lower than Q4_0, batch mode (perplexity) is even closer (but still slower). * Adding Q5_K - scalar, AVX2, CUDA Performance is ~20% lower compared to Q4_K on the CPU. This is to be expected, considering that we are memory bound on the CPU and the 5-bit model is ~22% larger than the 4-bit. On the GPU, single token prediction is about the same as Q4_0 for both, single token and batch prediction. * Per convention, all QX_K quantizations use Q5_K for output.weight * Adding quantization mixes * Quantization mixes: didn't quite get what I wanted in the last commit * Q4_K dot product for ARM_NEON * Q6_K dot product for ARM_NEON * Q5_K dot product for ARM_NEON * Adding Q3_K dot for ARM_NEON It is 22% slower than Q4_K, despite the smaller model size. On x86_64, where we are memory bound, the Q3_K model is quite a bit faster than Q4_K. * A very slightly faster ARM_NEON Q3_K dot * Adding Q2_K - just CUDA for now Token prediction is pretty good - about 15.5 ms on a RTX 4080. Perplexity is about the same as Q4_K. * Adding scalar and AVX2 Q2_K dot * Adding ARM_NEON Q2_K dot About the same performance as Q4_K. * A slightly faster ARM_NEON Q2_K dot Single token prediction is now ~36 ms on M2 Max. The code is much simpler too. * Fixed bug in Q2_K CUDA dot product kernel Stranegly enough, for the few prompts I tried with the 7B model the responses looked perfectly reasonable. Only realized something is not quite right when I tried the larger models and started getting nonse back. In any case, Q2_K single token evaluation time on an RTX 4080 in a Ryzen7950X box iusing CUDA and model fully loaded on the GPU are ~15.5 ms for 7B, ~25.4 ms for 13B, and ~55.8 ms for 30B. The max number of layers that fit in VRAM for The 65B is 32. With that, we get ~330 ms per token, which is not that much faster than just running on the CPU (~470 ms per token). * Don't print zeros/NaNs when no count histogram has been collected * A 10% faster CUDA vector dot kernel for Q3_K Q3_K is now running at ~18.5 ms / token on CUDA, so the gap to Q4_0 is only 10%. It seems memory acccess pattern is more important for performance than the amount of computation the kernel does. * A slightly daster Q4_K AVX2 dot product For perplexity, where we are less memory bound, time per pass drops by ~5%. Barely measurable difference for single token prediction. * A slightly faster ARM_NEON A4_K dot product * Minor * Fix quantization error test We cannot possibly be expecting rmse < 0.002 for 2- and 3-bit quantization variants. * Fix docker build I have been sloppy with vector reinterpret casts on ARM_NEON. It seems clang is very forgiving in that regard. * Added forgotten ggml.o dependence on k_quants.h to the Makefile * Had unintentionally committed the Makefile with -Ofast enabled * ggml : rename k_quants -> ggml-quants-k, use lowercase in code --------- Co-authored-by: Iwan Kawrakow <iwan.kawrakow@gmail.com> Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
2023-06-05 19:56:18 +00:00
ggml : quantization refactoring (#3833) * ggml : factor all quantization code in ggml-quants ggml-ci * ggml-quants : fix Zig and Swift builds + quantize tool ggml-ci * quantize : --pure option for disabling k-quant mixtures --------- Co-authored-by: cebtenzzre <cebtenzzre@gmail.com>
2023-10-29 16:32:28 +00:00
const uint8x16_t v0_0 = vld1q_u8(x0->qs);
const uint8x16_t v0_1 = vld1q_u8(x1->qs);
ggml : add SOTA 2,3,4,5,6 bit k-quantizations (#1684) * Starting to add k-quantization to ggml I think it is better to have quantization separate from ggml. For now just adding the k-quants there, but it would be better to also factor out the existing ggml quantizations. * Adding Q3_K and Q8_K (de)-quantization * Q3_K now working on CUDA and AVX2/scalar CUDA is not ideal - ~50% slower than Q4_0 for single token prediction, about the same in batch mode (perplexity). CPU single token is ~55 ms (on Ryzen 7950X). * Some improvement for Q3_K on CUDA It is now ~22.5 ms/token on my GPU, so ~30% slower than Q4_0. * Some more CUDA optimizations for Q3_K Single token is now 20.5 ms/token (~20% slower than Q4_0). Perplexity is on par with Q4_0. * Adding Q4_K - scalar, AVX2, CUDA Performance is the same or perhaps very slightly better than Q4_0 on the CPU. On the GPU, single token prediction is ~10% better than Q4_0, batch mode (perplexity is about the same). * Adding Q6_K - scalar, AVX2, CUDA Performance is ~40% lower compared to Q4_K on the CPU. This is to be expected, considering that we are memory bound on the CPU and the 6-bit model is ~44% larger than the 4-bit. On the GPU, single token prediction is ~6% lower than Q4_0, batch mode (perplexity) is even closer (but still slower). * Adding Q5_K - scalar, AVX2, CUDA Performance is ~20% lower compared to Q4_K on the CPU. This is to be expected, considering that we are memory bound on the CPU and the 5-bit model is ~22% larger than the 4-bit. On the GPU, single token prediction is about the same as Q4_0 for both, single token and batch prediction. * Per convention, all QX_K quantizations use Q5_K for output.weight * Adding quantization mixes * Quantization mixes: didn't quite get what I wanted in the last commit * Q4_K dot product for ARM_NEON * Q6_K dot product for ARM_NEON * Q5_K dot product for ARM_NEON * Adding Q3_K dot for ARM_NEON It is 22% slower than Q4_K, despite the smaller model size. On x86_64, where we are memory bound, the Q3_K model is quite a bit faster than Q4_K. * A very slightly faster ARM_NEON Q3_K dot * Adding Q2_K - just CUDA for now Token prediction is pretty good - about 15.5 ms on a RTX 4080. Perplexity is about the same as Q4_K. * Adding scalar and AVX2 Q2_K dot * Adding ARM_NEON Q2_K dot About the same performance as Q4_K. * A slightly faster ARM_NEON Q2_K dot Single token prediction is now ~36 ms on M2 Max. The code is much simpler too. * Fixed bug in Q2_K CUDA dot product kernel Stranegly enough, for the few prompts I tried with the 7B model the responses looked perfectly reasonable. Only realized something is not quite right when I tried the larger models and started getting nonse back. In any case, Q2_K single token evaluation time on an RTX 4080 in a Ryzen7950X box iusing CUDA and model fully loaded on the GPU are ~15.5 ms for 7B, ~25.4 ms for 13B, and ~55.8 ms for 30B. The max number of layers that fit in VRAM for The 65B is 32. With that, we get ~330 ms per token, which is not that much faster than just running on the CPU (~470 ms per token). * Don't print zeros/NaNs when no count histogram has been collected * A 10% faster CUDA vector dot kernel for Q3_K Q3_K is now running at ~18.5 ms / token on CUDA, so the gap to Q4_0 is only 10%. It seems memory acccess pattern is more important for performance than the amount of computation the kernel does. * A slightly daster Q4_K AVX2 dot product For perplexity, where we are less memory bound, time per pass drops by ~5%. Barely measurable difference for single token prediction. * A slightly faster ARM_NEON A4_K dot product * Minor * Fix quantization error test We cannot possibly be expecting rmse < 0.002 for 2- and 3-bit quantization variants. * Fix docker build I have been sloppy with vector reinterpret casts on ARM_NEON. It seems clang is very forgiving in that regard. * Added forgotten ggml.o dependence on k_quants.h to the Makefile * Had unintentionally committed the Makefile with -Ofast enabled * ggml : rename k_quants -> ggml-quants-k, use lowercase in code --------- Co-authored-by: Iwan Kawrakow <iwan.kawrakow@gmail.com> Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
2023-06-05 19:56:18 +00:00
ggml : quantization refactoring (#3833) * ggml : factor all quantization code in ggml-quants ggml-ci * ggml-quants : fix Zig and Swift builds + quantize tool ggml-ci * quantize : --pure option for disabling k-quant mixtures --------- Co-authored-by: cebtenzzre <cebtenzzre@gmail.com>
2023-10-29 16:32:28 +00:00
// 4-bit -> 8-bit
const int8x16_t v0_0l = vreinterpretq_s8_u8(vandq_u8 (v0_0, m4b));
const int8x16_t v0_0h = vreinterpretq_s8_u8(vshrq_n_u8(v0_0, 4));
const int8x16_t v0_1l = vreinterpretq_s8_u8(vandq_u8 (v0_1, m4b));
const int8x16_t v0_1h = vreinterpretq_s8_u8(vshrq_n_u8(v0_1, 4));
ggml : add SOTA 2,3,4,5,6 bit k-quantizations (#1684) * Starting to add k-quantization to ggml I think it is better to have quantization separate from ggml. For now just adding the k-quants there, but it would be better to also factor out the existing ggml quantizations. * Adding Q3_K and Q8_K (de)-quantization * Q3_K now working on CUDA and AVX2/scalar CUDA is not ideal - ~50% slower than Q4_0 for single token prediction, about the same in batch mode (perplexity). CPU single token is ~55 ms (on Ryzen 7950X). * Some improvement for Q3_K on CUDA It is now ~22.5 ms/token on my GPU, so ~30% slower than Q4_0. * Some more CUDA optimizations for Q3_K Single token is now 20.5 ms/token (~20% slower than Q4_0). Perplexity is on par with Q4_0. * Adding Q4_K - scalar, AVX2, CUDA Performance is the same or perhaps very slightly better than Q4_0 on the CPU. On the GPU, single token prediction is ~10% better than Q4_0, batch mode (perplexity is about the same). * Adding Q6_K - scalar, AVX2, CUDA Performance is ~40% lower compared to Q4_K on the CPU. This is to be expected, considering that we are memory bound on the CPU and the 6-bit model is ~44% larger than the 4-bit. On the GPU, single token prediction is ~6% lower than Q4_0, batch mode (perplexity) is even closer (but still slower). * Adding Q5_K - scalar, AVX2, CUDA Performance is ~20% lower compared to Q4_K on the CPU. This is to be expected, considering that we are memory bound on the CPU and the 5-bit model is ~22% larger than the 4-bit. On the GPU, single token prediction is about the same as Q4_0 for both, single token and batch prediction. * Per convention, all QX_K quantizations use Q5_K for output.weight * Adding quantization mixes * Quantization mixes: didn't quite get what I wanted in the last commit * Q4_K dot product for ARM_NEON * Q6_K dot product for ARM_NEON * Q5_K dot product for ARM_NEON * Adding Q3_K dot for ARM_NEON It is 22% slower than Q4_K, despite the smaller model size. On x86_64, where we are memory bound, the Q3_K model is quite a bit faster than Q4_K. * A very slightly faster ARM_NEON Q3_K dot * Adding Q2_K - just CUDA for now Token prediction is pretty good - about 15.5 ms on a RTX 4080. Perplexity is about the same as Q4_K. * Adding scalar and AVX2 Q2_K dot * Adding ARM_NEON Q2_K dot About the same performance as Q4_K. * A slightly faster ARM_NEON Q2_K dot Single token prediction is now ~36 ms on M2 Max. The code is much simpler too. * Fixed bug in Q2_K CUDA dot product kernel Stranegly enough, for the few prompts I tried with the 7B model the responses looked perfectly reasonable. Only realized something is not quite right when I tried the larger models and started getting nonse back. In any case, Q2_K single token evaluation time on an RTX 4080 in a Ryzen7950X box iusing CUDA and model fully loaded on the GPU are ~15.5 ms for 7B, ~25.4 ms for 13B, and ~55.8 ms for 30B. The max number of layers that fit in VRAM for The 65B is 32. With that, we get ~330 ms per token, which is not that much faster than just running on the CPU (~470 ms per token). * Don't print zeros/NaNs when no count histogram has been collected * A 10% faster CUDA vector dot kernel for Q3_K Q3_K is now running at ~18.5 ms / token on CUDA, so the gap to Q4_0 is only 10%. It seems memory acccess pattern is more important for performance than the amount of computation the kernel does. * A slightly daster Q4_K AVX2 dot product For perplexity, where we are less memory bound, time per pass drops by ~5%. Barely measurable difference for single token prediction. * A slightly faster ARM_NEON A4_K dot product * Minor * Fix quantization error test We cannot possibly be expecting rmse < 0.002 for 2- and 3-bit quantization variants. * Fix docker build I have been sloppy with vector reinterpret casts on ARM_NEON. It seems clang is very forgiving in that regard. * Added forgotten ggml.o dependence on k_quants.h to the Makefile * Had unintentionally committed the Makefile with -Ofast enabled * ggml : rename k_quants -> ggml-quants-k, use lowercase in code --------- Co-authored-by: Iwan Kawrakow <iwan.kawrakow@gmail.com> Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
2023-06-05 19:56:18 +00:00
ggml : quantization refactoring (#3833) * ggml : factor all quantization code in ggml-quants ggml-ci * ggml-quants : fix Zig and Swift builds + quantize tool ggml-ci * quantize : --pure option for disabling k-quant mixtures --------- Co-authored-by: cebtenzzre <cebtenzzre@gmail.com>
2023-10-29 16:32:28 +00:00
// load y
const int8x16_t v1_0l = vld1q_s8(y0->qs);
const int8x16_t v1_0h = vld1q_s8(y0->qs + 16);
const int8x16_t v1_1l = vld1q_s8(y1->qs);
const int8x16_t v1_1h = vld1q_s8(y1->qs + 16);
// dot product into int32x4_t
ggml : add ggml_vdotq_s32 alias (#4715) ggml-ci
2023-12-31 09:43:31 +00:00
const int32x4_t p_0 = ggml_vdotq_s32(ggml_vdotq_s32(vdupq_n_s32(0), v0_0l, v1_0l), v0_0h, v1_0h);
const int32x4_t p_1 = ggml_vdotq_s32(ggml_vdotq_s32(vdupq_n_s32(0), v0_1l, v1_1l), v0_1h, v1_1h);
ggml : quantization refactoring (#3833) * ggml : factor all quantization code in ggml-quants ggml-ci * ggml-quants : fix Zig and Swift builds + quantize tool ggml-ci * quantize : --pure option for disabling k-quant mixtures --------- Co-authored-by: cebtenzzre <cebtenzzre@gmail.com>
2023-10-29 16:32:28 +00:00
ggml : move FP16 <-> FP32 code to ggml-impl.h (#3861) * ggml : move FP16 <-> FP32 stuff to ggml-impl.h ggml-ci * tests : fix ARM build * ggml : explicitly initialize deprecated type traits * ggml : add math.h to ggml-impl.h * ggml : remove duplicate static assert macros * ggml : prefix lookup tables with ggml_ ggml-ci * ggml-impl : move extern "C" to start of file
2023-10-30 17:19:15 +00:00
sumv0 = vmlaq_n_f32(sumv0, vcvtq_f32_s32(p_0), GGML_FP16_TO_FP32(x0->d)*y0->d);
sumv1 = vmlaq_n_f32(sumv1, vcvtq_f32_s32(p_1), GGML_FP16_TO_FP32(x1->d)*y1->d);
ggml : add SOTA 2,3,4,5,6 bit k-quantizations (#1684) * Starting to add k-quantization to ggml I think it is better to have quantization separate from ggml. For now just adding the k-quants there, but it would be better to also factor out the existing ggml quantizations. * Adding Q3_K and Q8_K (de)-quantization * Q3_K now working on CUDA and AVX2/scalar CUDA is not ideal - ~50% slower than Q4_0 for single token prediction, about the same in batch mode (perplexity). CPU single token is ~55 ms (on Ryzen 7950X). * Some improvement for Q3_K on CUDA It is now ~22.5 ms/token on my GPU, so ~30% slower than Q4_0. * Some more CUDA optimizations for Q3_K Single token is now 20.5 ms/token (~20% slower than Q4_0). Perplexity is on par with Q4_0. * Adding Q4_K - scalar, AVX2, CUDA Performance is the same or perhaps very slightly better than Q4_0 on the CPU. On the GPU, single token prediction is ~10% better than Q4_0, batch mode (perplexity is about the same). * Adding Q6_K - scalar, AVX2, CUDA Performance is ~40% lower compared to Q4_K on the CPU. This is to be expected, considering that we are memory bound on the CPU and the 6-bit model is ~44% larger than the 4-bit. On the GPU, single token prediction is ~6% lower than Q4_0, batch mode (perplexity) is even closer (but still slower). * Adding Q5_K - scalar, AVX2, CUDA Performance is ~20% lower compared to Q4_K on the CPU. This is to be expected, considering that we are memory bound on the CPU and the 5-bit model is ~22% larger than the 4-bit. On the GPU, single token prediction is about the same as Q4_0 for both, single token and batch prediction. * Per convention, all QX_K quantizations use Q5_K for output.weight * Adding quantization mixes * Quantization mixes: didn't quite get what I wanted in the last commit * Q4_K dot product for ARM_NEON * Q6_K dot product for ARM_NEON * Q5_K dot product for ARM_NEON * Adding Q3_K dot for ARM_NEON It is 22% slower than Q4_K, despite the smaller model size. On x86_64, where we are memory bound, the Q3_K model is quite a bit faster than Q4_K. * A very slightly faster ARM_NEON Q3_K dot * Adding Q2_K - just CUDA for now Token prediction is pretty good - about 15.5 ms on a RTX 4080. Perplexity is about the same as Q4_K. * Adding scalar and AVX2 Q2_K dot * Adding ARM_NEON Q2_K dot About the same performance as Q4_K. * A slightly faster ARM_NEON Q2_K dot Single token prediction is now ~36 ms on M2 Max. The code is much simpler too. * Fixed bug in Q2_K CUDA dot product kernel Stranegly enough, for the few prompts I tried with the 7B model the responses looked perfectly reasonable. Only realized something is not quite right when I tried the larger models and started getting nonse back. In any case, Q2_K single token evaluation time on an RTX 4080 in a Ryzen7950X box iusing CUDA and model fully loaded on the GPU are ~15.5 ms for 7B, ~25.4 ms for 13B, and ~55.8 ms for 30B. The max number of layers that fit in VRAM for The 65B is 32. With that, we get ~330 ms per token, which is not that much faster than just running on the CPU (~470 ms per token). * Don't print zeros/NaNs when no count histogram has been collected * A 10% faster CUDA vector dot kernel for Q3_K Q3_K is now running at ~18.5 ms / token on CUDA, so the gap to Q4_0 is only 10%. It seems memory acccess pattern is more important for performance than the amount of computation the kernel does. * A slightly daster Q4_K AVX2 dot product For perplexity, where we are less memory bound, time per pass drops by ~5%. Barely measurable difference for single token prediction. * A slightly faster ARM_NEON A4_K dot product * Minor * Fix quantization error test We cannot possibly be expecting rmse < 0.002 for 2- and 3-bit quantization variants. * Fix docker build I have been sloppy with vector reinterpret casts on ARM_NEON. It seems clang is very forgiving in that regard. * Added forgotten ggml.o dependence on k_quants.h to the Makefile * Had unintentionally committed the Makefile with -Ofast enabled * ggml : rename k_quants -> ggml-quants-k, use lowercase in code --------- Co-authored-by: Iwan Kawrakow <iwan.kawrakow@gmail.com> Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
2023-06-05 19:56:18 +00:00
}
ggml : quantization refactoring (#3833) * ggml : factor all quantization code in ggml-quants ggml-ci * ggml-quants : fix Zig and Swift builds + quantize tool ggml-ci * quantize : --pure option for disabling k-quant mixtures --------- Co-authored-by: cebtenzzre <cebtenzzre@gmail.com>
2023-10-29 16:32:28 +00:00
*s = vaddvq_f32(sumv0) + vaddvq_f32(sumv1) + summs;
#elif defined(__AVX2__) || defined(__AVX__)
// Initialize accumulator with zeros
__m256 acc = _mm256_setzero_ps();
float summs = 0;
// Main loop
for (int i = 0; i < nb; ++i) {
ggml : move FP16 <-> FP32 code to ggml-impl.h (#3861) * ggml : move FP16 <-> FP32 stuff to ggml-impl.h ggml-ci * tests : fix ARM build * ggml : explicitly initialize deprecated type traits * ggml : add math.h to ggml-impl.h * ggml : remove duplicate static assert macros * ggml : prefix lookup tables with ggml_ ggml-ci * ggml-impl : move extern "C" to start of file
2023-10-30 17:19:15 +00:00
const float d0 = GGML_FP16_TO_FP32(x[i].d);
ggml : quantization refactoring (#3833) * ggml : factor all quantization code in ggml-quants ggml-ci * ggml-quants : fix Zig and Swift builds + quantize tool ggml-ci * quantize : --pure option for disabling k-quant mixtures --------- Co-authored-by: cebtenzzre <cebtenzzre@gmail.com>
2023-10-29 16:32:28 +00:00
const float d1 = y[i].d;
ggml : move FP16 <-> FP32 code to ggml-impl.h (#3861) * ggml : move FP16 <-> FP32 stuff to ggml-impl.h ggml-ci * tests : fix ARM build * ggml : explicitly initialize deprecated type traits * ggml : add math.h to ggml-impl.h * ggml : remove duplicate static assert macros * ggml : prefix lookup tables with ggml_ ggml-ci * ggml-impl : move extern "C" to start of file
2023-10-30 17:19:15 +00:00
summs += GGML_FP16_TO_FP32(x[i].m) * y[i].s;
ggml : quantization refactoring (#3833) * ggml : factor all quantization code in ggml-quants ggml-ci * ggml-quants : fix Zig and Swift builds + quantize tool ggml-ci * quantize : --pure option for disabling k-quant mixtures --------- Co-authored-by: cebtenzzre <cebtenzzre@gmail.com>
2023-10-29 16:32:28 +00:00
const __m256 d0v = _mm256_set1_ps( d0 );
const __m256 d1v = _mm256_set1_ps( d1 );
// Compute combined scales
const __m256 d0d1 = _mm256_mul_ps( d0v, d1v );
// Load 16 bytes, and unpack 4 bit fields into bytes, making 32 bytes
const __m256i bx = bytes_from_nibbles_32(x[i].qs);
const __m256i by = _mm256_loadu_si256( (const __m256i *)y[i].qs );
const __m256 xy = mul_sum_us8_pairs_float(bx, by);
// Accumulate d0*d1*x*y
#if defined(__AVX2__)
acc = _mm256_fmadd_ps( d0d1, xy, acc );
k-quants : support for super-block size of 64 (#2001) * k_quants: WIP super-blocks with 64 weights * k_quants: WIP super-blocks with 64 weights Q6_K scalar and AVX2 works * k_quants: WIP super-blocks with 64 weights Q4_K scalar and AVX2 works * k_quants: WIP super-blocks with 64 weights Q2_K scalar and AVX2 works. Q2_K is way too slow (it is actually slower than the scalar implementation) * k_quants: WIP super-blocks with 64 weights Q3_K scalar and AVX2 works. * k_quants: WIP super-blocks with 64 weights Q5_K scalar and AVX2 works, and with that all k_quants are done on AVX2 and scalar * k_quants: WIP super-blocks with 64 weights Q6_K working on CUDA. Cannot make it run quite as gast as with super-blocks with 256 weigths: 8% slower on 4080, 20% slower on the 1660 (but there we fit 1 less layer on the GPU because pf the larger model size), so some fraction of these 20% is due to that, * k_quants: WIP super-blocks with 64 weights Q4_K working on CUDA. ~10% slower on GTX-1660, 16% slower on 4080. * k_quants: WIP super-blocks with 64 weights Q2_K working on CUDA. ~3% slower on GTX-1660, 10% slower on 4080. * k_quants: WIP super-blocks with 64 weights Q3_K working on CUDA. * k_quants: WIP super-blocks with 64 weights Q5_K working on CUDA, and with this CUDA is done. * k_quants: WIP super-blocks with 64 weights Q6_K working on ARM_NEON * k_quants: WIP super-blocks with 64 weights Q4_K working on ARM_NEON, but quite a bit slower than 256 weights * k_quants: WIP super-blocks with 64 weights Q2_K working on ARM_NEON, but quite a bit slower than 256 weights * k_quants: WIP super-blocks with 64 weights Q3_K working on ARM_NEON, but quite a bit slower than 256 weights. * k_quants: WIP super-blocks with 64 weights Q5_K working on ARM_NEON, but quite a bit slower than 256 weights. With that, we have full support for ARM_NEON, although performance is not quite there. * k_quants: WIP super-blocks with 64 weights Slightly more efficient Q3_K and Q5_K * k_quants: WIP super-blocks with 64 weights Another small improvement for Q3_K and Q5_K on ARM_NEON * k_quants: WIP super-blocks with 64 weights Yet another speedup for Q5_K on ARM_NEON. We are now within 10% of the QK_K = 256 version. * k_quants: WIP super-blocks with 64 weights * We are able to pass preprocessor macros to the Metal compiler * Q6_K works and is actually slightly more efficient than the QK_K = 256 version (25.2 ms vs 25.8 ms) * k_quants: WIP super-blocks with 64 weights Q4_K works on Metal and is actually slightly faster than QK_K = 256 (21.95 ms vs 24.0 ms). * k_quants: WIP super-blocks with 64 weights Q2_K works on Metal and is very slightly faster than QK_K = 256 (23.8 ms vs 24.2 ms). * k_quants: WIP super-blocks with 64 weights Q3_K works on Metal and is slightly faster than QK_K = 256 (26.6 ms vs 28.3 ms). * k_quants: WIP super-blocks with 64 weights Q5_K works on Metal and is slightly faster than QK_K = 256 (23.7 ms vs 26.3 ms). * k_quants: call them _K, not _k, also on Metal * k_quants: correctly define QK_K in llama.cpp * Fixed bug in q4_K quantization added with the 64-block addition * Simplify via lambda * k_quants: swicth Q3_K to 4-bit scales when QK_K = 64 Otherwise there isn't much benefit from this quantization type. There is some very slight loss in accuracy, but we reduce size by ~7%. E.g., for OpenLLaMA-3B, Q3_K_S perplexity is 8.6131 with 8-bit scales and 8.6352 with 4-bit, while file size decreases from 1.53G to 1.44G. * k_quants: switch Q4_K to 4-bit scales when QK_K = 64 Here the loss in accuracy is greater than for Q3_K, but the Q4_K points still move further to the left on the perplexity vs size curve. * k_quants: forgot to add the Metal changes in last commit * k_quants: change Q5_K to be type 0 when QK_K = 64 Still needs AVX2 implementation * k_quants: AVX2 implementation for new 64-weight Q5_K * k_quants: 10% faster ARM_NEON Q5_K dot product * k_quants: fixed issue caused by merging with master --------- Co-authored-by: Iwan Kawrakow <iwan.kawrakow@gmail.com>
2023-06-26 16:43:07 +00:00
#else
ggml : quantization refactoring (#3833) * ggml : factor all quantization code in ggml-quants ggml-ci * ggml-quants : fix Zig and Swift builds + quantize tool ggml-ci * quantize : --pure option for disabling k-quant mixtures --------- Co-authored-by: cebtenzzre <cebtenzzre@gmail.com>
2023-10-29 16:32:28 +00:00
acc = _mm256_add_ps( _mm256_mul_ps( d0d1, xy ), acc );
#endif
}
*s = hsum_float_8(acc) + summs;
#elif defined(__riscv_v_intrinsic)
float sumf = 0.0;
size_t vl = __riscv_vsetvl_e8m1(qk/2);
k-quants : support for super-block size of 64 (#2001) * k_quants: WIP super-blocks with 64 weights * k_quants: WIP super-blocks with 64 weights Q6_K scalar and AVX2 works * k_quants: WIP super-blocks with 64 weights Q4_K scalar and AVX2 works * k_quants: WIP super-blocks with 64 weights Q2_K scalar and AVX2 works. Q2_K is way too slow (it is actually slower than the scalar implementation) * k_quants: WIP super-blocks with 64 weights Q3_K scalar and AVX2 works. * k_quants: WIP super-blocks with 64 weights Q5_K scalar and AVX2 works, and with that all k_quants are done on AVX2 and scalar * k_quants: WIP super-blocks with 64 weights Q6_K working on CUDA. Cannot make it run quite as gast as with super-blocks with 256 weigths: 8% slower on 4080, 20% slower on the 1660 (but there we fit 1 less layer on the GPU because pf the larger model size), so some fraction of these 20% is due to that, * k_quants: WIP super-blocks with 64 weights Q4_K working on CUDA. ~10% slower on GTX-1660, 16% slower on 4080. * k_quants: WIP super-blocks with 64 weights Q2_K working on CUDA. ~3% slower on GTX-1660, 10% slower on 4080. * k_quants: WIP super-blocks with 64 weights Q3_K working on CUDA. * k_quants: WIP super-blocks with 64 weights Q5_K working on CUDA, and with this CUDA is done. * k_quants: WIP super-blocks with 64 weights Q6_K working on ARM_NEON * k_quants: WIP super-blocks with 64 weights Q4_K working on ARM_NEON, but quite a bit slower than 256 weights * k_quants: WIP super-blocks with 64 weights Q2_K working on ARM_NEON, but quite a bit slower than 256 weights * k_quants: WIP super-blocks with 64 weights Q3_K working on ARM_NEON, but quite a bit slower than 256 weights. * k_quants: WIP super-blocks with 64 weights Q5_K working on ARM_NEON, but quite a bit slower than 256 weights. With that, we have full support for ARM_NEON, although performance is not quite there. * k_quants: WIP super-blocks with 64 weights Slightly more efficient Q3_K and Q5_K * k_quants: WIP super-blocks with 64 weights Another small improvement for Q3_K and Q5_K on ARM_NEON * k_quants: WIP super-blocks with 64 weights Yet another speedup for Q5_K on ARM_NEON. We are now within 10% of the QK_K = 256 version. * k_quants: WIP super-blocks with 64 weights * We are able to pass preprocessor macros to the Metal compiler * Q6_K works and is actually slightly more efficient than the QK_K = 256 version (25.2 ms vs 25.8 ms) * k_quants: WIP super-blocks with 64 weights Q4_K works on Metal and is actually slightly faster than QK_K = 256 (21.95 ms vs 24.0 ms). * k_quants: WIP super-blocks with 64 weights Q2_K works on Metal and is very slightly faster than QK_K = 256 (23.8 ms vs 24.2 ms). * k_quants: WIP super-blocks with 64 weights Q3_K works on Metal and is slightly faster than QK_K = 256 (26.6 ms vs 28.3 ms). * k_quants: WIP super-blocks with 64 weights Q5_K works on Metal and is slightly faster than QK_K = 256 (23.7 ms vs 26.3 ms). * k_quants: call them _K, not _k, also on Metal * k_quants: correctly define QK_K in llama.cpp * Fixed bug in q4_K quantization added with the 64-block addition * Simplify via lambda * k_quants: swicth Q3_K to 4-bit scales when QK_K = 64 Otherwise there isn't much benefit from this quantization type. There is some very slight loss in accuracy, but we reduce size by ~7%. E.g., for OpenLLaMA-3B, Q3_K_S perplexity is 8.6131 with 8-bit scales and 8.6352 with 4-bit, while file size decreases from 1.53G to 1.44G. * k_quants: switch Q4_K to 4-bit scales when QK_K = 64 Here the loss in accuracy is greater than for Q3_K, but the Q4_K points still move further to the left on the perplexity vs size curve. * k_quants: forgot to add the Metal changes in last commit * k_quants: change Q5_K to be type 0 when QK_K = 64 Still needs AVX2 implementation * k_quants: AVX2 implementation for new 64-weight Q5_K * k_quants: 10% faster ARM_NEON Q5_K dot product * k_quants: fixed issue caused by merging with master --------- Co-authored-by: Iwan Kawrakow <iwan.kawrakow@gmail.com>
2023-06-26 16:43:07 +00:00
for (int i = 0; i < nb; i++) {
ggml : quantization refactoring (#3833) * ggml : factor all quantization code in ggml-quants ggml-ci * ggml-quants : fix Zig and Swift builds + quantize tool ggml-ci * quantize : --pure option for disabling k-quant mixtures --------- Co-authored-by: cebtenzzre <cebtenzzre@gmail.com>
2023-10-29 16:32:28 +00:00
// load elements
vuint8mf2_t tx = __riscv_vle8_v_u8mf2(x[i].qs, vl);
k-quants : support for super-block size of 64 (#2001) * k_quants: WIP super-blocks with 64 weights * k_quants: WIP super-blocks with 64 weights Q6_K scalar and AVX2 works * k_quants: WIP super-blocks with 64 weights Q4_K scalar and AVX2 works * k_quants: WIP super-blocks with 64 weights Q2_K scalar and AVX2 works. Q2_K is way too slow (it is actually slower than the scalar implementation) * k_quants: WIP super-blocks with 64 weights Q3_K scalar and AVX2 works. * k_quants: WIP super-blocks with 64 weights Q5_K scalar and AVX2 works, and with that all k_quants are done on AVX2 and scalar * k_quants: WIP super-blocks with 64 weights Q6_K working on CUDA. Cannot make it run quite as gast as with super-blocks with 256 weigths: 8% slower on 4080, 20% slower on the 1660 (but there we fit 1 less layer on the GPU because pf the larger model size), so some fraction of these 20% is due to that, * k_quants: WIP super-blocks with 64 weights Q4_K working on CUDA. ~10% slower on GTX-1660, 16% slower on 4080. * k_quants: WIP super-blocks with 64 weights Q2_K working on CUDA. ~3% slower on GTX-1660, 10% slower on 4080. * k_quants: WIP super-blocks with 64 weights Q3_K working on CUDA. * k_quants: WIP super-blocks with 64 weights Q5_K working on CUDA, and with this CUDA is done. * k_quants: WIP super-blocks with 64 weights Q6_K working on ARM_NEON * k_quants: WIP super-blocks with 64 weights Q4_K working on ARM_NEON, but quite a bit slower than 256 weights * k_quants: WIP super-blocks with 64 weights Q2_K working on ARM_NEON, but quite a bit slower than 256 weights * k_quants: WIP super-blocks with 64 weights Q3_K working on ARM_NEON, but quite a bit slower than 256 weights. * k_quants: WIP super-blocks with 64 weights Q5_K working on ARM_NEON, but quite a bit slower than 256 weights. With that, we have full support for ARM_NEON, although performance is not quite there. * k_quants: WIP super-blocks with 64 weights Slightly more efficient Q3_K and Q5_K * k_quants: WIP super-blocks with 64 weights Another small improvement for Q3_K and Q5_K on ARM_NEON * k_quants: WIP super-blocks with 64 weights Yet another speedup for Q5_K on ARM_NEON. We are now within 10% of the QK_K = 256 version. * k_quants: WIP super-blocks with 64 weights * We are able to pass preprocessor macros to the Metal compiler * Q6_K works and is actually slightly more efficient than the QK_K = 256 version (25.2 ms vs 25.8 ms) * k_quants: WIP super-blocks with 64 weights Q4_K works on Metal and is actually slightly faster than QK_K = 256 (21.95 ms vs 24.0 ms). * k_quants: WIP super-blocks with 64 weights Q2_K works on Metal and is very slightly faster than QK_K = 256 (23.8 ms vs 24.2 ms). * k_quants: WIP super-blocks with 64 weights Q3_K works on Metal and is slightly faster than QK_K = 256 (26.6 ms vs 28.3 ms). * k_quants: WIP super-blocks with 64 weights Q5_K works on Metal and is slightly faster than QK_K = 256 (23.7 ms vs 26.3 ms). * k_quants: call them _K, not _k, also on Metal * k_quants: correctly define QK_K in llama.cpp * Fixed bug in q4_K quantization added with the 64-block addition * Simplify via lambda * k_quants: swicth Q3_K to 4-bit scales when QK_K = 64 Otherwise there isn't much benefit from this quantization type. There is some very slight loss in accuracy, but we reduce size by ~7%. E.g., for OpenLLaMA-3B, Q3_K_S perplexity is 8.6131 with 8-bit scales and 8.6352 with 4-bit, while file size decreases from 1.53G to 1.44G. * k_quants: switch Q4_K to 4-bit scales when QK_K = 64 Here the loss in accuracy is greater than for Q3_K, but the Q4_K points still move further to the left on the perplexity vs size curve. * k_quants: forgot to add the Metal changes in last commit * k_quants: change Q5_K to be type 0 when QK_K = 64 Still needs AVX2 implementation * k_quants: AVX2 implementation for new 64-weight Q5_K * k_quants: 10% faster ARM_NEON Q5_K dot product * k_quants: fixed issue caused by merging with master --------- Co-authored-by: Iwan Kawrakow <iwan.kawrakow@gmail.com>
2023-06-26 16:43:07 +00:00
ggml : quantization refactoring (#3833) * ggml : factor all quantization code in ggml-quants ggml-ci * ggml-quants : fix Zig and Swift builds + quantize tool ggml-ci * quantize : --pure option for disabling k-quant mixtures --------- Co-authored-by: cebtenzzre <cebtenzzre@gmail.com>
2023-10-29 16:32:28 +00:00
vint8mf2_t y0 = __riscv_vle8_v_i8mf2(y[i].qs, vl);
vint8mf2_t y1 = __riscv_vle8_v_i8mf2(y[i].qs+16, vl);
k-quants : support for super-block size of 64 (#2001) * k_quants: WIP super-blocks with 64 weights * k_quants: WIP super-blocks with 64 weights Q6_K scalar and AVX2 works * k_quants: WIP super-blocks with 64 weights Q4_K scalar and AVX2 works * k_quants: WIP super-blocks with 64 weights Q2_K scalar and AVX2 works. Q2_K is way too slow (it is actually slower than the scalar implementation) * k_quants: WIP super-blocks with 64 weights Q3_K scalar and AVX2 works. * k_quants: WIP super-blocks with 64 weights Q5_K scalar and AVX2 works, and with that all k_quants are done on AVX2 and scalar * k_quants: WIP super-blocks with 64 weights Q6_K working on CUDA. Cannot make it run quite as gast as with super-blocks with 256 weigths: 8% slower on 4080, 20% slower on the 1660 (but there we fit 1 less layer on the GPU because pf the larger model size), so some fraction of these 20% is due to that, * k_quants: WIP super-blocks with 64 weights Q4_K working on CUDA. ~10% slower on GTX-1660, 16% slower on 4080. * k_quants: WIP super-blocks with 64 weights Q2_K working on CUDA. ~3% slower on GTX-1660, 10% slower on 4080. * k_quants: WIP super-blocks with 64 weights Q3_K working on CUDA. * k_quants: WIP super-blocks with 64 weights Q5_K working on CUDA, and with this CUDA is done. * k_quants: WIP super-blocks with 64 weights Q6_K working on ARM_NEON * k_quants: WIP super-blocks with 64 weights Q4_K working on ARM_NEON, but quite a bit slower than 256 weights * k_quants: WIP super-blocks with 64 weights Q2_K working on ARM_NEON, but quite a bit slower than 256 weights * k_quants: WIP super-blocks with 64 weights Q3_K working on ARM_NEON, but quite a bit slower than 256 weights. * k_quants: WIP super-blocks with 64 weights Q5_K working on ARM_NEON, but quite a bit slower than 256 weights. With that, we have full support for ARM_NEON, although performance is not quite there. * k_quants: WIP super-blocks with 64 weights Slightly more efficient Q3_K and Q5_K * k_quants: WIP super-blocks with 64 weights Another small improvement for Q3_K and Q5_K on ARM_NEON * k_quants: WIP super-blocks with 64 weights Yet another speedup for Q5_K on ARM_NEON. We are now within 10% of the QK_K = 256 version. * k_quants: WIP super-blocks with 64 weights * We are able to pass preprocessor macros to the Metal compiler * Q6_K works and is actually slightly more efficient than the QK_K = 256 version (25.2 ms vs 25.8 ms) * k_quants: WIP super-blocks with 64 weights Q4_K works on Metal and is actually slightly faster than QK_K = 256 (21.95 ms vs 24.0 ms). * k_quants: WIP super-blocks with 64 weights Q2_K works on Metal and is very slightly faster than QK_K = 256 (23.8 ms vs 24.2 ms). * k_quants: WIP super-blocks with 64 weights Q3_K works on Metal and is slightly faster than QK_K = 256 (26.6 ms vs 28.3 ms). * k_quants: WIP super-blocks with 64 weights Q5_K works on Metal and is slightly faster than QK_K = 256 (23.7 ms vs 26.3 ms). * k_quants: call them _K, not _k, also on Metal * k_quants: correctly define QK_K in llama.cpp * Fixed bug in q4_K quantization added with the 64-block addition * Simplify via lambda * k_quants: swicth Q3_K to 4-bit scales when QK_K = 64 Otherwise there isn't much benefit from this quantization type. There is some very slight loss in accuracy, but we reduce size by ~7%. E.g., for OpenLLaMA-3B, Q3_K_S perplexity is 8.6131 with 8-bit scales and 8.6352 with 4-bit, while file size decreases from 1.53G to 1.44G. * k_quants: switch Q4_K to 4-bit scales when QK_K = 64 Here the loss in accuracy is greater than for Q3_K, but the Q4_K points still move further to the left on the perplexity vs size curve. * k_quants: forgot to add the Metal changes in last commit * k_quants: change Q5_K to be type 0 when QK_K = 64 Still needs AVX2 implementation * k_quants: AVX2 implementation for new 64-weight Q5_K * k_quants: 10% faster ARM_NEON Q5_K dot product * k_quants: fixed issue caused by merging with master --------- Co-authored-by: Iwan Kawrakow <iwan.kawrakow@gmail.com>
2023-06-26 16:43:07 +00:00
ggml : quantization refactoring (#3833) * ggml : factor all quantization code in ggml-quants ggml-ci * ggml-quants : fix Zig and Swift builds + quantize tool ggml-ci * quantize : --pure option for disabling k-quant mixtures --------- Co-authored-by: cebtenzzre <cebtenzzre@gmail.com>
2023-10-29 16:32:28 +00:00
// mask and store lower part of x, and then upper part
vuint8mf2_t x_a = __riscv_vand_vx_u8mf2(tx, 0x0F, vl);
vuint8mf2_t x_l = __riscv_vsrl_vx_u8mf2(tx, 0x04, vl);
k-quants : support for super-block size of 64 (#2001) * k_quants: WIP super-blocks with 64 weights * k_quants: WIP super-blocks with 64 weights Q6_K scalar and AVX2 works * k_quants: WIP super-blocks with 64 weights Q4_K scalar and AVX2 works * k_quants: WIP super-blocks with 64 weights Q2_K scalar and AVX2 works. Q2_K is way too slow (it is actually slower than the scalar implementation) * k_quants: WIP super-blocks with 64 weights Q3_K scalar and AVX2 works. * k_quants: WIP super-blocks with 64 weights Q5_K scalar and AVX2 works, and with that all k_quants are done on AVX2 and scalar * k_quants: WIP super-blocks with 64 weights Q6_K working on CUDA. Cannot make it run quite as gast as with super-blocks with 256 weigths: 8% slower on 4080, 20% slower on the 1660 (but there we fit 1 less layer on the GPU because pf the larger model size), so some fraction of these 20% is due to that, * k_quants: WIP super-blocks with 64 weights Q4_K working on CUDA. ~10% slower on GTX-1660, 16% slower on 4080. * k_quants: WIP super-blocks with 64 weights Q2_K working on CUDA. ~3% slower on GTX-1660, 10% slower on 4080. * k_quants: WIP super-blocks with 64 weights Q3_K working on CUDA. * k_quants: WIP super-blocks with 64 weights Q5_K working on CUDA, and with this CUDA is done. * k_quants: WIP super-blocks with 64 weights Q6_K working on ARM_NEON * k_quants: WIP super-blocks with 64 weights Q4_K working on ARM_NEON, but quite a bit slower than 256 weights * k_quants: WIP super-blocks with 64 weights Q2_K working on ARM_NEON, but quite a bit slower than 256 weights * k_quants: WIP super-blocks with 64 weights Q3_K working on ARM_NEON, but quite a bit slower than 256 weights. * k_quants: WIP super-blocks with 64 weights Q5_K working on ARM_NEON, but quite a bit slower than 256 weights. With that, we have full support for ARM_NEON, although performance is not quite there. * k_quants: WIP super-blocks with 64 weights Slightly more efficient Q3_K and Q5_K * k_quants: WIP super-blocks with 64 weights Another small improvement for Q3_K and Q5_K on ARM_NEON * k_quants: WIP super-blocks with 64 weights Yet another speedup for Q5_K on ARM_NEON. We are now within 10% of the QK_K = 256 version. * k_quants: WIP super-blocks with 64 weights * We are able to pass preprocessor macros to the Metal compiler * Q6_K works and is actually slightly more efficient than the QK_K = 256 version (25.2 ms vs 25.8 ms) * k_quants: WIP super-blocks with 64 weights Q4_K works on Metal and is actually slightly faster than QK_K = 256 (21.95 ms vs 24.0 ms). * k_quants: WIP super-blocks with 64 weights Q2_K works on Metal and is very slightly faster than QK_K = 256 (23.8 ms vs 24.2 ms). * k_quants: WIP super-blocks with 64 weights Q3_K works on Metal and is slightly faster than QK_K = 256 (26.6 ms vs 28.3 ms). * k_quants: WIP super-blocks with 64 weights Q5_K works on Metal and is slightly faster than QK_K = 256 (23.7 ms vs 26.3 ms). * k_quants: call them _K, not _k, also on Metal * k_quants: correctly define QK_K in llama.cpp * Fixed bug in q4_K quantization added with the 64-block addition * Simplify via lambda * k_quants: swicth Q3_K to 4-bit scales when QK_K = 64 Otherwise there isn't much benefit from this quantization type. There is some very slight loss in accuracy, but we reduce size by ~7%. E.g., for OpenLLaMA-3B, Q3_K_S perplexity is 8.6131 with 8-bit scales and 8.6352 with 4-bit, while file size decreases from 1.53G to 1.44G. * k_quants: switch Q4_K to 4-bit scales when QK_K = 64 Here the loss in accuracy is greater than for Q3_K, but the Q4_K points still move further to the left on the perplexity vs size curve. * k_quants: forgot to add the Metal changes in last commit * k_quants: change Q5_K to be type 0 when QK_K = 64 Still needs AVX2 implementation * k_quants: AVX2 implementation for new 64-weight Q5_K * k_quants: 10% faster ARM_NEON Q5_K dot product * k_quants: fixed issue caused by merging with master --------- Co-authored-by: Iwan Kawrakow <iwan.kawrakow@gmail.com>
2023-06-26 16:43:07 +00:00
ggml : quantization refactoring (#3833) * ggml : factor all quantization code in ggml-quants ggml-ci * ggml-quants : fix Zig and Swift builds + quantize tool ggml-ci * quantize : --pure option for disabling k-quant mixtures --------- Co-authored-by: cebtenzzre <cebtenzzre@gmail.com>
2023-10-29 16:32:28 +00:00
vint8mf2_t v0 = __riscv_vreinterpret_v_u8mf2_i8mf2(x_a);
vint8mf2_t v1 = __riscv_vreinterpret_v_u8mf2_i8mf2(x_l);
k-quants : support for super-block size of 64 (#2001) * k_quants: WIP super-blocks with 64 weights * k_quants: WIP super-blocks with 64 weights Q6_K scalar and AVX2 works * k_quants: WIP super-blocks with 64 weights Q4_K scalar and AVX2 works * k_quants: WIP super-blocks with 64 weights Q2_K scalar and AVX2 works. Q2_K is way too slow (it is actually slower than the scalar implementation) * k_quants: WIP super-blocks with 64 weights Q3_K scalar and AVX2 works. * k_quants: WIP super-blocks with 64 weights Q5_K scalar and AVX2 works, and with that all k_quants are done on AVX2 and scalar * k_quants: WIP super-blocks with 64 weights Q6_K working on CUDA. Cannot make it run quite as gast as with super-blocks with 256 weigths: 8% slower on 4080, 20% slower on the 1660 (but there we fit 1 less layer on the GPU because pf the larger model size), so some fraction of these 20% is due to that, * k_quants: WIP super-blocks with 64 weights Q4_K working on CUDA. ~10% slower on GTX-1660, 16% slower on 4080. * k_quants: WIP super-blocks with 64 weights Q2_K working on CUDA. ~3% slower on GTX-1660, 10% slower on 4080. * k_quants: WIP super-blocks with 64 weights Q3_K working on CUDA. * k_quants: WIP super-blocks with 64 weights Q5_K working on CUDA, and with this CUDA is done. * k_quants: WIP super-blocks with 64 weights Q6_K working on ARM_NEON * k_quants: WIP super-blocks with 64 weights Q4_K working on ARM_NEON, but quite a bit slower than 256 weights * k_quants: WIP super-blocks with 64 weights Q2_K working on ARM_NEON, but quite a bit slower than 256 weights * k_quants: WIP super-blocks with 64 weights Q3_K working on ARM_NEON, but quite a bit slower than 256 weights. * k_quants: WIP super-blocks with 64 weights Q5_K working on ARM_NEON, but quite a bit slower than 256 weights. With that, we have full support for ARM_NEON, although performance is not quite there. * k_quants: WIP super-blocks with 64 weights Slightly more efficient Q3_K and Q5_K * k_quants: WIP super-blocks with 64 weights Another small improvement for Q3_K and Q5_K on ARM_NEON * k_quants: WIP super-blocks with 64 weights Yet another speedup for Q5_K on ARM_NEON. We are now within 10% of the QK_K = 256 version. * k_quants: WIP super-blocks with 64 weights * We are able to pass preprocessor macros to the Metal compiler * Q6_K works and is actually slightly more efficient than the QK_K = 256 version (25.2 ms vs 25.8 ms) * k_quants: WIP super-blocks with 64 weights Q4_K works on Metal and is actually slightly faster than QK_K = 256 (21.95 ms vs 24.0 ms). * k_quants: WIP super-blocks with 64 weights Q2_K works on Metal and is very slightly faster than QK_K = 256 (23.8 ms vs 24.2 ms). * k_quants: WIP super-blocks with 64 weights Q3_K works on Metal and is slightly faster than QK_K = 256 (26.6 ms vs 28.3 ms). * k_quants: WIP super-blocks with 64 weights Q5_K works on Metal and is slightly faster than QK_K = 256 (23.7 ms vs 26.3 ms). * k_quants: call them _K, not _k, also on Metal * k_quants: correctly define QK_K in llama.cpp * Fixed bug in q4_K quantization added with the 64-block addition * Simplify via lambda * k_quants: swicth Q3_K to 4-bit scales when QK_K = 64 Otherwise there isn't much benefit from this quantization type. There is some very slight loss in accuracy, but we reduce size by ~7%. E.g., for OpenLLaMA-3B, Q3_K_S perplexity is 8.6131 with 8-bit scales and 8.6352 with 4-bit, while file size decreases from 1.53G to 1.44G. * k_quants: switch Q4_K to 4-bit scales when QK_K = 64 Here the loss in accuracy is greater than for Q3_K, but the Q4_K points still move further to the left on the perplexity vs size curve. * k_quants: forgot to add the Metal changes in last commit * k_quants: change Q5_K to be type 0 when QK_K = 64 Still needs AVX2 implementation * k_quants: AVX2 implementation for new 64-weight Q5_K * k_quants: 10% faster ARM_NEON Q5_K dot product * k_quants: fixed issue caused by merging with master --------- Co-authored-by: Iwan Kawrakow <iwan.kawrakow@gmail.com>
2023-06-26 16:43:07 +00:00
ggml : quantization refactoring (#3833) * ggml : factor all quantization code in ggml-quants ggml-ci * ggml-quants : fix Zig and Swift builds + quantize tool ggml-ci * quantize : --pure option for disabling k-quant mixtures --------- Co-authored-by: cebtenzzre <cebtenzzre@gmail.com>
2023-10-29 16:32:28 +00:00
vint16m1_t vec_mul1 = __riscv_vwmul_vv_i16m1(v0, y0, vl);
vint16m1_t vec_mul2 = __riscv_vwmul_vv_i16m1(v1, y1, vl);
vint32m1_t vec_zero = __riscv_vmv_v_x_i32m1(0, vl);
vint32m1_t vs1 = __riscv_vwredsum_vs_i16m1_i32m1(vec_mul1, vec_zero, vl);
vint32m1_t vs2 = __riscv_vwredsum_vs_i16m1_i32m1(vec_mul2, vs1, vl);
int sumi = __riscv_vmv_x_s_i32m1_i32(vs2);
ggml : move FP16 <-> FP32 code to ggml-impl.h (#3861) * ggml : move FP16 <-> FP32 stuff to ggml-impl.h ggml-ci * tests : fix ARM build * ggml : explicitly initialize deprecated type traits * ggml : add math.h to ggml-impl.h * ggml : remove duplicate static assert macros * ggml : prefix lookup tables with ggml_ ggml-ci * ggml-impl : move extern "C" to start of file
2023-10-30 17:19:15 +00:00
sumf += (GGML_FP16_TO_FP32(x[i].d)*y[i].d)*sumi + GGML_FP16_TO_FP32(x[i].m)*y[i].s;
ggml : quantization refactoring (#3833) * ggml : factor all quantization code in ggml-quants ggml-ci * ggml-quants : fix Zig and Swift builds + quantize tool ggml-ci * quantize : --pure option for disabling k-quant mixtures --------- Co-authored-by: cebtenzzre <cebtenzzre@gmail.com>
2023-10-29 16:32:28 +00:00
}
*s = sumf;
#else
// scalar
float sumf = 0.0;
for (int i = 0; i < nb; i++) {
int sumi = 0;
for (int j = 0; j < qk/2; ++j) {
const int v0 = (x[i].qs[j] & 0x0F);
const int v1 = (x[i].qs[j] >> 4);
sumi += (v0 * y[i].qs[j]) + (v1 * y[i].qs[j + qk/2]);
k-quants : support for super-block size of 64 (#2001) * k_quants: WIP super-blocks with 64 weights * k_quants: WIP super-blocks with 64 weights Q6_K scalar and AVX2 works * k_quants: WIP super-blocks with 64 weights Q4_K scalar and AVX2 works * k_quants: WIP super-blocks with 64 weights Q2_K scalar and AVX2 works. Q2_K is way too slow (it is actually slower than the scalar implementation) * k_quants: WIP super-blocks with 64 weights Q3_K scalar and AVX2 works. * k_quants: WIP super-blocks with 64 weights Q5_K scalar and AVX2 works, and with that all k_quants are done on AVX2 and scalar * k_quants: WIP super-blocks with 64 weights Q6_K working on CUDA. Cannot make it run quite as gast as with super-blocks with 256 weigths: 8% slower on 4080, 20% slower on the 1660 (but there we fit 1 less layer on the GPU because pf the larger model size), so some fraction of these 20% is due to that, * k_quants: WIP super-blocks with 64 weights Q4_K working on CUDA. ~10% slower on GTX-1660, 16% slower on 4080. * k_quants: WIP super-blocks with 64 weights Q2_K working on CUDA. ~3% slower on GTX-1660, 10% slower on 4080. * k_quants: WIP super-blocks with 64 weights Q3_K working on CUDA. * k_quants: WIP super-blocks with 64 weights Q5_K working on CUDA, and with this CUDA is done. * k_quants: WIP super-blocks with 64 weights Q6_K working on ARM_NEON * k_quants: WIP super-blocks with 64 weights Q4_K working on ARM_NEON, but quite a bit slower than 256 weights * k_quants: WIP super-blocks with 64 weights Q2_K working on ARM_NEON, but quite a bit slower than 256 weights * k_quants: WIP super-blocks with 64 weights Q3_K working on ARM_NEON, but quite a bit slower than 256 weights. * k_quants: WIP super-blocks with 64 weights Q5_K working on ARM_NEON, but quite a bit slower than 256 weights. With that, we have full support for ARM_NEON, although performance is not quite there. * k_quants: WIP super-blocks with 64 weights Slightly more efficient Q3_K and Q5_K * k_quants: WIP super-blocks with 64 weights Another small improvement for Q3_K and Q5_K on ARM_NEON * k_quants: WIP super-blocks with 64 weights Yet another speedup for Q5_K on ARM_NEON. We are now within 10% of the QK_K = 256 version. * k_quants: WIP super-blocks with 64 weights * We are able to pass preprocessor macros to the Metal compiler * Q6_K works and is actually slightly more efficient than the QK_K = 256 version (25.2 ms vs 25.8 ms) * k_quants: WIP super-blocks with 64 weights Q4_K works on Metal and is actually slightly faster than QK_K = 256 (21.95 ms vs 24.0 ms). * k_quants: WIP super-blocks with 64 weights Q2_K works on Metal and is very slightly faster than QK_K = 256 (23.8 ms vs 24.2 ms). * k_quants: WIP super-blocks with 64 weights Q3_K works on Metal and is slightly faster than QK_K = 256 (26.6 ms vs 28.3 ms). * k_quants: WIP super-blocks with 64 weights Q5_K works on Metal and is slightly faster than QK_K = 256 (23.7 ms vs 26.3 ms). * k_quants: call them _K, not _k, also on Metal * k_quants: correctly define QK_K in llama.cpp * Fixed bug in q4_K quantization added with the 64-block addition * Simplify via lambda * k_quants: swicth Q3_K to 4-bit scales when QK_K = 64 Otherwise there isn't much benefit from this quantization type. There is some very slight loss in accuracy, but we reduce size by ~7%. E.g., for OpenLLaMA-3B, Q3_K_S perplexity is 8.6131 with 8-bit scales and 8.6352 with 4-bit, while file size decreases from 1.53G to 1.44G. * k_quants: switch Q4_K to 4-bit scales when QK_K = 64 Here the loss in accuracy is greater than for Q3_K, but the Q4_K points still move further to the left on the perplexity vs size curve. * k_quants: forgot to add the Metal changes in last commit * k_quants: change Q5_K to be type 0 when QK_K = 64 Still needs AVX2 implementation * k_quants: AVX2 implementation for new 64-weight Q5_K * k_quants: 10% faster ARM_NEON Q5_K dot product * k_quants: fixed issue caused by merging with master --------- Co-authored-by: Iwan Kawrakow <iwan.kawrakow@gmail.com>
2023-06-26 16:43:07 +00:00
}
ggml : quantization refactoring (#3833) * ggml : factor all quantization code in ggml-quants ggml-ci * ggml-quants : fix Zig and Swift builds + quantize tool ggml-ci * quantize : --pure option for disabling k-quant mixtures --------- Co-authored-by: cebtenzzre <cebtenzzre@gmail.com>
2023-10-29 16:32:28 +00:00
ggml : move FP16 <-> FP32 code to ggml-impl.h (#3861) * ggml : move FP16 <-> FP32 stuff to ggml-impl.h ggml-ci * tests : fix ARM build * ggml : explicitly initialize deprecated type traits * ggml : add math.h to ggml-impl.h * ggml : remove duplicate static assert macros * ggml : prefix lookup tables with ggml_ ggml-ci * ggml-impl : move extern "C" to start of file
2023-10-30 17:19:15 +00:00
sumf += (GGML_FP16_TO_FP32(x[i].d)*y[i].d)*sumi + GGML_FP16_TO_FP32(x[i].m)*y[i].s;
k-quants : support for super-block size of 64 (#2001) * k_quants: WIP super-blocks with 64 weights * k_quants: WIP super-blocks with 64 weights Q6_K scalar and AVX2 works * k_quants: WIP super-blocks with 64 weights Q4_K scalar and AVX2 works * k_quants: WIP super-blocks with 64 weights Q2_K scalar and AVX2 works. Q2_K is way too slow (it is actually slower than the scalar implementation) * k_quants: WIP super-blocks with 64 weights Q3_K scalar and AVX2 works. * k_quants: WIP super-blocks with 64 weights Q5_K scalar and AVX2 works, and with that all k_quants are done on AVX2 and scalar * k_quants: WIP super-blocks with 64 weights Q6_K working on CUDA. Cannot make it run quite as gast as with super-blocks with 256 weigths: 8% slower on 4080, 20% slower on the 1660 (but there we fit 1 less layer on the GPU because pf the larger model size), so some fraction of these 20% is due to that, * k_quants: WIP super-blocks with 64 weights Q4_K working on CUDA. ~10% slower on GTX-1660, 16% slower on 4080. * k_quants: WIP super-blocks with 64 weights Q2_K working on CUDA. ~3% slower on GTX-1660, 10% slower on 4080. * k_quants: WIP super-blocks with 64 weights Q3_K working on CUDA. * k_quants: WIP super-blocks with 64 weights Q5_K working on CUDA, and with this CUDA is done. * k_quants: WIP super-blocks with 64 weights Q6_K working on ARM_NEON * k_quants: WIP super-blocks with 64 weights Q4_K working on ARM_NEON, but quite a bit slower than 256 weights * k_quants: WIP super-blocks with 64 weights Q2_K working on ARM_NEON, but quite a bit slower than 256 weights * k_quants: WIP super-blocks with 64 weights Q3_K working on ARM_NEON, but quite a bit slower than 256 weights. * k_quants: WIP super-blocks with 64 weights Q5_K working on ARM_NEON, but quite a bit slower than 256 weights. With that, we have full support for ARM_NEON, although performance is not quite there. * k_quants: WIP super-blocks with 64 weights Slightly more efficient Q3_K and Q5_K * k_quants: WIP super-blocks with 64 weights Another small improvement for Q3_K and Q5_K on ARM_NEON * k_quants: WIP super-blocks with 64 weights Yet another speedup for Q5_K on ARM_NEON. We are now within 10% of the QK_K = 256 version. * k_quants: WIP super-blocks with 64 weights * We are able to pass preprocessor macros to the Metal compiler * Q6_K works and is actually slightly more efficient than the QK_K = 256 version (25.2 ms vs 25.8 ms) * k_quants: WIP super-blocks with 64 weights Q4_K works on Metal and is actually slightly faster than QK_K = 256 (21.95 ms vs 24.0 ms). * k_quants: WIP super-blocks with 64 weights Q2_K works on Metal and is very slightly faster than QK_K = 256 (23.8 ms vs 24.2 ms). * k_quants: WIP super-blocks with 64 weights Q3_K works on Metal and is slightly faster than QK_K = 256 (26.6 ms vs 28.3 ms). * k_quants: WIP super-blocks with 64 weights Q5_K works on Metal and is slightly faster than QK_K = 256 (23.7 ms vs 26.3 ms). * k_quants: call them _K, not _k, also on Metal * k_quants: correctly define QK_K in llama.cpp * Fixed bug in q4_K quantization added with the 64-block addition * Simplify via lambda * k_quants: swicth Q3_K to 4-bit scales when QK_K = 64 Otherwise there isn't much benefit from this quantization type. There is some very slight loss in accuracy, but we reduce size by ~7%. E.g., for OpenLLaMA-3B, Q3_K_S perplexity is 8.6131 with 8-bit scales and 8.6352 with 4-bit, while file size decreases from 1.53G to 1.44G. * k_quants: switch Q4_K to 4-bit scales when QK_K = 64 Here the loss in accuracy is greater than for Q3_K, but the Q4_K points still move further to the left on the perplexity vs size curve. * k_quants: forgot to add the Metal changes in last commit * k_quants: change Q5_K to be type 0 when QK_K = 64 Still needs AVX2 implementation * k_quants: AVX2 implementation for new 64-weight Q5_K * k_quants: 10% faster ARM_NEON Q5_K dot product * k_quants: fixed issue caused by merging with master --------- Co-authored-by: Iwan Kawrakow <iwan.kawrakow@gmail.com>
2023-06-26 16:43:07 +00:00
}
ggml : add SOTA 2,3,4,5,6 bit k-quantizations (#1684) * Starting to add k-quantization to ggml I think it is better to have quantization separate from ggml. For now just adding the k-quants there, but it would be better to also factor out the existing ggml quantizations. * Adding Q3_K and Q8_K (de)-quantization * Q3_K now working on CUDA and AVX2/scalar CUDA is not ideal - ~50% slower than Q4_0 for single token prediction, about the same in batch mode (perplexity). CPU single token is ~55 ms (on Ryzen 7950X). * Some improvement for Q3_K on CUDA It is now ~22.5 ms/token on my GPU, so ~30% slower than Q4_0. * Some more CUDA optimizations for Q3_K Single token is now 20.5 ms/token (~20% slower than Q4_0). Perplexity is on par with Q4_0. * Adding Q4_K - scalar, AVX2, CUDA Performance is the same or perhaps very slightly better than Q4_0 on the CPU. On the GPU, single token prediction is ~10% better than Q4_0, batch mode (perplexity is about the same). * Adding Q6_K - scalar, AVX2, CUDA Performance is ~40% lower compared to Q4_K on the CPU. This is to be expected, considering that we are memory bound on the CPU and the 6-bit model is ~44% larger than the 4-bit. On the GPU, single token prediction is ~6% lower than Q4_0, batch mode (perplexity) is even closer (but still slower). * Adding Q5_K - scalar, AVX2, CUDA Performance is ~20% lower compared to Q4_K on the CPU. This is to be expected, considering that we are memory bound on the CPU and the 5-bit model is ~22% larger than the 4-bit. On the GPU, single token prediction is about the same as Q4_0 for both, single token and batch prediction. * Per convention, all QX_K quantizations use Q5_K for output.weight * Adding quantization mixes * Quantization mixes: didn't quite get what I wanted in the last commit * Q4_K dot product for ARM_NEON * Q6_K dot product for ARM_NEON * Q5_K dot product for ARM_NEON * Adding Q3_K dot for ARM_NEON It is 22% slower than Q4_K, despite the smaller model size. On x86_64, where we are memory bound, the Q3_K model is quite a bit faster than Q4_K. * A very slightly faster ARM_NEON Q3_K dot * Adding Q2_K - just CUDA for now Token prediction is pretty good - about 15.5 ms on a RTX 4080. Perplexity is about the same as Q4_K. * Adding scalar and AVX2 Q2_K dot * Adding ARM_NEON Q2_K dot About the same performance as Q4_K. * A slightly faster ARM_NEON Q2_K dot Single token prediction is now ~36 ms on M2 Max. The code is much simpler too. * Fixed bug in Q2_K CUDA dot product kernel Stranegly enough, for the few prompts I tried with the 7B model the responses looked perfectly reasonable. Only realized something is not quite right when I tried the larger models and started getting nonse back. In any case, Q2_K single token evaluation time on an RTX 4080 in a Ryzen7950X box iusing CUDA and model fully loaded on the GPU are ~15.5 ms for 7B, ~25.4 ms for 13B, and ~55.8 ms for 30B. The max number of layers that fit in VRAM for The 65B is 32. With that, we get ~330 ms per token, which is not that much faster than just running on the CPU (~470 ms per token). * Don't print zeros/NaNs when no count histogram has been collected * A 10% faster CUDA vector dot kernel for Q3_K Q3_K is now running at ~18.5 ms / token on CUDA, so the gap to Q4_0 is only 10%. It seems memory acccess pattern is more important for performance than the amount of computation the kernel does. * A slightly daster Q4_K AVX2 dot product For perplexity, where we are less memory bound, time per pass drops by ~5%. Barely measurable difference for single token prediction. * A slightly faster ARM_NEON A4_K dot product * Minor * Fix quantization error test We cannot possibly be expecting rmse < 0.002 for 2- and 3-bit quantization variants. * Fix docker build I have been sloppy with vector reinterpret casts on ARM_NEON. It seems clang is very forgiving in that regard. * Added forgotten ggml.o dependence on k_quants.h to the Makefile * Had unintentionally committed the Makefile with -Ofast enabled * ggml : rename k_quants -> ggml-quants-k, use lowercase in code --------- Co-authored-by: Iwan Kawrakow <iwan.kawrakow@gmail.com> Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
2023-06-05 19:56:18 +00:00
ggml : quantization refactoring (#3833) * ggml : factor all quantization code in ggml-quants ggml-ci * ggml-quants : fix Zig and Swift builds + quantize tool ggml-ci * quantize : --pure option for disabling k-quant mixtures --------- Co-authored-by: cebtenzzre <cebtenzzre@gmail.com>
2023-10-29 16:32:28 +00:00
*s = sumf;
#endif
ggml : add SOTA 2,3,4,5,6 bit k-quantizations (#1684) * Starting to add k-quantization to ggml I think it is better to have quantization separate from ggml. For now just adding the k-quants there, but it would be better to also factor out the existing ggml quantizations. * Adding Q3_K and Q8_K (de)-quantization * Q3_K now working on CUDA and AVX2/scalar CUDA is not ideal - ~50% slower than Q4_0 for single token prediction, about the same in batch mode (perplexity). CPU single token is ~55 ms (on Ryzen 7950X). * Some improvement for Q3_K on CUDA It is now ~22.5 ms/token on my GPU, so ~30% slower than Q4_0. * Some more CUDA optimizations for Q3_K Single token is now 20.5 ms/token (~20% slower than Q4_0). Perplexity is on par with Q4_0. * Adding Q4_K - scalar, AVX2, CUDA Performance is the same or perhaps very slightly better than Q4_0 on the CPU. On the GPU, single token prediction is ~10% better than Q4_0, batch mode (perplexity is about the same). * Adding Q6_K - scalar, AVX2, CUDA Performance is ~40% lower compared to Q4_K on the CPU. This is to be expected, considering that we are memory bound on the CPU and the 6-bit model is ~44% larger than the 4-bit. On the GPU, single token prediction is ~6% lower than Q4_0, batch mode (perplexity) is even closer (but still slower). * Adding Q5_K - scalar, AVX2, CUDA Performance is ~20% lower compared to Q4_K on the CPU. This is to be expected, considering that we are memory bound on the CPU and the 5-bit model is ~22% larger than the 4-bit. On the GPU, single token prediction is about the same as Q4_0 for both, single token and batch prediction. * Per convention, all QX_K quantizations use Q5_K for output.weight * Adding quantization mixes * Quantization mixes: didn't quite get what I wanted in the last commit * Q4_K dot product for ARM_NEON * Q6_K dot product for ARM_NEON * Q5_K dot product for ARM_NEON * Adding Q3_K dot for ARM_NEON It is 22% slower than Q4_K, despite the smaller model size. On x86_64, where we are memory bound, the Q3_K model is quite a bit faster than Q4_K. * A very slightly faster ARM_NEON Q3_K dot * Adding Q2_K - just CUDA for now Token prediction is pretty good - about 15.5 ms on a RTX 4080. Perplexity is about the same as Q4_K. * Adding scalar and AVX2 Q2_K dot * Adding ARM_NEON Q2_K dot About the same performance as Q4_K. * A slightly faster ARM_NEON Q2_K dot Single token prediction is now ~36 ms on M2 Max. The code is much simpler too. * Fixed bug in Q2_K CUDA dot product kernel Stranegly enough, for the few prompts I tried with the 7B model the responses looked perfectly reasonable. Only realized something is not quite right when I tried the larger models and started getting nonse back. In any case, Q2_K single token evaluation time on an RTX 4080 in a Ryzen7950X box iusing CUDA and model fully loaded on the GPU are ~15.5 ms for 7B, ~25.4 ms for 13B, and ~55.8 ms for 30B. The max number of layers that fit in VRAM for The 65B is 32. With that, we get ~330 ms per token, which is not that much faster than just running on the CPU (~470 ms per token). * Don't print zeros/NaNs when no count histogram has been collected * A 10% faster CUDA vector dot kernel for Q3_K Q3_K is now running at ~18.5 ms / token on CUDA, so the gap to Q4_0 is only 10%. It seems memory acccess pattern is more important for performance than the amount of computation the kernel does. * A slightly daster Q4_K AVX2 dot product For perplexity, where we are less memory bound, time per pass drops by ~5%. Barely measurable difference for single token prediction. * A slightly faster ARM_NEON A4_K dot product * Minor * Fix quantization error test We cannot possibly be expecting rmse < 0.002 for 2- and 3-bit quantization variants. * Fix docker build I have been sloppy with vector reinterpret casts on ARM_NEON. It seems clang is very forgiving in that regard. * Added forgotten ggml.o dependence on k_quants.h to the Makefile * Had unintentionally committed the Makefile with -Ofast enabled * ggml : rename k_quants -> ggml-quants-k, use lowercase in code --------- Co-authored-by: Iwan Kawrakow <iwan.kawrakow@gmail.com> Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
2023-06-05 19:56:18 +00:00
}
ggml : quantization refactoring (#3833) * ggml : factor all quantization code in ggml-quants ggml-ci * ggml-quants : fix Zig and Swift builds + quantize tool ggml-ci * quantize : --pure option for disabling k-quant mixtures --------- Co-authored-by: cebtenzzre <cebtenzzre@gmail.com>
2023-10-29 16:32:28 +00:00
void ggml_vec_dot_q5_0_q8_0(const int n, float * restrict s, const void * restrict vx, const void * restrict vy) {
const int qk = QK8_0;
const int nb = n / qk;
ggml : add SOTA 2,3,4,5,6 bit k-quantizations (#1684) * Starting to add k-quantization to ggml I think it is better to have quantization separate from ggml. For now just adding the k-quants there, but it would be better to also factor out the existing ggml quantizations. * Adding Q3_K and Q8_K (de)-quantization * Q3_K now working on CUDA and AVX2/scalar CUDA is not ideal - ~50% slower than Q4_0 for single token prediction, about the same in batch mode (perplexity). CPU single token is ~55 ms (on Ryzen 7950X). * Some improvement for Q3_K on CUDA It is now ~22.5 ms/token on my GPU, so ~30% slower than Q4_0. * Some more CUDA optimizations for Q3_K Single token is now 20.5 ms/token (~20% slower than Q4_0). Perplexity is on par with Q4_0. * Adding Q4_K - scalar, AVX2, CUDA Performance is the same or perhaps very slightly better than Q4_0 on the CPU. On the GPU, single token prediction is ~10% better than Q4_0, batch mode (perplexity is about the same). * Adding Q6_K - scalar, AVX2, CUDA Performance is ~40% lower compared to Q4_K on the CPU. This is to be expected, considering that we are memory bound on the CPU and the 6-bit model is ~44% larger than the 4-bit. On the GPU, single token prediction is ~6% lower than Q4_0, batch mode (perplexity) is even closer (but still slower). * Adding Q5_K - scalar, AVX2, CUDA Performance is ~20% lower compared to Q4_K on the CPU. This is to be expected, considering that we are memory bound on the CPU and the 5-bit model is ~22% larger than the 4-bit. On the GPU, single token prediction is about the same as Q4_0 for both, single token and batch prediction. * Per convention, all QX_K quantizations use Q5_K for output.weight * Adding quantization mixes * Quantization mixes: didn't quite get what I wanted in the last commit * Q4_K dot product for ARM_NEON * Q6_K dot product for ARM_NEON * Q5_K dot product for ARM_NEON * Adding Q3_K dot for ARM_NEON It is 22% slower than Q4_K, despite the smaller model size. On x86_64, where we are memory bound, the Q3_K model is quite a bit faster than Q4_K. * A very slightly faster ARM_NEON Q3_K dot * Adding Q2_K - just CUDA for now Token prediction is pretty good - about 15.5 ms on a RTX 4080. Perplexity is about the same as Q4_K. * Adding scalar and AVX2 Q2_K dot * Adding ARM_NEON Q2_K dot About the same performance as Q4_K. * A slightly faster ARM_NEON Q2_K dot Single token prediction is now ~36 ms on M2 Max. The code is much simpler too. * Fixed bug in Q2_K CUDA dot product kernel Stranegly enough, for the few prompts I tried with the 7B model the responses looked perfectly reasonable. Only realized something is not quite right when I tried the larger models and started getting nonse back. In any case, Q2_K single token evaluation time on an RTX 4080 in a Ryzen7950X box iusing CUDA and model fully loaded on the GPU are ~15.5 ms for 7B, ~25.4 ms for 13B, and ~55.8 ms for 30B. The max number of layers that fit in VRAM for The 65B is 32. With that, we get ~330 ms per token, which is not that much faster than just running on the CPU (~470 ms per token). * Don't print zeros/NaNs when no count histogram has been collected * A 10% faster CUDA vector dot kernel for Q3_K Q3_K is now running at ~18.5 ms / token on CUDA, so the gap to Q4_0 is only 10%. It seems memory acccess pattern is more important for performance than the amount of computation the kernel does. * A slightly daster Q4_K AVX2 dot product For perplexity, where we are less memory bound, time per pass drops by ~5%. Barely measurable difference for single token prediction. * A slightly faster ARM_NEON A4_K dot product * Minor * Fix quantization error test We cannot possibly be expecting rmse < 0.002 for 2- and 3-bit quantization variants. * Fix docker build I have been sloppy with vector reinterpret casts on ARM_NEON. It seems clang is very forgiving in that regard. * Added forgotten ggml.o dependence on k_quants.h to the Makefile * Had unintentionally committed the Makefile with -Ofast enabled * ggml : rename k_quants -> ggml-quants-k, use lowercase in code --------- Co-authored-by: Iwan Kawrakow <iwan.kawrakow@gmail.com> Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
2023-06-05 19:56:18 +00:00
ggml : quantization refactoring (#3833) * ggml : factor all quantization code in ggml-quants ggml-ci * ggml-quants : fix Zig and Swift builds + quantize tool ggml-ci * quantize : --pure option for disabling k-quant mixtures --------- Co-authored-by: cebtenzzre <cebtenzzre@gmail.com>
2023-10-29 16:32:28 +00:00
assert(n % qk == 0);
assert(qk == QK5_0);
const block_q5_0 * restrict x = vx;
const block_q8_0 * restrict y = vy;
#if defined(__ARM_NEON)
float32x4_t sumv0 = vdupq_n_f32(0.0f);
float32x4_t sumv1 = vdupq_n_f32(0.0f);
uint32_t qh0;
uint32_t qh1;
uint64_t tmp0[4];
uint64_t tmp1[4];
assert(nb % 2 == 0); // TODO: handle odd nb
for (int i = 0; i < nb; i += 2) {
const block_q5_0 * restrict x0 = &x[i];
const block_q5_0 * restrict x1 = &x[i + 1];
const block_q8_0 * restrict y0 = &y[i];
const block_q8_0 * restrict y1 = &y[i + 1];
const uint8x16_t m4b = vdupq_n_u8(0x0F);
// extract the 5th bit via lookup table ((!b) << 4)
memcpy(&qh0, x0->qh, sizeof(qh0));
memcpy(&qh1, x1->qh, sizeof(qh1));
tmp0[0] = table_b2b_1[(qh0 >> 0) & 0xFF];
tmp0[1] = table_b2b_1[(qh0 >> 8) & 0xFF];
tmp0[2] = table_b2b_1[(qh0 >> 16) & 0xFF];
tmp0[3] = table_b2b_1[(qh0 >> 24) ];
tmp1[0] = table_b2b_1[(qh1 >> 0) & 0xFF];
tmp1[1] = table_b2b_1[(qh1 >> 8) & 0xFF];
tmp1[2] = table_b2b_1[(qh1 >> 16) & 0xFF];
tmp1[3] = table_b2b_1[(qh1 >> 24) ];
const int8x16_t qhl0 = vld1q_s8((const int8_t *)(tmp0 + 0));
const int8x16_t qhh0 = vld1q_s8((const int8_t *)(tmp0 + 2));
const int8x16_t qhl1 = vld1q_s8((const int8_t *)(tmp1 + 0));
const int8x16_t qhh1 = vld1q_s8((const int8_t *)(tmp1 + 2));
const uint8x16_t v0_0 = vld1q_u8(x0->qs);
const uint8x16_t v0_1 = vld1q_u8(x1->qs);
// 4-bit -> 8-bit
int8x16_t v0_0l = vreinterpretq_s8_u8(vandq_u8 (v0_0, m4b));
int8x16_t v0_0h = vreinterpretq_s8_u8(vshrq_n_u8(v0_0, 4));
int8x16_t v0_1l = vreinterpretq_s8_u8(vandq_u8 (v0_1, m4b));
int8x16_t v0_1h = vreinterpretq_s8_u8(vshrq_n_u8(v0_1, 4));
// add high bit and sub 16 (equivalent to sub 0x10 when bit is zero)
const int8x16_t v0_0lf = vsubq_s8(v0_0l, qhl0);
const int8x16_t v0_0hf = vsubq_s8(v0_0h, qhh0);
const int8x16_t v0_1lf = vsubq_s8(v0_1l, qhl1);
const int8x16_t v0_1hf = vsubq_s8(v0_1h, qhh1);
// load y
const int8x16_t v1_0l = vld1q_s8(y0->qs);
const int8x16_t v1_0h = vld1q_s8(y0->qs + 16);
const int8x16_t v1_1l = vld1q_s8(y1->qs);
const int8x16_t v1_1h = vld1q_s8(y1->qs + 16);
sumv0 = vmlaq_n_f32(sumv0, vcvtq_f32_s32(vaddq_s32(
ggml : add ggml_vdotq_s32 alias (#4715) ggml-ci
2023-12-31 09:43:31 +00:00
ggml_vdotq_s32(vdupq_n_s32(0), v0_0lf, v1_0l),
ggml_vdotq_s32(vdupq_n_s32(0), v0_0hf, v1_0h))), GGML_FP16_TO_FP32(x0->d)*GGML_FP16_TO_FP32(y0->d));
ggml : quantization refactoring (#3833) * ggml : factor all quantization code in ggml-quants ggml-ci * ggml-quants : fix Zig and Swift builds + quantize tool ggml-ci * quantize : --pure option for disabling k-quant mixtures --------- Co-authored-by: cebtenzzre <cebtenzzre@gmail.com>
2023-10-29 16:32:28 +00:00
sumv1 = vmlaq_n_f32(sumv1, vcvtq_f32_s32(vaddq_s32(
ggml : add ggml_vdotq_s32 alias (#4715) ggml-ci
2023-12-31 09:43:31 +00:00
ggml_vdotq_s32(vdupq_n_s32(0), v0_1lf, v1_1l),
ggml_vdotq_s32(vdupq_n_s32(0), v0_1hf, v1_1h))), GGML_FP16_TO_FP32(x1->d)*GGML_FP16_TO_FP32(y1->d));
ggml : add SOTA 2,3,4,5,6 bit k-quantizations (#1684) * Starting to add k-quantization to ggml I think it is better to have quantization separate from ggml. For now just adding the k-quants there, but it would be better to also factor out the existing ggml quantizations. * Adding Q3_K and Q8_K (de)-quantization * Q3_K now working on CUDA and AVX2/scalar CUDA is not ideal - ~50% slower than Q4_0 for single token prediction, about the same in batch mode (perplexity). CPU single token is ~55 ms (on Ryzen 7950X). * Some improvement for Q3_K on CUDA It is now ~22.5 ms/token on my GPU, so ~30% slower than Q4_0. * Some more CUDA optimizations for Q3_K Single token is now 20.5 ms/token (~20% slower than Q4_0). Perplexity is on par with Q4_0. * Adding Q4_K - scalar, AVX2, CUDA Performance is the same or perhaps very slightly better than Q4_0 on the CPU. On the GPU, single token prediction is ~10% better than Q4_0, batch mode (perplexity is about the same). * Adding Q6_K - scalar, AVX2, CUDA Performance is ~40% lower compared to Q4_K on the CPU. This is to be expected, considering that we are memory bound on the CPU and the 6-bit model is ~44% larger than the 4-bit. On the GPU, single token prediction is ~6% lower than Q4_0, batch mode (perplexity) is even closer (but still slower). * Adding Q5_K - scalar, AVX2, CUDA Performance is ~20% lower compared to Q4_K on the CPU. This is to be expected, considering that we are memory bound on the CPU and the 5-bit model is ~22% larger than the 4-bit. On the GPU, single token prediction is about the same as Q4_0 for both, single token and batch prediction. * Per convention, all QX_K quantizations use Q5_K for output.weight * Adding quantization mixes * Quantization mixes: didn't quite get what I wanted in the last commit * Q4_K dot product for ARM_NEON * Q6_K dot product for ARM_NEON * Q5_K dot product for ARM_NEON * Adding Q3_K dot for ARM_NEON It is 22% slower than Q4_K, despite the smaller model size. On x86_64, where we are memory bound, the Q3_K model is quite a bit faster than Q4_K. * A very slightly faster ARM_NEON Q3_K dot * Adding Q2_K - just CUDA for now Token prediction is pretty good - about 15.5 ms on a RTX 4080. Perplexity is about the same as Q4_K. * Adding scalar and AVX2 Q2_K dot * Adding ARM_NEON Q2_K dot About the same performance as Q4_K. * A slightly faster ARM_NEON Q2_K dot Single token prediction is now ~36 ms on M2 Max. The code is much simpler too. * Fixed bug in Q2_K CUDA dot product kernel Stranegly enough, for the few prompts I tried with the 7B model the responses looked perfectly reasonable. Only realized something is not quite right when I tried the larger models and started getting nonse back. In any case, Q2_K single token evaluation time on an RTX 4080 in a Ryzen7950X box iusing CUDA and model fully loaded on the GPU are ~15.5 ms for 7B, ~25.4 ms for 13B, and ~55.8 ms for 30B. The max number of layers that fit in VRAM for The 65B is 32. With that, we get ~330 ms per token, which is not that much faster than just running on the CPU (~470 ms per token). * Don't print zeros/NaNs when no count histogram has been collected * A 10% faster CUDA vector dot kernel for Q3_K Q3_K is now running at ~18.5 ms / token on CUDA, so the gap to Q4_0 is only 10%. It seems memory acccess pattern is more important for performance than the amount of computation the kernel does. * A slightly daster Q4_K AVX2 dot product For perplexity, where we are less memory bound, time per pass drops by ~5%. Barely measurable difference for single token prediction. * A slightly faster ARM_NEON A4_K dot product * Minor * Fix quantization error test We cannot possibly be expecting rmse < 0.002 for 2- and 3-bit quantization variants. * Fix docker build I have been sloppy with vector reinterpret casts on ARM_NEON. It seems clang is very forgiving in that regard. * Added forgotten ggml.o dependence on k_quants.h to the Makefile * Had unintentionally committed the Makefile with -Ofast enabled * ggml : rename k_quants -> ggml-quants-k, use lowercase in code --------- Co-authored-by: Iwan Kawrakow <iwan.kawrakow@gmail.com> Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
2023-06-05 19:56:18 +00:00
}
ggml : quantization refactoring (#3833) * ggml : factor all quantization code in ggml-quants ggml-ci * ggml-quants : fix Zig and Swift builds + quantize tool ggml-ci * quantize : --pure option for disabling k-quant mixtures --------- Co-authored-by: cebtenzzre <cebtenzzre@gmail.com>
2023-10-29 16:32:28 +00:00
*s = vaddvq_f32(sumv0) + vaddvq_f32(sumv1);
#elif defined(__wasm_simd128__)
v128_t sumv = wasm_f32x4_splat(0.0f);
ggml : add SOTA 2,3,4,5,6 bit k-quantizations (#1684) * Starting to add k-quantization to ggml I think it is better to have quantization separate from ggml. For now just adding the k-quants there, but it would be better to also factor out the existing ggml quantizations. * Adding Q3_K and Q8_K (de)-quantization * Q3_K now working on CUDA and AVX2/scalar CUDA is not ideal - ~50% slower than Q4_0 for single token prediction, about the same in batch mode (perplexity). CPU single token is ~55 ms (on Ryzen 7950X). * Some improvement for Q3_K on CUDA It is now ~22.5 ms/token on my GPU, so ~30% slower than Q4_0. * Some more CUDA optimizations for Q3_K Single token is now 20.5 ms/token (~20% slower than Q4_0). Perplexity is on par with Q4_0. * Adding Q4_K - scalar, AVX2, CUDA Performance is the same or perhaps very slightly better than Q4_0 on the CPU. On the GPU, single token prediction is ~10% better than Q4_0, batch mode (perplexity is about the same). * Adding Q6_K - scalar, AVX2, CUDA Performance is ~40% lower compared to Q4_K on the CPU. This is to be expected, considering that we are memory bound on the CPU and the 6-bit model is ~44% larger than the 4-bit. On the GPU, single token prediction is ~6% lower than Q4_0, batch mode (perplexity) is even closer (but still slower). * Adding Q5_K - scalar, AVX2, CUDA Performance is ~20% lower compared to Q4_K on the CPU. This is to be expected, considering that we are memory bound on the CPU and the 5-bit model is ~22% larger than the 4-bit. On the GPU, single token prediction is about the same as Q4_0 for both, single token and batch prediction. * Per convention, all QX_K quantizations use Q5_K for output.weight * Adding quantization mixes * Quantization mixes: didn't quite get what I wanted in the last commit * Q4_K dot product for ARM_NEON * Q6_K dot product for ARM_NEON * Q5_K dot product for ARM_NEON * Adding Q3_K dot for ARM_NEON It is 22% slower than Q4_K, despite the smaller model size. On x86_64, where we are memory bound, the Q3_K model is quite a bit faster than Q4_K. * A very slightly faster ARM_NEON Q3_K dot * Adding Q2_K - just CUDA for now Token prediction is pretty good - about 15.5 ms on a RTX 4080. Perplexity is about the same as Q4_K. * Adding scalar and AVX2 Q2_K dot * Adding ARM_NEON Q2_K dot About the same performance as Q4_K. * A slightly faster ARM_NEON Q2_K dot Single token prediction is now ~36 ms on M2 Max. The code is much simpler too. * Fixed bug in Q2_K CUDA dot product kernel Stranegly enough, for the few prompts I tried with the 7B model the responses looked perfectly reasonable. Only realized something is not quite right when I tried the larger models and started getting nonse back. In any case, Q2_K single token evaluation time on an RTX 4080 in a Ryzen7950X box iusing CUDA and model fully loaded on the GPU are ~15.5 ms for 7B, ~25.4 ms for 13B, and ~55.8 ms for 30B. The max number of layers that fit in VRAM for The 65B is 32. With that, we get ~330 ms per token, which is not that much faster than just running on the CPU (~470 ms per token). * Don't print zeros/NaNs when no count histogram has been collected * A 10% faster CUDA vector dot kernel for Q3_K Q3_K is now running at ~18.5 ms / token on CUDA, so the gap to Q4_0 is only 10%. It seems memory acccess pattern is more important for performance than the amount of computation the kernel does. * A slightly daster Q4_K AVX2 dot product For perplexity, where we are less memory bound, time per pass drops by ~5%. Barely measurable difference for single token prediction. * A slightly faster ARM_NEON A4_K dot product * Minor * Fix quantization error test We cannot possibly be expecting rmse < 0.002 for 2- and 3-bit quantization variants. * Fix docker build I have been sloppy with vector reinterpret casts on ARM_NEON. It seems clang is very forgiving in that regard. * Added forgotten ggml.o dependence on k_quants.h to the Makefile * Had unintentionally committed the Makefile with -Ofast enabled * ggml : rename k_quants -> ggml-quants-k, use lowercase in code --------- Co-authored-by: Iwan Kawrakow <iwan.kawrakow@gmail.com> Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
2023-06-05 19:56:18 +00:00
ggml : quantization refactoring (#3833) * ggml : factor all quantization code in ggml-quants ggml-ci * ggml-quants : fix Zig and Swift builds + quantize tool ggml-ci * quantize : --pure option for disabling k-quant mixtures --------- Co-authored-by: cebtenzzre <cebtenzzre@gmail.com>
2023-10-29 16:32:28 +00:00
uint32_t qh;
uint64_t tmp[4];
ggml : add SOTA 2,3,4,5,6 bit k-quantizations (#1684) * Starting to add k-quantization to ggml I think it is better to have quantization separate from ggml. For now just adding the k-quants there, but it would be better to also factor out the existing ggml quantizations. * Adding Q3_K and Q8_K (de)-quantization * Q3_K now working on CUDA and AVX2/scalar CUDA is not ideal - ~50% slower than Q4_0 for single token prediction, about the same in batch mode (perplexity). CPU single token is ~55 ms (on Ryzen 7950X). * Some improvement for Q3_K on CUDA It is now ~22.5 ms/token on my GPU, so ~30% slower than Q4_0. * Some more CUDA optimizations for Q3_K Single token is now 20.5 ms/token (~20% slower than Q4_0). Perplexity is on par with Q4_0. * Adding Q4_K - scalar, AVX2, CUDA Performance is the same or perhaps very slightly better than Q4_0 on the CPU. On the GPU, single token prediction is ~10% better than Q4_0, batch mode (perplexity is about the same). * Adding Q6_K - scalar, AVX2, CUDA Performance is ~40% lower compared to Q4_K on the CPU. This is to be expected, considering that we are memory bound on the CPU and the 6-bit model is ~44% larger than the 4-bit. On the GPU, single token prediction is ~6% lower than Q4_0, batch mode (perplexity) is even closer (but still slower). * Adding Q5_K - scalar, AVX2, CUDA Performance is ~20% lower compared to Q4_K on the CPU. This is to be expected, considering that we are memory bound on the CPU and the 5-bit model is ~22% larger than the 4-bit. On the GPU, single token prediction is about the same as Q4_0 for both, single token and batch prediction. * Per convention, all QX_K quantizations use Q5_K for output.weight * Adding quantization mixes * Quantization mixes: didn't quite get what I wanted in the last commit * Q4_K dot product for ARM_NEON * Q6_K dot product for ARM_NEON * Q5_K dot product for ARM_NEON * Adding Q3_K dot for ARM_NEON It is 22% slower than Q4_K, despite the smaller model size. On x86_64, where we are memory bound, the Q3_K model is quite a bit faster than Q4_K. * A very slightly faster ARM_NEON Q3_K dot * Adding Q2_K - just CUDA for now Token prediction is pretty good - about 15.5 ms on a RTX 4080. Perplexity is about the same as Q4_K. * Adding scalar and AVX2 Q2_K dot * Adding ARM_NEON Q2_K dot About the same performance as Q4_K. * A slightly faster ARM_NEON Q2_K dot Single token prediction is now ~36 ms on M2 Max. The code is much simpler too. * Fixed bug in Q2_K CUDA dot product kernel Stranegly enough, for the few prompts I tried with the 7B model the responses looked perfectly reasonable. Only realized something is not quite right when I tried the larger models and started getting nonse back. In any case, Q2_K single token evaluation time on an RTX 4080 in a Ryzen7950X box iusing CUDA and model fully loaded on the GPU are ~15.5 ms for 7B, ~25.4 ms for 13B, and ~55.8 ms for 30B. The max number of layers that fit in VRAM for The 65B is 32. With that, we get ~330 ms per token, which is not that much faster than just running on the CPU (~470 ms per token). * Don't print zeros/NaNs when no count histogram has been collected * A 10% faster CUDA vector dot kernel for Q3_K Q3_K is now running at ~18.5 ms / token on CUDA, so the gap to Q4_0 is only 10%. It seems memory acccess pattern is more important for performance than the amount of computation the kernel does. * A slightly daster Q4_K AVX2 dot product For perplexity, where we are less memory bound, time per pass drops by ~5%. Barely measurable difference for single token prediction. * A slightly faster ARM_NEON A4_K dot product * Minor * Fix quantization error test We cannot possibly be expecting rmse < 0.002 for 2- and 3-bit quantization variants. * Fix docker build I have been sloppy with vector reinterpret casts on ARM_NEON. It seems clang is very forgiving in that regard. * Added forgotten ggml.o dependence on k_quants.h to the Makefile * Had unintentionally committed the Makefile with -Ofast enabled * ggml : rename k_quants -> ggml-quants-k, use lowercase in code --------- Co-authored-by: Iwan Kawrakow <iwan.kawrakow@gmail.com> Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
2023-06-05 19:56:18 +00:00
ggml : quantization refactoring (#3833) * ggml : factor all quantization code in ggml-quants ggml-ci * ggml-quants : fix Zig and Swift builds + quantize tool ggml-ci * quantize : --pure option for disabling k-quant mixtures --------- Co-authored-by: cebtenzzre <cebtenzzre@gmail.com>
2023-10-29 16:32:28 +00:00
// TODO: check if unrolling this is better
for (int i = 0; i < nb; ++i) {
const block_q5_0 * restrict x0 = &x[i];
const block_q8_0 * restrict y0 = &y[i];
const v128_t m4b = wasm_i8x16_splat(0x0F);
// extract the 5th bit
memcpy(&qh, x0->qh, sizeof(qh));
tmp[0] = table_b2b_1[(qh >> 0) & 0xFF];
tmp[1] = table_b2b_1[(qh >> 8) & 0xFF];
tmp[2] = table_b2b_1[(qh >> 16) & 0xFF];
tmp[3] = table_b2b_1[(qh >> 24) ];
const v128_t qhl = wasm_v128_load(tmp + 0);
const v128_t qhh = wasm_v128_load(tmp + 2);
const v128_t v0 = wasm_v128_load(x0->qs);
// 4-bit -> 8-bit
const v128_t v0l = wasm_v128_and (v0, m4b);
const v128_t v0h = wasm_u8x16_shr(v0, 4);
// add high bit and sub 16 (equivalent to sub 0x10 when bit is zero)
const v128_t v0lf = wasm_i8x16_sub(v0l, qhl);
const v128_t v0hf = wasm_i8x16_sub(v0h, qhh);
// load y
const v128_t v1l = wasm_v128_load(y0->qs);
const v128_t v1h = wasm_v128_load(y0->qs + 16);
// int8x16 -> int16x8
const v128_t v0lfl = wasm_i16x8_extend_low_i8x16 (v0lf);
const v128_t v0lfh = wasm_i16x8_extend_high_i8x16(v0lf);
const v128_t v0hfl = wasm_i16x8_extend_low_i8x16 (v0hf);
const v128_t v0hfh = wasm_i16x8_extend_high_i8x16(v0hf);
const v128_t v1ll = wasm_i16x8_extend_low_i8x16 (v1l);
const v128_t v1lh = wasm_i16x8_extend_high_i8x16(v1l);
const v128_t v1hl = wasm_i16x8_extend_low_i8x16 (v1h);
const v128_t v1hh = wasm_i16x8_extend_high_i8x16(v1h);
// dot product
sumv = wasm_f32x4_add(sumv, wasm_f32x4_mul(wasm_f32x4_convert_i32x4(
wasm_i32x4_add(
wasm_i32x4_add(wasm_i32x4_dot_i16x8(v0lfl, v1ll),
wasm_i32x4_dot_i16x8(v0lfh, v1lh)),
wasm_i32x4_add(wasm_i32x4_dot_i16x8(v0hfl, v1hl),
wasm_i32x4_dot_i16x8(v0hfh, v1hh)))),
ggml : move FP16 <-> FP32 code to ggml-impl.h (#3861) * ggml : move FP16 <-> FP32 stuff to ggml-impl.h ggml-ci * tests : fix ARM build * ggml : explicitly initialize deprecated type traits * ggml : add math.h to ggml-impl.h * ggml : remove duplicate static assert macros * ggml : prefix lookup tables with ggml_ ggml-ci * ggml-impl : move extern "C" to start of file
2023-10-30 17:19:15 +00:00
wasm_f32x4_splat(GGML_FP16_TO_FP32(x0->d) * GGML_FP16_TO_FP32(y0->d))));
ggml : quantization refactoring (#3833) * ggml : factor all quantization code in ggml-quants ggml-ci * ggml-quants : fix Zig and Swift builds + quantize tool ggml-ci * quantize : --pure option for disabling k-quant mixtures --------- Co-authored-by: cebtenzzre <cebtenzzre@gmail.com>
2023-10-29 16:32:28 +00:00
}
ggml : add SOTA 2,3,4,5,6 bit k-quantizations (#1684) * Starting to add k-quantization to ggml I think it is better to have quantization separate from ggml. For now just adding the k-quants there, but it would be better to also factor out the existing ggml quantizations. * Adding Q3_K and Q8_K (de)-quantization * Q3_K now working on CUDA and AVX2/scalar CUDA is not ideal - ~50% slower than Q4_0 for single token prediction, about the same in batch mode (perplexity). CPU single token is ~55 ms (on Ryzen 7950X). * Some improvement for Q3_K on CUDA It is now ~22.5 ms/token on my GPU, so ~30% slower than Q4_0. * Some more CUDA optimizations for Q3_K Single token is now 20.5 ms/token (~20% slower than Q4_0). Perplexity is on par with Q4_0. * Adding Q4_K - scalar, AVX2, CUDA Performance is the same or perhaps very slightly better than Q4_0 on the CPU. On the GPU, single token prediction is ~10% better than Q4_0, batch mode (perplexity is about the same). * Adding Q6_K - scalar, AVX2, CUDA Performance is ~40% lower compared to Q4_K on the CPU. This is to be expected, considering that we are memory bound on the CPU and the 6-bit model is ~44% larger than the 4-bit. On the GPU, single token prediction is ~6% lower than Q4_0, batch mode (perplexity) is even closer (but still slower). * Adding Q5_K - scalar, AVX2, CUDA Performance is ~20% lower compared to Q4_K on the CPU. This is to be expected, considering that we are memory bound on the CPU and the 5-bit model is ~22% larger than the 4-bit. On the GPU, single token prediction is about the same as Q4_0 for both, single token and batch prediction. * Per convention, all QX_K quantizations use Q5_K for output.weight * Adding quantization mixes * Quantization mixes: didn't quite get what I wanted in the last commit * Q4_K dot product for ARM_NEON * Q6_K dot product for ARM_NEON * Q5_K dot product for ARM_NEON * Adding Q3_K dot for ARM_NEON It is 22% slower than Q4_K, despite the smaller model size. On x86_64, where we are memory bound, the Q3_K model is quite a bit faster than Q4_K. * A very slightly faster ARM_NEON Q3_K dot * Adding Q2_K - just CUDA for now Token prediction is pretty good - about 15.5 ms on a RTX 4080. Perplexity is about the same as Q4_K. * Adding scalar and AVX2 Q2_K dot * Adding ARM_NEON Q2_K dot About the same performance as Q4_K. * A slightly faster ARM_NEON Q2_K dot Single token prediction is now ~36 ms on M2 Max. The code is much simpler too. * Fixed bug in Q2_K CUDA dot product kernel Stranegly enough, for the few prompts I tried with the 7B model the responses looked perfectly reasonable. Only realized something is not quite right when I tried the larger models and started getting nonse back. In any case, Q2_K single token evaluation time on an RTX 4080 in a Ryzen7950X box iusing CUDA and model fully loaded on the GPU are ~15.5 ms for 7B, ~25.4 ms for 13B, and ~55.8 ms for 30B. The max number of layers that fit in VRAM for The 65B is 32. With that, we get ~330 ms per token, which is not that much faster than just running on the CPU (~470 ms per token). * Don't print zeros/NaNs when no count histogram has been collected * A 10% faster CUDA vector dot kernel for Q3_K Q3_K is now running at ~18.5 ms / token on CUDA, so the gap to Q4_0 is only 10%. It seems memory acccess pattern is more important for performance than the amount of computation the kernel does. * A slightly daster Q4_K AVX2 dot product For perplexity, where we are less memory bound, time per pass drops by ~5%. Barely measurable difference for single token prediction. * A slightly faster ARM_NEON A4_K dot product * Minor * Fix quantization error test We cannot possibly be expecting rmse < 0.002 for 2- and 3-bit quantization variants. * Fix docker build I have been sloppy with vector reinterpret casts on ARM_NEON. It seems clang is very forgiving in that regard. * Added forgotten ggml.o dependence on k_quants.h to the Makefile * Had unintentionally committed the Makefile with -Ofast enabled * ggml : rename k_quants -> ggml-quants-k, use lowercase in code --------- Co-authored-by: Iwan Kawrakow <iwan.kawrakow@gmail.com> Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
2023-06-05 19:56:18 +00:00
ggml : quantization refactoring (#3833) * ggml : factor all quantization code in ggml-quants ggml-ci * ggml-quants : fix Zig and Swift builds + quantize tool ggml-ci * quantize : --pure option for disabling k-quant mixtures --------- Co-authored-by: cebtenzzre <cebtenzzre@gmail.com>
2023-10-29 16:32:28 +00:00
*s = wasm_f32x4_extract_lane(sumv, 0) + wasm_f32x4_extract_lane(sumv, 1) +
wasm_f32x4_extract_lane(sumv, 2) + wasm_f32x4_extract_lane(sumv, 3);
#elif defined(__AVX2__)
// Initialize accumulator with zeros
__m256 acc = _mm256_setzero_ps();
// Main loop
ggml : add SOTA 2,3,4,5,6 bit k-quantizations (#1684) * Starting to add k-quantization to ggml I think it is better to have quantization separate from ggml. For now just adding the k-quants there, but it would be better to also factor out the existing ggml quantizations. * Adding Q3_K and Q8_K (de)-quantization * Q3_K now working on CUDA and AVX2/scalar CUDA is not ideal - ~50% slower than Q4_0 for single token prediction, about the same in batch mode (perplexity). CPU single token is ~55 ms (on Ryzen 7950X). * Some improvement for Q3_K on CUDA It is now ~22.5 ms/token on my GPU, so ~30% slower than Q4_0. * Some more CUDA optimizations for Q3_K Single token is now 20.5 ms/token (~20% slower than Q4_0). Perplexity is on par with Q4_0. * Adding Q4_K - scalar, AVX2, CUDA Performance is the same or perhaps very slightly better than Q4_0 on the CPU. On the GPU, single token prediction is ~10% better than Q4_0, batch mode (perplexity is about the same). * Adding Q6_K - scalar, AVX2, CUDA Performance is ~40% lower compared to Q4_K on the CPU. This is to be expected, considering that we are memory bound on the CPU and the 6-bit model is ~44% larger than the 4-bit. On the GPU, single token prediction is ~6% lower than Q4_0, batch mode (perplexity) is even closer (but still slower). * Adding Q5_K - scalar, AVX2, CUDA Performance is ~20% lower compared to Q4_K on the CPU. This is to be expected, considering that we are memory bound on the CPU and the 5-bit model is ~22% larger than the 4-bit. On the GPU, single token prediction is about the same as Q4_0 for both, single token and batch prediction. * Per convention, all QX_K quantizations use Q5_K for output.weight * Adding quantization mixes * Quantization mixes: didn't quite get what I wanted in the last commit * Q4_K dot product for ARM_NEON * Q6_K dot product for ARM_NEON * Q5_K dot product for ARM_NEON * Adding Q3_K dot for ARM_NEON It is 22% slower than Q4_K, despite the smaller model size. On x86_64, where we are memory bound, the Q3_K model is quite a bit faster than Q4_K. * A very slightly faster ARM_NEON Q3_K dot * Adding Q2_K - just CUDA for now Token prediction is pretty good - about 15.5 ms on a RTX 4080. Perplexity is about the same as Q4_K. * Adding scalar and AVX2 Q2_K dot * Adding ARM_NEON Q2_K dot About the same performance as Q4_K. * A slightly faster ARM_NEON Q2_K dot Single token prediction is now ~36 ms on M2 Max. The code is much simpler too. * Fixed bug in Q2_K CUDA dot product kernel Stranegly enough, for the few prompts I tried with the 7B model the responses looked perfectly reasonable. Only realized something is not quite right when I tried the larger models and started getting nonse back. In any case, Q2_K single token evaluation time on an RTX 4080 in a Ryzen7950X box iusing CUDA and model fully loaded on the GPU are ~15.5 ms for 7B, ~25.4 ms for 13B, and ~55.8 ms for 30B. The max number of layers that fit in VRAM for The 65B is 32. With that, we get ~330 ms per token, which is not that much faster than just running on the CPU (~470 ms per token). * Don't print zeros/NaNs when no count histogram has been collected * A 10% faster CUDA vector dot kernel for Q3_K Q3_K is now running at ~18.5 ms / token on CUDA, so the gap to Q4_0 is only 10%. It seems memory acccess pattern is more important for performance than the amount of computation the kernel does. * A slightly daster Q4_K AVX2 dot product For perplexity, where we are less memory bound, time per pass drops by ~5%. Barely measurable difference for single token prediction. * A slightly faster ARM_NEON A4_K dot product * Minor * Fix quantization error test We cannot possibly be expecting rmse < 0.002 for 2- and 3-bit quantization variants. * Fix docker build I have been sloppy with vector reinterpret casts on ARM_NEON. It seems clang is very forgiving in that regard. * Added forgotten ggml.o dependence on k_quants.h to the Makefile * Had unintentionally committed the Makefile with -Ofast enabled * ggml : rename k_quants -> ggml-quants-k, use lowercase in code --------- Co-authored-by: Iwan Kawrakow <iwan.kawrakow@gmail.com> Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
2023-06-05 19:56:18 +00:00
for (int i = 0; i < nb; i++) {
ggml : quantization refactoring (#3833) * ggml : factor all quantization code in ggml-quants ggml-ci * ggml-quants : fix Zig and Swift builds + quantize tool ggml-ci * quantize : --pure option for disabling k-quant mixtures --------- Co-authored-by: cebtenzzre <cebtenzzre@gmail.com>
2023-10-29 16:32:28 +00:00
/* Compute combined scale for the block */
ggml : move FP16 <-> FP32 code to ggml-impl.h (#3861) * ggml : move FP16 <-> FP32 stuff to ggml-impl.h ggml-ci * tests : fix ARM build * ggml : explicitly initialize deprecated type traits * ggml : add math.h to ggml-impl.h * ggml : remove duplicate static assert macros * ggml : prefix lookup tables with ggml_ ggml-ci * ggml-impl : move extern "C" to start of file
2023-10-30 17:19:15 +00:00
const __m256 d = _mm256_set1_ps(GGML_FP16_TO_FP32(x[i].d) * GGML_FP16_TO_FP32(y[i].d));
ggml : add SOTA 2,3,4,5,6 bit k-quantizations (#1684) * Starting to add k-quantization to ggml I think it is better to have quantization separate from ggml. For now just adding the k-quants there, but it would be better to also factor out the existing ggml quantizations. * Adding Q3_K and Q8_K (de)-quantization * Q3_K now working on CUDA and AVX2/scalar CUDA is not ideal - ~50% slower than Q4_0 for single token prediction, about the same in batch mode (perplexity). CPU single token is ~55 ms (on Ryzen 7950X). * Some improvement for Q3_K on CUDA It is now ~22.5 ms/token on my GPU, so ~30% slower than Q4_0. * Some more CUDA optimizations for Q3_K Single token is now 20.5 ms/token (~20% slower than Q4_0). Perplexity is on par with Q4_0. * Adding Q4_K - scalar, AVX2, CUDA Performance is the same or perhaps very slightly better than Q4_0 on the CPU. On the GPU, single token prediction is ~10% better than Q4_0, batch mode (perplexity is about the same). * Adding Q6_K - scalar, AVX2, CUDA Performance is ~40% lower compared to Q4_K on the CPU. This is to be expected, considering that we are memory bound on the CPU and the 6-bit model is ~44% larger than the 4-bit. On the GPU, single token prediction is ~6% lower than Q4_0, batch mode (perplexity) is even closer (but still slower). * Adding Q5_K - scalar, AVX2, CUDA Performance is ~20% lower compared to Q4_K on the CPU. This is to be expected, considering that we are memory bound on the CPU and the 5-bit model is ~22% larger than the 4-bit. On the GPU, single token prediction is about the same as Q4_0 for both, single token and batch prediction. * Per convention, all QX_K quantizations use Q5_K for output.weight * Adding quantization mixes * Quantization mixes: didn't quite get what I wanted in the last commit * Q4_K dot product for ARM_NEON * Q6_K dot product for ARM_NEON * Q5_K dot product for ARM_NEON * Adding Q3_K dot for ARM_NEON It is 22% slower than Q4_K, despite the smaller model size. On x86_64, where we are memory bound, the Q3_K model is quite a bit faster than Q4_K. * A very slightly faster ARM_NEON Q3_K dot * Adding Q2_K - just CUDA for now Token prediction is pretty good - about 15.5 ms on a RTX 4080. Perplexity is about the same as Q4_K. * Adding scalar and AVX2 Q2_K dot * Adding ARM_NEON Q2_K dot About the same performance as Q4_K. * A slightly faster ARM_NEON Q2_K dot Single token prediction is now ~36 ms on M2 Max. The code is much simpler too. * Fixed bug in Q2_K CUDA dot product kernel Stranegly enough, for the few prompts I tried with the 7B model the responses looked perfectly reasonable. Only realized something is not quite right when I tried the larger models and started getting nonse back. In any case, Q2_K single token evaluation time on an RTX 4080 in a Ryzen7950X box iusing CUDA and model fully loaded on the GPU are ~15.5 ms for 7B, ~25.4 ms for 13B, and ~55.8 ms for 30B. The max number of layers that fit in VRAM for The 65B is 32. With that, we get ~330 ms per token, which is not that much faster than just running on the CPU (~470 ms per token). * Don't print zeros/NaNs when no count histogram has been collected * A 10% faster CUDA vector dot kernel for Q3_K Q3_K is now running at ~18.5 ms / token on CUDA, so the gap to Q4_0 is only 10%. It seems memory acccess pattern is more important for performance than the amount of computation the kernel does. * A slightly daster Q4_K AVX2 dot product For perplexity, where we are less memory bound, time per pass drops by ~5%. Barely measurable difference for single token prediction. * A slightly faster ARM_NEON A4_K dot product * Minor * Fix quantization error test We cannot possibly be expecting rmse < 0.002 for 2- and 3-bit quantization variants. * Fix docker build I have been sloppy with vector reinterpret casts on ARM_NEON. It seems clang is very forgiving in that regard. * Added forgotten ggml.o dependence on k_quants.h to the Makefile * Had unintentionally committed the Makefile with -Ofast enabled * ggml : rename k_quants -> ggml-quants-k, use lowercase in code --------- Co-authored-by: Iwan Kawrakow <iwan.kawrakow@gmail.com> Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
2023-06-05 19:56:18 +00:00
ggml : quantization refactoring (#3833) * ggml : factor all quantization code in ggml-quants ggml-ci * ggml-quants : fix Zig and Swift builds + quantize tool ggml-ci * quantize : --pure option for disabling k-quant mixtures --------- Co-authored-by: cebtenzzre <cebtenzzre@gmail.com>
2023-10-29 16:32:28 +00:00
__m256i bx = bytes_from_nibbles_32(x[i].qs);
__m256i bxhi = bytes_from_bits_32(x[i].qh);
bxhi = _mm256_andnot_si256(bxhi, _mm256_set1_epi8((char)0xF0));
bx = _mm256_or_si256(bx, bxhi);
ggml : add SOTA 2,3,4,5,6 bit k-quantizations (#1684) * Starting to add k-quantization to ggml I think it is better to have quantization separate from ggml. For now just adding the k-quants there, but it would be better to also factor out the existing ggml quantizations. * Adding Q3_K and Q8_K (de)-quantization * Q3_K now working on CUDA and AVX2/scalar CUDA is not ideal - ~50% slower than Q4_0 for single token prediction, about the same in batch mode (perplexity). CPU single token is ~55 ms (on Ryzen 7950X). * Some improvement for Q3_K on CUDA It is now ~22.5 ms/token on my GPU, so ~30% slower than Q4_0. * Some more CUDA optimizations for Q3_K Single token is now 20.5 ms/token (~20% slower than Q4_0). Perplexity is on par with Q4_0. * Adding Q4_K - scalar, AVX2, CUDA Performance is the same or perhaps very slightly better than Q4_0 on the CPU. On the GPU, single token prediction is ~10% better than Q4_0, batch mode (perplexity is about the same). * Adding Q6_K - scalar, AVX2, CUDA Performance is ~40% lower compared to Q4_K on the CPU. This is to be expected, considering that we are memory bound on the CPU and the 6-bit model is ~44% larger than the 4-bit. On the GPU, single token prediction is ~6% lower than Q4_0, batch mode (perplexity) is even closer (but still slower). * Adding Q5_K - scalar, AVX2, CUDA Performance is ~20% lower compared to Q4_K on the CPU. This is to be expected, considering that we are memory bound on the CPU and the 5-bit model is ~22% larger than the 4-bit. On the GPU, single token prediction is about the same as Q4_0 for both, single token and batch prediction. * Per convention, all QX_K quantizations use Q5_K for output.weight * Adding quantization mixes * Quantization mixes: didn't quite get what I wanted in the last commit * Q4_K dot product for ARM_NEON * Q6_K dot product for ARM_NEON * Q5_K dot product for ARM_NEON * Adding Q3_K dot for ARM_NEON It is 22% slower than Q4_K, despite the smaller model size. On x86_64, where we are memory bound, the Q3_K model is quite a bit faster than Q4_K. * A very slightly faster ARM_NEON Q3_K dot * Adding Q2_K - just CUDA for now Token prediction is pretty good - about 15.5 ms on a RTX 4080. Perplexity is about the same as Q4_K. * Adding scalar and AVX2 Q2_K dot * Adding ARM_NEON Q2_K dot About the same performance as Q4_K. * A slightly faster ARM_NEON Q2_K dot Single token prediction is now ~36 ms on M2 Max. The code is much simpler too. * Fixed bug in Q2_K CUDA dot product kernel Stranegly enough, for the few prompts I tried with the 7B model the responses looked perfectly reasonable. Only realized something is not quite right when I tried the larger models and started getting nonse back. In any case, Q2_K single token evaluation time on an RTX 4080 in a Ryzen7950X box iusing CUDA and model fully loaded on the GPU are ~15.5 ms for 7B, ~25.4 ms for 13B, and ~55.8 ms for 30B. The max number of layers that fit in VRAM for The 65B is 32. With that, we get ~330 ms per token, which is not that much faster than just running on the CPU (~470 ms per token). * Don't print zeros/NaNs when no count histogram has been collected * A 10% faster CUDA vector dot kernel for Q3_K Q3_K is now running at ~18.5 ms / token on CUDA, so the gap to Q4_0 is only 10%. It seems memory acccess pattern is more important for performance than the amount of computation the kernel does. * A slightly daster Q4_K AVX2 dot product For perplexity, where we are less memory bound, time per pass drops by ~5%. Barely measurable difference for single token prediction. * A slightly faster ARM_NEON A4_K dot product * Minor * Fix quantization error test We cannot possibly be expecting rmse < 0.002 for 2- and 3-bit quantization variants. * Fix docker build I have been sloppy with vector reinterpret casts on ARM_NEON. It seems clang is very forgiving in that regard. * Added forgotten ggml.o dependence on k_quants.h to the Makefile * Had unintentionally committed the Makefile with -Ofast enabled * ggml : rename k_quants -> ggml-quants-k, use lowercase in code --------- Co-authored-by: Iwan Kawrakow <iwan.kawrakow@gmail.com> Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
2023-06-05 19:56:18 +00:00
ggml : quantization refactoring (#3833) * ggml : factor all quantization code in ggml-quants ggml-ci * ggml-quants : fix Zig and Swift builds + quantize tool ggml-ci * quantize : --pure option for disabling k-quant mixtures --------- Co-authored-by: cebtenzzre <cebtenzzre@gmail.com>
2023-10-29 16:32:28 +00:00
__m256i by = _mm256_loadu_si256((const __m256i *)y[i].qs);
ggml : add SOTA 2,3,4,5,6 bit k-quantizations (#1684) * Starting to add k-quantization to ggml I think it is better to have quantization separate from ggml. For now just adding the k-quants there, but it would be better to also factor out the existing ggml quantizations. * Adding Q3_K and Q8_K (de)-quantization * Q3_K now working on CUDA and AVX2/scalar CUDA is not ideal - ~50% slower than Q4_0 for single token prediction, about the same in batch mode (perplexity). CPU single token is ~55 ms (on Ryzen 7950X). * Some improvement for Q3_K on CUDA It is now ~22.5 ms/token on my GPU, so ~30% slower than Q4_0. * Some more CUDA optimizations for Q3_K Single token is now 20.5 ms/token (~20% slower than Q4_0). Perplexity is on par with Q4_0. * Adding Q4_K - scalar, AVX2, CUDA Performance is the same or perhaps very slightly better than Q4_0 on the CPU. On the GPU, single token prediction is ~10% better than Q4_0, batch mode (perplexity is about the same). * Adding Q6_K - scalar, AVX2, CUDA Performance is ~40% lower compared to Q4_K on the CPU. This is to be expected, considering that we are memory bound on the CPU and the 6-bit model is ~44% larger than the 4-bit. On the GPU, single token prediction is ~6% lower than Q4_0, batch mode (perplexity) is even closer (but still slower). * Adding Q5_K - scalar, AVX2, CUDA Performance is ~20% lower compared to Q4_K on the CPU. This is to be expected, considering that we are memory bound on the CPU and the 5-bit model is ~22% larger than the 4-bit. On the GPU, single token prediction is about the same as Q4_0 for both, single token and batch prediction. * Per convention, all QX_K quantizations use Q5_K for output.weight * Adding quantization mixes * Quantization mixes: didn't quite get what I wanted in the last commit * Q4_K dot product for ARM_NEON * Q6_K dot product for ARM_NEON * Q5_K dot product for ARM_NEON * Adding Q3_K dot for ARM_NEON It is 22% slower than Q4_K, despite the smaller model size. On x86_64, where we are memory bound, the Q3_K model is quite a bit faster than Q4_K. * A very slightly faster ARM_NEON Q3_K dot * Adding Q2_K - just CUDA for now Token prediction is pretty good - about 15.5 ms on a RTX 4080. Perplexity is about the same as Q4_K. * Adding scalar and AVX2 Q2_K dot * Adding ARM_NEON Q2_K dot About the same performance as Q4_K. * A slightly faster ARM_NEON Q2_K dot Single token prediction is now ~36 ms on M2 Max. The code is much simpler too. * Fixed bug in Q2_K CUDA dot product kernel Stranegly enough, for the few prompts I tried with the 7B model the responses looked perfectly reasonable. Only realized something is not quite right when I tried the larger models and started getting nonse back. In any case, Q2_K single token evaluation time on an RTX 4080 in a Ryzen7950X box iusing CUDA and model fully loaded on the GPU are ~15.5 ms for 7B, ~25.4 ms for 13B, and ~55.8 ms for 30B. The max number of layers that fit in VRAM for The 65B is 32. With that, we get ~330 ms per token, which is not that much faster than just running on the CPU (~470 ms per token). * Don't print zeros/NaNs when no count histogram has been collected * A 10% faster CUDA vector dot kernel for Q3_K Q3_K is now running at ~18.5 ms / token on CUDA, so the gap to Q4_0 is only 10%. It seems memory acccess pattern is more important for performance than the amount of computation the kernel does. * A slightly daster Q4_K AVX2 dot product For perplexity, where we are less memory bound, time per pass drops by ~5%. Barely measurable difference for single token prediction. * A slightly faster ARM_NEON A4_K dot product * Minor * Fix quantization error test We cannot possibly be expecting rmse < 0.002 for 2- and 3-bit quantization variants. * Fix docker build I have been sloppy with vector reinterpret casts on ARM_NEON. It seems clang is very forgiving in that regard. * Added forgotten ggml.o dependence on k_quants.h to the Makefile * Had unintentionally committed the Makefile with -Ofast enabled * ggml : rename k_quants -> ggml-quants-k, use lowercase in code --------- Co-authored-by: Iwan Kawrakow <iwan.kawrakow@gmail.com> Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
2023-06-05 19:56:18 +00:00
ggml : quantization refactoring (#3833) * ggml : factor all quantization code in ggml-quants ggml-ci * ggml-quants : fix Zig and Swift builds + quantize tool ggml-ci * quantize : --pure option for disabling k-quant mixtures --------- Co-authored-by: cebtenzzre <cebtenzzre@gmail.com>
2023-10-29 16:32:28 +00:00
const __m256 q = mul_sum_i8_pairs_float(bx, by);
k-quants : support for super-block size of 64 (#2001) * k_quants: WIP super-blocks with 64 weights * k_quants: WIP super-blocks with 64 weights Q6_K scalar and AVX2 works * k_quants: WIP super-blocks with 64 weights Q4_K scalar and AVX2 works * k_quants: WIP super-blocks with 64 weights Q2_K scalar and AVX2 works. Q2_K is way too slow (it is actually slower than the scalar implementation) * k_quants: WIP super-blocks with 64 weights Q3_K scalar and AVX2 works. * k_quants: WIP super-blocks with 64 weights Q5_K scalar and AVX2 works, and with that all k_quants are done on AVX2 and scalar * k_quants: WIP super-blocks with 64 weights Q6_K working on CUDA. Cannot make it run quite as gast as with super-blocks with 256 weigths: 8% slower on 4080, 20% slower on the 1660 (but there we fit 1 less layer on the GPU because pf the larger model size), so some fraction of these 20% is due to that, * k_quants: WIP super-blocks with 64 weights Q4_K working on CUDA. ~10% slower on GTX-1660, 16% slower on 4080. * k_quants: WIP super-blocks with 64 weights Q2_K working on CUDA. ~3% slower on GTX-1660, 10% slower on 4080. * k_quants: WIP super-blocks with 64 weights Q3_K working on CUDA. * k_quants: WIP super-blocks with 64 weights Q5_K working on CUDA, and with this CUDA is done. * k_quants: WIP super-blocks with 64 weights Q6_K working on ARM_NEON * k_quants: WIP super-blocks with 64 weights Q4_K working on ARM_NEON, but quite a bit slower than 256 weights * k_quants: WIP super-blocks with 64 weights Q2_K working on ARM_NEON, but quite a bit slower than 256 weights * k_quants: WIP super-blocks with 64 weights Q3_K working on ARM_NEON, but quite a bit slower than 256 weights. * k_quants: WIP super-blocks with 64 weights Q5_K working on ARM_NEON, but quite a bit slower than 256 weights. With that, we have full support for ARM_NEON, although performance is not quite there. * k_quants: WIP super-blocks with 64 weights Slightly more efficient Q3_K and Q5_K * k_quants: WIP super-blocks with 64 weights Another small improvement for Q3_K and Q5_K on ARM_NEON * k_quants: WIP super-blocks with 64 weights Yet another speedup for Q5_K on ARM_NEON. We are now within 10% of the QK_K = 256 version. * k_quants: WIP super-blocks with 64 weights * We are able to pass preprocessor macros to the Metal compiler * Q6_K works and is actually slightly more efficient than the QK_K = 256 version (25.2 ms vs 25.8 ms) * k_quants: WIP super-blocks with 64 weights Q4_K works on Metal and is actually slightly faster than QK_K = 256 (21.95 ms vs 24.0 ms). * k_quants: WIP super-blocks with 64 weights Q2_K works on Metal and is very slightly faster than QK_K = 256 (23.8 ms vs 24.2 ms). * k_quants: WIP super-blocks with 64 weights Q3_K works on Metal and is slightly faster than QK_K = 256 (26.6 ms vs 28.3 ms). * k_quants: WIP super-blocks with 64 weights Q5_K works on Metal and is slightly faster than QK_K = 256 (23.7 ms vs 26.3 ms). * k_quants: call them _K, not _k, also on Metal * k_quants: correctly define QK_K in llama.cpp * Fixed bug in q4_K quantization added with the 64-block addition * Simplify via lambda * k_quants: swicth Q3_K to 4-bit scales when QK_K = 64 Otherwise there isn't much benefit from this quantization type. There is some very slight loss in accuracy, but we reduce size by ~7%. E.g., for OpenLLaMA-3B, Q3_K_S perplexity is 8.6131 with 8-bit scales and 8.6352 with 4-bit, while file size decreases from 1.53G to 1.44G. * k_quants: switch Q4_K to 4-bit scales when QK_K = 64 Here the loss in accuracy is greater than for Q3_K, but the Q4_K points still move further to the left on the perplexity vs size curve. * k_quants: forgot to add the Metal changes in last commit * k_quants: change Q5_K to be type 0 when QK_K = 64 Still needs AVX2 implementation * k_quants: AVX2 implementation for new 64-weight Q5_K * k_quants: 10% faster ARM_NEON Q5_K dot product * k_quants: fixed issue caused by merging with master --------- Co-authored-by: Iwan Kawrakow <iwan.kawrakow@gmail.com>
2023-06-26 16:43:07 +00:00
ggml : quantization refactoring (#3833) * ggml : factor all quantization code in ggml-quants ggml-ci * ggml-quants : fix Zig and Swift builds + quantize tool ggml-ci * quantize : --pure option for disabling k-quant mixtures --------- Co-authored-by: cebtenzzre <cebtenzzre@gmail.com>
2023-10-29 16:32:28 +00:00
/* Multiply q with scale and accumulate */
acc = _mm256_fmadd_ps(d, q, acc);
}
*s = hsum_float_8(acc);
#elif defined(__AVX__)
// Initialize accumulator with zeros
__m256 acc = _mm256_setzero_ps();
__m128i mask = _mm_set1_epi8((char)0xF0);
// Main loop
for (int i = 0; i < nb; i++) {
/* Compute combined scale for the block */
ggml : move FP16 <-> FP32 code to ggml-impl.h (#3861) * ggml : move FP16 <-> FP32 stuff to ggml-impl.h ggml-ci * tests : fix ARM build * ggml : explicitly initialize deprecated type traits * ggml : add math.h to ggml-impl.h * ggml : remove duplicate static assert macros * ggml : prefix lookup tables with ggml_ ggml-ci * ggml-impl : move extern "C" to start of file
2023-10-30 17:19:15 +00:00
const __m256 d = _mm256_set1_ps(GGML_FP16_TO_FP32(x[i].d) * GGML_FP16_TO_FP32(y[i].d));
ggml : quantization refactoring (#3833) * ggml : factor all quantization code in ggml-quants ggml-ci * ggml-quants : fix Zig and Swift builds + quantize tool ggml-ci * quantize : --pure option for disabling k-quant mixtures --------- Co-authored-by: cebtenzzre <cebtenzzre@gmail.com>
2023-10-29 16:32:28 +00:00
__m256i bx = bytes_from_nibbles_32(x[i].qs);
const __m256i bxhi = bytes_from_bits_32(x[i].qh);
__m128i bxhil = _mm256_castsi256_si128(bxhi);
__m128i bxhih = _mm256_extractf128_si256(bxhi, 1);
bxhil = _mm_andnot_si128(bxhil, mask);
bxhih = _mm_andnot_si128(bxhih, mask);
__m128i bxl = _mm256_castsi256_si128(bx);
__m128i bxh = _mm256_extractf128_si256(bx, 1);
bxl = _mm_or_si128(bxl, bxhil);
bxh = _mm_or_si128(bxh, bxhih);
bx = MM256_SET_M128I(bxh, bxl);
const __m256i by = _mm256_loadu_si256((const __m256i *)y[i].qs);
const __m256 q = mul_sum_i8_pairs_float(bx, by);
/* Multiply q with scale and accumulate */
acc = _mm256_add_ps(_mm256_mul_ps(d, q), acc);
}
ggml : add SOTA 2,3,4,5,6 bit k-quantizations (#1684) * Starting to add k-quantization to ggml I think it is better to have quantization separate from ggml. For now just adding the k-quants there, but it would be better to also factor out the existing ggml quantizations. * Adding Q3_K and Q8_K (de)-quantization * Q3_K now working on CUDA and AVX2/scalar CUDA is not ideal - ~50% slower than Q4_0 for single token prediction, about the same in batch mode (perplexity). CPU single token is ~55 ms (on Ryzen 7950X). * Some improvement for Q3_K on CUDA It is now ~22.5 ms/token on my GPU, so ~30% slower than Q4_0. * Some more CUDA optimizations for Q3_K Single token is now 20.5 ms/token (~20% slower than Q4_0). Perplexity is on par with Q4_0. * Adding Q4_K - scalar, AVX2, CUDA Performance is the same or perhaps very slightly better than Q4_0 on the CPU. On the GPU, single token prediction is ~10% better than Q4_0, batch mode (perplexity is about the same). * Adding Q6_K - scalar, AVX2, CUDA Performance is ~40% lower compared to Q4_K on the CPU. This is to be expected, considering that we are memory bound on the CPU and the 6-bit model is ~44% larger than the 4-bit. On the GPU, single token prediction is ~6% lower than Q4_0, batch mode (perplexity) is even closer (but still slower). * Adding Q5_K - scalar, AVX2, CUDA Performance is ~20% lower compared to Q4_K on the CPU. This is to be expected, considering that we are memory bound on the CPU and the 5-bit model is ~22% larger than the 4-bit. On the GPU, single token prediction is about the same as Q4_0 for both, single token and batch prediction. * Per convention, all QX_K quantizations use Q5_K for output.weight * Adding quantization mixes * Quantization mixes: didn't quite get what I wanted in the last commit * Q4_K dot product for ARM_NEON * Q6_K dot product for ARM_NEON * Q5_K dot product for ARM_NEON * Adding Q3_K dot for ARM_NEON It is 22% slower than Q4_K, despite the smaller model size. On x86_64, where we are memory bound, the Q3_K model is quite a bit faster than Q4_K. * A very slightly faster ARM_NEON Q3_K dot * Adding Q2_K - just CUDA for now Token prediction is pretty good - about 15.5 ms on a RTX 4080. Perplexity is about the same as Q4_K. * Adding scalar and AVX2 Q2_K dot * Adding ARM_NEON Q2_K dot About the same performance as Q4_K. * A slightly faster ARM_NEON Q2_K dot Single token prediction is now ~36 ms on M2 Max. The code is much simpler too. * Fixed bug in Q2_K CUDA dot product kernel Stranegly enough, for the few prompts I tried with the 7B model the responses looked perfectly reasonable. Only realized something is not quite right when I tried the larger models and started getting nonse back. In any case, Q2_K single token evaluation time on an RTX 4080 in a Ryzen7950X box iusing CUDA and model fully loaded on the GPU are ~15.5 ms for 7B, ~25.4 ms for 13B, and ~55.8 ms for 30B. The max number of layers that fit in VRAM for The 65B is 32. With that, we get ~330 ms per token, which is not that much faster than just running on the CPU (~470 ms per token). * Don't print zeros/NaNs when no count histogram has been collected * A 10% faster CUDA vector dot kernel for Q3_K Q3_K is now running at ~18.5 ms / token on CUDA, so the gap to Q4_0 is only 10%. It seems memory acccess pattern is more important for performance than the amount of computation the kernel does. * A slightly daster Q4_K AVX2 dot product For perplexity, where we are less memory bound, time per pass drops by ~5%. Barely measurable difference for single token prediction. * A slightly faster ARM_NEON A4_K dot product * Minor * Fix quantization error test We cannot possibly be expecting rmse < 0.002 for 2- and 3-bit quantization variants. * Fix docker build I have been sloppy with vector reinterpret casts on ARM_NEON. It seems clang is very forgiving in that regard. * Added forgotten ggml.o dependence on k_quants.h to the Makefile * Had unintentionally committed the Makefile with -Ofast enabled * ggml : rename k_quants -> ggml-quants-k, use lowercase in code --------- Co-authored-by: Iwan Kawrakow <iwan.kawrakow@gmail.com> Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
2023-06-05 19:56:18 +00:00
ggml : quantization refactoring (#3833) * ggml : factor all quantization code in ggml-quants ggml-ci * ggml-quants : fix Zig and Swift builds + quantize tool ggml-ci * quantize : --pure option for disabling k-quant mixtures --------- Co-authored-by: cebtenzzre <cebtenzzre@gmail.com>
2023-10-29 16:32:28 +00:00
*s = hsum_float_8(acc);
#elif defined(__riscv_v_intrinsic)
float sumf = 0.0;
ggml : add SOTA 2,3,4,5,6 bit k-quantizations (#1684) * Starting to add k-quantization to ggml I think it is better to have quantization separate from ggml. For now just adding the k-quants there, but it would be better to also factor out the existing ggml quantizations. * Adding Q3_K and Q8_K (de)-quantization * Q3_K now working on CUDA and AVX2/scalar CUDA is not ideal - ~50% slower than Q4_0 for single token prediction, about the same in batch mode (perplexity). CPU single token is ~55 ms (on Ryzen 7950X). * Some improvement for Q3_K on CUDA It is now ~22.5 ms/token on my GPU, so ~30% slower than Q4_0. * Some more CUDA optimizations for Q3_K Single token is now 20.5 ms/token (~20% slower than Q4_0). Perplexity is on par with Q4_0. * Adding Q4_K - scalar, AVX2, CUDA Performance is the same or perhaps very slightly better than Q4_0 on the CPU. On the GPU, single token prediction is ~10% better than Q4_0, batch mode (perplexity is about the same). * Adding Q6_K - scalar, AVX2, CUDA Performance is ~40% lower compared to Q4_K on the CPU. This is to be expected, considering that we are memory bound on the CPU and the 6-bit model is ~44% larger than the 4-bit. On the GPU, single token prediction is ~6% lower than Q4_0, batch mode (perplexity) is even closer (but still slower). * Adding Q5_K - scalar, AVX2, CUDA Performance is ~20% lower compared to Q4_K on the CPU. This is to be expected, considering that we are memory bound on the CPU and the 5-bit model is ~22% larger than the 4-bit. On the GPU, single token prediction is about the same as Q4_0 for both, single token and batch prediction. * Per convention, all QX_K quantizations use Q5_K for output.weight * Adding quantization mixes * Quantization mixes: didn't quite get what I wanted in the last commit * Q4_K dot product for ARM_NEON * Q6_K dot product for ARM_NEON * Q5_K dot product for ARM_NEON * Adding Q3_K dot for ARM_NEON It is 22% slower than Q4_K, despite the smaller model size. On x86_64, where we are memory bound, the Q3_K model is quite a bit faster than Q4_K. * A very slightly faster ARM_NEON Q3_K dot * Adding Q2_K - just CUDA for now Token prediction is pretty good - about 15.5 ms on a RTX 4080. Perplexity is about the same as Q4_K. * Adding scalar and AVX2 Q2_K dot * Adding ARM_NEON Q2_K dot About the same performance as Q4_K. * A slightly faster ARM_NEON Q2_K dot Single token prediction is now ~36 ms on M2 Max. The code is much simpler too. * Fixed bug in Q2_K CUDA dot product kernel Stranegly enough, for the few prompts I tried with the 7B model the responses looked perfectly reasonable. Only realized something is not quite right when I tried the larger models and started getting nonse back. In any case, Q2_K single token evaluation time on an RTX 4080 in a Ryzen7950X box iusing CUDA and model fully loaded on the GPU are ~15.5 ms for 7B, ~25.4 ms for 13B, and ~55.8 ms for 30B. The max number of layers that fit in VRAM for The 65B is 32. With that, we get ~330 ms per token, which is not that much faster than just running on the CPU (~470 ms per token). * Don't print zeros/NaNs when no count histogram has been collected * A 10% faster CUDA vector dot kernel for Q3_K Q3_K is now running at ~18.5 ms / token on CUDA, so the gap to Q4_0 is only 10%. It seems memory acccess pattern is more important for performance than the amount of computation the kernel does. * A slightly daster Q4_K AVX2 dot product For perplexity, where we are less memory bound, time per pass drops by ~5%. Barely measurable difference for single token prediction. * A slightly faster ARM_NEON A4_K dot product * Minor * Fix quantization error test We cannot possibly be expecting rmse < 0.002 for 2- and 3-bit quantization variants. * Fix docker build I have been sloppy with vector reinterpret casts on ARM_NEON. It seems clang is very forgiving in that regard. * Added forgotten ggml.o dependence on k_quants.h to the Makefile * Had unintentionally committed the Makefile with -Ofast enabled * ggml : rename k_quants -> ggml-quants-k, use lowercase in code --------- Co-authored-by: Iwan Kawrakow <iwan.kawrakow@gmail.com> Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
2023-06-05 19:56:18 +00:00
ggml : quantization refactoring (#3833) * ggml : factor all quantization code in ggml-quants ggml-ci * ggml-quants : fix Zig and Swift builds + quantize tool ggml-ci * quantize : --pure option for disabling k-quant mixtures --------- Co-authored-by: cebtenzzre <cebtenzzre@gmail.com>
2023-10-29 16:32:28 +00:00
uint32_t qh;
ggml : add SOTA 2,3,4,5,6 bit k-quantizations (#1684) * Starting to add k-quantization to ggml I think it is better to have quantization separate from ggml. For now just adding the k-quants there, but it would be better to also factor out the existing ggml quantizations. * Adding Q3_K and Q8_K (de)-quantization * Q3_K now working on CUDA and AVX2/scalar CUDA is not ideal - ~50% slower than Q4_0 for single token prediction, about the same in batch mode (perplexity). CPU single token is ~55 ms (on Ryzen 7950X). * Some improvement for Q3_K on CUDA It is now ~22.5 ms/token on my GPU, so ~30% slower than Q4_0. * Some more CUDA optimizations for Q3_K Single token is now 20.5 ms/token (~20% slower than Q4_0). Perplexity is on par with Q4_0. * Adding Q4_K - scalar, AVX2, CUDA Performance is the same or perhaps very slightly better than Q4_0 on the CPU. On the GPU, single token prediction is ~10% better than Q4_0, batch mode (perplexity is about the same). * Adding Q6_K - scalar, AVX2, CUDA Performance is ~40% lower compared to Q4_K on the CPU. This is to be expected, considering that we are memory bound on the CPU and the 6-bit model is ~44% larger than the 4-bit. On the GPU, single token prediction is ~6% lower than Q4_0, batch mode (perplexity) is even closer (but still slower). * Adding Q5_K - scalar, AVX2, CUDA Performance is ~20% lower compared to Q4_K on the CPU. This is to be expected, considering that we are memory bound on the CPU and the 5-bit model is ~22% larger than the 4-bit. On the GPU, single token prediction is about the same as Q4_0 for both, single token and batch prediction. * Per convention, all QX_K quantizations use Q5_K for output.weight * Adding quantization mixes * Quantization mixes: didn't quite get what I wanted in the last commit * Q4_K dot product for ARM_NEON * Q6_K dot product for ARM_NEON * Q5_K dot product for ARM_NEON * Adding Q3_K dot for ARM_NEON It is 22% slower than Q4_K, despite the smaller model size. On x86_64, where we are memory bound, the Q3_K model is quite a bit faster than Q4_K. * A very slightly faster ARM_NEON Q3_K dot * Adding Q2_K - just CUDA for now Token prediction is pretty good - about 15.5 ms on a RTX 4080. Perplexity is about the same as Q4_K. * Adding scalar and AVX2 Q2_K dot * Adding ARM_NEON Q2_K dot About the same performance as Q4_K. * A slightly faster ARM_NEON Q2_K dot Single token prediction is now ~36 ms on M2 Max. The code is much simpler too. * Fixed bug in Q2_K CUDA dot product kernel Stranegly enough, for the few prompts I tried with the 7B model the responses looked perfectly reasonable. Only realized something is not quite right when I tried the larger models and started getting nonse back. In any case, Q2_K single token evaluation time on an RTX 4080 in a Ryzen7950X box iusing CUDA and model fully loaded on the GPU are ~15.5 ms for 7B, ~25.4 ms for 13B, and ~55.8 ms for 30B. The max number of layers that fit in VRAM for The 65B is 32. With that, we get ~330 ms per token, which is not that much faster than just running on the CPU (~470 ms per token). * Don't print zeros/NaNs when no count histogram has been collected * A 10% faster CUDA vector dot kernel for Q3_K Q3_K is now running at ~18.5 ms / token on CUDA, so the gap to Q4_0 is only 10%. It seems memory acccess pattern is more important for performance than the amount of computation the kernel does. * A slightly daster Q4_K AVX2 dot product For perplexity, where we are less memory bound, time per pass drops by ~5%. Barely measurable difference for single token prediction. * A slightly faster ARM_NEON A4_K dot product * Minor * Fix quantization error test We cannot possibly be expecting rmse < 0.002 for 2- and 3-bit quantization variants. * Fix docker build I have been sloppy with vector reinterpret casts on ARM_NEON. It seems clang is very forgiving in that regard. * Added forgotten ggml.o dependence on k_quants.h to the Makefile * Had unintentionally committed the Makefile with -Ofast enabled * ggml : rename k_quants -> ggml-quants-k, use lowercase in code --------- Co-authored-by: Iwan Kawrakow <iwan.kawrakow@gmail.com> Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
2023-06-05 19:56:18 +00:00
ggml : quantization refactoring (#3833) * ggml : factor all quantization code in ggml-quants ggml-ci * ggml-quants : fix Zig and Swift builds + quantize tool ggml-ci * quantize : --pure option for disabling k-quant mixtures --------- Co-authored-by: cebtenzzre <cebtenzzre@gmail.com>
2023-10-29 16:32:28 +00:00
size_t vl = __riscv_vsetvl_e8m1(qk/2);
english : use `typos` to fix comments and logs (#4354)
2023-12-12 09:53:36 +00:00
// These temporary registers are for masking and shift operations
ggml : quantization refactoring (#3833) * ggml : factor all quantization code in ggml-quants ggml-ci * ggml-quants : fix Zig and Swift builds + quantize tool ggml-ci * quantize : --pure option for disabling k-quant mixtures --------- Co-authored-by: cebtenzzre <cebtenzzre@gmail.com>
2023-10-29 16:32:28 +00:00
vuint32m2_t vt_1 = __riscv_vid_v_u32m2(vl);
vuint32m2_t vt_2 = __riscv_vsll_vv_u32m2(__riscv_vmv_v_x_u32m2(1, vl), vt_1, vl);
vuint32m2_t vt_3 = __riscv_vsll_vx_u32m2(vt_2, 16, vl);
vuint32m2_t vt_4 = __riscv_vadd_vx_u32m2(vt_1, 12, vl);
ggml : add SOTA 2,3,4,5,6 bit k-quantizations (#1684) * Starting to add k-quantization to ggml I think it is better to have quantization separate from ggml. For now just adding the k-quants there, but it would be better to also factor out the existing ggml quantizations. * Adding Q3_K and Q8_K (de)-quantization * Q3_K now working on CUDA and AVX2/scalar CUDA is not ideal - ~50% slower than Q4_0 for single token prediction, about the same in batch mode (perplexity). CPU single token is ~55 ms (on Ryzen 7950X). * Some improvement for Q3_K on CUDA It is now ~22.5 ms/token on my GPU, so ~30% slower than Q4_0. * Some more CUDA optimizations for Q3_K Single token is now 20.5 ms/token (~20% slower than Q4_0). Perplexity is on par with Q4_0. * Adding Q4_K - scalar, AVX2, CUDA Performance is the same or perhaps very slightly better than Q4_0 on the CPU. On the GPU, single token prediction is ~10% better than Q4_0, batch mode (perplexity is about the same). * Adding Q6_K - scalar, AVX2, CUDA Performance is ~40% lower compared to Q4_K on the CPU. This is to be expected, considering that we are memory bound on the CPU and the 6-bit model is ~44% larger than the 4-bit. On the GPU, single token prediction is ~6% lower than Q4_0, batch mode (perplexity) is even closer (but still slower). * Adding Q5_K - scalar, AVX2, CUDA Performance is ~20% lower compared to Q4_K on the CPU. This is to be expected, considering that we are memory bound on the CPU and the 5-bit model is ~22% larger than the 4-bit. On the GPU, single token prediction is about the same as Q4_0 for both, single token and batch prediction. * Per convention, all QX_K quantizations use Q5_K for output.weight * Adding quantization mixes * Quantization mixes: didn't quite get what I wanted in the last commit * Q4_K dot product for ARM_NEON * Q6_K dot product for ARM_NEON * Q5_K dot product for ARM_NEON * Adding Q3_K dot for ARM_NEON It is 22% slower than Q4_K, despite the smaller model size. On x86_64, where we are memory bound, the Q3_K model is quite a bit faster than Q4_K. * A very slightly faster ARM_NEON Q3_K dot * Adding Q2_K - just CUDA for now Token prediction is pretty good - about 15.5 ms on a RTX 4080. Perplexity is about the same as Q4_K. * Adding scalar and AVX2 Q2_K dot * Adding ARM_NEON Q2_K dot About the same performance as Q4_K. * A slightly faster ARM_NEON Q2_K dot Single token prediction is now ~36 ms on M2 Max. The code is much simpler too. * Fixed bug in Q2_K CUDA dot product kernel Stranegly enough, for the few prompts I tried with the 7B model the responses looked perfectly reasonable. Only realized something is not quite right when I tried the larger models and started getting nonse back. In any case, Q2_K single token evaluation time on an RTX 4080 in a Ryzen7950X box iusing CUDA and model fully loaded on the GPU are ~15.5 ms for 7B, ~25.4 ms for 13B, and ~55.8 ms for 30B. The max number of layers that fit in VRAM for The 65B is 32. With that, we get ~330 ms per token, which is not that much faster than just running on the CPU (~470 ms per token). * Don't print zeros/NaNs when no count histogram has been collected * A 10% faster CUDA vector dot kernel for Q3_K Q3_K is now running at ~18.5 ms / token on CUDA, so the gap to Q4_0 is only 10%. It seems memory acccess pattern is more important for performance than the amount of computation the kernel does. * A slightly daster Q4_K AVX2 dot product For perplexity, where we are less memory bound, time per pass drops by ~5%. Barely measurable difference for single token prediction. * A slightly faster ARM_NEON A4_K dot product * Minor * Fix quantization error test We cannot possibly be expecting rmse < 0.002 for 2- and 3-bit quantization variants. * Fix docker build I have been sloppy with vector reinterpret casts on ARM_NEON. It seems clang is very forgiving in that regard. * Added forgotten ggml.o dependence on k_quants.h to the Makefile * Had unintentionally committed the Makefile with -Ofast enabled * ggml : rename k_quants -> ggml-quants-k, use lowercase in code --------- Co-authored-by: Iwan Kawrakow <iwan.kawrakow@gmail.com> Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
2023-06-05 19:56:18 +00:00
for (int i = 0; i < nb; i++) {
ggml : quantization refactoring (#3833) * ggml : factor all quantization code in ggml-quants ggml-ci * ggml-quants : fix Zig and Swift builds + quantize tool ggml-ci * quantize : --pure option for disabling k-quant mixtures --------- Co-authored-by: cebtenzzre <cebtenzzre@gmail.com>
2023-10-29 16:32:28 +00:00
memcpy(&qh, x[i].qh, sizeof(uint32_t));
ggml : add SOTA 2,3,4,5,6 bit k-quantizations (#1684) * Starting to add k-quantization to ggml I think it is better to have quantization separate from ggml. For now just adding the k-quants there, but it would be better to also factor out the existing ggml quantizations. * Adding Q3_K and Q8_K (de)-quantization * Q3_K now working on CUDA and AVX2/scalar CUDA is not ideal - ~50% slower than Q4_0 for single token prediction, about the same in batch mode (perplexity). CPU single token is ~55 ms (on Ryzen 7950X). * Some improvement for Q3_K on CUDA It is now ~22.5 ms/token on my GPU, so ~30% slower than Q4_0. * Some more CUDA optimizations for Q3_K Single token is now 20.5 ms/token (~20% slower than Q4_0). Perplexity is on par with Q4_0. * Adding Q4_K - scalar, AVX2, CUDA Performance is the same or perhaps very slightly better than Q4_0 on the CPU. On the GPU, single token prediction is ~10% better than Q4_0, batch mode (perplexity is about the same). * Adding Q6_K - scalar, AVX2, CUDA Performance is ~40% lower compared to Q4_K on the CPU. This is to be expected, considering that we are memory bound on the CPU and the 6-bit model is ~44% larger than the 4-bit. On the GPU, single token prediction is ~6% lower than Q4_0, batch mode (perplexity) is even closer (but still slower). * Adding Q5_K - scalar, AVX2, CUDA Performance is ~20% lower compared to Q4_K on the CPU. This is to be expected, considering that we are memory bound on the CPU and the 5-bit model is ~22% larger than the 4-bit. On the GPU, single token prediction is about the same as Q4_0 for both, single token and batch prediction. * Per convention, all QX_K quantizations use Q5_K for output.weight * Adding quantization mixes * Quantization mixes: didn't quite get what I wanted in the last commit * Q4_K dot product for ARM_NEON * Q6_K dot product for ARM_NEON * Q5_K dot product for ARM_NEON * Adding Q3_K dot for ARM_NEON It is 22% slower than Q4_K, despite the smaller model size. On x86_64, where we are memory bound, the Q3_K model is quite a bit faster than Q4_K. * A very slightly faster ARM_NEON Q3_K dot * Adding Q2_K - just CUDA for now Token prediction is pretty good - about 15.5 ms on a RTX 4080. Perplexity is about the same as Q4_K. * Adding scalar and AVX2 Q2_K dot * Adding ARM_NEON Q2_K dot About the same performance as Q4_K. * A slightly faster ARM_NEON Q2_K dot Single token prediction is now ~36 ms on M2 Max. The code is much simpler too. * Fixed bug in Q2_K CUDA dot product kernel Stranegly enough, for the few prompts I tried with the 7B model the responses looked perfectly reasonable. Only realized something is not quite right when I tried the larger models and started getting nonse back. In any case, Q2_K single token evaluation time on an RTX 4080 in a Ryzen7950X box iusing CUDA and model fully loaded on the GPU are ~15.5 ms for 7B, ~25.4 ms for 13B, and ~55.8 ms for 30B. The max number of layers that fit in VRAM for The 65B is 32. With that, we get ~330 ms per token, which is not that much faster than just running on the CPU (~470 ms per token). * Don't print zeros/NaNs when no count histogram has been collected * A 10% faster CUDA vector dot kernel for Q3_K Q3_K is now running at ~18.5 ms / token on CUDA, so the gap to Q4_0 is only 10%. It seems memory acccess pattern is more important for performance than the amount of computation the kernel does. * A slightly daster Q4_K AVX2 dot product For perplexity, where we are less memory bound, time per pass drops by ~5%. Barely measurable difference for single token prediction. * A slightly faster ARM_NEON A4_K dot product * Minor * Fix quantization error test We cannot possibly be expecting rmse < 0.002 for 2- and 3-bit quantization variants. * Fix docker build I have been sloppy with vector reinterpret casts on ARM_NEON. It seems clang is very forgiving in that regard. * Added forgotten ggml.o dependence on k_quants.h to the Makefile * Had unintentionally committed the Makefile with -Ofast enabled * ggml : rename k_quants -> ggml-quants-k, use lowercase in code --------- Co-authored-by: Iwan Kawrakow <iwan.kawrakow@gmail.com> Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
2023-06-05 19:56:18 +00:00
ggml : quantization refactoring (#3833) * ggml : factor all quantization code in ggml-quants ggml-ci * ggml-quants : fix Zig and Swift builds + quantize tool ggml-ci * quantize : --pure option for disabling k-quant mixtures --------- Co-authored-by: cebtenzzre <cebtenzzre@gmail.com>
2023-10-29 16:32:28 +00:00
// ((qh & (1u << (j + 0 ))) >> (j + 0 )) << 4;
vuint32m2_t xha_0 = __riscv_vand_vx_u32m2(vt_2, qh, vl);
vuint32m2_t xhr_0 = __riscv_vsrl_vv_u32m2(xha_0, vt_1, vl);
vuint32m2_t xhl_0 = __riscv_vsll_vx_u32m2(xhr_0, 4, vl);
ggml : add SOTA 2,3,4,5,6 bit k-quantizations (#1684) * Starting to add k-quantization to ggml I think it is better to have quantization separate from ggml. For now just adding the k-quants there, but it would be better to also factor out the existing ggml quantizations. * Adding Q3_K and Q8_K (de)-quantization * Q3_K now working on CUDA and AVX2/scalar CUDA is not ideal - ~50% slower than Q4_0 for single token prediction, about the same in batch mode (perplexity). CPU single token is ~55 ms (on Ryzen 7950X). * Some improvement for Q3_K on CUDA It is now ~22.5 ms/token on my GPU, so ~30% slower than Q4_0. * Some more CUDA optimizations for Q3_K Single token is now 20.5 ms/token (~20% slower than Q4_0). Perplexity is on par with Q4_0. * Adding Q4_K - scalar, AVX2, CUDA Performance is the same or perhaps very slightly better than Q4_0 on the CPU. On the GPU, single token prediction is ~10% better than Q4_0, batch mode (perplexity is about the same). * Adding Q6_K - scalar, AVX2, CUDA Performance is ~40% lower compared to Q4_K on the CPU. This is to be expected, considering that we are memory bound on the CPU and the 6-bit model is ~44% larger than the 4-bit. On the GPU, single token prediction is ~6% lower than Q4_0, batch mode (perplexity) is even closer (but still slower). * Adding Q5_K - scalar, AVX2, CUDA Performance is ~20% lower compared to Q4_K on the CPU. This is to be expected, considering that we are memory bound on the CPU and the 5-bit model is ~22% larger than the 4-bit. On the GPU, single token prediction is about the same as Q4_0 for both, single token and batch prediction. * Per convention, all QX_K quantizations use Q5_K for output.weight * Adding quantization mixes * Quantization mixes: didn't quite get what I wanted in the last commit * Q4_K dot product for ARM_NEON * Q6_K dot product for ARM_NEON * Q5_K dot product for ARM_NEON * Adding Q3_K dot for ARM_NEON It is 22% slower than Q4_K, despite the smaller model size. On x86_64, where we are memory bound, the Q3_K model is quite a bit faster than Q4_K. * A very slightly faster ARM_NEON Q3_K dot * Adding Q2_K - just CUDA for now Token prediction is pretty good - about 15.5 ms on a RTX 4080. Perplexity is about the same as Q4_K. * Adding scalar and AVX2 Q2_K dot * Adding ARM_NEON Q2_K dot About the same performance as Q4_K. * A slightly faster ARM_NEON Q2_K dot Single token prediction is now ~36 ms on M2 Max. The code is much simpler too. * Fixed bug in Q2_K CUDA dot product kernel Stranegly enough, for the few prompts I tried with the 7B model the responses looked perfectly reasonable. Only realized something is not quite right when I tried the larger models and started getting nonse back. In any case, Q2_K single token evaluation time on an RTX 4080 in a Ryzen7950X box iusing CUDA and model fully loaded on the GPU are ~15.5 ms for 7B, ~25.4 ms for 13B, and ~55.8 ms for 30B. The max number of layers that fit in VRAM for The 65B is 32. With that, we get ~330 ms per token, which is not that much faster than just running on the CPU (~470 ms per token). * Don't print zeros/NaNs when no count histogram has been collected * A 10% faster CUDA vector dot kernel for Q3_K Q3_K is now running at ~18.5 ms / token on CUDA, so the gap to Q4_0 is only 10%. It seems memory acccess pattern is more important for performance than the amount of computation the kernel does. * A slightly daster Q4_K AVX2 dot product For perplexity, where we are less memory bound, time per pass drops by ~5%. Barely measurable difference for single token prediction. * A slightly faster ARM_NEON A4_K dot product * Minor * Fix quantization error test We cannot possibly be expecting rmse < 0.002 for 2- and 3-bit quantization variants. * Fix docker build I have been sloppy with vector reinterpret casts on ARM_NEON. It seems clang is very forgiving in that regard. * Added forgotten ggml.o dependence on k_quants.h to the Makefile * Had unintentionally committed the Makefile with -Ofast enabled * ggml : rename k_quants -> ggml-quants-k, use lowercase in code --------- Co-authored-by: Iwan Kawrakow <iwan.kawrakow@gmail.com> Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
2023-06-05 19:56:18 +00:00
ggml : quantization refactoring (#3833) * ggml : factor all quantization code in ggml-quants ggml-ci * ggml-quants : fix Zig and Swift builds + quantize tool ggml-ci * quantize : --pure option for disabling k-quant mixtures --------- Co-authored-by: cebtenzzre <cebtenzzre@gmail.com>
2023-10-29 16:32:28 +00:00
// ((qh & (1u << (j + 16))) >> (j + 12));
vuint32m2_t xha_1 = __riscv_vand_vx_u32m2(vt_3, qh, vl);
vuint32m2_t xhl_1 = __riscv_vsrl_vv_u32m2(xha_1, vt_4, vl);
k-quants : support for super-block size of 64 (#2001) * k_quants: WIP super-blocks with 64 weights * k_quants: WIP super-blocks with 64 weights Q6_K scalar and AVX2 works * k_quants: WIP super-blocks with 64 weights Q4_K scalar and AVX2 works * k_quants: WIP super-blocks with 64 weights Q2_K scalar and AVX2 works. Q2_K is way too slow (it is actually slower than the scalar implementation) * k_quants: WIP super-blocks with 64 weights Q3_K scalar and AVX2 works. * k_quants: WIP super-blocks with 64 weights Q5_K scalar and AVX2 works, and with that all k_quants are done on AVX2 and scalar * k_quants: WIP super-blocks with 64 weights Q6_K working on CUDA. Cannot make it run quite as gast as with super-blocks with 256 weigths: 8% slower on 4080, 20% slower on the 1660 (but there we fit 1 less layer on the GPU because pf the larger model size), so some fraction of these 20% is due to that, * k_quants: WIP super-blocks with 64 weights Q4_K working on CUDA. ~10% slower on GTX-1660, 16% slower on 4080. * k_quants: WIP super-blocks with 64 weights Q2_K working on CUDA. ~3% slower on GTX-1660, 10% slower on 4080. * k_quants: WIP super-blocks with 64 weights Q3_K working on CUDA. * k_quants: WIP super-blocks with 64 weights Q5_K working on CUDA, and with this CUDA is done. * k_quants: WIP super-blocks with 64 weights Q6_K working on ARM_NEON * k_quants: WIP super-blocks with 64 weights Q4_K working on ARM_NEON, but quite a bit slower than 256 weights * k_quants: WIP super-blocks with 64 weights Q2_K working on ARM_NEON, but quite a bit slower than 256 weights * k_quants: WIP super-blocks with 64 weights Q3_K working on ARM_NEON, but quite a bit slower than 256 weights. * k_quants: WIP super-blocks with 64 weights Q5_K working on ARM_NEON, but quite a bit slower than 256 weights. With that, we have full support for ARM_NEON, although performance is not quite there. * k_quants: WIP super-blocks with 64 weights Slightly more efficient Q3_K and Q5_K * k_quants: WIP super-blocks with 64 weights Another small improvement for Q3_K and Q5_K on ARM_NEON * k_quants: WIP super-blocks with 64 weights Yet another speedup for Q5_K on ARM_NEON. We are now within 10% of the QK_K = 256 version. * k_quants: WIP super-blocks with 64 weights * We are able to pass preprocessor macros to the Metal compiler * Q6_K works and is actually slightly more efficient than the QK_K = 256 version (25.2 ms vs 25.8 ms) * k_quants: WIP super-blocks with 64 weights Q4_K works on Metal and is actually slightly faster than QK_K = 256 (21.95 ms vs 24.0 ms). * k_quants: WIP super-blocks with 64 weights Q2_K works on Metal and is very slightly faster than QK_K = 256 (23.8 ms vs 24.2 ms). * k_quants: WIP super-blocks with 64 weights Q3_K works on Metal and is slightly faster than QK_K = 256 (26.6 ms vs 28.3 ms). * k_quants: WIP super-blocks with 64 weights Q5_K works on Metal and is slightly faster than QK_K = 256 (23.7 ms vs 26.3 ms). * k_quants: call them _K, not _k, also on Metal * k_quants: correctly define QK_K in llama.cpp * Fixed bug in q4_K quantization added with the 64-block addition * Simplify via lambda * k_quants: swicth Q3_K to 4-bit scales when QK_K = 64 Otherwise there isn't much benefit from this quantization type. There is some very slight loss in accuracy, but we reduce size by ~7%. E.g., for OpenLLaMA-3B, Q3_K_S perplexity is 8.6131 with 8-bit scales and 8.6352 with 4-bit, while file size decreases from 1.53G to 1.44G. * k_quants: switch Q4_K to 4-bit scales when QK_K = 64 Here the loss in accuracy is greater than for Q3_K, but the Q4_K points still move further to the left on the perplexity vs size curve. * k_quants: forgot to add the Metal changes in last commit * k_quants: change Q5_K to be type 0 when QK_K = 64 Still needs AVX2 implementation * k_quants: AVX2 implementation for new 64-weight Q5_K * k_quants: 10% faster ARM_NEON Q5_K dot product * k_quants: fixed issue caused by merging with master --------- Co-authored-by: Iwan Kawrakow <iwan.kawrakow@gmail.com>
2023-06-26 16:43:07 +00:00
ggml : quantization refactoring (#3833) * ggml : factor all quantization code in ggml-quants ggml-ci * ggml-quants : fix Zig and Swift builds + quantize tool ggml-ci * quantize : --pure option for disabling k-quant mixtures --------- Co-authored-by: cebtenzzre <cebtenzzre@gmail.com>
2023-10-29 16:32:28 +00:00
// narrowing
vuint16m1_t xhc_0 = __riscv_vncvt_x_x_w_u16m1(xhl_0, vl);
vuint8mf2_t xh_0 = __riscv_vncvt_x_x_w_u8mf2(xhc_0, vl);
k-quants : support for super-block size of 64 (#2001) * k_quants: WIP super-blocks with 64 weights * k_quants: WIP super-blocks with 64 weights Q6_K scalar and AVX2 works * k_quants: WIP super-blocks with 64 weights Q4_K scalar and AVX2 works * k_quants: WIP super-blocks with 64 weights Q2_K scalar and AVX2 works. Q2_K is way too slow (it is actually slower than the scalar implementation) * k_quants: WIP super-blocks with 64 weights Q3_K scalar and AVX2 works. * k_quants: WIP super-blocks with 64 weights Q5_K scalar and AVX2 works, and with that all k_quants are done on AVX2 and scalar * k_quants: WIP super-blocks with 64 weights Q6_K working on CUDA. Cannot make it run quite as gast as with super-blocks with 256 weigths: 8% slower on 4080, 20% slower on the 1660 (but there we fit 1 less layer on the GPU because pf the larger model size), so some fraction of these 20% is due to that, * k_quants: WIP super-blocks with 64 weights Q4_K working on CUDA. ~10% slower on GTX-1660, 16% slower on 4080. * k_quants: WIP super-blocks with 64 weights Q2_K working on CUDA. ~3% slower on GTX-1660, 10% slower on 4080. * k_quants: WIP super-blocks with 64 weights Q3_K working on CUDA. * k_quants: WIP super-blocks with 64 weights Q5_K working on CUDA, and with this CUDA is done. * k_quants: WIP super-blocks with 64 weights Q6_K working on ARM_NEON * k_quants: WIP super-blocks with 64 weights Q4_K working on ARM_NEON, but quite a bit slower than 256 weights * k_quants: WIP super-blocks with 64 weights Q2_K working on ARM_NEON, but quite a bit slower than 256 weights * k_quants: WIP super-blocks with 64 weights Q3_K working on ARM_NEON, but quite a bit slower than 256 weights. * k_quants: WIP super-blocks with 64 weights Q5_K working on ARM_NEON, but quite a bit slower than 256 weights. With that, we have full support for ARM_NEON, although performance is not quite there. * k_quants: WIP super-blocks with 64 weights Slightly more efficient Q3_K and Q5_K * k_quants: WIP super-blocks with 64 weights Another small improvement for Q3_K and Q5_K on ARM_NEON * k_quants: WIP super-blocks with 64 weights Yet another speedup for Q5_K on ARM_NEON. We are now within 10% of the QK_K = 256 version. * k_quants: WIP super-blocks with 64 weights * We are able to pass preprocessor macros to the Metal compiler * Q6_K works and is actually slightly more efficient than the QK_K = 256 version (25.2 ms vs 25.8 ms) * k_quants: WIP super-blocks with 64 weights Q4_K works on Metal and is actually slightly faster than QK_K = 256 (21.95 ms vs 24.0 ms). * k_quants: WIP super-blocks with 64 weights Q2_K works on Metal and is very slightly faster than QK_K = 256 (23.8 ms vs 24.2 ms). * k_quants: WIP super-blocks with 64 weights Q3_K works on Metal and is slightly faster than QK_K = 256 (26.6 ms vs 28.3 ms). * k_quants: WIP super-blocks with 64 weights Q5_K works on Metal and is slightly faster than QK_K = 256 (23.7 ms vs 26.3 ms). * k_quants: call them _K, not _k, also on Metal * k_quants: correctly define QK_K in llama.cpp * Fixed bug in q4_K quantization added with the 64-block addition * Simplify via lambda * k_quants: swicth Q3_K to 4-bit scales when QK_K = 64 Otherwise there isn't much benefit from this quantization type. There is some very slight loss in accuracy, but we reduce size by ~7%. E.g., for OpenLLaMA-3B, Q3_K_S perplexity is 8.6131 with 8-bit scales and 8.6352 with 4-bit, while file size decreases from 1.53G to 1.44G. * k_quants: switch Q4_K to 4-bit scales when QK_K = 64 Here the loss in accuracy is greater than for Q3_K, but the Q4_K points still move further to the left on the perplexity vs size curve. * k_quants: forgot to add the Metal changes in last commit * k_quants: change Q5_K to be type 0 when QK_K = 64 Still needs AVX2 implementation * k_quants: AVX2 implementation for new 64-weight Q5_K * k_quants: 10% faster ARM_NEON Q5_K dot product * k_quants: fixed issue caused by merging with master --------- Co-authored-by: Iwan Kawrakow <iwan.kawrakow@gmail.com>
2023-06-26 16:43:07 +00:00
ggml : quantization refactoring (#3833) * ggml : factor all quantization code in ggml-quants ggml-ci * ggml-quants : fix Zig and Swift builds + quantize tool ggml-ci * quantize : --pure option for disabling k-quant mixtures --------- Co-authored-by: cebtenzzre <cebtenzzre@gmail.com>
2023-10-29 16:32:28 +00:00
vuint16m1_t xhc_1 = __riscv_vncvt_x_x_w_u16m1(xhl_1, vl);
vuint8mf2_t xh_1 = __riscv_vncvt_x_x_w_u8mf2(xhc_1, vl);
// load
vuint8mf2_t tx = __riscv_vle8_v_u8mf2(x[i].qs, vl);
vint8mf2_t y0 = __riscv_vle8_v_i8mf2(y[i].qs, vl);
vint8mf2_t y1 = __riscv_vle8_v_i8mf2(y[i].qs+16, vl);
vuint8mf2_t x_at = __riscv_vand_vx_u8mf2(tx, 0x0F, vl);
vuint8mf2_t x_lt = __riscv_vsrl_vx_u8mf2(tx, 0x04, vl);
vuint8mf2_t x_a = __riscv_vor_vv_u8mf2(x_at, xh_0, vl);
vuint8mf2_t x_l = __riscv_vor_vv_u8mf2(x_lt, xh_1, vl);
vint8mf2_t x_ai = __riscv_vreinterpret_v_u8mf2_i8mf2(x_a);
vint8mf2_t x_li = __riscv_vreinterpret_v_u8mf2_i8mf2(x_l);
vint8mf2_t v0 = __riscv_vsub_vx_i8mf2(x_ai, 16, vl);
vint8mf2_t v1 = __riscv_vsub_vx_i8mf2(x_li, 16, vl);
vint16m1_t vec_mul1 = __riscv_vwmul_vv_i16m1(v0, y0, vl);
vint16m1_t vec_mul2 = __riscv_vwmul_vv_i16m1(v1, y1, vl);
vint32m1_t vec_zero = __riscv_vmv_v_x_i32m1(0, vl);
vint32m1_t vs1 = __riscv_vwredsum_vs_i16m1_i32m1(vec_mul1, vec_zero, vl);
vint32m1_t vs2 = __riscv_vwredsum_vs_i16m1_i32m1(vec_mul2, vs1, vl);
int sumi = __riscv_vmv_x_s_i32m1_i32(vs2);
ggml : move FP16 <-> FP32 code to ggml-impl.h (#3861) * ggml : move FP16 <-> FP32 stuff to ggml-impl.h ggml-ci * tests : fix ARM build * ggml : explicitly initialize deprecated type traits * ggml : add math.h to ggml-impl.h * ggml : remove duplicate static assert macros * ggml : prefix lookup tables with ggml_ ggml-ci * ggml-impl : move extern "C" to start of file
2023-10-30 17:19:15 +00:00
sumf += (GGML_FP16_TO_FP32(x[i].d)*GGML_FP16_TO_FP32(y[i].d)) * sumi;
ggml : quantization refactoring (#3833) * ggml : factor all quantization code in ggml-quants ggml-ci * ggml-quants : fix Zig and Swift builds + quantize tool ggml-ci * quantize : --pure option for disabling k-quant mixtures --------- Co-authored-by: cebtenzzre <cebtenzzre@gmail.com>
2023-10-29 16:32:28 +00:00
}
*s = sumf;
k-quants : support for super-block size of 64 (#2001) * k_quants: WIP super-blocks with 64 weights * k_quants: WIP super-blocks with 64 weights Q6_K scalar and AVX2 works * k_quants: WIP super-blocks with 64 weights Q4_K scalar and AVX2 works * k_quants: WIP super-blocks with 64 weights Q2_K scalar and AVX2 works. Q2_K is way too slow (it is actually slower than the scalar implementation) * k_quants: WIP super-blocks with 64 weights Q3_K scalar and AVX2 works. * k_quants: WIP super-blocks with 64 weights Q5_K scalar and AVX2 works, and with that all k_quants are done on AVX2 and scalar * k_quants: WIP super-blocks with 64 weights Q6_K working on CUDA. Cannot make it run quite as gast as with super-blocks with 256 weigths: 8% slower on 4080, 20% slower on the 1660 (but there we fit 1 less layer on the GPU because pf the larger model size), so some fraction of these 20% is due to that, * k_quants: WIP super-blocks with 64 weights Q4_K working on CUDA. ~10% slower on GTX-1660, 16% slower on 4080. * k_quants: WIP super-blocks with 64 weights Q2_K working on CUDA. ~3% slower on GTX-1660, 10% slower on 4080. * k_quants: WIP super-blocks with 64 weights Q3_K working on CUDA. * k_quants: WIP super-blocks with 64 weights Q5_K working on CUDA, and with this CUDA is done. * k_quants: WIP super-blocks with 64 weights Q6_K working on ARM_NEON * k_quants: WIP super-blocks with 64 weights Q4_K working on ARM_NEON, but quite a bit slower than 256 weights * k_quants: WIP super-blocks with 64 weights Q2_K working on ARM_NEON, but quite a bit slower than 256 weights * k_quants: WIP super-blocks with 64 weights Q3_K working on ARM_NEON, but quite a bit slower than 256 weights. * k_quants: WIP super-blocks with 64 weights Q5_K working on ARM_NEON, but quite a bit slower than 256 weights. With that, we have full support for ARM_NEON, although performance is not quite there. * k_quants: WIP super-blocks with 64 weights Slightly more efficient Q3_K and Q5_K * k_quants: WIP super-blocks with 64 weights Another small improvement for Q3_K and Q5_K on ARM_NEON * k_quants: WIP super-blocks with 64 weights Yet another speedup for Q5_K on ARM_NEON. We are now within 10% of the QK_K = 256 version. * k_quants: WIP super-blocks with 64 weights * We are able to pass preprocessor macros to the Metal compiler * Q6_K works and is actually slightly more efficient than the QK_K = 256 version (25.2 ms vs 25.8 ms) * k_quants: WIP super-blocks with 64 weights Q4_K works on Metal and is actually slightly faster than QK_K = 256 (21.95 ms vs 24.0 ms). * k_quants: WIP super-blocks with 64 weights Q2_K works on Metal and is very slightly faster than QK_K = 256 (23.8 ms vs 24.2 ms). * k_quants: WIP super-blocks with 64 weights Q3_K works on Metal and is slightly faster than QK_K = 256 (26.6 ms vs 28.3 ms). * k_quants: WIP super-blocks with 64 weights Q5_K works on Metal and is slightly faster than QK_K = 256 (23.7 ms vs 26.3 ms). * k_quants: call them _K, not _k, also on Metal * k_quants: correctly define QK_K in llama.cpp * Fixed bug in q4_K quantization added with the 64-block addition * Simplify via lambda * k_quants: swicth Q3_K to 4-bit scales when QK_K = 64 Otherwise there isn't much benefit from this quantization type. There is some very slight loss in accuracy, but we reduce size by ~7%. E.g., for OpenLLaMA-3B, Q3_K_S perplexity is 8.6131 with 8-bit scales and 8.6352 with 4-bit, while file size decreases from 1.53G to 1.44G. * k_quants: switch Q4_K to 4-bit scales when QK_K = 64 Here the loss in accuracy is greater than for Q3_K, but the Q4_K points still move further to the left on the perplexity vs size curve. * k_quants: forgot to add the Metal changes in last commit * k_quants: change Q5_K to be type 0 when QK_K = 64 Still needs AVX2 implementation * k_quants: AVX2 implementation for new 64-weight Q5_K * k_quants: 10% faster ARM_NEON Q5_K dot product * k_quants: fixed issue caused by merging with master --------- Co-authored-by: Iwan Kawrakow <iwan.kawrakow@gmail.com>
2023-06-26 16:43:07 +00:00
#else
ggml : quantization refactoring (#3833) * ggml : factor all quantization code in ggml-quants ggml-ci * ggml-quants : fix Zig and Swift builds + quantize tool ggml-ci * quantize : --pure option for disabling k-quant mixtures --------- Co-authored-by: cebtenzzre <cebtenzzre@gmail.com>
2023-10-29 16:32:28 +00:00
// scalar
float sumf = 0.0;
for (int i = 0; i < nb; i++) {
uint32_t qh;
memcpy(&qh, x[i].qh, sizeof(qh));
int sumi = 0;
for (int j = 0; j < qk/2; ++j) {
const uint8_t xh_0 = ((qh & (1u << (j + 0 ))) >> (j + 0 )) << 4;
const uint8_t xh_1 = ((qh & (1u << (j + 16))) >> (j + 12));
const int32_t x0 = ((x[i].qs[j] & 0x0F) | xh_0) - 16;
const int32_t x1 = ((x[i].qs[j] >> 4) | xh_1) - 16;
sumi += (x0 * y[i].qs[j]) + (x1 * y[i].qs[j + qk/2]);
k-quants : support for super-block size of 64 (#2001) * k_quants: WIP super-blocks with 64 weights * k_quants: WIP super-blocks with 64 weights Q6_K scalar and AVX2 works * k_quants: WIP super-blocks with 64 weights Q4_K scalar and AVX2 works * k_quants: WIP super-blocks with 64 weights Q2_K scalar and AVX2 works. Q2_K is way too slow (it is actually slower than the scalar implementation) * k_quants: WIP super-blocks with 64 weights Q3_K scalar and AVX2 works. * k_quants: WIP super-blocks with 64 weights Q5_K scalar and AVX2 works, and with that all k_quants are done on AVX2 and scalar * k_quants: WIP super-blocks with 64 weights Q6_K working on CUDA. Cannot make it run quite as gast as with super-blocks with 256 weigths: 8% slower on 4080, 20% slower on the 1660 (but there we fit 1 less layer on the GPU because pf the larger model size), so some fraction of these 20% is due to that, * k_quants: WIP super-blocks with 64 weights Q4_K working on CUDA. ~10% slower on GTX-1660, 16% slower on 4080. * k_quants: WIP super-blocks with 64 weights Q2_K working on CUDA. ~3% slower on GTX-1660, 10% slower on 4080. * k_quants: WIP super-blocks with 64 weights Q3_K working on CUDA. * k_quants: WIP super-blocks with 64 weights Q5_K working on CUDA, and with this CUDA is done. * k_quants: WIP super-blocks with 64 weights Q6_K working on ARM_NEON * k_quants: WIP super-blocks with 64 weights Q4_K working on ARM_NEON, but quite a bit slower than 256 weights * k_quants: WIP super-blocks with 64 weights Q2_K working on ARM_NEON, but quite a bit slower than 256 weights * k_quants: WIP super-blocks with 64 weights Q3_K working on ARM_NEON, but quite a bit slower than 256 weights. * k_quants: WIP super-blocks with 64 weights Q5_K working on ARM_NEON, but quite a bit slower than 256 weights. With that, we have full support for ARM_NEON, although performance is not quite there. * k_quants: WIP super-blocks with 64 weights Slightly more efficient Q3_K and Q5_K * k_quants: WIP super-blocks with 64 weights Another small improvement for Q3_K and Q5_K on ARM_NEON * k_quants: WIP super-blocks with 64 weights Yet another speedup for Q5_K on ARM_NEON. We are now within 10% of the QK_K = 256 version. * k_quants: WIP super-blocks with 64 weights * We are able to pass preprocessor macros to the Metal compiler * Q6_K works and is actually slightly more efficient than the QK_K = 256 version (25.2 ms vs 25.8 ms) * k_quants: WIP super-blocks with 64 weights Q4_K works on Metal and is actually slightly faster than QK_K = 256 (21.95 ms vs 24.0 ms). * k_quants: WIP super-blocks with 64 weights Q2_K works on Metal and is very slightly faster than QK_K = 256 (23.8 ms vs 24.2 ms). * k_quants: WIP super-blocks with 64 weights Q3_K works on Metal and is slightly faster than QK_K = 256 (26.6 ms vs 28.3 ms). * k_quants: WIP super-blocks with 64 weights Q5_K works on Metal and is slightly faster than QK_K = 256 (23.7 ms vs 26.3 ms). * k_quants: call them _K, not _k, also on Metal * k_quants: correctly define QK_K in llama.cpp * Fixed bug in q4_K quantization added with the 64-block addition * Simplify via lambda * k_quants: swicth Q3_K to 4-bit scales when QK_K = 64 Otherwise there isn't much benefit from this quantization type. There is some very slight loss in accuracy, but we reduce size by ~7%. E.g., for OpenLLaMA-3B, Q3_K_S perplexity is 8.6131 with 8-bit scales and 8.6352 with 4-bit, while file size decreases from 1.53G to 1.44G. * k_quants: switch Q4_K to 4-bit scales when QK_K = 64 Here the loss in accuracy is greater than for Q3_K, but the Q4_K points still move further to the left on the perplexity vs size curve. * k_quants: forgot to add the Metal changes in last commit * k_quants: change Q5_K to be type 0 when QK_K = 64 Still needs AVX2 implementation * k_quants: AVX2 implementation for new 64-weight Q5_K * k_quants: 10% faster ARM_NEON Q5_K dot product * k_quants: fixed issue caused by merging with master --------- Co-authored-by: Iwan Kawrakow <iwan.kawrakow@gmail.com>
2023-06-26 16:43:07 +00:00
}
ggml : add SOTA 2,3,4,5,6 bit k-quantizations (#1684) * Starting to add k-quantization to ggml I think it is better to have quantization separate from ggml. For now just adding the k-quants there, but it would be better to also factor out the existing ggml quantizations. * Adding Q3_K and Q8_K (de)-quantization * Q3_K now working on CUDA and AVX2/scalar CUDA is not ideal - ~50% slower than Q4_0 for single token prediction, about the same in batch mode (perplexity). CPU single token is ~55 ms (on Ryzen 7950X). * Some improvement for Q3_K on CUDA It is now ~22.5 ms/token on my GPU, so ~30% slower than Q4_0. * Some more CUDA optimizations for Q3_K Single token is now 20.5 ms/token (~20% slower than Q4_0). Perplexity is on par with Q4_0. * Adding Q4_K - scalar, AVX2, CUDA Performance is the same or perhaps very slightly better than Q4_0 on the CPU. On the GPU, single token prediction is ~10% better than Q4_0, batch mode (perplexity is about the same). * Adding Q6_K - scalar, AVX2, CUDA Performance is ~40% lower compared to Q4_K on the CPU. This is to be expected, considering that we are memory bound on the CPU and the 6-bit model is ~44% larger than the 4-bit. On the GPU, single token prediction is ~6% lower than Q4_0, batch mode (perplexity) is even closer (but still slower). * Adding Q5_K - scalar, AVX2, CUDA Performance is ~20% lower compared to Q4_K on the CPU. This is to be expected, considering that we are memory bound on the CPU and the 5-bit model is ~22% larger than the 4-bit. On the GPU, single token prediction is about the same as Q4_0 for both, single token and batch prediction. * Per convention, all QX_K quantizations use Q5_K for output.weight * Adding quantization mixes * Quantization mixes: didn't quite get what I wanted in the last commit * Q4_K dot product for ARM_NEON * Q6_K dot product for ARM_NEON * Q5_K dot product for ARM_NEON * Adding Q3_K dot for ARM_NEON It is 22% slower than Q4_K, despite the smaller model size. On x86_64, where we are memory bound, the Q3_K model is quite a bit faster than Q4_K. * A very slightly faster ARM_NEON Q3_K dot * Adding Q2_K - just CUDA for now Token prediction is pretty good - about 15.5 ms on a RTX 4080. Perplexity is about the same as Q4_K. * Adding scalar and AVX2 Q2_K dot * Adding ARM_NEON Q2_K dot About the same performance as Q4_K. * A slightly faster ARM_NEON Q2_K dot Single token prediction is now ~36 ms on M2 Max. The code is much simpler too. * Fixed bug in Q2_K CUDA dot product kernel Stranegly enough, for the few prompts I tried with the 7B model the responses looked perfectly reasonable. Only realized something is not quite right when I tried the larger models and started getting nonse back. In any case, Q2_K single token evaluation time on an RTX 4080 in a Ryzen7950X box iusing CUDA and model fully loaded on the GPU are ~15.5 ms for 7B, ~25.4 ms for 13B, and ~55.8 ms for 30B. The max number of layers that fit in VRAM for The 65B is 32. With that, we get ~330 ms per token, which is not that much faster than just running on the CPU (~470 ms per token). * Don't print zeros/NaNs when no count histogram has been collected * A 10% faster CUDA vector dot kernel for Q3_K Q3_K is now running at ~18.5 ms / token on CUDA, so the gap to Q4_0 is only 10%. It seems memory acccess pattern is more important for performance than the amount of computation the kernel does. * A slightly daster Q4_K AVX2 dot product For perplexity, where we are less memory bound, time per pass drops by ~5%. Barely measurable difference for single token prediction. * A slightly faster ARM_NEON A4_K dot product * Minor * Fix quantization error test We cannot possibly be expecting rmse < 0.002 for 2- and 3-bit quantization variants. * Fix docker build I have been sloppy with vector reinterpret casts on ARM_NEON. It seems clang is very forgiving in that regard. * Added forgotten ggml.o dependence on k_quants.h to the Makefile * Had unintentionally committed the Makefile with -Ofast enabled * ggml : rename k_quants -> ggml-quants-k, use lowercase in code --------- Co-authored-by: Iwan Kawrakow <iwan.kawrakow@gmail.com> Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
2023-06-05 19:56:18 +00:00
ggml : move FP16 <-> FP32 code to ggml-impl.h (#3861) * ggml : move FP16 <-> FP32 stuff to ggml-impl.h ggml-ci * tests : fix ARM build * ggml : explicitly initialize deprecated type traits * ggml : add math.h to ggml-impl.h * ggml : remove duplicate static assert macros * ggml : prefix lookup tables with ggml_ ggml-ci * ggml-impl : move extern "C" to start of file
2023-10-30 17:19:15 +00:00
sumf += (GGML_FP16_TO_FP32(x[i].d)*GGML_FP16_TO_FP32(y[i].d)) * sumi;
ggml : add SOTA 2,3,4,5,6 bit k-quantizations (#1684) * Starting to add k-quantization to ggml I think it is better to have quantization separate from ggml. For now just adding the k-quants there, but it would be better to also factor out the existing ggml quantizations. * Adding Q3_K and Q8_K (de)-quantization * Q3_K now working on CUDA and AVX2/scalar CUDA is not ideal - ~50% slower than Q4_0 for single token prediction, about the same in batch mode (perplexity). CPU single token is ~55 ms (on Ryzen 7950X). * Some improvement for Q3_K on CUDA It is now ~22.5 ms/token on my GPU, so ~30% slower than Q4_0. * Some more CUDA optimizations for Q3_K Single token is now 20.5 ms/token (~20% slower than Q4_0). Perplexity is on par with Q4_0. * Adding Q4_K - scalar, AVX2, CUDA Performance is the same or perhaps very slightly better than Q4_0 on the CPU. On the GPU, single token prediction is ~10% better than Q4_0, batch mode (perplexity is about the same). * Adding Q6_K - scalar, AVX2, CUDA Performance is ~40% lower compared to Q4_K on the CPU. This is to be expected, considering that we are memory bound on the CPU and the 6-bit model is ~44% larger than the 4-bit. On the GPU, single token prediction is ~6% lower than Q4_0, batch mode (perplexity) is even closer (but still slower). * Adding Q5_K - scalar, AVX2, CUDA Performance is ~20% lower compared to Q4_K on the CPU. This is to be expected, considering that we are memory bound on the CPU and the 5-bit model is ~22% larger than the 4-bit. On the GPU, single token prediction is about the same as Q4_0 for both, single token and batch prediction. * Per convention, all QX_K quantizations use Q5_K for output.weight * Adding quantization mixes * Quantization mixes: didn't quite get what I wanted in the last commit * Q4_K dot product for ARM_NEON * Q6_K dot product for ARM_NEON * Q5_K dot product for ARM_NEON * Adding Q3_K dot for ARM_NEON It is 22% slower than Q4_K, despite the smaller model size. On x86_64, where we are memory bound, the Q3_K model is quite a bit faster than Q4_K. * A very slightly faster ARM_NEON Q3_K dot * Adding Q2_K - just CUDA for now Token prediction is pretty good - about 15.5 ms on a RTX 4080. Perplexity is about the same as Q4_K. * Adding scalar and AVX2 Q2_K dot * Adding ARM_NEON Q2_K dot About the same performance as Q4_K. * A slightly faster ARM_NEON Q2_K dot Single token prediction is now ~36 ms on M2 Max. The code is much simpler too. * Fixed bug in Q2_K CUDA dot product kernel Stranegly enough, for the few prompts I tried with the 7B model the responses looked perfectly reasonable. Only realized something is not quite right when I tried the larger models and started getting nonse back. In any case, Q2_K single token evaluation time on an RTX 4080 in a Ryzen7950X box iusing CUDA and model fully loaded on the GPU are ~15.5 ms for 7B, ~25.4 ms for 13B, and ~55.8 ms for 30B. The max number of layers that fit in VRAM for The 65B is 32. With that, we get ~330 ms per token, which is not that much faster than just running on the CPU (~470 ms per token). * Don't print zeros/NaNs when no count histogram has been collected * A 10% faster CUDA vector dot kernel for Q3_K Q3_K is now running at ~18.5 ms / token on CUDA, so the gap to Q4_0 is only 10%. It seems memory acccess pattern is more important for performance than the amount of computation the kernel does. * A slightly daster Q4_K AVX2 dot product For perplexity, where we are less memory bound, time per pass drops by ~5%. Barely measurable difference for single token prediction. * A slightly faster ARM_NEON A4_K dot product * Minor * Fix quantization error test We cannot possibly be expecting rmse < 0.002 for 2- and 3-bit quantization variants. * Fix docker build I have been sloppy with vector reinterpret casts on ARM_NEON. It seems clang is very forgiving in that regard. * Added forgotten ggml.o dependence on k_quants.h to the Makefile * Had unintentionally committed the Makefile with -Ofast enabled * ggml : rename k_quants -> ggml-quants-k, use lowercase in code --------- Co-authored-by: Iwan Kawrakow <iwan.kawrakow@gmail.com> Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
2023-06-05 19:56:18 +00:00
}
ggml : quantization refactoring (#3833) * ggml : factor all quantization code in ggml-quants ggml-ci * ggml-quants : fix Zig and Swift builds + quantize tool ggml-ci * quantize : --pure option for disabling k-quant mixtures --------- Co-authored-by: cebtenzzre <cebtenzzre@gmail.com>
2023-10-29 16:32:28 +00:00
*s = sumf;
#endif
ggml : add SOTA 2,3,4,5,6 bit k-quantizations (#1684) * Starting to add k-quantization to ggml I think it is better to have quantization separate from ggml. For now just adding the k-quants there, but it would be better to also factor out the existing ggml quantizations. * Adding Q3_K and Q8_K (de)-quantization * Q3_K now working on CUDA and AVX2/scalar CUDA is not ideal - ~50% slower than Q4_0 for single token prediction, about the same in batch mode (perplexity). CPU single token is ~55 ms (on Ryzen 7950X). * Some improvement for Q3_K on CUDA It is now ~22.5 ms/token on my GPU, so ~30% slower than Q4_0. * Some more CUDA optimizations for Q3_K Single token is now 20.5 ms/token (~20% slower than Q4_0). Perplexity is on par with Q4_0. * Adding Q4_K - scalar, AVX2, CUDA Performance is the same or perhaps very slightly better than Q4_0 on the CPU. On the GPU, single token prediction is ~10% better than Q4_0, batch mode (perplexity is about the same). * Adding Q6_K - scalar, AVX2, CUDA Performance is ~40% lower compared to Q4_K on the CPU. This is to be expected, considering that we are memory bound on the CPU and the 6-bit model is ~44% larger than the 4-bit. On the GPU, single token prediction is ~6% lower than Q4_0, batch mode (perplexity) is even closer (but still slower). * Adding Q5_K - scalar, AVX2, CUDA Performance is ~20% lower compared to Q4_K on the CPU. This is to be expected, considering that we are memory bound on the CPU and the 5-bit model is ~22% larger than the 4-bit. On the GPU, single token prediction is about the same as Q4_0 for both, single token and batch prediction. * Per convention, all QX_K quantizations use Q5_K for output.weight * Adding quantization mixes * Quantization mixes: didn't quite get what I wanted in the last commit * Q4_K dot product for ARM_NEON * Q6_K dot product for ARM_NEON * Q5_K dot product for ARM_NEON * Adding Q3_K dot for ARM_NEON It is 22% slower than Q4_K, despite the smaller model size. On x86_64, where we are memory bound, the Q3_K model is quite a bit faster than Q4_K. * A very slightly faster ARM_NEON Q3_K dot * Adding Q2_K - just CUDA for now Token prediction is pretty good - about 15.5 ms on a RTX 4080. Perplexity is about the same as Q4_K. * Adding scalar and AVX2 Q2_K dot * Adding ARM_NEON Q2_K dot About the same performance as Q4_K. * A slightly faster ARM_NEON Q2_K dot Single token prediction is now ~36 ms on M2 Max. The code is much simpler too. * Fixed bug in Q2_K CUDA dot product kernel Stranegly enough, for the few prompts I tried with the 7B model the responses looked perfectly reasonable. Only realized something is not quite right when I tried the larger models and started getting nonse back. In any case, Q2_K single token evaluation time on an RTX 4080 in a Ryzen7950X box iusing CUDA and model fully loaded on the GPU are ~15.5 ms for 7B, ~25.4 ms for 13B, and ~55.8 ms for 30B. The max number of layers that fit in VRAM for The 65B is 32. With that, we get ~330 ms per token, which is not that much faster than just running on the CPU (~470 ms per token). * Don't print zeros/NaNs when no count histogram has been collected * A 10% faster CUDA vector dot kernel for Q3_K Q3_K is now running at ~18.5 ms / token on CUDA, so the gap to Q4_0 is only 10%. It seems memory acccess pattern is more important for performance than the amount of computation the kernel does. * A slightly daster Q4_K AVX2 dot product For perplexity, where we are less memory bound, time per pass drops by ~5%. Barely measurable difference for single token prediction. * A slightly faster ARM_NEON A4_K dot product * Minor * Fix quantization error test We cannot possibly be expecting rmse < 0.002 for 2- and 3-bit quantization variants. * Fix docker build I have been sloppy with vector reinterpret casts on ARM_NEON. It seems clang is very forgiving in that regard. * Added forgotten ggml.o dependence on k_quants.h to the Makefile * Had unintentionally committed the Makefile with -Ofast enabled * ggml : rename k_quants -> ggml-quants-k, use lowercase in code --------- Co-authored-by: Iwan Kawrakow <iwan.kawrakow@gmail.com> Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
2023-06-05 19:56:18 +00:00
}
ggml : quantization refactoring (#3833) * ggml : factor all quantization code in ggml-quants ggml-ci * ggml-quants : fix Zig and Swift builds + quantize tool ggml-ci * quantize : --pure option for disabling k-quant mixtures --------- Co-authored-by: cebtenzzre <cebtenzzre@gmail.com>
2023-10-29 16:32:28 +00:00
void ggml_vec_dot_q5_1_q8_1(const int n, float * restrict s, const void * restrict vx, const void * restrict vy) {
const int qk = QK8_1;
const int nb = n / qk;
k-quants : fix quantization ranges (#3646)
2023-10-17 06:19:28 +00:00
ggml : quantization refactoring (#3833) * ggml : factor all quantization code in ggml-quants ggml-ci * ggml-quants : fix Zig and Swift builds + quantize tool ggml-ci * quantize : --pure option for disabling k-quant mixtures --------- Co-authored-by: cebtenzzre <cebtenzzre@gmail.com>
2023-10-29 16:32:28 +00:00
assert(n % qk == 0);
assert(qk == QK5_1);
ggml : add SOTA 2,3,4,5,6 bit k-quantizations (#1684) * Starting to add k-quantization to ggml I think it is better to have quantization separate from ggml. For now just adding the k-quants there, but it would be better to also factor out the existing ggml quantizations. * Adding Q3_K and Q8_K (de)-quantization * Q3_K now working on CUDA and AVX2/scalar CUDA is not ideal - ~50% slower than Q4_0 for single token prediction, about the same in batch mode (perplexity). CPU single token is ~55 ms (on Ryzen 7950X). * Some improvement for Q3_K on CUDA It is now ~22.5 ms/token on my GPU, so ~30% slower than Q4_0. * Some more CUDA optimizations for Q3_K Single token is now 20.5 ms/token (~20% slower than Q4_0). Perplexity is on par with Q4_0. * Adding Q4_K - scalar, AVX2, CUDA Performance is the same or perhaps very slightly better than Q4_0 on the CPU. On the GPU, single token prediction is ~10% better than Q4_0, batch mode (perplexity is about the same). * Adding Q6_K - scalar, AVX2, CUDA Performance is ~40% lower compared to Q4_K on the CPU. This is to be expected, considering that we are memory bound on the CPU and the 6-bit model is ~44% larger than the 4-bit. On the GPU, single token prediction is ~6% lower than Q4_0, batch mode (perplexity) is even closer (but still slower). * Adding Q5_K - scalar, AVX2, CUDA Performance is ~20% lower compared to Q4_K on the CPU. This is to be expected, considering that we are memory bound on the CPU and the 5-bit model is ~22% larger than the 4-bit. On the GPU, single token prediction is about the same as Q4_0 for both, single token and batch prediction. * Per convention, all QX_K quantizations use Q5_K for output.weight * Adding quantization mixes * Quantization mixes: didn't quite get what I wanted in the last commit * Q4_K dot product for ARM_NEON * Q6_K dot product for ARM_NEON * Q5_K dot product for ARM_NEON * Adding Q3_K dot for ARM_NEON It is 22% slower than Q4_K, despite the smaller model size. On x86_64, where we are memory bound, the Q3_K model is quite a bit faster than Q4_K. * A very slightly faster ARM_NEON Q3_K dot * Adding Q2_K - just CUDA for now Token prediction is pretty good - about 15.5 ms on a RTX 4080. Perplexity is about the same as Q4_K. * Adding scalar and AVX2 Q2_K dot * Adding ARM_NEON Q2_K dot About the same performance as Q4_K. * A slightly faster ARM_NEON Q2_K dot Single token prediction is now ~36 ms on M2 Max. The code is much simpler too. * Fixed bug in Q2_K CUDA dot product kernel Stranegly enough, for the few prompts I tried with the 7B model the responses looked perfectly reasonable. Only realized something is not quite right when I tried the larger models and started getting nonse back. In any case, Q2_K single token evaluation time on an RTX 4080 in a Ryzen7950X box iusing CUDA and model fully loaded on the GPU are ~15.5 ms for 7B, ~25.4 ms for 13B, and ~55.8 ms for 30B. The max number of layers that fit in VRAM for The 65B is 32. With that, we get ~330 ms per token, which is not that much faster than just running on the CPU (~470 ms per token). * Don't print zeros/NaNs when no count histogram has been collected * A 10% faster CUDA vector dot kernel for Q3_K Q3_K is now running at ~18.5 ms / token on CUDA, so the gap to Q4_0 is only 10%. It seems memory acccess pattern is more important for performance than the amount of computation the kernel does. * A slightly daster Q4_K AVX2 dot product For perplexity, where we are less memory bound, time per pass drops by ~5%. Barely measurable difference for single token prediction. * A slightly faster ARM_NEON A4_K dot product * Minor * Fix quantization error test We cannot possibly be expecting rmse < 0.002 for 2- and 3-bit quantization variants. * Fix docker build I have been sloppy with vector reinterpret casts on ARM_NEON. It seems clang is very forgiving in that regard. * Added forgotten ggml.o dependence on k_quants.h to the Makefile * Had unintentionally committed the Makefile with -Ofast enabled * ggml : rename k_quants -> ggml-quants-k, use lowercase in code --------- Co-authored-by: Iwan Kawrakow <iwan.kawrakow@gmail.com> Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
2023-06-05 19:56:18 +00:00
ggml : quantization refactoring (#3833) * ggml : factor all quantization code in ggml-quants ggml-ci * ggml-quants : fix Zig and Swift builds + quantize tool ggml-ci * quantize : --pure option for disabling k-quant mixtures --------- Co-authored-by: cebtenzzre <cebtenzzre@gmail.com>
2023-10-29 16:32:28 +00:00
const block_q5_1 * restrict x = vx;
const block_q8_1 * restrict y = vy;
ggml : add SOTA 2,3,4,5,6 bit k-quantizations (#1684) * Starting to add k-quantization to ggml I think it is better to have quantization separate from ggml. For now just adding the k-quants there, but it would be better to also factor out the existing ggml quantizations. * Adding Q3_K and Q8_K (de)-quantization * Q3_K now working on CUDA and AVX2/scalar CUDA is not ideal - ~50% slower than Q4_0 for single token prediction, about the same in batch mode (perplexity). CPU single token is ~55 ms (on Ryzen 7950X). * Some improvement for Q3_K on CUDA It is now ~22.5 ms/token on my GPU, so ~30% slower than Q4_0. * Some more CUDA optimizations for Q3_K Single token is now 20.5 ms/token (~20% slower than Q4_0). Perplexity is on par with Q4_0. * Adding Q4_K - scalar, AVX2, CUDA Performance is the same or perhaps very slightly better than Q4_0 on the CPU. On the GPU, single token prediction is ~10% better than Q4_0, batch mode (perplexity is about the same). * Adding Q6_K - scalar, AVX2, CUDA Performance is ~40% lower compared to Q4_K on the CPU. This is to be expected, considering that we are memory bound on the CPU and the 6-bit model is ~44% larger than the 4-bit. On the GPU, single token prediction is ~6% lower than Q4_0, batch mode (perplexity) is even closer (but still slower). * Adding Q5_K - scalar, AVX2, CUDA Performance is ~20% lower compared to Q4_K on the CPU. This is to be expected, considering that we are memory bound on the CPU and the 5-bit model is ~22% larger than the 4-bit. On the GPU, single token prediction is about the same as Q4_0 for both, single token and batch prediction. * Per convention, all QX_K quantizations use Q5_K for output.weight * Adding quantization mixes * Quantization mixes: didn't quite get what I wanted in the last commit * Q4_K dot product for ARM_NEON * Q6_K dot product for ARM_NEON * Q5_K dot product for ARM_NEON * Adding Q3_K dot for ARM_NEON It is 22% slower than Q4_K, despite the smaller model size. On x86_64, where we are memory bound, the Q3_K model is quite a bit faster than Q4_K. * A very slightly faster ARM_NEON Q3_K dot * Adding Q2_K - just CUDA for now Token prediction is pretty good - about 15.5 ms on a RTX 4080. Perplexity is about the same as Q4_K. * Adding scalar and AVX2 Q2_K dot * Adding ARM_NEON Q2_K dot About the same performance as Q4_K. * A slightly faster ARM_NEON Q2_K dot Single token prediction is now ~36 ms on M2 Max. The code is much simpler too. * Fixed bug in Q2_K CUDA dot product kernel Stranegly enough, for the few prompts I tried with the 7B model the responses looked perfectly reasonable. Only realized something is not quite right when I tried the larger models and started getting nonse back. In any case, Q2_K single token evaluation time on an RTX 4080 in a Ryzen7950X box iusing CUDA and model fully loaded on the GPU are ~15.5 ms for 7B, ~25.4 ms for 13B, and ~55.8 ms for 30B. The max number of layers that fit in VRAM for The 65B is 32. With that, we get ~330 ms per token, which is not that much faster than just running on the CPU (~470 ms per token). * Don't print zeros/NaNs when no count histogram has been collected * A 10% faster CUDA vector dot kernel for Q3_K Q3_K is now running at ~18.5 ms / token on CUDA, so the gap to Q4_0 is only 10%. It seems memory acccess pattern is more important for performance than the amount of computation the kernel does. * A slightly daster Q4_K AVX2 dot product For perplexity, where we are less memory bound, time per pass drops by ~5%. Barely measurable difference for single token prediction. * A slightly faster ARM_NEON A4_K dot product * Minor * Fix quantization error test We cannot possibly be expecting rmse < 0.002 for 2- and 3-bit quantization variants. * Fix docker build I have been sloppy with vector reinterpret casts on ARM_NEON. It seems clang is very forgiving in that regard. * Added forgotten ggml.o dependence on k_quants.h to the Makefile * Had unintentionally committed the Makefile with -Ofast enabled * ggml : rename k_quants -> ggml-quants-k, use lowercase in code --------- Co-authored-by: Iwan Kawrakow <iwan.kawrakow@gmail.com> Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
2023-06-05 19:56:18 +00:00
ggml : quantization refactoring (#3833) * ggml : factor all quantization code in ggml-quants ggml-ci * ggml-quants : fix Zig and Swift builds + quantize tool ggml-ci * quantize : --pure option for disabling k-quant mixtures --------- Co-authored-by: cebtenzzre <cebtenzzre@gmail.com>
2023-10-29 16:32:28 +00:00
#if defined(__ARM_NEON)
float32x4_t sumv0 = vdupq_n_f32(0.0f);
float32x4_t sumv1 = vdupq_n_f32(0.0f);
ggml : add SOTA 2,3,4,5,6 bit k-quantizations (#1684) * Starting to add k-quantization to ggml I think it is better to have quantization separate from ggml. For now just adding the k-quants there, but it would be better to also factor out the existing ggml quantizations. * Adding Q3_K and Q8_K (de)-quantization * Q3_K now working on CUDA and AVX2/scalar CUDA is not ideal - ~50% slower than Q4_0 for single token prediction, about the same in batch mode (perplexity). CPU single token is ~55 ms (on Ryzen 7950X). * Some improvement for Q3_K on CUDA It is now ~22.5 ms/token on my GPU, so ~30% slower than Q4_0. * Some more CUDA optimizations for Q3_K Single token is now 20.5 ms/token (~20% slower than Q4_0). Perplexity is on par with Q4_0. * Adding Q4_K - scalar, AVX2, CUDA Performance is the same or perhaps very slightly better than Q4_0 on the CPU. On the GPU, single token prediction is ~10% better than Q4_0, batch mode (perplexity is about the same). * Adding Q6_K - scalar, AVX2, CUDA Performance is ~40% lower compared to Q4_K on the CPU. This is to be expected, considering that we are memory bound on the CPU and the 6-bit model is ~44% larger than the 4-bit. On the GPU, single token prediction is ~6% lower than Q4_0, batch mode (perplexity) is even closer (but still slower). * Adding Q5_K - scalar, AVX2, CUDA Performance is ~20% lower compared to Q4_K on the CPU. This is to be expected, considering that we are memory bound on the CPU and the 5-bit model is ~22% larger than the 4-bit. On the GPU, single token prediction is about the same as Q4_0 for both, single token and batch prediction. * Per convention, all QX_K quantizations use Q5_K for output.weight * Adding quantization mixes * Quantization mixes: didn't quite get what I wanted in the last commit * Q4_K dot product for ARM_NEON * Q6_K dot product for ARM_NEON * Q5_K dot product for ARM_NEON * Adding Q3_K dot for ARM_NEON It is 22% slower than Q4_K, despite the smaller model size. On x86_64, where we are memory bound, the Q3_K model is quite a bit faster than Q4_K. * A very slightly faster ARM_NEON Q3_K dot * Adding Q2_K - just CUDA for now Token prediction is pretty good - about 15.5 ms on a RTX 4080. Perplexity is about the same as Q4_K. * Adding scalar and AVX2 Q2_K dot * Adding ARM_NEON Q2_K dot About the same performance as Q4_K. * A slightly faster ARM_NEON Q2_K dot Single token prediction is now ~36 ms on M2 Max. The code is much simpler too. * Fixed bug in Q2_K CUDA dot product kernel Stranegly enough, for the few prompts I tried with the 7B model the responses looked perfectly reasonable. Only realized something is not quite right when I tried the larger models and started getting nonse back. In any case, Q2_K single token evaluation time on an RTX 4080 in a Ryzen7950X box iusing CUDA and model fully loaded on the GPU are ~15.5 ms for 7B, ~25.4 ms for 13B, and ~55.8 ms for 30B. The max number of layers that fit in VRAM for The 65B is 32. With that, we get ~330 ms per token, which is not that much faster than just running on the CPU (~470 ms per token). * Don't print zeros/NaNs when no count histogram has been collected * A 10% faster CUDA vector dot kernel for Q3_K Q3_K is now running at ~18.5 ms / token on CUDA, so the gap to Q4_0 is only 10%. It seems memory acccess pattern is more important for performance than the amount of computation the kernel does. * A slightly daster Q4_K AVX2 dot product For perplexity, where we are less memory bound, time per pass drops by ~5%. Barely measurable difference for single token prediction. * A slightly faster ARM_NEON A4_K dot product * Minor * Fix quantization error test We cannot possibly be expecting rmse < 0.002 for 2- and 3-bit quantization variants. * Fix docker build I have been sloppy with vector reinterpret casts on ARM_NEON. It seems clang is very forgiving in that regard. * Added forgotten ggml.o dependence on k_quants.h to the Makefile * Had unintentionally committed the Makefile with -Ofast enabled * ggml : rename k_quants -> ggml-quants-k, use lowercase in code --------- Co-authored-by: Iwan Kawrakow <iwan.kawrakow@gmail.com> Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
2023-06-05 19:56:18 +00:00
ggml : quantization refactoring (#3833) * ggml : factor all quantization code in ggml-quants ggml-ci * ggml-quants : fix Zig and Swift builds + quantize tool ggml-ci * quantize : --pure option for disabling k-quant mixtures --------- Co-authored-by: cebtenzzre <cebtenzzre@gmail.com>
2023-10-29 16:32:28 +00:00
float summs0 = 0.0f;
float summs1 = 0.0f;
uint32_t qh0;
uint32_t qh1;
uint64_t tmp0[4];
uint64_t tmp1[4];
assert(nb % 2 == 0); // TODO: handle odd nb
for (int i = 0; i < nb; i += 2) {
const block_q5_1 * restrict x0 = &x[i];
const block_q5_1 * restrict x1 = &x[i + 1];
const block_q8_1 * restrict y0 = &y[i];
const block_q8_1 * restrict y1 = &y[i + 1];
const uint8x16_t m4b = vdupq_n_u8(0x0F);
ggml : move FP16 <-> FP32 code to ggml-impl.h (#3861) * ggml : move FP16 <-> FP32 stuff to ggml-impl.h ggml-ci * tests : fix ARM build * ggml : explicitly initialize deprecated type traits * ggml : add math.h to ggml-impl.h * ggml : remove duplicate static assert macros * ggml : prefix lookup tables with ggml_ ggml-ci * ggml-impl : move extern "C" to start of file
2023-10-30 17:19:15 +00:00
summs0 += GGML_FP16_TO_FP32(x0->m) * y0->s;
summs1 += GGML_FP16_TO_FP32(x1->m) * y1->s;
ggml : quantization refactoring (#3833) * ggml : factor all quantization code in ggml-quants ggml-ci * ggml-quants : fix Zig and Swift builds + quantize tool ggml-ci * quantize : --pure option for disabling k-quant mixtures --------- Co-authored-by: cebtenzzre <cebtenzzre@gmail.com>
2023-10-29 16:32:28 +00:00
// extract the 5th bit via lookup table ((b) << 4)
memcpy(&qh0, x0->qh, sizeof(qh0));
memcpy(&qh1, x1->qh, sizeof(qh1));
tmp0[0] = table_b2b_0[(qh0 >> 0) & 0xFF];
tmp0[1] = table_b2b_0[(qh0 >> 8) & 0xFF];
tmp0[2] = table_b2b_0[(qh0 >> 16) & 0xFF];
tmp0[3] = table_b2b_0[(qh0 >> 24) ];
tmp1[0] = table_b2b_0[(qh1 >> 0) & 0xFF];
tmp1[1] = table_b2b_0[(qh1 >> 8) & 0xFF];
tmp1[2] = table_b2b_0[(qh1 >> 16) & 0xFF];
tmp1[3] = table_b2b_0[(qh1 >> 24) ];
const int8x16_t qhl0 = vld1q_s8((const int8_t *)(tmp0 + 0));
const int8x16_t qhh0 = vld1q_s8((const int8_t *)(tmp0 + 2));
const int8x16_t qhl1 = vld1q_s8((const int8_t *)(tmp1 + 0));
const int8x16_t qhh1 = vld1q_s8((const int8_t *)(tmp1 + 2));
const uint8x16_t v0_0 = vld1q_u8(x0->qs);
const uint8x16_t v0_1 = vld1q_u8(x1->qs);
// 4-bit -> 8-bit
const int8x16_t v0_0l = vreinterpretq_s8_u8(vandq_u8 (v0_0, m4b));
const int8x16_t v0_0h = vreinterpretq_s8_u8(vshrq_n_u8(v0_0, 4));
const int8x16_t v0_1l = vreinterpretq_s8_u8(vandq_u8 (v0_1, m4b));
const int8x16_t v0_1h = vreinterpretq_s8_u8(vshrq_n_u8(v0_1, 4));
// add high bit
const int8x16_t v0_0lf = vorrq_s8(v0_0l, qhl0);
const int8x16_t v0_0hf = vorrq_s8(v0_0h, qhh0);
const int8x16_t v0_1lf = vorrq_s8(v0_1l, qhl1);
const int8x16_t v0_1hf = vorrq_s8(v0_1h, qhh1);
// load y
const int8x16_t v1_0l = vld1q_s8(y0->qs);
const int8x16_t v1_0h = vld1q_s8(y0->qs + 16);
const int8x16_t v1_1l = vld1q_s8(y1->qs);
const int8x16_t v1_1h = vld1q_s8(y1->qs + 16);
sumv0 = vmlaq_n_f32(sumv0, vcvtq_f32_s32(vaddq_s32(
ggml : add ggml_vdotq_s32 alias (#4715) ggml-ci
2023-12-31 09:43:31 +00:00
ggml_vdotq_s32(vdupq_n_s32(0), v0_0lf, v1_0l),
ggml_vdotq_s32(vdupq_n_s32(0), v0_0hf, v1_0h))), GGML_FP16_TO_FP32(x0->d)*y0->d);
ggml : quantization refactoring (#3833) * ggml : factor all quantization code in ggml-quants ggml-ci * ggml-quants : fix Zig and Swift builds + quantize tool ggml-ci * quantize : --pure option for disabling k-quant mixtures --------- Co-authored-by: cebtenzzre <cebtenzzre@gmail.com>
2023-10-29 16:32:28 +00:00
sumv1 = vmlaq_n_f32(sumv1, vcvtq_f32_s32(vaddq_s32(
ggml : add ggml_vdotq_s32 alias (#4715) ggml-ci
2023-12-31 09:43:31 +00:00
ggml_vdotq_s32(vdupq_n_s32(0), v0_1lf, v1_1l),
ggml_vdotq_s32(vdupq_n_s32(0), v0_1hf, v1_1h))), GGML_FP16_TO_FP32(x1->d)*y1->d);
ggml : quantization refactoring (#3833) * ggml : factor all quantization code in ggml-quants ggml-ci * ggml-quants : fix Zig and Swift builds + quantize tool ggml-ci * quantize : --pure option for disabling k-quant mixtures --------- Co-authored-by: cebtenzzre <cebtenzzre@gmail.com>
2023-10-29 16:32:28 +00:00
}
ggml : add SOTA 2,3,4,5,6 bit k-quantizations (#1684) * Starting to add k-quantization to ggml I think it is better to have quantization separate from ggml. For now just adding the k-quants there, but it would be better to also factor out the existing ggml quantizations. * Adding Q3_K and Q8_K (de)-quantization * Q3_K now working on CUDA and AVX2/scalar CUDA is not ideal - ~50% slower than Q4_0 for single token prediction, about the same in batch mode (perplexity). CPU single token is ~55 ms (on Ryzen 7950X). * Some improvement for Q3_K on CUDA It is now ~22.5 ms/token on my GPU, so ~30% slower than Q4_0. * Some more CUDA optimizations for Q3_K Single token is now 20.5 ms/token (~20% slower than Q4_0). Perplexity is on par with Q4_0. * Adding Q4_K - scalar, AVX2, CUDA Performance is the same or perhaps very slightly better than Q4_0 on the CPU. On the GPU, single token prediction is ~10% better than Q4_0, batch mode (perplexity is about the same). * Adding Q6_K - scalar, AVX2, CUDA Performance is ~40% lower compared to Q4_K on the CPU. This is to be expected, considering that we are memory bound on the CPU and the 6-bit model is ~44% larger than the 4-bit. On the GPU, single token prediction is ~6% lower than Q4_0, batch mode (perplexity) is even closer (but still slower). * Adding Q5_K - scalar, AVX2, CUDA Performance is ~20% lower compared to Q4_K on the CPU. This is to be expected, considering that we are memory bound on the CPU and the 5-bit model is ~22% larger than the 4-bit. On the GPU, single token prediction is about the same as Q4_0 for both, single token and batch prediction. * Per convention, all QX_K quantizations use Q5_K for output.weight * Adding quantization mixes * Quantization mixes: didn't quite get what I wanted in the last commit * Q4_K dot product for ARM_NEON * Q6_K dot product for ARM_NEON * Q5_K dot product for ARM_NEON * Adding Q3_K dot for ARM_NEON It is 22% slower than Q4_K, despite the smaller model size. On x86_64, where we are memory bound, the Q3_K model is quite a bit faster than Q4_K. * A very slightly faster ARM_NEON Q3_K dot * Adding Q2_K - just CUDA for now Token prediction is pretty good - about 15.5 ms on a RTX 4080. Perplexity is about the same as Q4_K. * Adding scalar and AVX2 Q2_K dot * Adding ARM_NEON Q2_K dot About the same performance as Q4_K. * A slightly faster ARM_NEON Q2_K dot Single token prediction is now ~36 ms on M2 Max. The code is much simpler too. * Fixed bug in Q2_K CUDA dot product kernel Stranegly enough, for the few prompts I tried with the 7B model the responses looked perfectly reasonable. Only realized something is not quite right when I tried the larger models and started getting nonse back. In any case, Q2_K single token evaluation time on an RTX 4080 in a Ryzen7950X box iusing CUDA and model fully loaded on the GPU are ~15.5 ms for 7B, ~25.4 ms for 13B, and ~55.8 ms for 30B. The max number of layers that fit in VRAM for The 65B is 32. With that, we get ~330 ms per token, which is not that much faster than just running on the CPU (~470 ms per token). * Don't print zeros/NaNs when no count histogram has been collected * A 10% faster CUDA vector dot kernel for Q3_K Q3_K is now running at ~18.5 ms / token on CUDA, so the gap to Q4_0 is only 10%. It seems memory acccess pattern is more important for performance than the amount of computation the kernel does. * A slightly daster Q4_K AVX2 dot product For perplexity, where we are less memory bound, time per pass drops by ~5%. Barely measurable difference for single token prediction. * A slightly faster ARM_NEON A4_K dot product * Minor * Fix quantization error test We cannot possibly be expecting rmse < 0.002 for 2- and 3-bit quantization variants. * Fix docker build I have been sloppy with vector reinterpret casts on ARM_NEON. It seems clang is very forgiving in that regard. * Added forgotten ggml.o dependence on k_quants.h to the Makefile * Had unintentionally committed the Makefile with -Ofast enabled * ggml : rename k_quants -> ggml-quants-k, use lowercase in code --------- Co-authored-by: Iwan Kawrakow <iwan.kawrakow@gmail.com> Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
2023-06-05 19:56:18 +00:00
ggml : quantization refactoring (#3833) * ggml : factor all quantization code in ggml-quants ggml-ci * ggml-quants : fix Zig and Swift builds + quantize tool ggml-ci * quantize : --pure option for disabling k-quant mixtures --------- Co-authored-by: cebtenzzre <cebtenzzre@gmail.com>
2023-10-29 16:32:28 +00:00
*s = vaddvq_f32(sumv0) + vaddvq_f32(sumv1) + summs0 + summs1;
#elif defined(__wasm_simd128__)
v128_t sumv = wasm_f32x4_splat(0.0f);
ggml : add SOTA 2,3,4,5,6 bit k-quantizations (#1684) * Starting to add k-quantization to ggml I think it is better to have quantization separate from ggml. For now just adding the k-quants there, but it would be better to also factor out the existing ggml quantizations. * Adding Q3_K and Q8_K (de)-quantization * Q3_K now working on CUDA and AVX2/scalar CUDA is not ideal - ~50% slower than Q4_0 for single token prediction, about the same in batch mode (perplexity). CPU single token is ~55 ms (on Ryzen 7950X). * Some improvement for Q3_K on CUDA It is now ~22.5 ms/token on my GPU, so ~30% slower than Q4_0. * Some more CUDA optimizations for Q3_K Single token is now 20.5 ms/token (~20% slower than Q4_0). Perplexity is on par with Q4_0. * Adding Q4_K - scalar, AVX2, CUDA Performance is the same or perhaps very slightly better than Q4_0 on the CPU. On the GPU, single token prediction is ~10% better than Q4_0, batch mode (perplexity is about the same). * Adding Q6_K - scalar, AVX2, CUDA Performance is ~40% lower compared to Q4_K on the CPU. This is to be expected, considering that we are memory bound on the CPU and the 6-bit model is ~44% larger than the 4-bit. On the GPU, single token prediction is ~6% lower than Q4_0, batch mode (perplexity) is even closer (but still slower). * Adding Q5_K - scalar, AVX2, CUDA Performance is ~20% lower compared to Q4_K on the CPU. This is to be expected, considering that we are memory bound on the CPU and the 5-bit model is ~22% larger than the 4-bit. On the GPU, single token prediction is about the same as Q4_0 for both, single token and batch prediction. * Per convention, all QX_K quantizations use Q5_K for output.weight * Adding quantization mixes * Quantization mixes: didn't quite get what I wanted in the last commit * Q4_K dot product for ARM_NEON * Q6_K dot product for ARM_NEON * Q5_K dot product for ARM_NEON * Adding Q3_K dot for ARM_NEON It is 22% slower than Q4_K, despite the smaller model size. On x86_64, where we are memory bound, the Q3_K model is quite a bit faster than Q4_K. * A very slightly faster ARM_NEON Q3_K dot * Adding Q2_K - just CUDA for now Token prediction is pretty good - about 15.5 ms on a RTX 4080. Perplexity is about the same as Q4_K. * Adding scalar and AVX2 Q2_K dot * Adding ARM_NEON Q2_K dot About the same performance as Q4_K. * A slightly faster ARM_NEON Q2_K dot Single token prediction is now ~36 ms on M2 Max. The code is much simpler too. * Fixed bug in Q2_K CUDA dot product kernel Stranegly enough, for the few prompts I tried with the 7B model the responses looked perfectly reasonable. Only realized something is not quite right when I tried the larger models and started getting nonse back. In any case, Q2_K single token evaluation time on an RTX 4080 in a Ryzen7950X box iusing CUDA and model fully loaded on the GPU are ~15.5 ms for 7B, ~25.4 ms for 13B, and ~55.8 ms for 30B. The max number of layers that fit in VRAM for The 65B is 32. With that, we get ~330 ms per token, which is not that much faster than just running on the CPU (~470 ms per token). * Don't print zeros/NaNs when no count histogram has been collected * A 10% faster CUDA vector dot kernel for Q3_K Q3_K is now running at ~18.5 ms / token on CUDA, so the gap to Q4_0 is only 10%. It seems memory acccess pattern is more important for performance than the amount of computation the kernel does. * A slightly daster Q4_K AVX2 dot product For perplexity, where we are less memory bound, time per pass drops by ~5%. Barely measurable difference for single token prediction. * A slightly faster ARM_NEON A4_K dot product * Minor * Fix quantization error test We cannot possibly be expecting rmse < 0.002 for 2- and 3-bit quantization variants. * Fix docker build I have been sloppy with vector reinterpret casts on ARM_NEON. It seems clang is very forgiving in that regard. * Added forgotten ggml.o dependence on k_quants.h to the Makefile * Had unintentionally committed the Makefile with -Ofast enabled * ggml : rename k_quants -> ggml-quants-k, use lowercase in code --------- Co-authored-by: Iwan Kawrakow <iwan.kawrakow@gmail.com> Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
2023-06-05 19:56:18 +00:00
ggml : quantization refactoring (#3833) * ggml : factor all quantization code in ggml-quants ggml-ci * ggml-quants : fix Zig and Swift builds + quantize tool ggml-ci * quantize : --pure option for disabling k-quant mixtures --------- Co-authored-by: cebtenzzre <cebtenzzre@gmail.com>
2023-10-29 16:32:28 +00:00
float summs = 0.0f;
k-quants : support for super-block size of 64 (#2001) * k_quants: WIP super-blocks with 64 weights * k_quants: WIP super-blocks with 64 weights Q6_K scalar and AVX2 works * k_quants: WIP super-blocks with 64 weights Q4_K scalar and AVX2 works * k_quants: WIP super-blocks with 64 weights Q2_K scalar and AVX2 works. Q2_K is way too slow (it is actually slower than the scalar implementation) * k_quants: WIP super-blocks with 64 weights Q3_K scalar and AVX2 works. * k_quants: WIP super-blocks with 64 weights Q5_K scalar and AVX2 works, and with that all k_quants are done on AVX2 and scalar * k_quants: WIP super-blocks with 64 weights Q6_K working on CUDA. Cannot make it run quite as gast as with super-blocks with 256 weigths: 8% slower on 4080, 20% slower on the 1660 (but there we fit 1 less layer on the GPU because pf the larger model size), so some fraction of these 20% is due to that, * k_quants: WIP super-blocks with 64 weights Q4_K working on CUDA. ~10% slower on GTX-1660, 16% slower on 4080. * k_quants: WIP super-blocks with 64 weights Q2_K working on CUDA. ~3% slower on GTX-1660, 10% slower on 4080. * k_quants: WIP super-blocks with 64 weights Q3_K working on CUDA. * k_quants: WIP super-blocks with 64 weights Q5_K working on CUDA, and with this CUDA is done. * k_quants: WIP super-blocks with 64 weights Q6_K working on ARM_NEON * k_quants: WIP super-blocks with 64 weights Q4_K working on ARM_NEON, but quite a bit slower than 256 weights * k_quants: WIP super-blocks with 64 weights Q2_K working on ARM_NEON, but quite a bit slower than 256 weights * k_quants: WIP super-blocks with 64 weights Q3_K working on ARM_NEON, but quite a bit slower than 256 weights. * k_quants: WIP super-blocks with 64 weights Q5_K working on ARM_NEON, but quite a bit slower than 256 weights. With that, we have full support for ARM_NEON, although performance is not quite there. * k_quants: WIP super-blocks with 64 weights Slightly more efficient Q3_K and Q5_K * k_quants: WIP super-blocks with 64 weights Another small improvement for Q3_K and Q5_K on ARM_NEON * k_quants: WIP super-blocks with 64 weights Yet another speedup for Q5_K on ARM_NEON. We are now within 10% of the QK_K = 256 version. * k_quants: WIP super-blocks with 64 weights * We are able to pass preprocessor macros to the Metal compiler * Q6_K works and is actually slightly more efficient than the QK_K = 256 version (25.2 ms vs 25.8 ms) * k_quants: WIP super-blocks with 64 weights Q4_K works on Metal and is actually slightly faster than QK_K = 256 (21.95 ms vs 24.0 ms). * k_quants: WIP super-blocks with 64 weights Q2_K works on Metal and is very slightly faster than QK_K = 256 (23.8 ms vs 24.2 ms). * k_quants: WIP super-blocks with 64 weights Q3_K works on Metal and is slightly faster than QK_K = 256 (26.6 ms vs 28.3 ms). * k_quants: WIP super-blocks with 64 weights Q5_K works on Metal and is slightly faster than QK_K = 256 (23.7 ms vs 26.3 ms). * k_quants: call them _K, not _k, also on Metal * k_quants: correctly define QK_K in llama.cpp * Fixed bug in q4_K quantization added with the 64-block addition * Simplify via lambda * k_quants: swicth Q3_K to 4-bit scales when QK_K = 64 Otherwise there isn't much benefit from this quantization type. There is some very slight loss in accuracy, but we reduce size by ~7%. E.g., for OpenLLaMA-3B, Q3_K_S perplexity is 8.6131 with 8-bit scales and 8.6352 with 4-bit, while file size decreases from 1.53G to 1.44G. * k_quants: switch Q4_K to 4-bit scales when QK_K = 64 Here the loss in accuracy is greater than for Q3_K, but the Q4_K points still move further to the left on the perplexity vs size curve. * k_quants: forgot to add the Metal changes in last commit * k_quants: change Q5_K to be type 0 when QK_K = 64 Still needs AVX2 implementation * k_quants: AVX2 implementation for new 64-weight Q5_K * k_quants: 10% faster ARM_NEON Q5_K dot product * k_quants: fixed issue caused by merging with master --------- Co-authored-by: Iwan Kawrakow <iwan.kawrakow@gmail.com>
2023-06-26 16:43:07 +00:00
ggml : quantization refactoring (#3833) * ggml : factor all quantization code in ggml-quants ggml-ci * ggml-quants : fix Zig and Swift builds + quantize tool ggml-ci * quantize : --pure option for disabling k-quant mixtures --------- Co-authored-by: cebtenzzre <cebtenzzre@gmail.com>
2023-10-29 16:32:28 +00:00
uint32_t qh;
uint64_t tmp[4];
ggml : add SOTA 2,3,4,5,6 bit k-quantizations (#1684) * Starting to add k-quantization to ggml I think it is better to have quantization separate from ggml. For now just adding the k-quants there, but it would be better to also factor out the existing ggml quantizations. * Adding Q3_K and Q8_K (de)-quantization * Q3_K now working on CUDA and AVX2/scalar CUDA is not ideal - ~50% slower than Q4_0 for single token prediction, about the same in batch mode (perplexity). CPU single token is ~55 ms (on Ryzen 7950X). * Some improvement for Q3_K on CUDA It is now ~22.5 ms/token on my GPU, so ~30% slower than Q4_0. * Some more CUDA optimizations for Q3_K Single token is now 20.5 ms/token (~20% slower than Q4_0). Perplexity is on par with Q4_0. * Adding Q4_K - scalar, AVX2, CUDA Performance is the same or perhaps very slightly better than Q4_0 on the CPU. On the GPU, single token prediction is ~10% better than Q4_0, batch mode (perplexity is about the same). * Adding Q6_K - scalar, AVX2, CUDA Performance is ~40% lower compared to Q4_K on the CPU. This is to be expected, considering that we are memory bound on the CPU and the 6-bit model is ~44% larger than the 4-bit. On the GPU, single token prediction is ~6% lower than Q4_0, batch mode (perplexity) is even closer (but still slower). * Adding Q5_K - scalar, AVX2, CUDA Performance is ~20% lower compared to Q4_K on the CPU. This is to be expected, considering that we are memory bound on the CPU and the 5-bit model is ~22% larger than the 4-bit. On the GPU, single token prediction is about the same as Q4_0 for both, single token and batch prediction. * Per convention, all QX_K quantizations use Q5_K for output.weight * Adding quantization mixes * Quantization mixes: didn't quite get what I wanted in the last commit * Q4_K dot product for ARM_NEON * Q6_K dot product for ARM_NEON * Q5_K dot product for ARM_NEON * Adding Q3_K dot for ARM_NEON It is 22% slower than Q4_K, despite the smaller model size. On x86_64, where we are memory bound, the Q3_K model is quite a bit faster than Q4_K. * A very slightly faster ARM_NEON Q3_K dot * Adding Q2_K - just CUDA for now Token prediction is pretty good - about 15.5 ms on a RTX 4080. Perplexity is about the same as Q4_K. * Adding scalar and AVX2 Q2_K dot * Adding ARM_NEON Q2_K dot About the same performance as Q4_K. * A slightly faster ARM_NEON Q2_K dot Single token prediction is now ~36 ms on M2 Max. The code is much simpler too. * Fixed bug in Q2_K CUDA dot product kernel Stranegly enough, for the few prompts I tried with the 7B model the responses looked perfectly reasonable. Only realized something is not quite right when I tried the larger models and started getting nonse back. In any case, Q2_K single token evaluation time on an RTX 4080 in a Ryzen7950X box iusing CUDA and model fully loaded on the GPU are ~15.5 ms for 7B, ~25.4 ms for 13B, and ~55.8 ms for 30B. The max number of layers that fit in VRAM for The 65B is 32. With that, we get ~330 ms per token, which is not that much faster than just running on the CPU (~470 ms per token). * Don't print zeros/NaNs when no count histogram has been collected * A 10% faster CUDA vector dot kernel for Q3_K Q3_K is now running at ~18.5 ms / token on CUDA, so the gap to Q4_0 is only 10%. It seems memory acccess pattern is more important for performance than the amount of computation the kernel does. * A slightly daster Q4_K AVX2 dot product For perplexity, where we are less memory bound, time per pass drops by ~5%. Barely measurable difference for single token prediction. * A slightly faster ARM_NEON A4_K dot product * Minor * Fix quantization error test We cannot possibly be expecting rmse < 0.002 for 2- and 3-bit quantization variants. * Fix docker build I have been sloppy with vector reinterpret casts on ARM_NEON. It seems clang is very forgiving in that regard. * Added forgotten ggml.o dependence on k_quants.h to the Makefile * Had unintentionally committed the Makefile with -Ofast enabled * ggml : rename k_quants -> ggml-quants-k, use lowercase in code --------- Co-authored-by: Iwan Kawrakow <iwan.kawrakow@gmail.com> Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
2023-06-05 19:56:18 +00:00
ggml : quantization refactoring (#3833) * ggml : factor all quantization code in ggml-quants ggml-ci * ggml-quants : fix Zig and Swift builds + quantize tool ggml-ci * quantize : --pure option for disabling k-quant mixtures --------- Co-authored-by: cebtenzzre <cebtenzzre@gmail.com>
2023-10-29 16:32:28 +00:00
// TODO: check if unrolling this is better
for (int i = 0; i < nb; ++i) {
const block_q5_1 * restrict x0 = &x[i];
const block_q8_1 * restrict y0 = &y[i];
ggml : add SOTA 2,3,4,5,6 bit k-quantizations (#1684) * Starting to add k-quantization to ggml I think it is better to have quantization separate from ggml. For now just adding the k-quants there, but it would be better to also factor out the existing ggml quantizations. * Adding Q3_K and Q8_K (de)-quantization * Q3_K now working on CUDA and AVX2/scalar CUDA is not ideal - ~50% slower than Q4_0 for single token prediction, about the same in batch mode (perplexity). CPU single token is ~55 ms (on Ryzen 7950X). * Some improvement for Q3_K on CUDA It is now ~22.5 ms/token on my GPU, so ~30% slower than Q4_0. * Some more CUDA optimizations for Q3_K Single token is now 20.5 ms/token (~20% slower than Q4_0). Perplexity is on par with Q4_0. * Adding Q4_K - scalar, AVX2, CUDA Performance is the same or perhaps very slightly better than Q4_0 on the CPU. On the GPU, single token prediction is ~10% better than Q4_0, batch mode (perplexity is about the same). * Adding Q6_K - scalar, AVX2, CUDA Performance is ~40% lower compared to Q4_K on the CPU. This is to be expected, considering that we are memory bound on the CPU and the 6-bit model is ~44% larger than the 4-bit. On the GPU, single token prediction is ~6% lower than Q4_0, batch mode (perplexity) is even closer (but still slower). * Adding Q5_K - scalar, AVX2, CUDA Performance is ~20% lower compared to Q4_K on the CPU. This is to be expected, considering that we are memory bound on the CPU and the 5-bit model is ~22% larger than the 4-bit. On the GPU, single token prediction is about the same as Q4_0 for both, single token and batch prediction. * Per convention, all QX_K quantizations use Q5_K for output.weight * Adding quantization mixes * Quantization mixes: didn't quite get what I wanted in the last commit * Q4_K dot product for ARM_NEON * Q6_K dot product for ARM_NEON * Q5_K dot product for ARM_NEON * Adding Q3_K dot for ARM_NEON It is 22% slower than Q4_K, despite the smaller model size. On x86_64, where we are memory bound, the Q3_K model is quite a bit faster than Q4_K. * A very slightly faster ARM_NEON Q3_K dot * Adding Q2_K - just CUDA for now Token prediction is pretty good - about 15.5 ms on a RTX 4080. Perplexity is about the same as Q4_K. * Adding scalar and AVX2 Q2_K dot * Adding ARM_NEON Q2_K dot About the same performance as Q4_K. * A slightly faster ARM_NEON Q2_K dot Single token prediction is now ~36 ms on M2 Max. The code is much simpler too. * Fixed bug in Q2_K CUDA dot product kernel Stranegly enough, for the few prompts I tried with the 7B model the responses looked perfectly reasonable. Only realized something is not quite right when I tried the larger models and started getting nonse back. In any case, Q2_K single token evaluation time on an RTX 4080 in a Ryzen7950X box iusing CUDA and model fully loaded on the GPU are ~15.5 ms for 7B, ~25.4 ms for 13B, and ~55.8 ms for 30B. The max number of layers that fit in VRAM for The 65B is 32. With that, we get ~330 ms per token, which is not that much faster than just running on the CPU (~470 ms per token). * Don't print zeros/NaNs when no count histogram has been collected * A 10% faster CUDA vector dot kernel for Q3_K Q3_K is now running at ~18.5 ms / token on CUDA, so the gap to Q4_0 is only 10%. It seems memory acccess pattern is more important for performance than the amount of computation the kernel does. * A slightly daster Q4_K AVX2 dot product For perplexity, where we are less memory bound, time per pass drops by ~5%. Barely measurable difference for single token prediction. * A slightly faster ARM_NEON A4_K dot product * Minor * Fix quantization error test We cannot possibly be expecting rmse < 0.002 for 2- and 3-bit quantization variants. * Fix docker build I have been sloppy with vector reinterpret casts on ARM_NEON. It seems clang is very forgiving in that regard. * Added forgotten ggml.o dependence on k_quants.h to the Makefile * Had unintentionally committed the Makefile with -Ofast enabled * ggml : rename k_quants -> ggml-quants-k, use lowercase in code --------- Co-authored-by: Iwan Kawrakow <iwan.kawrakow@gmail.com> Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
2023-06-05 19:56:18 +00:00
ggml : move FP16 <-> FP32 code to ggml-impl.h (#3861) * ggml : move FP16 <-> FP32 stuff to ggml-impl.h ggml-ci * tests : fix ARM build * ggml : explicitly initialize deprecated type traits * ggml : add math.h to ggml-impl.h * ggml : remove duplicate static assert macros * ggml : prefix lookup tables with ggml_ ggml-ci * ggml-impl : move extern "C" to start of file
2023-10-30 17:19:15 +00:00
summs += GGML_FP16_TO_FP32(x0->m) * y0->s;
ggml : add SOTA 2,3,4,5,6 bit k-quantizations (#1684) * Starting to add k-quantization to ggml I think it is better to have quantization separate from ggml. For now just adding the k-quants there, but it would be better to also factor out the existing ggml quantizations. * Adding Q3_K and Q8_K (de)-quantization * Q3_K now working on CUDA and AVX2/scalar CUDA is not ideal - ~50% slower than Q4_0 for single token prediction, about the same in batch mode (perplexity). CPU single token is ~55 ms (on Ryzen 7950X). * Some improvement for Q3_K on CUDA It is now ~22.5 ms/token on my GPU, so ~30% slower than Q4_0. * Some more CUDA optimizations for Q3_K Single token is now 20.5 ms/token (~20% slower than Q4_0). Perplexity is on par with Q4_0. * Adding Q4_K - scalar, AVX2, CUDA Performance is the same or perhaps very slightly better than Q4_0 on the CPU. On the GPU, single token prediction is ~10% better than Q4_0, batch mode (perplexity is about the same). * Adding Q6_K - scalar, AVX2, CUDA Performance is ~40% lower compared to Q4_K on the CPU. This is to be expected, considering that we are memory bound on the CPU and the 6-bit model is ~44% larger than the 4-bit. On the GPU, single token prediction is ~6% lower than Q4_0, batch mode (perplexity) is even closer (but still slower). * Adding Q5_K - scalar, AVX2, CUDA Performance is ~20% lower compared to Q4_K on the CPU. This is to be expected, considering that we are memory bound on the CPU and the 5-bit model is ~22% larger than the 4-bit. On the GPU, single token prediction is about the same as Q4_0 for both, single token and batch prediction. * Per convention, all QX_K quantizations use Q5_K for output.weight * Adding quantization mixes * Quantization mixes: didn't quite get what I wanted in the last commit * Q4_K dot product for ARM_NEON * Q6_K dot product for ARM_NEON * Q5_K dot product for ARM_NEON * Adding Q3_K dot for ARM_NEON It is 22% slower than Q4_K, despite the smaller model size. On x86_64, where we are memory bound, the Q3_K model is quite a bit faster than Q4_K. * A very slightly faster ARM_NEON Q3_K dot * Adding Q2_K - just CUDA for now Token prediction is pretty good - about 15.5 ms on a RTX 4080. Perplexity is about the same as Q4_K. * Adding scalar and AVX2 Q2_K dot * Adding ARM_NEON Q2_K dot About the same performance as Q4_K. * A slightly faster ARM_NEON Q2_K dot Single token prediction is now ~36 ms on M2 Max. The code is much simpler too. * Fixed bug in Q2_K CUDA dot product kernel Stranegly enough, for the few prompts I tried with the 7B model the responses looked perfectly reasonable. Only realized something is not quite right when I tried the larger models and started getting nonse back. In any case, Q2_K single token evaluation time on an RTX 4080 in a Ryzen7950X box iusing CUDA and model fully loaded on the GPU are ~15.5 ms for 7B, ~25.4 ms for 13B, and ~55.8 ms for 30B. The max number of layers that fit in VRAM for The 65B is 32. With that, we get ~330 ms per token, which is not that much faster than just running on the CPU (~470 ms per token). * Don't print zeros/NaNs when no count histogram has been collected * A 10% faster CUDA vector dot kernel for Q3_K Q3_K is now running at ~18.5 ms / token on CUDA, so the gap to Q4_0 is only 10%. It seems memory acccess pattern is more important for performance than the amount of computation the kernel does. * A slightly daster Q4_K AVX2 dot product For perplexity, where we are less memory bound, time per pass drops by ~5%. Barely measurable difference for single token prediction. * A slightly faster ARM_NEON A4_K dot product * Minor * Fix quantization error test We cannot possibly be expecting rmse < 0.002 for 2- and 3-bit quantization variants. * Fix docker build I have been sloppy with vector reinterpret casts on ARM_NEON. It seems clang is very forgiving in that regard. * Added forgotten ggml.o dependence on k_quants.h to the Makefile * Had unintentionally committed the Makefile with -Ofast enabled * ggml : rename k_quants -> ggml-quants-k, use lowercase in code --------- Co-authored-by: Iwan Kawrakow <iwan.kawrakow@gmail.com> Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
2023-06-05 19:56:18 +00:00
ggml : quantization refactoring (#3833) * ggml : factor all quantization code in ggml-quants ggml-ci * ggml-quants : fix Zig and Swift builds + quantize tool ggml-ci * quantize : --pure option for disabling k-quant mixtures --------- Co-authored-by: cebtenzzre <cebtenzzre@gmail.com>
2023-10-29 16:32:28 +00:00
const v128_t m4b = wasm_i8x16_splat(0x0F);
ggml : add SOTA 2,3,4,5,6 bit k-quantizations (#1684) * Starting to add k-quantization to ggml I think it is better to have quantization separate from ggml. For now just adding the k-quants there, but it would be better to also factor out the existing ggml quantizations. * Adding Q3_K and Q8_K (de)-quantization * Q3_K now working on CUDA and AVX2/scalar CUDA is not ideal - ~50% slower than Q4_0 for single token prediction, about the same in batch mode (perplexity). CPU single token is ~55 ms (on Ryzen 7950X). * Some improvement for Q3_K on CUDA It is now ~22.5 ms/token on my GPU, so ~30% slower than Q4_0. * Some more CUDA optimizations for Q3_K Single token is now 20.5 ms/token (~20% slower than Q4_0). Perplexity is on par with Q4_0. * Adding Q4_K - scalar, AVX2, CUDA Performance is the same or perhaps very slightly better than Q4_0 on the CPU. On the GPU, single token prediction is ~10% better than Q4_0, batch mode (perplexity is about the same). * Adding Q6_K - scalar, AVX2, CUDA Performance is ~40% lower compared to Q4_K on the CPU. This is to be expected, considering that we are memory bound on the CPU and the 6-bit model is ~44% larger than the 4-bit. On the GPU, single token prediction is ~6% lower than Q4_0, batch mode (perplexity) is even closer (but still slower). * Adding Q5_K - scalar, AVX2, CUDA Performance is ~20% lower compared to Q4_K on the CPU. This is to be expected, considering that we are memory bound on the CPU and the 5-bit model is ~22% larger than the 4-bit. On the GPU, single token prediction is about the same as Q4_0 for both, single token and batch prediction. * Per convention, all QX_K quantizations use Q5_K for output.weight * Adding quantization mixes * Quantization mixes: didn't quite get what I wanted in the last commit * Q4_K dot product for ARM_NEON * Q6_K dot product for ARM_NEON * Q5_K dot product for ARM_NEON * Adding Q3_K dot for ARM_NEON It is 22% slower than Q4_K, despite the smaller model size. On x86_64, where we are memory bound, the Q3_K model is quite a bit faster than Q4_K. * A very slightly faster ARM_NEON Q3_K dot * Adding Q2_K - just CUDA for now Token prediction is pretty good - about 15.5 ms on a RTX 4080. Perplexity is about the same as Q4_K. * Adding scalar and AVX2 Q2_K dot * Adding ARM_NEON Q2_K dot About the same performance as Q4_K. * A slightly faster ARM_NEON Q2_K dot Single token prediction is now ~36 ms on M2 Max. The code is much simpler too. * Fixed bug in Q2_K CUDA dot product kernel Stranegly enough, for the few prompts I tried with the 7B model the responses looked perfectly reasonable. Only realized something is not quite right when I tried the larger models and started getting nonse back. In any case, Q2_K single token evaluation time on an RTX 4080 in a Ryzen7950X box iusing CUDA and model fully loaded on the GPU are ~15.5 ms for 7B, ~25.4 ms for 13B, and ~55.8 ms for 30B. The max number of layers that fit in VRAM for The 65B is 32. With that, we get ~330 ms per token, which is not that much faster than just running on the CPU (~470 ms per token). * Don't print zeros/NaNs when no count histogram has been collected * A 10% faster CUDA vector dot kernel for Q3_K Q3_K is now running at ~18.5 ms / token on CUDA, so the gap to Q4_0 is only 10%. It seems memory acccess pattern is more important for performance than the amount of computation the kernel does. * A slightly daster Q4_K AVX2 dot product For perplexity, where we are less memory bound, time per pass drops by ~5%. Barely measurable difference for single token prediction. * A slightly faster ARM_NEON A4_K dot product * Minor * Fix quantization error test We cannot possibly be expecting rmse < 0.002 for 2- and 3-bit quantization variants. * Fix docker build I have been sloppy with vector reinterpret casts on ARM_NEON. It seems clang is very forgiving in that regard. * Added forgotten ggml.o dependence on k_quants.h to the Makefile * Had unintentionally committed the Makefile with -Ofast enabled * ggml : rename k_quants -> ggml-quants-k, use lowercase in code --------- Co-authored-by: Iwan Kawrakow <iwan.kawrakow@gmail.com> Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
2023-06-05 19:56:18 +00:00
ggml : quantization refactoring (#3833) * ggml : factor all quantization code in ggml-quants ggml-ci * ggml-quants : fix Zig and Swift builds + quantize tool ggml-ci * quantize : --pure option for disabling k-quant mixtures --------- Co-authored-by: cebtenzzre <cebtenzzre@gmail.com>
2023-10-29 16:32:28 +00:00
// extract the 5th bit
memcpy(&qh, x0->qh, sizeof(qh));
k-quants : support for super-block size of 64 (#2001) * k_quants: WIP super-blocks with 64 weights * k_quants: WIP super-blocks with 64 weights Q6_K scalar and AVX2 works * k_quants: WIP super-blocks with 64 weights Q4_K scalar and AVX2 works * k_quants: WIP super-blocks with 64 weights Q2_K scalar and AVX2 works. Q2_K is way too slow (it is actually slower than the scalar implementation) * k_quants: WIP super-blocks with 64 weights Q3_K scalar and AVX2 works. * k_quants: WIP super-blocks with 64 weights Q5_K scalar and AVX2 works, and with that all k_quants are done on AVX2 and scalar * k_quants: WIP super-blocks with 64 weights Q6_K working on CUDA. Cannot make it run quite as gast as with super-blocks with 256 weigths: 8% slower on 4080, 20% slower on the 1660 (but there we fit 1 less layer on the GPU because pf the larger model size), so some fraction of these 20% is due to that, * k_quants: WIP super-blocks with 64 weights Q4_K working on CUDA. ~10% slower on GTX-1660, 16% slower on 4080. * k_quants: WIP super-blocks with 64 weights Q2_K working on CUDA. ~3% slower on GTX-1660, 10% slower on 4080. * k_quants: WIP super-blocks with 64 weights Q3_K working on CUDA. * k_quants: WIP super-blocks with 64 weights Q5_K working on CUDA, and with this CUDA is done. * k_quants: WIP super-blocks with 64 weights Q6_K working on ARM_NEON * k_quants: WIP super-blocks with 64 weights Q4_K working on ARM_NEON, but quite a bit slower than 256 weights * k_quants: WIP super-blocks with 64 weights Q2_K working on ARM_NEON, but quite a bit slower than 256 weights * k_quants: WIP super-blocks with 64 weights Q3_K working on ARM_NEON, but quite a bit slower than 256 weights. * k_quants: WIP super-blocks with 64 weights Q5_K working on ARM_NEON, but quite a bit slower than 256 weights. With that, we have full support for ARM_NEON, although performance is not quite there. * k_quants: WIP super-blocks with 64 weights Slightly more efficient Q3_K and Q5_K * k_quants: WIP super-blocks with 64 weights Another small improvement for Q3_K and Q5_K on ARM_NEON * k_quants: WIP super-blocks with 64 weights Yet another speedup for Q5_K on ARM_NEON. We are now within 10% of the QK_K = 256 version. * k_quants: WIP super-blocks with 64 weights * We are able to pass preprocessor macros to the Metal compiler * Q6_K works and is actually slightly more efficient than the QK_K = 256 version (25.2 ms vs 25.8 ms) * k_quants: WIP super-blocks with 64 weights Q4_K works on Metal and is actually slightly faster than QK_K = 256 (21.95 ms vs 24.0 ms). * k_quants: WIP super-blocks with 64 weights Q2_K works on Metal and is very slightly faster than QK_K = 256 (23.8 ms vs 24.2 ms). * k_quants: WIP super-blocks with 64 weights Q3_K works on Metal and is slightly faster than QK_K = 256 (26.6 ms vs 28.3 ms). * k_quants: WIP super-blocks with 64 weights Q5_K works on Metal and is slightly faster than QK_K = 256 (23.7 ms vs 26.3 ms). * k_quants: call them _K, not _k, also on Metal * k_quants: correctly define QK_K in llama.cpp * Fixed bug in q4_K quantization added with the 64-block addition * Simplify via lambda * k_quants: swicth Q3_K to 4-bit scales when QK_K = 64 Otherwise there isn't much benefit from this quantization type. There is some very slight loss in accuracy, but we reduce size by ~7%. E.g., for OpenLLaMA-3B, Q3_K_S perplexity is 8.6131 with 8-bit scales and 8.6352 with 4-bit, while file size decreases from 1.53G to 1.44G. * k_quants: switch Q4_K to 4-bit scales when QK_K = 64 Here the loss in accuracy is greater than for Q3_K, but the Q4_K points still move further to the left on the perplexity vs size curve. * k_quants: forgot to add the Metal changes in last commit * k_quants: change Q5_K to be type 0 when QK_K = 64 Still needs AVX2 implementation * k_quants: AVX2 implementation for new 64-weight Q5_K * k_quants: 10% faster ARM_NEON Q5_K dot product * k_quants: fixed issue caused by merging with master --------- Co-authored-by: Iwan Kawrakow <iwan.kawrakow@gmail.com>
2023-06-26 16:43:07 +00:00
ggml : quantization refactoring (#3833) * ggml : factor all quantization code in ggml-quants ggml-ci * ggml-quants : fix Zig and Swift builds + quantize tool ggml-ci * quantize : --pure option for disabling k-quant mixtures --------- Co-authored-by: cebtenzzre <cebtenzzre@gmail.com>
2023-10-29 16:32:28 +00:00
tmp[0] = table_b2b_0[(qh >> 0) & 0xFF];
tmp[1] = table_b2b_0[(qh >> 8) & 0xFF];
tmp[2] = table_b2b_0[(qh >> 16) & 0xFF];
tmp[3] = table_b2b_0[(qh >> 24) ];
k-quants : support for super-block size of 64 (#2001) * k_quants: WIP super-blocks with 64 weights * k_quants: WIP super-blocks with 64 weights Q6_K scalar and AVX2 works * k_quants: WIP super-blocks with 64 weights Q4_K scalar and AVX2 works * k_quants: WIP super-blocks with 64 weights Q2_K scalar and AVX2 works. Q2_K is way too slow (it is actually slower than the scalar implementation) * k_quants: WIP super-blocks with 64 weights Q3_K scalar and AVX2 works. * k_quants: WIP super-blocks with 64 weights Q5_K scalar and AVX2 works, and with that all k_quants are done on AVX2 and scalar * k_quants: WIP super-blocks with 64 weights Q6_K working on CUDA. Cannot make it run quite as gast as with super-blocks with 256 weigths: 8% slower on 4080, 20% slower on the 1660 (but there we fit 1 less layer on the GPU because pf the larger model size), so some fraction of these 20% is due to that, * k_quants: WIP super-blocks with 64 weights Q4_K working on CUDA. ~10% slower on GTX-1660, 16% slower on 4080. * k_quants: WIP super-blocks with 64 weights Q2_K working on CUDA. ~3% slower on GTX-1660, 10% slower on 4080. * k_quants: WIP super-blocks with 64 weights Q3_K working on CUDA. * k_quants: WIP super-blocks with 64 weights Q5_K working on CUDA, and with this CUDA is done. * k_quants: WIP super-blocks with 64 weights Q6_K working on ARM_NEON * k_quants: WIP super-blocks with 64 weights Q4_K working on ARM_NEON, but quite a bit slower than 256 weights * k_quants: WIP super-blocks with 64 weights Q2_K working on ARM_NEON, but quite a bit slower than 256 weights * k_quants: WIP super-blocks with 64 weights Q3_K working on ARM_NEON, but quite a bit slower than 256 weights. * k_quants: WIP super-blocks with 64 weights Q5_K working on ARM_NEON, but quite a bit slower than 256 weights. With that, we have full support for ARM_NEON, although performance is not quite there. * k_quants: WIP super-blocks with 64 weights Slightly more efficient Q3_K and Q5_K * k_quants: WIP super-blocks with 64 weights Another small improvement for Q3_K and Q5_K on ARM_NEON * k_quants: WIP super-blocks with 64 weights Yet another speedup for Q5_K on ARM_NEON. We are now within 10% of the QK_K = 256 version. * k_quants: WIP super-blocks with 64 weights * We are able to pass preprocessor macros to the Metal compiler * Q6_K works and is actually slightly more efficient than the QK_K = 256 version (25.2 ms vs 25.8 ms) * k_quants: WIP super-blocks with 64 weights Q4_K works on Metal and is actually slightly faster than QK_K = 256 (21.95 ms vs 24.0 ms). * k_quants: WIP super-blocks with 64 weights Q2_K works on Metal and is very slightly faster than QK_K = 256 (23.8 ms vs 24.2 ms). * k_quants: WIP super-blocks with 64 weights Q3_K works on Metal and is slightly faster than QK_K = 256 (26.6 ms vs 28.3 ms). * k_quants: WIP super-blocks with 64 weights Q5_K works on Metal and is slightly faster than QK_K = 256 (23.7 ms vs 26.3 ms). * k_quants: call them _K, not _k, also on Metal * k_quants: correctly define QK_K in llama.cpp * Fixed bug in q4_K quantization added with the 64-block addition * Simplify via lambda * k_quants: swicth Q3_K to 4-bit scales when QK_K = 64 Otherwise there isn't much benefit from this quantization type. There is some very slight loss in accuracy, but we reduce size by ~7%. E.g., for OpenLLaMA-3B, Q3_K_S perplexity is 8.6131 with 8-bit scales and 8.6352 with 4-bit, while file size decreases from 1.53G to 1.44G. * k_quants: switch Q4_K to 4-bit scales when QK_K = 64 Here the loss in accuracy is greater than for Q3_K, but the Q4_K points still move further to the left on the perplexity vs size curve. * k_quants: forgot to add the Metal changes in last commit * k_quants: change Q5_K to be type 0 when QK_K = 64 Still needs AVX2 implementation * k_quants: AVX2 implementation for new 64-weight Q5_K * k_quants: 10% faster ARM_NEON Q5_K dot product * k_quants: fixed issue caused by merging with master --------- Co-authored-by: Iwan Kawrakow <iwan.kawrakow@gmail.com>
2023-06-26 16:43:07 +00:00
ggml : quantization refactoring (#3833) * ggml : factor all quantization code in ggml-quants ggml-ci * ggml-quants : fix Zig and Swift builds + quantize tool ggml-ci * quantize : --pure option for disabling k-quant mixtures --------- Co-authored-by: cebtenzzre <cebtenzzre@gmail.com>
2023-10-29 16:32:28 +00:00
const v128_t qhl = wasm_v128_load(tmp + 0);
const v128_t qhh = wasm_v128_load(tmp + 2);
k-quants : support for super-block size of 64 (#2001) * k_quants: WIP super-blocks with 64 weights * k_quants: WIP super-blocks with 64 weights Q6_K scalar and AVX2 works * k_quants: WIP super-blocks with 64 weights Q4_K scalar and AVX2 works * k_quants: WIP super-blocks with 64 weights Q2_K scalar and AVX2 works. Q2_K is way too slow (it is actually slower than the scalar implementation) * k_quants: WIP super-blocks with 64 weights Q3_K scalar and AVX2 works. * k_quants: WIP super-blocks with 64 weights Q5_K scalar and AVX2 works, and with that all k_quants are done on AVX2 and scalar * k_quants: WIP super-blocks with 64 weights Q6_K working on CUDA. Cannot make it run quite as gast as with super-blocks with 256 weigths: 8% slower on 4080, 20% slower on the 1660 (but there we fit 1 less layer on the GPU because pf the larger model size), so some fraction of these 20% is due to that, * k_quants: WIP super-blocks with 64 weights Q4_K working on CUDA. ~10% slower on GTX-1660, 16% slower on 4080. * k_quants: WIP super-blocks with 64 weights Q2_K working on CUDA. ~3% slower on GTX-1660, 10% slower on 4080. * k_quants: WIP super-blocks with 64 weights Q3_K working on CUDA. * k_quants: WIP super-blocks with 64 weights Q5_K working on CUDA, and with this CUDA is done. * k_quants: WIP super-blocks with 64 weights Q6_K working on ARM_NEON * k_quants: WIP super-blocks with 64 weights Q4_K working on ARM_NEON, but quite a bit slower than 256 weights * k_quants: WIP super-blocks with 64 weights Q2_K working on ARM_NEON, but quite a bit slower than 256 weights * k_quants: WIP super-blocks with 64 weights Q3_K working on ARM_NEON, but quite a bit slower than 256 weights. * k_quants: WIP super-blocks with 64 weights Q5_K working on ARM_NEON, but quite a bit slower than 256 weights. With that, we have full support for ARM_NEON, although performance is not quite there. * k_quants: WIP super-blocks with 64 weights Slightly more efficient Q3_K and Q5_K * k_quants: WIP super-blocks with 64 weights Another small improvement for Q3_K and Q5_K on ARM_NEON * k_quants: WIP super-blocks with 64 weights Yet another speedup for Q5_K on ARM_NEON. We are now within 10% of the QK_K = 256 version. * k_quants: WIP super-blocks with 64 weights * We are able to pass preprocessor macros to the Metal compiler * Q6_K works and is actually slightly more efficient than the QK_K = 256 version (25.2 ms vs 25.8 ms) * k_quants: WIP super-blocks with 64 weights Q4_K works on Metal and is actually slightly faster than QK_K = 256 (21.95 ms vs 24.0 ms). * k_quants: WIP super-blocks with 64 weights Q2_K works on Metal and is very slightly faster than QK_K = 256 (23.8 ms vs 24.2 ms). * k_quants: WIP super-blocks with 64 weights Q3_K works on Metal and is slightly faster than QK_K = 256 (26.6 ms vs 28.3 ms). * k_quants: WIP super-blocks with 64 weights Q5_K works on Metal and is slightly faster than QK_K = 256 (23.7 ms vs 26.3 ms). * k_quants: call them _K, not _k, also on Metal * k_quants: correctly define QK_K in llama.cpp * Fixed bug in q4_K quantization added with the 64-block addition * Simplify via lambda * k_quants: swicth Q3_K to 4-bit scales when QK_K = 64 Otherwise there isn't much benefit from this quantization type. There is some very slight loss in accuracy, but we reduce size by ~7%. E.g., for OpenLLaMA-3B, Q3_K_S perplexity is 8.6131 with 8-bit scales and 8.6352 with 4-bit, while file size decreases from 1.53G to 1.44G. * k_quants: switch Q4_K to 4-bit scales when QK_K = 64 Here the loss in accuracy is greater than for Q3_K, but the Q4_K points still move further to the left on the perplexity vs size curve. * k_quants: forgot to add the Metal changes in last commit * k_quants: change Q5_K to be type 0 when QK_K = 64 Still needs AVX2 implementation * k_quants: AVX2 implementation for new 64-weight Q5_K * k_quants: 10% faster ARM_NEON Q5_K dot product * k_quants: fixed issue caused by merging with master --------- Co-authored-by: Iwan Kawrakow <iwan.kawrakow@gmail.com>
2023-06-26 16:43:07 +00:00
ggml : quantization refactoring (#3833) * ggml : factor all quantization code in ggml-quants ggml-ci * ggml-quants : fix Zig and Swift builds + quantize tool ggml-ci * quantize : --pure option for disabling k-quant mixtures --------- Co-authored-by: cebtenzzre <cebtenzzre@gmail.com>
2023-10-29 16:32:28 +00:00
const v128_t v0 = wasm_v128_load(x0->qs);
k-quants : support for super-block size of 64 (#2001) * k_quants: WIP super-blocks with 64 weights * k_quants: WIP super-blocks with 64 weights Q6_K scalar and AVX2 works * k_quants: WIP super-blocks with 64 weights Q4_K scalar and AVX2 works * k_quants: WIP super-blocks with 64 weights Q2_K scalar and AVX2 works. Q2_K is way too slow (it is actually slower than the scalar implementation) * k_quants: WIP super-blocks with 64 weights Q3_K scalar and AVX2 works. * k_quants: WIP super-blocks with 64 weights Q5_K scalar and AVX2 works, and with that all k_quants are done on AVX2 and scalar * k_quants: WIP super-blocks with 64 weights Q6_K working on CUDA. Cannot make it run quite as gast as with super-blocks with 256 weigths: 8% slower on 4080, 20% slower on the 1660 (but there we fit 1 less layer on the GPU because pf the larger model size), so some fraction of these 20% is due to that, * k_quants: WIP super-blocks with 64 weights Q4_K working on CUDA. ~10% slower on GTX-1660, 16% slower on 4080. * k_quants: WIP super-blocks with 64 weights Q2_K working on CUDA. ~3% slower on GTX-1660, 10% slower on 4080. * k_quants: WIP super-blocks with 64 weights Q3_K working on CUDA. * k_quants: WIP super-blocks with 64 weights Q5_K working on CUDA, and with this CUDA is done. * k_quants: WIP super-blocks with 64 weights Q6_K working on ARM_NEON * k_quants: WIP super-blocks with 64 weights Q4_K working on ARM_NEON, but quite a bit slower than 256 weights * k_quants: WIP super-blocks with 64 weights Q2_K working on ARM_NEON, but quite a bit slower than 256 weights * k_quants: WIP super-blocks with 64 weights Q3_K working on ARM_NEON, but quite a bit slower than 256 weights. * k_quants: WIP super-blocks with 64 weights Q5_K working on ARM_NEON, but quite a bit slower than 256 weights. With that, we have full support for ARM_NEON, although performance is not quite there. * k_quants: WIP super-blocks with 64 weights Slightly more efficient Q3_K and Q5_K * k_quants: WIP super-blocks with 64 weights Another small improvement for Q3_K and Q5_K on ARM_NEON * k_quants: WIP super-blocks with 64 weights Yet another speedup for Q5_K on ARM_NEON. We are now within 10% of the QK_K = 256 version. * k_quants: WIP super-blocks with 64 weights * We are able to pass preprocessor macros to the Metal compiler * Q6_K works and is actually slightly more efficient than the QK_K = 256 version (25.2 ms vs 25.8 ms) * k_quants: WIP super-blocks with 64 weights Q4_K works on Metal and is actually slightly faster than QK_K = 256 (21.95 ms vs 24.0 ms). * k_quants: WIP super-blocks with 64 weights Q2_K works on Metal and is very slightly faster than QK_K = 256 (23.8 ms vs 24.2 ms). * k_quants: WIP super-blocks with 64 weights Q3_K works on Metal and is slightly faster than QK_K = 256 (26.6 ms vs 28.3 ms). * k_quants: WIP super-blocks with 64 weights Q5_K works on Metal and is slightly faster than QK_K = 256 (23.7 ms vs 26.3 ms). * k_quants: call them _K, not _k, also on Metal * k_quants: correctly define QK_K in llama.cpp * Fixed bug in q4_K quantization added with the 64-block addition * Simplify via lambda * k_quants: swicth Q3_K to 4-bit scales when QK_K = 64 Otherwise there isn't much benefit from this quantization type. There is some very slight loss in accuracy, but we reduce size by ~7%. E.g., for OpenLLaMA-3B, Q3_K_S perplexity is 8.6131 with 8-bit scales and 8.6352 with 4-bit, while file size decreases from 1.53G to 1.44G. * k_quants: switch Q4_K to 4-bit scales when QK_K = 64 Here the loss in accuracy is greater than for Q3_K, but the Q4_K points still move further to the left on the perplexity vs size curve. * k_quants: forgot to add the Metal changes in last commit * k_quants: change Q5_K to be type 0 when QK_K = 64 Still needs AVX2 implementation * k_quants: AVX2 implementation for new 64-weight Q5_K * k_quants: 10% faster ARM_NEON Q5_K dot product * k_quants: fixed issue caused by merging with master --------- Co-authored-by: Iwan Kawrakow <iwan.kawrakow@gmail.com>
2023-06-26 16:43:07 +00:00
ggml : quantization refactoring (#3833) * ggml : factor all quantization code in ggml-quants ggml-ci * ggml-quants : fix Zig and Swift builds + quantize tool ggml-ci * quantize : --pure option for disabling k-quant mixtures --------- Co-authored-by: cebtenzzre <cebtenzzre@gmail.com>
2023-10-29 16:32:28 +00:00
// 4-bit -> 8-bit
const v128_t v0l = wasm_v128_and (v0, m4b);
const v128_t v0h = wasm_u8x16_shr(v0, 4);
ggml : add SOTA 2,3,4,5,6 bit k-quantizations (#1684) * Starting to add k-quantization to ggml I think it is better to have quantization separate from ggml. For now just adding the k-quants there, but it would be better to also factor out the existing ggml quantizations. * Adding Q3_K and Q8_K (de)-quantization * Q3_K now working on CUDA and AVX2/scalar CUDA is not ideal - ~50% slower than Q4_0 for single token prediction, about the same in batch mode (perplexity). CPU single token is ~55 ms (on Ryzen 7950X). * Some improvement for Q3_K on CUDA It is now ~22.5 ms/token on my GPU, so ~30% slower than Q4_0. * Some more CUDA optimizations for Q3_K Single token is now 20.5 ms/token (~20% slower than Q4_0). Perplexity is on par with Q4_0. * Adding Q4_K - scalar, AVX2, CUDA Performance is the same or perhaps very slightly better than Q4_0 on the CPU. On the GPU, single token prediction is ~10% better than Q4_0, batch mode (perplexity is about the same). * Adding Q6_K - scalar, AVX2, CUDA Performance is ~40% lower compared to Q4_K on the CPU. This is to be expected, considering that we are memory bound on the CPU and the 6-bit model is ~44% larger than the 4-bit. On the GPU, single token prediction is ~6% lower than Q4_0, batch mode (perplexity) is even closer (but still slower). * Adding Q5_K - scalar, AVX2, CUDA Performance is ~20% lower compared to Q4_K on the CPU. This is to be expected, considering that we are memory bound on the CPU and the 5-bit model is ~22% larger than the 4-bit. On the GPU, single token prediction is about the same as Q4_0 for both, single token and batch prediction. * Per convention, all QX_K quantizations use Q5_K for output.weight * Adding quantization mixes * Quantization mixes: didn't quite get what I wanted in the last commit * Q4_K dot product for ARM_NEON * Q6_K dot product for ARM_NEON * Q5_K dot product for ARM_NEON * Adding Q3_K dot for ARM_NEON It is 22% slower than Q4_K, despite the smaller model size. On x86_64, where we are memory bound, the Q3_K model is quite a bit faster than Q4_K. * A very slightly faster ARM_NEON Q3_K dot * Adding Q2_K - just CUDA for now Token prediction is pretty good - about 15.5 ms on a RTX 4080. Perplexity is about the same as Q4_K. * Adding scalar and AVX2 Q2_K dot * Adding ARM_NEON Q2_K dot About the same performance as Q4_K. * A slightly faster ARM_NEON Q2_K dot Single token prediction is now ~36 ms on M2 Max. The code is much simpler too. * Fixed bug in Q2_K CUDA dot product kernel Stranegly enough, for the few prompts I tried with the 7B model the responses looked perfectly reasonable. Only realized something is not quite right when I tried the larger models and started getting nonse back. In any case, Q2_K single token evaluation time on an RTX 4080 in a Ryzen7950X box iusing CUDA and model fully loaded on the GPU are ~15.5 ms for 7B, ~25.4 ms for 13B, and ~55.8 ms for 30B. The max number of layers that fit in VRAM for The 65B is 32. With that, we get ~330 ms per token, which is not that much faster than just running on the CPU (~470 ms per token). * Don't print zeros/NaNs when no count histogram has been collected * A 10% faster CUDA vector dot kernel for Q3_K Q3_K is now running at ~18.5 ms / token on CUDA, so the gap to Q4_0 is only 10%. It seems memory acccess pattern is more important for performance than the amount of computation the kernel does. * A slightly daster Q4_K AVX2 dot product For perplexity, where we are less memory bound, time per pass drops by ~5%. Barely measurable difference for single token prediction. * A slightly faster ARM_NEON A4_K dot product * Minor * Fix quantization error test We cannot possibly be expecting rmse < 0.002 for 2- and 3-bit quantization variants. * Fix docker build I have been sloppy with vector reinterpret casts on ARM_NEON. It seems clang is very forgiving in that regard. * Added forgotten ggml.o dependence on k_quants.h to the Makefile * Had unintentionally committed the Makefile with -Ofast enabled * ggml : rename k_quants -> ggml-quants-k, use lowercase in code --------- Co-authored-by: Iwan Kawrakow <iwan.kawrakow@gmail.com> Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
2023-06-05 19:56:18 +00:00
ggml : quantization refactoring (#3833) * ggml : factor all quantization code in ggml-quants ggml-ci * ggml-quants : fix Zig and Swift builds + quantize tool ggml-ci * quantize : --pure option for disabling k-quant mixtures --------- Co-authored-by: cebtenzzre <cebtenzzre@gmail.com>
2023-10-29 16:32:28 +00:00
// add high bit
const v128_t v0lf = wasm_v128_or(v0l, qhl);
const v128_t v0hf = wasm_v128_or(v0h, qhh);
ggml : add SOTA 2,3,4,5,6 bit k-quantizations (#1684) * Starting to add k-quantization to ggml I think it is better to have quantization separate from ggml. For now just adding the k-quants there, but it would be better to also factor out the existing ggml quantizations. * Adding Q3_K and Q8_K (de)-quantization * Q3_K now working on CUDA and AVX2/scalar CUDA is not ideal - ~50% slower than Q4_0 for single token prediction, about the same in batch mode (perplexity). CPU single token is ~55 ms (on Ryzen 7950X). * Some improvement for Q3_K on CUDA It is now ~22.5 ms/token on my GPU, so ~30% slower than Q4_0. * Some more CUDA optimizations for Q3_K Single token is now 20.5 ms/token (~20% slower than Q4_0). Perplexity is on par with Q4_0. * Adding Q4_K - scalar, AVX2, CUDA Performance is the same or perhaps very slightly better than Q4_0 on the CPU. On the GPU, single token prediction is ~10% better than Q4_0, batch mode (perplexity is about the same). * Adding Q6_K - scalar, AVX2, CUDA Performance is ~40% lower compared to Q4_K on the CPU. This is to be expected, considering that we are memory bound on the CPU and the 6-bit model is ~44% larger than the 4-bit. On the GPU, single token prediction is ~6% lower than Q4_0, batch mode (perplexity) is even closer (but still slower). * Adding Q5_K - scalar, AVX2, CUDA Performance is ~20% lower compared to Q4_K on the CPU. This is to be expected, considering that we are memory bound on the CPU and the 5-bit model is ~22% larger than the 4-bit. On the GPU, single token prediction is about the same as Q4_0 for both, single token and batch prediction. * Per convention, all QX_K quantizations use Q5_K for output.weight * Adding quantization mixes * Quantization mixes: didn't quite get what I wanted in the last commit * Q4_K dot product for ARM_NEON * Q6_K dot product for ARM_NEON * Q5_K dot product for ARM_NEON * Adding Q3_K dot for ARM_NEON It is 22% slower than Q4_K, despite the smaller model size. On x86_64, where we are memory bound, the Q3_K model is quite a bit faster than Q4_K. * A very slightly faster ARM_NEON Q3_K dot * Adding Q2_K - just CUDA for now Token prediction is pretty good - about 15.5 ms on a RTX 4080. Perplexity is about the same as Q4_K. * Adding scalar and AVX2 Q2_K dot * Adding ARM_NEON Q2_K dot About the same performance as Q4_K. * A slightly faster ARM_NEON Q2_K dot Single token prediction is now ~36 ms on M2 Max. The code is much simpler too. * Fixed bug in Q2_K CUDA dot product kernel Stranegly enough, for the few prompts I tried with the 7B model the responses looked perfectly reasonable. Only realized something is not quite right when I tried the larger models and started getting nonse back. In any case, Q2_K single token evaluation time on an RTX 4080 in a Ryzen7950X box iusing CUDA and model fully loaded on the GPU are ~15.5 ms for 7B, ~25.4 ms for 13B, and ~55.8 ms for 30B. The max number of layers that fit in VRAM for The 65B is 32. With that, we get ~330 ms per token, which is not that much faster than just running on the CPU (~470 ms per token). * Don't print zeros/NaNs when no count histogram has been collected * A 10% faster CUDA vector dot kernel for Q3_K Q3_K is now running at ~18.5 ms / token on CUDA, so the gap to Q4_0 is only 10%. It seems memory acccess pattern is more important for performance than the amount of computation the kernel does. * A slightly daster Q4_K AVX2 dot product For perplexity, where we are less memory bound, time per pass drops by ~5%. Barely measurable difference for single token prediction. * A slightly faster ARM_NEON A4_K dot product * Minor * Fix quantization error test We cannot possibly be expecting rmse < 0.002 for 2- and 3-bit quantization variants. * Fix docker build I have been sloppy with vector reinterpret casts on ARM_NEON. It seems clang is very forgiving in that regard. * Added forgotten ggml.o dependence on k_quants.h to the Makefile * Had unintentionally committed the Makefile with -Ofast enabled * ggml : rename k_quants -> ggml-quants-k, use lowercase in code --------- Co-authored-by: Iwan Kawrakow <iwan.kawrakow@gmail.com> Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
2023-06-05 19:56:18 +00:00
ggml : quantization refactoring (#3833) * ggml : factor all quantization code in ggml-quants ggml-ci * ggml-quants : fix Zig and Swift builds + quantize tool ggml-ci * quantize : --pure option for disabling k-quant mixtures --------- Co-authored-by: cebtenzzre <cebtenzzre@gmail.com>
2023-10-29 16:32:28 +00:00
// load y
const v128_t v1l = wasm_v128_load(y0->qs);
const v128_t v1h = wasm_v128_load(y0->qs + 16);
// int8x16 -> int16x8
const v128_t v0lfl = wasm_i16x8_extend_low_i8x16 (v0lf);
const v128_t v0lfh = wasm_i16x8_extend_high_i8x16(v0lf);
const v128_t v0hfl = wasm_i16x8_extend_low_i8x16 (v0hf);
const v128_t v0hfh = wasm_i16x8_extend_high_i8x16(v0hf);
const v128_t v1ll = wasm_i16x8_extend_low_i8x16 (v1l);
const v128_t v1lh = wasm_i16x8_extend_high_i8x16(v1l);
const v128_t v1hl = wasm_i16x8_extend_low_i8x16 (v1h);
const v128_t v1hh = wasm_i16x8_extend_high_i8x16(v1h);
// dot product
sumv = wasm_f32x4_add(sumv,
wasm_f32x4_mul(wasm_f32x4_convert_i32x4(wasm_i32x4_add(
wasm_i32x4_add(wasm_i32x4_dot_i16x8(v0lfl, v1ll),
wasm_i32x4_dot_i16x8(v0lfh, v1lh)),
wasm_i32x4_add(wasm_i32x4_dot_i16x8(v0hfl, v1hl),
wasm_i32x4_dot_i16x8(v0hfh, v1hh)))),
ggml : move FP16 <-> FP32 code to ggml-impl.h (#3861) * ggml : move FP16 <-> FP32 stuff to ggml-impl.h ggml-ci * tests : fix ARM build * ggml : explicitly initialize deprecated type traits * ggml : add math.h to ggml-impl.h * ggml : remove duplicate static assert macros * ggml : prefix lookup tables with ggml_ ggml-ci * ggml-impl : move extern "C" to start of file
2023-10-30 17:19:15 +00:00
wasm_f32x4_splat(GGML_FP16_TO_FP32(x0->d) * y0->d)));
ggml : add SOTA 2,3,4,5,6 bit k-quantizations (#1684) * Starting to add k-quantization to ggml I think it is better to have quantization separate from ggml. For now just adding the k-quants there, but it would be better to also factor out the existing ggml quantizations. * Adding Q3_K and Q8_K (de)-quantization * Q3_K now working on CUDA and AVX2/scalar CUDA is not ideal - ~50% slower than Q4_0 for single token prediction, about the same in batch mode (perplexity). CPU single token is ~55 ms (on Ryzen 7950X). * Some improvement for Q3_K on CUDA It is now ~22.5 ms/token on my GPU, so ~30% slower than Q4_0. * Some more CUDA optimizations for Q3_K Single token is now 20.5 ms/token (~20% slower than Q4_0). Perplexity is on par with Q4_0. * Adding Q4_K - scalar, AVX2, CUDA Performance is the same or perhaps very slightly better than Q4_0 on the CPU. On the GPU, single token prediction is ~10% better than Q4_0, batch mode (perplexity is about the same). * Adding Q6_K - scalar, AVX2, CUDA Performance is ~40% lower compared to Q4_K on the CPU. This is to be expected, considering that we are memory bound on the CPU and the 6-bit model is ~44% larger than the 4-bit. On the GPU, single token prediction is ~6% lower than Q4_0, batch mode (perplexity) is even closer (but still slower). * Adding Q5_K - scalar, AVX2, CUDA Performance is ~20% lower compared to Q4_K on the CPU. This is to be expected, considering that we are memory bound on the CPU and the 5-bit model is ~22% larger than the 4-bit. On the GPU, single token prediction is about the same as Q4_0 for both, single token and batch prediction. * Per convention, all QX_K quantizations use Q5_K for output.weight * Adding quantization mixes * Quantization mixes: didn't quite get what I wanted in the last commit * Q4_K dot product for ARM_NEON * Q6_K dot product for ARM_NEON * Q5_K dot product for ARM_NEON * Adding Q3_K dot for ARM_NEON It is 22% slower than Q4_K, despite the smaller model size. On x86_64, where we are memory bound, the Q3_K model is quite a bit faster than Q4_K. * A very slightly faster ARM_NEON Q3_K dot * Adding Q2_K - just CUDA for now Token prediction is pretty good - about 15.5 ms on a RTX 4080. Perplexity is about the same as Q4_K. * Adding scalar and AVX2 Q2_K dot * Adding ARM_NEON Q2_K dot About the same performance as Q4_K. * A slightly faster ARM_NEON Q2_K dot Single token prediction is now ~36 ms on M2 Max. The code is much simpler too. * Fixed bug in Q2_K CUDA dot product kernel Stranegly enough, for the few prompts I tried with the 7B model the responses looked perfectly reasonable. Only realized something is not quite right when I tried the larger models and started getting nonse back. In any case, Q2_K single token evaluation time on an RTX 4080 in a Ryzen7950X box iusing CUDA and model fully loaded on the GPU are ~15.5 ms for 7B, ~25.4 ms for 13B, and ~55.8 ms for 30B. The max number of layers that fit in VRAM for The 65B is 32. With that, we get ~330 ms per token, which is not that much faster than just running on the CPU (~470 ms per token). * Don't print zeros/NaNs when no count histogram has been collected * A 10% faster CUDA vector dot kernel for Q3_K Q3_K is now running at ~18.5 ms / token on CUDA, so the gap to Q4_0 is only 10%. It seems memory acccess pattern is more important for performance than the amount of computation the kernel does. * A slightly daster Q4_K AVX2 dot product For perplexity, where we are less memory bound, time per pass drops by ~5%. Barely measurable difference for single token prediction. * A slightly faster ARM_NEON A4_K dot product * Minor * Fix quantization error test We cannot possibly be expecting rmse < 0.002 for 2- and 3-bit quantization variants. * Fix docker build I have been sloppy with vector reinterpret casts on ARM_NEON. It seems clang is very forgiving in that regard. * Added forgotten ggml.o dependence on k_quants.h to the Makefile * Had unintentionally committed the Makefile with -Ofast enabled * ggml : rename k_quants -> ggml-quants-k, use lowercase in code --------- Co-authored-by: Iwan Kawrakow <iwan.kawrakow@gmail.com> Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
2023-06-05 19:56:18 +00:00
}
ggml : quantization refactoring (#3833) * ggml : factor all quantization code in ggml-quants ggml-ci * ggml-quants : fix Zig and Swift builds + quantize tool ggml-ci * quantize : --pure option for disabling k-quant mixtures --------- Co-authored-by: cebtenzzre <cebtenzzre@gmail.com>
2023-10-29 16:32:28 +00:00
*s = wasm_f32x4_extract_lane(sumv, 0) + wasm_f32x4_extract_lane(sumv, 1) +
wasm_f32x4_extract_lane(sumv, 2) + wasm_f32x4_extract_lane(sumv, 3) + summs;
#elif defined(__AVX2__)
// Initialize accumulator with zeros
__m256 acc = _mm256_setzero_ps();
float summs = 0.0f;
ggml : add SOTA 2,3,4,5,6 bit k-quantizations (#1684) * Starting to add k-quantization to ggml I think it is better to have quantization separate from ggml. For now just adding the k-quants there, but it would be better to also factor out the existing ggml quantizations. * Adding Q3_K and Q8_K (de)-quantization * Q3_K now working on CUDA and AVX2/scalar CUDA is not ideal - ~50% slower than Q4_0 for single token prediction, about the same in batch mode (perplexity). CPU single token is ~55 ms (on Ryzen 7950X). * Some improvement for Q3_K on CUDA It is now ~22.5 ms/token on my GPU, so ~30% slower than Q4_0. * Some more CUDA optimizations for Q3_K Single token is now 20.5 ms/token (~20% slower than Q4_0). Perplexity is on par with Q4_0. * Adding Q4_K - scalar, AVX2, CUDA Performance is the same or perhaps very slightly better than Q4_0 on the CPU. On the GPU, single token prediction is ~10% better than Q4_0, batch mode (perplexity is about the same). * Adding Q6_K - scalar, AVX2, CUDA Performance is ~40% lower compared to Q4_K on the CPU. This is to be expected, considering that we are memory bound on the CPU and the 6-bit model is ~44% larger than the 4-bit. On the GPU, single token prediction is ~6% lower than Q4_0, batch mode (perplexity) is even closer (but still slower). * Adding Q5_K - scalar, AVX2, CUDA Performance is ~20% lower compared to Q4_K on the CPU. This is to be expected, considering that we are memory bound on the CPU and the 5-bit model is ~22% larger than the 4-bit. On the GPU, single token prediction is about the same as Q4_0 for both, single token and batch prediction. * Per convention, all QX_K quantizations use Q5_K for output.weight * Adding quantization mixes * Quantization mixes: didn't quite get what I wanted in the last commit * Q4_K dot product for ARM_NEON * Q6_K dot product for ARM_NEON * Q5_K dot product for ARM_NEON * Adding Q3_K dot for ARM_NEON It is 22% slower than Q4_K, despite the smaller model size. On x86_64, where we are memory bound, the Q3_K model is quite a bit faster than Q4_K. * A very slightly faster ARM_NEON Q3_K dot * Adding Q2_K - just CUDA for now Token prediction is pretty good - about 15.5 ms on a RTX 4080. Perplexity is about the same as Q4_K. * Adding scalar and AVX2 Q2_K dot * Adding ARM_NEON Q2_K dot About the same performance as Q4_K. * A slightly faster ARM_NEON Q2_K dot Single token prediction is now ~36 ms on M2 Max. The code is much simpler too. * Fixed bug in Q2_K CUDA dot product kernel Stranegly enough, for the few prompts I tried with the 7B model the responses looked perfectly reasonable. Only realized something is not quite right when I tried the larger models and started getting nonse back. In any case, Q2_K single token evaluation time on an RTX 4080 in a Ryzen7950X box iusing CUDA and model fully loaded on the GPU are ~15.5 ms for 7B, ~25.4 ms for 13B, and ~55.8 ms for 30B. The max number of layers that fit in VRAM for The 65B is 32. With that, we get ~330 ms per token, which is not that much faster than just running on the CPU (~470 ms per token). * Don't print zeros/NaNs when no count histogram has been collected * A 10% faster CUDA vector dot kernel for Q3_K Q3_K is now running at ~18.5 ms / token on CUDA, so the gap to Q4_0 is only 10%. It seems memory acccess pattern is more important for performance than the amount of computation the kernel does. * A slightly daster Q4_K AVX2 dot product For perplexity, where we are less memory bound, time per pass drops by ~5%. Barely measurable difference for single token prediction. * A slightly faster ARM_NEON A4_K dot product * Minor * Fix quantization error test We cannot possibly be expecting rmse < 0.002 for 2- and 3-bit quantization variants. * Fix docker build I have been sloppy with vector reinterpret casts on ARM_NEON. It seems clang is very forgiving in that regard. * Added forgotten ggml.o dependence on k_quants.h to the Makefile * Had unintentionally committed the Makefile with -Ofast enabled * ggml : rename k_quants -> ggml-quants-k, use lowercase in code --------- Co-authored-by: Iwan Kawrakow <iwan.kawrakow@gmail.com> Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
2023-06-05 19:56:18 +00:00
ggml : quantization refactoring (#3833) * ggml : factor all quantization code in ggml-quants ggml-ci * ggml-quants : fix Zig and Swift builds + quantize tool ggml-ci * quantize : --pure option for disabling k-quant mixtures --------- Co-authored-by: cebtenzzre <cebtenzzre@gmail.com>
2023-10-29 16:32:28 +00:00
// Main loop
ggml : add SOTA 2,3,4,5,6 bit k-quantizations (#1684) * Starting to add k-quantization to ggml I think it is better to have quantization separate from ggml. For now just adding the k-quants there, but it would be better to also factor out the existing ggml quantizations. * Adding Q3_K and Q8_K (de)-quantization * Q3_K now working on CUDA and AVX2/scalar CUDA is not ideal - ~50% slower than Q4_0 for single token prediction, about the same in batch mode (perplexity). CPU single token is ~55 ms (on Ryzen 7950X). * Some improvement for Q3_K on CUDA It is now ~22.5 ms/token on my GPU, so ~30% slower than Q4_0. * Some more CUDA optimizations for Q3_K Single token is now 20.5 ms/token (~20% slower than Q4_0). Perplexity is on par with Q4_0. * Adding Q4_K - scalar, AVX2, CUDA Performance is the same or perhaps very slightly better than Q4_0 on the CPU. On the GPU, single token prediction is ~10% better than Q4_0, batch mode (perplexity is about the same). * Adding Q6_K - scalar, AVX2, CUDA Performance is ~40% lower compared to Q4_K on the CPU. This is to be expected, considering that we are memory bound on the CPU and the 6-bit model is ~44% larger than the 4-bit. On the GPU, single token prediction is ~6% lower than Q4_0, batch mode (perplexity) is even closer (but still slower). * Adding Q5_K - scalar, AVX2, CUDA Performance is ~20% lower compared to Q4_K on the CPU. This is to be expected, considering that we are memory bound on the CPU and the 5-bit model is ~22% larger than the 4-bit. On the GPU, single token prediction is about the same as Q4_0 for both, single token and batch prediction. * Per convention, all QX_K quantizations use Q5_K for output.weight * Adding quantization mixes * Quantization mixes: didn't quite get what I wanted in the last commit * Q4_K dot product for ARM_NEON * Q6_K dot product for ARM_NEON * Q5_K dot product for ARM_NEON * Adding Q3_K dot for ARM_NEON It is 22% slower than Q4_K, despite the smaller model size. On x86_64, where we are memory bound, the Q3_K model is quite a bit faster than Q4_K. * A very slightly faster ARM_NEON Q3_K dot * Adding Q2_K - just CUDA for now Token prediction is pretty good - about 15.5 ms on a RTX 4080. Perplexity is about the same as Q4_K. * Adding scalar and AVX2 Q2_K dot * Adding ARM_NEON Q2_K dot About the same performance as Q4_K. * A slightly faster ARM_NEON Q2_K dot Single token prediction is now ~36 ms on M2 Max. The code is much simpler too. * Fixed bug in Q2_K CUDA dot product kernel Stranegly enough, for the few prompts I tried with the 7B model the responses looked perfectly reasonable. Only realized something is not quite right when I tried the larger models and started getting nonse back. In any case, Q2_K single token evaluation time on an RTX 4080 in a Ryzen7950X box iusing CUDA and model fully loaded on the GPU are ~15.5 ms for 7B, ~25.4 ms for 13B, and ~55.8 ms for 30B. The max number of layers that fit in VRAM for The 65B is 32. With that, we get ~330 ms per token, which is not that much faster than just running on the CPU (~470 ms per token). * Don't print zeros/NaNs when no count histogram has been collected * A 10% faster CUDA vector dot kernel for Q3_K Q3_K is now running at ~18.5 ms / token on CUDA, so the gap to Q4_0 is only 10%. It seems memory acccess pattern is more important for performance than the amount of computation the kernel does. * A slightly daster Q4_K AVX2 dot product For perplexity, where we are less memory bound, time per pass drops by ~5%. Barely measurable difference for single token prediction. * A slightly faster ARM_NEON A4_K dot product * Minor * Fix quantization error test We cannot possibly be expecting rmse < 0.002 for 2- and 3-bit quantization variants. * Fix docker build I have been sloppy with vector reinterpret casts on ARM_NEON. It seems clang is very forgiving in that regard. * Added forgotten ggml.o dependence on k_quants.h to the Makefile * Had unintentionally committed the Makefile with -Ofast enabled * ggml : rename k_quants -> ggml-quants-k, use lowercase in code --------- Co-authored-by: Iwan Kawrakow <iwan.kawrakow@gmail.com> Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
2023-06-05 19:56:18 +00:00
for (int i = 0; i < nb; i++) {
ggml : move FP16 <-> FP32 code to ggml-impl.h (#3861) * ggml : move FP16 <-> FP32 stuff to ggml-impl.h ggml-ci * tests : fix ARM build * ggml : explicitly initialize deprecated type traits * ggml : add math.h to ggml-impl.h * ggml : remove duplicate static assert macros * ggml : prefix lookup tables with ggml_ ggml-ci * ggml-impl : move extern "C" to start of file
2023-10-30 17:19:15 +00:00
const __m256 dx = _mm256_set1_ps(GGML_FP16_TO_FP32(x[i].d));
ggml : add SOTA 2,3,4,5,6 bit k-quantizations (#1684) * Starting to add k-quantization to ggml I think it is better to have quantization separate from ggml. For now just adding the k-quants there, but it would be better to also factor out the existing ggml quantizations. * Adding Q3_K and Q8_K (de)-quantization * Q3_K now working on CUDA and AVX2/scalar CUDA is not ideal - ~50% slower than Q4_0 for single token prediction, about the same in batch mode (perplexity). CPU single token is ~55 ms (on Ryzen 7950X). * Some improvement for Q3_K on CUDA It is now ~22.5 ms/token on my GPU, so ~30% slower than Q4_0. * Some more CUDA optimizations for Q3_K Single token is now 20.5 ms/token (~20% slower than Q4_0). Perplexity is on par with Q4_0. * Adding Q4_K - scalar, AVX2, CUDA Performance is the same or perhaps very slightly better than Q4_0 on the CPU. On the GPU, single token prediction is ~10% better than Q4_0, batch mode (perplexity is about the same). * Adding Q6_K - scalar, AVX2, CUDA Performance is ~40% lower compared to Q4_K on the CPU. This is to be expected, considering that we are memory bound on the CPU and the 6-bit model is ~44% larger than the 4-bit. On the GPU, single token prediction is ~6% lower than Q4_0, batch mode (perplexity) is even closer (but still slower). * Adding Q5_K - scalar, AVX2, CUDA Performance is ~20% lower compared to Q4_K on the CPU. This is to be expected, considering that we are memory bound on the CPU and the 5-bit model is ~22% larger than the 4-bit. On the GPU, single token prediction is about the same as Q4_0 for both, single token and batch prediction. * Per convention, all QX_K quantizations use Q5_K for output.weight * Adding quantization mixes * Quantization mixes: didn't quite get what I wanted in the last commit * Q4_K dot product for ARM_NEON * Q6_K dot product for ARM_NEON * Q5_K dot product for ARM_NEON * Adding Q3_K dot for ARM_NEON It is 22% slower than Q4_K, despite the smaller model size. On x86_64, where we are memory bound, the Q3_K model is quite a bit faster than Q4_K. * A very slightly faster ARM_NEON Q3_K dot * Adding Q2_K - just CUDA for now Token prediction is pretty good - about 15.5 ms on a RTX 4080. Perplexity is about the same as Q4_K. * Adding scalar and AVX2 Q2_K dot * Adding ARM_NEON Q2_K dot About the same performance as Q4_K. * A slightly faster ARM_NEON Q2_K dot Single token prediction is now ~36 ms on M2 Max. The code is much simpler too. * Fixed bug in Q2_K CUDA dot product kernel Stranegly enough, for the few prompts I tried with the 7B model the responses looked perfectly reasonable. Only realized something is not quite right when I tried the larger models and started getting nonse back. In any case, Q2_K single token evaluation time on an RTX 4080 in a Ryzen7950X box iusing CUDA and model fully loaded on the GPU are ~15.5 ms for 7B, ~25.4 ms for 13B, and ~55.8 ms for 30B. The max number of layers that fit in VRAM for The 65B is 32. With that, we get ~330 ms per token, which is not that much faster than just running on the CPU (~470 ms per token). * Don't print zeros/NaNs when no count histogram has been collected * A 10% faster CUDA vector dot kernel for Q3_K Q3_K is now running at ~18.5 ms / token on CUDA, so the gap to Q4_0 is only 10%. It seems memory acccess pattern is more important for performance than the amount of computation the kernel does. * A slightly daster Q4_K AVX2 dot product For perplexity, where we are less memory bound, time per pass drops by ~5%. Barely measurable difference for single token prediction. * A slightly faster ARM_NEON A4_K dot product * Minor * Fix quantization error test We cannot possibly be expecting rmse < 0.002 for 2- and 3-bit quantization variants. * Fix docker build I have been sloppy with vector reinterpret casts on ARM_NEON. It seems clang is very forgiving in that regard. * Added forgotten ggml.o dependence on k_quants.h to the Makefile * Had unintentionally committed the Makefile with -Ofast enabled * ggml : rename k_quants -> ggml-quants-k, use lowercase in code --------- Co-authored-by: Iwan Kawrakow <iwan.kawrakow@gmail.com> Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
2023-06-05 19:56:18 +00:00
ggml : move FP16 <-> FP32 code to ggml-impl.h (#3861) * ggml : move FP16 <-> FP32 stuff to ggml-impl.h ggml-ci * tests : fix ARM build * ggml : explicitly initialize deprecated type traits * ggml : add math.h to ggml-impl.h * ggml : remove duplicate static assert macros * ggml : prefix lookup tables with ggml_ ggml-ci * ggml-impl : move extern "C" to start of file
2023-10-30 17:19:15 +00:00
summs += GGML_FP16_TO_FP32(x[i].m) * y[i].s;
ggml : add SOTA 2,3,4,5,6 bit k-quantizations (#1684) * Starting to add k-quantization to ggml I think it is better to have quantization separate from ggml. For now just adding the k-quants there, but it would be better to also factor out the existing ggml quantizations. * Adding Q3_K and Q8_K (de)-quantization * Q3_K now working on CUDA and AVX2/scalar CUDA is not ideal - ~50% slower than Q4_0 for single token prediction, about the same in batch mode (perplexity). CPU single token is ~55 ms (on Ryzen 7950X). * Some improvement for Q3_K on CUDA It is now ~22.5 ms/token on my GPU, so ~30% slower than Q4_0. * Some more CUDA optimizations for Q3_K Single token is now 20.5 ms/token (~20% slower than Q4_0). Perplexity is on par with Q4_0. * Adding Q4_K - scalar, AVX2, CUDA Performance is the same or perhaps very slightly better than Q4_0 on the CPU. On the GPU, single token prediction is ~10% better than Q4_0, batch mode (perplexity is about the same). * Adding Q6_K - scalar, AVX2, CUDA Performance is ~40% lower compared to Q4_K on the CPU. This is to be expected, considering that we are memory bound on the CPU and the 6-bit model is ~44% larger than the 4-bit. On the GPU, single token prediction is ~6% lower than Q4_0, batch mode (perplexity) is even closer (but still slower). * Adding Q5_K - scalar, AVX2, CUDA Performance is ~20% lower compared to Q4_K on the CPU. This is to be expected, considering that we are memory bound on the CPU and the 5-bit model is ~22% larger than the 4-bit. On the GPU, single token prediction is about the same as Q4_0 for both, single token and batch prediction. * Per convention, all QX_K quantizations use Q5_K for output.weight * Adding quantization mixes * Quantization mixes: didn't quite get what I wanted in the last commit * Q4_K dot product for ARM_NEON * Q6_K dot product for ARM_NEON * Q5_K dot product for ARM_NEON * Adding Q3_K dot for ARM_NEON It is 22% slower than Q4_K, despite the smaller model size. On x86_64, where we are memory bound, the Q3_K model is quite a bit faster than Q4_K. * A very slightly faster ARM_NEON Q3_K dot * Adding Q2_K - just CUDA for now Token prediction is pretty good - about 15.5 ms on a RTX 4080. Perplexity is about the same as Q4_K. * Adding scalar and AVX2 Q2_K dot * Adding ARM_NEON Q2_K dot About the same performance as Q4_K. * A slightly faster ARM_NEON Q2_K dot Single token prediction is now ~36 ms on M2 Max. The code is much simpler too. * Fixed bug in Q2_K CUDA dot product kernel Stranegly enough, for the few prompts I tried with the 7B model the responses looked perfectly reasonable. Only realized something is not quite right when I tried the larger models and started getting nonse back. In any case, Q2_K single token evaluation time on an RTX 4080 in a Ryzen7950X box iusing CUDA and model fully loaded on the GPU are ~15.5 ms for 7B, ~25.4 ms for 13B, and ~55.8 ms for 30B. The max number of layers that fit in VRAM for The 65B is 32. With that, we get ~330 ms per token, which is not that much faster than just running on the CPU (~470 ms per token). * Don't print zeros/NaNs when no count histogram has been collected * A 10% faster CUDA vector dot kernel for Q3_K Q3_K is now running at ~18.5 ms / token on CUDA, so the gap to Q4_0 is only 10%. It seems memory acccess pattern is more important for performance than the amount of computation the kernel does. * A slightly daster Q4_K AVX2 dot product For perplexity, where we are less memory bound, time per pass drops by ~5%. Barely measurable difference for single token prediction. * A slightly faster ARM_NEON A4_K dot product * Minor * Fix quantization error test We cannot possibly be expecting rmse < 0.002 for 2- and 3-bit quantization variants. * Fix docker build I have been sloppy with vector reinterpret casts on ARM_NEON. It seems clang is very forgiving in that regard. * Added forgotten ggml.o dependence on k_quants.h to the Makefile * Had unintentionally committed the Makefile with -Ofast enabled * ggml : rename k_quants -> ggml-quants-k, use lowercase in code --------- Co-authored-by: Iwan Kawrakow <iwan.kawrakow@gmail.com> Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
2023-06-05 19:56:18 +00:00
ggml : quantization refactoring (#3833) * ggml : factor all quantization code in ggml-quants ggml-ci * ggml-quants : fix Zig and Swift builds + quantize tool ggml-ci * quantize : --pure option for disabling k-quant mixtures --------- Co-authored-by: cebtenzzre <cebtenzzre@gmail.com>
2023-10-29 16:32:28 +00:00
__m256i bx = bytes_from_nibbles_32(x[i].qs);
__m256i bxhi = bytes_from_bits_32(x[i].qh);
bxhi = _mm256_and_si256(bxhi, _mm256_set1_epi8(0x10));
bx = _mm256_or_si256(bx, bxhi);
k-quants : support for super-block size of 64 (#2001) * k_quants: WIP super-blocks with 64 weights * k_quants: WIP super-blocks with 64 weights Q6_K scalar and AVX2 works * k_quants: WIP super-blocks with 64 weights Q4_K scalar and AVX2 works * k_quants: WIP super-blocks with 64 weights Q2_K scalar and AVX2 works. Q2_K is way too slow (it is actually slower than the scalar implementation) * k_quants: WIP super-blocks with 64 weights Q3_K scalar and AVX2 works. * k_quants: WIP super-blocks with 64 weights Q5_K scalar and AVX2 works, and with that all k_quants are done on AVX2 and scalar * k_quants: WIP super-blocks with 64 weights Q6_K working on CUDA. Cannot make it run quite as gast as with super-blocks with 256 weigths: 8% slower on 4080, 20% slower on the 1660 (but there we fit 1 less layer on the GPU because pf the larger model size), so some fraction of these 20% is due to that, * k_quants: WIP super-blocks with 64 weights Q4_K working on CUDA. ~10% slower on GTX-1660, 16% slower on 4080. * k_quants: WIP super-blocks with 64 weights Q2_K working on CUDA. ~3% slower on GTX-1660, 10% slower on 4080. * k_quants: WIP super-blocks with 64 weights Q3_K working on CUDA. * k_quants: WIP super-blocks with 64 weights Q5_K working on CUDA, and with this CUDA is done. * k_quants: WIP super-blocks with 64 weights Q6_K working on ARM_NEON * k_quants: WIP super-blocks with 64 weights Q4_K working on ARM_NEON, but quite a bit slower than 256 weights * k_quants: WIP super-blocks with 64 weights Q2_K working on ARM_NEON, but quite a bit slower than 256 weights * k_quants: WIP super-blocks with 64 weights Q3_K working on ARM_NEON, but quite a bit slower than 256 weights. * k_quants: WIP super-blocks with 64 weights Q5_K working on ARM_NEON, but quite a bit slower than 256 weights. With that, we have full support for ARM_NEON, although performance is not quite there. * k_quants: WIP super-blocks with 64 weights Slightly more efficient Q3_K and Q5_K * k_quants: WIP super-blocks with 64 weights Another small improvement for Q3_K and Q5_K on ARM_NEON * k_quants: WIP super-blocks with 64 weights Yet another speedup for Q5_K on ARM_NEON. We are now within 10% of the QK_K = 256 version. * k_quants: WIP super-blocks with 64 weights * We are able to pass preprocessor macros to the Metal compiler * Q6_K works and is actually slightly more efficient than the QK_K = 256 version (25.2 ms vs 25.8 ms) * k_quants: WIP super-blocks with 64 weights Q4_K works on Metal and is actually slightly faster than QK_K = 256 (21.95 ms vs 24.0 ms). * k_quants: WIP super-blocks with 64 weights Q2_K works on Metal and is very slightly faster than QK_K = 256 (23.8 ms vs 24.2 ms). * k_quants: WIP super-blocks with 64 weights Q3_K works on Metal and is slightly faster than QK_K = 256 (26.6 ms vs 28.3 ms). * k_quants: WIP super-blocks with 64 weights Q5_K works on Metal and is slightly faster than QK_K = 256 (23.7 ms vs 26.3 ms). * k_quants: call them _K, not _k, also on Metal * k_quants: correctly define QK_K in llama.cpp * Fixed bug in q4_K quantization added with the 64-block addition * Simplify via lambda * k_quants: swicth Q3_K to 4-bit scales when QK_K = 64 Otherwise there isn't much benefit from this quantization type. There is some very slight loss in accuracy, but we reduce size by ~7%. E.g., for OpenLLaMA-3B, Q3_K_S perplexity is 8.6131 with 8-bit scales and 8.6352 with 4-bit, while file size decreases from 1.53G to 1.44G. * k_quants: switch Q4_K to 4-bit scales when QK_K = 64 Here the loss in accuracy is greater than for Q3_K, but the Q4_K points still move further to the left on the perplexity vs size curve. * k_quants: forgot to add the Metal changes in last commit * k_quants: change Q5_K to be type 0 when QK_K = 64 Still needs AVX2 implementation * k_quants: AVX2 implementation for new 64-weight Q5_K * k_quants: 10% faster ARM_NEON Q5_K dot product * k_quants: fixed issue caused by merging with master --------- Co-authored-by: Iwan Kawrakow <iwan.kawrakow@gmail.com>
2023-06-26 16:43:07 +00:00
ggml : quantization refactoring (#3833) * ggml : factor all quantization code in ggml-quants ggml-ci * ggml-quants : fix Zig and Swift builds + quantize tool ggml-ci * quantize : --pure option for disabling k-quant mixtures --------- Co-authored-by: cebtenzzre <cebtenzzre@gmail.com>
2023-10-29 16:32:28 +00:00
const __m256 dy = _mm256_set1_ps(y[i].d);
const __m256i by = _mm256_loadu_si256((const __m256i *)y[i].qs);
k-quants : support for super-block size of 64 (#2001) * k_quants: WIP super-blocks with 64 weights * k_quants: WIP super-blocks with 64 weights Q6_K scalar and AVX2 works * k_quants: WIP super-blocks with 64 weights Q4_K scalar and AVX2 works * k_quants: WIP super-blocks with 64 weights Q2_K scalar and AVX2 works. Q2_K is way too slow (it is actually slower than the scalar implementation) * k_quants: WIP super-blocks with 64 weights Q3_K scalar and AVX2 works. * k_quants: WIP super-blocks with 64 weights Q5_K scalar and AVX2 works, and with that all k_quants are done on AVX2 and scalar * k_quants: WIP super-blocks with 64 weights Q6_K working on CUDA. Cannot make it run quite as gast as with super-blocks with 256 weigths: 8% slower on 4080, 20% slower on the 1660 (but there we fit 1 less layer on the GPU because pf the larger model size), so some fraction of these 20% is due to that, * k_quants: WIP super-blocks with 64 weights Q4_K working on CUDA. ~10% slower on GTX-1660, 16% slower on 4080. * k_quants: WIP super-blocks with 64 weights Q2_K working on CUDA. ~3% slower on GTX-1660, 10% slower on 4080. * k_quants: WIP super-blocks with 64 weights Q3_K working on CUDA. * k_quants: WIP super-blocks with 64 weights Q5_K working on CUDA, and with this CUDA is done. * k_quants: WIP super-blocks with 64 weights Q6_K working on ARM_NEON * k_quants: WIP super-blocks with 64 weights Q4_K working on ARM_NEON, but quite a bit slower than 256 weights * k_quants: WIP super-blocks with 64 weights Q2_K working on ARM_NEON, but quite a bit slower than 256 weights * k_quants: WIP super-blocks with 64 weights Q3_K working on ARM_NEON, but quite a bit slower than 256 weights. * k_quants: WIP super-blocks with 64 weights Q5_K working on ARM_NEON, but quite a bit slower than 256 weights. With that, we have full support for ARM_NEON, although performance is not quite there. * k_quants: WIP super-blocks with 64 weights Slightly more efficient Q3_K and Q5_K * k_quants: WIP super-blocks with 64 weights Another small improvement for Q3_K and Q5_K on ARM_NEON * k_quants: WIP super-blocks with 64 weights Yet another speedup for Q5_K on ARM_NEON. We are now within 10% of the QK_K = 256 version. * k_quants: WIP super-blocks with 64 weights * We are able to pass preprocessor macros to the Metal compiler * Q6_K works and is actually slightly more efficient than the QK_K = 256 version (25.2 ms vs 25.8 ms) * k_quants: WIP super-blocks with 64 weights Q4_K works on Metal and is actually slightly faster than QK_K = 256 (21.95 ms vs 24.0 ms). * k_quants: WIP super-blocks with 64 weights Q2_K works on Metal and is very slightly faster than QK_K = 256 (23.8 ms vs 24.2 ms). * k_quants: WIP super-blocks with 64 weights Q3_K works on Metal and is slightly faster than QK_K = 256 (26.6 ms vs 28.3 ms). * k_quants: WIP super-blocks with 64 weights Q5_K works on Metal and is slightly faster than QK_K = 256 (23.7 ms vs 26.3 ms). * k_quants: call them _K, not _k, also on Metal * k_quants: correctly define QK_K in llama.cpp * Fixed bug in q4_K quantization added with the 64-block addition * Simplify via lambda * k_quants: swicth Q3_K to 4-bit scales when QK_K = 64 Otherwise there isn't much benefit from this quantization type. There is some very slight loss in accuracy, but we reduce size by ~7%. E.g., for OpenLLaMA-3B, Q3_K_S perplexity is 8.6131 with 8-bit scales and 8.6352 with 4-bit, while file size decreases from 1.53G to 1.44G. * k_quants: switch Q4_K to 4-bit scales when QK_K = 64 Here the loss in accuracy is greater than for Q3_K, but the Q4_K points still move further to the left on the perplexity vs size curve. * k_quants: forgot to add the Metal changes in last commit * k_quants: change Q5_K to be type 0 when QK_K = 64 Still needs AVX2 implementation * k_quants: AVX2 implementation for new 64-weight Q5_K * k_quants: 10% faster ARM_NEON Q5_K dot product * k_quants: fixed issue caused by merging with master --------- Co-authored-by: Iwan Kawrakow <iwan.kawrakow@gmail.com>
2023-06-26 16:43:07 +00:00
ggml : quantization refactoring (#3833) * ggml : factor all quantization code in ggml-quants ggml-ci * ggml-quants : fix Zig and Swift builds + quantize tool ggml-ci * quantize : --pure option for disabling k-quant mixtures --------- Co-authored-by: cebtenzzre <cebtenzzre@gmail.com>
2023-10-29 16:32:28 +00:00
const __m256 q = mul_sum_us8_pairs_float(bx, by);
acc = _mm256_fmadd_ps(q, _mm256_mul_ps(dx, dy), acc);
ggml : add SOTA 2,3,4,5,6 bit k-quantizations (#1684) * Starting to add k-quantization to ggml I think it is better to have quantization separate from ggml. For now just adding the k-quants there, but it would be better to also factor out the existing ggml quantizations. * Adding Q3_K and Q8_K (de)-quantization * Q3_K now working on CUDA and AVX2/scalar CUDA is not ideal - ~50% slower than Q4_0 for single token prediction, about the same in batch mode (perplexity). CPU single token is ~55 ms (on Ryzen 7950X). * Some improvement for Q3_K on CUDA It is now ~22.5 ms/token on my GPU, so ~30% slower than Q4_0. * Some more CUDA optimizations for Q3_K Single token is now 20.5 ms/token (~20% slower than Q4_0). Perplexity is on par with Q4_0. * Adding Q4_K - scalar, AVX2, CUDA Performance is the same or perhaps very slightly better than Q4_0 on the CPU. On the GPU, single token prediction is ~10% better than Q4_0, batch mode (perplexity is about the same). * Adding Q6_K - scalar, AVX2, CUDA Performance is ~40% lower compared to Q4_K on the CPU. This is to be expected, considering that we are memory bound on the CPU and the 6-bit model is ~44% larger than the 4-bit. On the GPU, single token prediction is ~6% lower than Q4_0, batch mode (perplexity) is even closer (but still slower). * Adding Q5_K - scalar, AVX2, CUDA Performance is ~20% lower compared to Q4_K on the CPU. This is to be expected, considering that we are memory bound on the CPU and the 5-bit model is ~22% larger than the 4-bit. On the GPU, single token prediction is about the same as Q4_0 for both, single token and batch prediction. * Per convention, all QX_K quantizations use Q5_K for output.weight * Adding quantization mixes * Quantization mixes: didn't quite get what I wanted in the last commit * Q4_K dot product for ARM_NEON * Q6_K dot product for ARM_NEON * Q5_K dot product for ARM_NEON * Adding Q3_K dot for ARM_NEON It is 22% slower than Q4_K, despite the smaller model size. On x86_64, where we are memory bound, the Q3_K model is quite a bit faster than Q4_K. * A very slightly faster ARM_NEON Q3_K dot * Adding Q2_K - just CUDA for now Token prediction is pretty good - about 15.5 ms on a RTX 4080. Perplexity is about the same as Q4_K. * Adding scalar and AVX2 Q2_K dot * Adding ARM_NEON Q2_K dot About the same performance as Q4_K. * A slightly faster ARM_NEON Q2_K dot Single token prediction is now ~36 ms on M2 Max. The code is much simpler too. * Fixed bug in Q2_K CUDA dot product kernel Stranegly enough, for the few prompts I tried with the 7B model the responses looked perfectly reasonable. Only realized something is not quite right when I tried the larger models and started getting nonse back. In any case, Q2_K single token evaluation time on an RTX 4080 in a Ryzen7950X box iusing CUDA and model fully loaded on the GPU are ~15.5 ms for 7B, ~25.4 ms for 13B, and ~55.8 ms for 30B. The max number of layers that fit in VRAM for The 65B is 32. With that, we get ~330 ms per token, which is not that much faster than just running on the CPU (~470 ms per token). * Don't print zeros/NaNs when no count histogram has been collected * A 10% faster CUDA vector dot kernel for Q3_K Q3_K is now running at ~18.5 ms / token on CUDA, so the gap to Q4_0 is only 10%. It seems memory acccess pattern is more important for performance than the amount of computation the kernel does. * A slightly daster Q4_K AVX2 dot product For perplexity, where we are less memory bound, time per pass drops by ~5%. Barely measurable difference for single token prediction. * A slightly faster ARM_NEON A4_K dot product * Minor * Fix quantization error test We cannot possibly be expecting rmse < 0.002 for 2- and 3-bit quantization variants. * Fix docker build I have been sloppy with vector reinterpret casts on ARM_NEON. It seems clang is very forgiving in that regard. * Added forgotten ggml.o dependence on k_quants.h to the Makefile * Had unintentionally committed the Makefile with -Ofast enabled * ggml : rename k_quants -> ggml-quants-k, use lowercase in code --------- Co-authored-by: Iwan Kawrakow <iwan.kawrakow@gmail.com> Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
2023-06-05 19:56:18 +00:00
}
ggml : quantization refactoring (#3833) * ggml : factor all quantization code in ggml-quants ggml-ci * ggml-quants : fix Zig and Swift builds + quantize tool ggml-ci * quantize : --pure option for disabling k-quant mixtures --------- Co-authored-by: cebtenzzre <cebtenzzre@gmail.com>
2023-10-29 16:32:28 +00:00
*s = hsum_float_8(acc) + summs;
#elif defined(__AVX__)
// Initialize accumulator with zeros
__m256 acc = _mm256_setzero_ps();
__m128i mask = _mm_set1_epi8(0x10);
float summs = 0.0f;
// Main loop
for (int i = 0; i < nb; i++) {
ggml : move FP16 <-> FP32 code to ggml-impl.h (#3861) * ggml : move FP16 <-> FP32 stuff to ggml-impl.h ggml-ci * tests : fix ARM build * ggml : explicitly initialize deprecated type traits * ggml : add math.h to ggml-impl.h * ggml : remove duplicate static assert macros * ggml : prefix lookup tables with ggml_ ggml-ci * ggml-impl : move extern "C" to start of file
2023-10-30 17:19:15 +00:00
const __m256 dx = _mm256_set1_ps(GGML_FP16_TO_FP32(x[i].d));
ggml : quantization refactoring (#3833) * ggml : factor all quantization code in ggml-quants ggml-ci * ggml-quants : fix Zig and Swift builds + quantize tool ggml-ci * quantize : --pure option for disabling k-quant mixtures --------- Co-authored-by: cebtenzzre <cebtenzzre@gmail.com>
2023-10-29 16:32:28 +00:00
ggml : move FP16 <-> FP32 code to ggml-impl.h (#3861) * ggml : move FP16 <-> FP32 stuff to ggml-impl.h ggml-ci * tests : fix ARM build * ggml : explicitly initialize deprecated type traits * ggml : add math.h to ggml-impl.h * ggml : remove duplicate static assert macros * ggml : prefix lookup tables with ggml_ ggml-ci * ggml-impl : move extern "C" to start of file
2023-10-30 17:19:15 +00:00
summs += GGML_FP16_TO_FP32(x[i].m) * y[i].s;
ggml : quantization refactoring (#3833) * ggml : factor all quantization code in ggml-quants ggml-ci * ggml-quants : fix Zig and Swift builds + quantize tool ggml-ci * quantize : --pure option for disabling k-quant mixtures --------- Co-authored-by: cebtenzzre <cebtenzzre@gmail.com>
2023-10-29 16:32:28 +00:00
__m256i bx = bytes_from_nibbles_32(x[i].qs);
const __m256i bxhi = bytes_from_bits_32(x[i].qh);
__m128i bxhil = _mm256_castsi256_si128(bxhi);
__m128i bxhih = _mm256_extractf128_si256(bxhi, 1);
bxhil = _mm_and_si128(bxhil, mask);
bxhih = _mm_and_si128(bxhih, mask);
__m128i bxl = _mm256_castsi256_si128(bx);
__m128i bxh = _mm256_extractf128_si256(bx, 1);
bxl = _mm_or_si128(bxl, bxhil);
bxh = _mm_or_si128(bxh, bxhih);
bx = MM256_SET_M128I(bxh, bxl);
ggml : add SOTA 2,3,4,5,6 bit k-quantizations (#1684) * Starting to add k-quantization to ggml I think it is better to have quantization separate from ggml. For now just adding the k-quants there, but it would be better to also factor out the existing ggml quantizations. * Adding Q3_K and Q8_K (de)-quantization * Q3_K now working on CUDA and AVX2/scalar CUDA is not ideal - ~50% slower than Q4_0 for single token prediction, about the same in batch mode (perplexity). CPU single token is ~55 ms (on Ryzen 7950X). * Some improvement for Q3_K on CUDA It is now ~22.5 ms/token on my GPU, so ~30% slower than Q4_0. * Some more CUDA optimizations for Q3_K Single token is now 20.5 ms/token (~20% slower than Q4_0). Perplexity is on par with Q4_0. * Adding Q4_K - scalar, AVX2, CUDA Performance is the same or perhaps very slightly better than Q4_0 on the CPU. On the GPU, single token prediction is ~10% better than Q4_0, batch mode (perplexity is about the same). * Adding Q6_K - scalar, AVX2, CUDA Performance is ~40% lower compared to Q4_K on the CPU. This is to be expected, considering that we are memory bound on the CPU and the 6-bit model is ~44% larger than the 4-bit. On the GPU, single token prediction is ~6% lower than Q4_0, batch mode (perplexity) is even closer (but still slower). * Adding Q5_K - scalar, AVX2, CUDA Performance is ~20% lower compared to Q4_K on the CPU. This is to be expected, considering that we are memory bound on the CPU and the 5-bit model is ~22% larger than the 4-bit. On the GPU, single token prediction is about the same as Q4_0 for both, single token and batch prediction. * Per convention, all QX_K quantizations use Q5_K for output.weight * Adding quantization mixes * Quantization mixes: didn't quite get what I wanted in the last commit * Q4_K dot product for ARM_NEON * Q6_K dot product for ARM_NEON * Q5_K dot product for ARM_NEON * Adding Q3_K dot for ARM_NEON It is 22% slower than Q4_K, despite the smaller model size. On x86_64, where we are memory bound, the Q3_K model is quite a bit faster than Q4_K. * A very slightly faster ARM_NEON Q3_K dot * Adding Q2_K - just CUDA for now Token prediction is pretty good - about 15.5 ms on a RTX 4080. Perplexity is about the same as Q4_K. * Adding scalar and AVX2 Q2_K dot * Adding ARM_NEON Q2_K dot About the same performance as Q4_K. * A slightly faster ARM_NEON Q2_K dot Single token prediction is now ~36 ms on M2 Max. The code is much simpler too. * Fixed bug in Q2_K CUDA dot product kernel Stranegly enough, for the few prompts I tried with the 7B model the responses looked perfectly reasonable. Only realized something is not quite right when I tried the larger models and started getting nonse back. In any case, Q2_K single token evaluation time on an RTX 4080 in a Ryzen7950X box iusing CUDA and model fully loaded on the GPU are ~15.5 ms for 7B, ~25.4 ms for 13B, and ~55.8 ms for 30B. The max number of layers that fit in VRAM for The 65B is 32. With that, we get ~330 ms per token, which is not that much faster than just running on the CPU (~470 ms per token). * Don't print zeros/NaNs when no count histogram has been collected * A 10% faster CUDA vector dot kernel for Q3_K Q3_K is now running at ~18.5 ms / token on CUDA, so the gap to Q4_0 is only 10%. It seems memory acccess pattern is more important for performance than the amount of computation the kernel does. * A slightly daster Q4_K AVX2 dot product For perplexity, where we are less memory bound, time per pass drops by ~5%. Barely measurable difference for single token prediction. * A slightly faster ARM_NEON A4_K dot product * Minor * Fix quantization error test We cannot possibly be expecting rmse < 0.002 for 2- and 3-bit quantization variants. * Fix docker build I have been sloppy with vector reinterpret casts on ARM_NEON. It seems clang is very forgiving in that regard. * Added forgotten ggml.o dependence on k_quants.h to the Makefile * Had unintentionally committed the Makefile with -Ofast enabled * ggml : rename k_quants -> ggml-quants-k, use lowercase in code --------- Co-authored-by: Iwan Kawrakow <iwan.kawrakow@gmail.com> Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
2023-06-05 19:56:18 +00:00
ggml : quantization refactoring (#3833) * ggml : factor all quantization code in ggml-quants ggml-ci * ggml-quants : fix Zig and Swift builds + quantize tool ggml-ci * quantize : --pure option for disabling k-quant mixtures --------- Co-authored-by: cebtenzzre <cebtenzzre@gmail.com>
2023-10-29 16:32:28 +00:00
const __m256 dy = _mm256_set1_ps(y[i].d);
const __m256i by = _mm256_loadu_si256((const __m256i *)y[i].qs);
k-quants : fix quantization ranges (#3646)
2023-10-17 06:19:28 +00:00
ggml : quantization refactoring (#3833) * ggml : factor all quantization code in ggml-quants ggml-ci * ggml-quants : fix Zig and Swift builds + quantize tool ggml-ci * quantize : --pure option for disabling k-quant mixtures --------- Co-authored-by: cebtenzzre <cebtenzzre@gmail.com>
2023-10-29 16:32:28 +00:00
const __m256 q = mul_sum_us8_pairs_float(bx, by);
acc = _mm256_add_ps(_mm256_mul_ps(q, _mm256_mul_ps(dx, dy)), acc);
ggml : add SOTA 2,3,4,5,6 bit k-quantizations (#1684) * Starting to add k-quantization to ggml I think it is better to have quantization separate from ggml. For now just adding the k-quants there, but it would be better to also factor out the existing ggml quantizations. * Adding Q3_K and Q8_K (de)-quantization * Q3_K now working on CUDA and AVX2/scalar CUDA is not ideal - ~50% slower than Q4_0 for single token prediction, about the same in batch mode (perplexity). CPU single token is ~55 ms (on Ryzen 7950X). * Some improvement for Q3_K on CUDA It is now ~22.5 ms/token on my GPU, so ~30% slower than Q4_0. * Some more CUDA optimizations for Q3_K Single token is now 20.5 ms/token (~20% slower than Q4_0). Perplexity is on par with Q4_0. * Adding Q4_K - scalar, AVX2, CUDA Performance is the same or perhaps very slightly better than Q4_0 on the CPU. On the GPU, single token prediction is ~10% better than Q4_0, batch mode (perplexity is about the same). * Adding Q6_K - scalar, AVX2, CUDA Performance is ~40% lower compared to Q4_K on the CPU. This is to be expected, considering that we are memory bound on the CPU and the 6-bit model is ~44% larger than the 4-bit. On the GPU, single token prediction is ~6% lower than Q4_0, batch mode (perplexity) is even closer (but still slower). * Adding Q5_K - scalar, AVX2, CUDA Performance is ~20% lower compared to Q4_K on the CPU. This is to be expected, considering that we are memory bound on the CPU and the 5-bit model is ~22% larger than the 4-bit. On the GPU, single token prediction is about the same as Q4_0 for both, single token and batch prediction. * Per convention, all QX_K quantizations use Q5_K for output.weight * Adding quantization mixes * Quantization mixes: didn't quite get what I wanted in the last commit * Q4_K dot product for ARM_NEON * Q6_K dot product for ARM_NEON * Q5_K dot product for ARM_NEON * Adding Q3_K dot for ARM_NEON It is 22% slower than Q4_K, despite the smaller model size. On x86_64, where we are memory bound, the Q3_K model is quite a bit faster than Q4_K. * A very slightly faster ARM_NEON Q3_K dot * Adding Q2_K - just CUDA for now Token prediction is pretty good - about 15.5 ms on a RTX 4080. Perplexity is about the same as Q4_K. * Adding scalar and AVX2 Q2_K dot * Adding ARM_NEON Q2_K dot About the same performance as Q4_K. * A slightly faster ARM_NEON Q2_K dot Single token prediction is now ~36 ms on M2 Max. The code is much simpler too. * Fixed bug in Q2_K CUDA dot product kernel Stranegly enough, for the few prompts I tried with the 7B model the responses looked perfectly reasonable. Only realized something is not quite right when I tried the larger models and started getting nonse back. In any case, Q2_K single token evaluation time on an RTX 4080 in a Ryzen7950X box iusing CUDA and model fully loaded on the GPU are ~15.5 ms for 7B, ~25.4 ms for 13B, and ~55.8 ms for 30B. The max number of layers that fit in VRAM for The 65B is 32. With that, we get ~330 ms per token, which is not that much faster than just running on the CPU (~470 ms per token). * Don't print zeros/NaNs when no count histogram has been collected * A 10% faster CUDA vector dot kernel for Q3_K Q3_K is now running at ~18.5 ms / token on CUDA, so the gap to Q4_0 is only 10%. It seems memory acccess pattern is more important for performance than the amount of computation the kernel does. * A slightly daster Q4_K AVX2 dot product For perplexity, where we are less memory bound, time per pass drops by ~5%. Barely measurable difference for single token prediction. * A slightly faster ARM_NEON A4_K dot product * Minor * Fix quantization error test We cannot possibly be expecting rmse < 0.002 for 2- and 3-bit quantization variants. * Fix docker build I have been sloppy with vector reinterpret casts on ARM_NEON. It seems clang is very forgiving in that regard. * Added forgotten ggml.o dependence on k_quants.h to the Makefile * Had unintentionally committed the Makefile with -Ofast enabled * ggml : rename k_quants -> ggml-quants-k, use lowercase in code --------- Co-authored-by: Iwan Kawrakow <iwan.kawrakow@gmail.com> Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
2023-06-05 19:56:18 +00:00
}
ggml : quantization refactoring (#3833) * ggml : factor all quantization code in ggml-quants ggml-ci * ggml-quants : fix Zig and Swift builds + quantize tool ggml-ci * quantize : --pure option for disabling k-quant mixtures --------- Co-authored-by: cebtenzzre <cebtenzzre@gmail.com>
2023-10-29 16:32:28 +00:00
*s = hsum_float_8(acc) + summs;
#elif defined(__riscv_v_intrinsic)
float sumf = 0.0;
ggml : add SOTA 2,3,4,5,6 bit k-quantizations (#1684) * Starting to add k-quantization to ggml I think it is better to have quantization separate from ggml. For now just adding the k-quants there, but it would be better to also factor out the existing ggml quantizations. * Adding Q3_K and Q8_K (de)-quantization * Q3_K now working on CUDA and AVX2/scalar CUDA is not ideal - ~50% slower than Q4_0 for single token prediction, about the same in batch mode (perplexity). CPU single token is ~55 ms (on Ryzen 7950X). * Some improvement for Q3_K on CUDA It is now ~22.5 ms/token on my GPU, so ~30% slower than Q4_0. * Some more CUDA optimizations for Q3_K Single token is now 20.5 ms/token (~20% slower than Q4_0). Perplexity is on par with Q4_0. * Adding Q4_K - scalar, AVX2, CUDA Performance is the same or perhaps very slightly better than Q4_0 on the CPU. On the GPU, single token prediction is ~10% better than Q4_0, batch mode (perplexity is about the same). * Adding Q6_K - scalar, AVX2, CUDA Performance is ~40% lower compared to Q4_K on the CPU. This is to be expected, considering that we are memory bound on the CPU and the 6-bit model is ~44% larger than the 4-bit. On the GPU, single token prediction is ~6% lower than Q4_0, batch mode (perplexity) is even closer (but still slower). * Adding Q5_K - scalar, AVX2, CUDA Performance is ~20% lower compared to Q4_K on the CPU. This is to be expected, considering that we are memory bound on the CPU and the 5-bit model is ~22% larger than the 4-bit. On the GPU, single token prediction is about the same as Q4_0 for both, single token and batch prediction. * Per convention, all QX_K quantizations use Q5_K for output.weight * Adding quantization mixes * Quantization mixes: didn't quite get what I wanted in the last commit * Q4_K dot product for ARM_NEON * Q6_K dot product for ARM_NEON * Q5_K dot product for ARM_NEON * Adding Q3_K dot for ARM_NEON It is 22% slower than Q4_K, despite the smaller model size. On x86_64, where we are memory bound, the Q3_K model is quite a bit faster than Q4_K. * A very slightly faster ARM_NEON Q3_K dot * Adding Q2_K - just CUDA for now Token prediction is pretty good - about 15.5 ms on a RTX 4080. Perplexity is about the same as Q4_K. * Adding scalar and AVX2 Q2_K dot * Adding ARM_NEON Q2_K dot About the same performance as Q4_K. * A slightly faster ARM_NEON Q2_K dot Single token prediction is now ~36 ms on M2 Max. The code is much simpler too. * Fixed bug in Q2_K CUDA dot product kernel Stranegly enough, for the few prompts I tried with the 7B model the responses looked perfectly reasonable. Only realized something is not quite right when I tried the larger models and started getting nonse back. In any case, Q2_K single token evaluation time on an RTX 4080 in a Ryzen7950X box iusing CUDA and model fully loaded on the GPU are ~15.5 ms for 7B, ~25.4 ms for 13B, and ~55.8 ms for 30B. The max number of layers that fit in VRAM for The 65B is 32. With that, we get ~330 ms per token, which is not that much faster than just running on the CPU (~470 ms per token). * Don't print zeros/NaNs when no count histogram has been collected * A 10% faster CUDA vector dot kernel for Q3_K Q3_K is now running at ~18.5 ms / token on CUDA, so the gap to Q4_0 is only 10%. It seems memory acccess pattern is more important for performance than the amount of computation the kernel does. * A slightly daster Q4_K AVX2 dot product For perplexity, where we are less memory bound, time per pass drops by ~5%. Barely measurable difference for single token prediction. * A slightly faster ARM_NEON A4_K dot product * Minor * Fix quantization error test We cannot possibly be expecting rmse < 0.002 for 2- and 3-bit quantization variants. * Fix docker build I have been sloppy with vector reinterpret casts on ARM_NEON. It seems clang is very forgiving in that regard. * Added forgotten ggml.o dependence on k_quants.h to the Makefile * Had unintentionally committed the Makefile with -Ofast enabled * ggml : rename k_quants -> ggml-quants-k, use lowercase in code --------- Co-authored-by: Iwan Kawrakow <iwan.kawrakow@gmail.com> Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
2023-06-05 19:56:18 +00:00
ggml : quantization refactoring (#3833) * ggml : factor all quantization code in ggml-quants ggml-ci * ggml-quants : fix Zig and Swift builds + quantize tool ggml-ci * quantize : --pure option for disabling k-quant mixtures --------- Co-authored-by: cebtenzzre <cebtenzzre@gmail.com>
2023-10-29 16:32:28 +00:00
uint32_t qh;
ggml : add SOTA 2,3,4,5,6 bit k-quantizations (#1684) * Starting to add k-quantization to ggml I think it is better to have quantization separate from ggml. For now just adding the k-quants there, but it would be better to also factor out the existing ggml quantizations. * Adding Q3_K and Q8_K (de)-quantization * Q3_K now working on CUDA and AVX2/scalar CUDA is not ideal - ~50% slower than Q4_0 for single token prediction, about the same in batch mode (perplexity). CPU single token is ~55 ms (on Ryzen 7950X). * Some improvement for Q3_K on CUDA It is now ~22.5 ms/token on my GPU, so ~30% slower than Q4_0. * Some more CUDA optimizations for Q3_K Single token is now 20.5 ms/token (~20% slower than Q4_0). Perplexity is on par with Q4_0. * Adding Q4_K - scalar, AVX2, CUDA Performance is the same or perhaps very slightly better than Q4_0 on the CPU. On the GPU, single token prediction is ~10% better than Q4_0, batch mode (perplexity is about the same). * Adding Q6_K - scalar, AVX2, CUDA Performance is ~40% lower compared to Q4_K on the CPU. This is to be expected, considering that we are memory bound on the CPU and the 6-bit model is ~44% larger than the 4-bit. On the GPU, single token prediction is ~6% lower than Q4_0, batch mode (perplexity) is even closer (but still slower). * Adding Q5_K - scalar, AVX2, CUDA Performance is ~20% lower compared to Q4_K on the CPU. This is to be expected, considering that we are memory bound on the CPU and the 5-bit model is ~22% larger than the 4-bit. On the GPU, single token prediction is about the same as Q4_0 for both, single token and batch prediction. * Per convention, all QX_K quantizations use Q5_K for output.weight * Adding quantization mixes * Quantization mixes: didn't quite get what I wanted in the last commit * Q4_K dot product for ARM_NEON * Q6_K dot product for ARM_NEON * Q5_K dot product for ARM_NEON * Adding Q3_K dot for ARM_NEON It is 22% slower than Q4_K, despite the smaller model size. On x86_64, where we are memory bound, the Q3_K model is quite a bit faster than Q4_K. * A very slightly faster ARM_NEON Q3_K dot * Adding Q2_K - just CUDA for now Token prediction is pretty good - about 15.5 ms on a RTX 4080. Perplexity is about the same as Q4_K. * Adding scalar and AVX2 Q2_K dot * Adding ARM_NEON Q2_K dot About the same performance as Q4_K. * A slightly faster ARM_NEON Q2_K dot Single token prediction is now ~36 ms on M2 Max. The code is much simpler too. * Fixed bug in Q2_K CUDA dot product kernel Stranegly enough, for the few prompts I tried with the 7B model the responses looked perfectly reasonable. Only realized something is not quite right when I tried the larger models and started getting nonse back. In any case, Q2_K single token evaluation time on an RTX 4080 in a Ryzen7950X box iusing CUDA and model fully loaded on the GPU are ~15.5 ms for 7B, ~25.4 ms for 13B, and ~55.8 ms for 30B. The max number of layers that fit in VRAM for The 65B is 32. With that, we get ~330 ms per token, which is not that much faster than just running on the CPU (~470 ms per token). * Don't print zeros/NaNs when no count histogram has been collected * A 10% faster CUDA vector dot kernel for Q3_K Q3_K is now running at ~18.5 ms / token on CUDA, so the gap to Q4_0 is only 10%. It seems memory acccess pattern is more important for performance than the amount of computation the kernel does. * A slightly daster Q4_K AVX2 dot product For perplexity, where we are less memory bound, time per pass drops by ~5%. Barely measurable difference for single token prediction. * A slightly faster ARM_NEON A4_K dot product * Minor * Fix quantization error test We cannot possibly be expecting rmse < 0.002 for 2- and 3-bit quantization variants. * Fix docker build I have been sloppy with vector reinterpret casts on ARM_NEON. It seems clang is very forgiving in that regard. * Added forgotten ggml.o dependence on k_quants.h to the Makefile * Had unintentionally committed the Makefile with -Ofast enabled * ggml : rename k_quants -> ggml-quants-k, use lowercase in code --------- Co-authored-by: Iwan Kawrakow <iwan.kawrakow@gmail.com> Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
2023-06-05 19:56:18 +00:00
ggml : quantization refactoring (#3833) * ggml : factor all quantization code in ggml-quants ggml-ci * ggml-quants : fix Zig and Swift builds + quantize tool ggml-ci * quantize : --pure option for disabling k-quant mixtures --------- Co-authored-by: cebtenzzre <cebtenzzre@gmail.com>
2023-10-29 16:32:28 +00:00
size_t vl = __riscv_vsetvl_e8m1(qk/2);
// temporary registers for shift operations
vuint32m2_t vt_1 = __riscv_vid_v_u32m2(vl);
vuint32m2_t vt_2 = __riscv_vadd_vx_u32m2(vt_1, 12, vl);
ggml : add SOTA 2,3,4,5,6 bit k-quantizations (#1684) * Starting to add k-quantization to ggml I think it is better to have quantization separate from ggml. For now just adding the k-quants there, but it would be better to also factor out the existing ggml quantizations. * Adding Q3_K and Q8_K (de)-quantization * Q3_K now working on CUDA and AVX2/scalar CUDA is not ideal - ~50% slower than Q4_0 for single token prediction, about the same in batch mode (perplexity). CPU single token is ~55 ms (on Ryzen 7950X). * Some improvement for Q3_K on CUDA It is now ~22.5 ms/token on my GPU, so ~30% slower than Q4_0. * Some more CUDA optimizations for Q3_K Single token is now 20.5 ms/token (~20% slower than Q4_0). Perplexity is on par with Q4_0. * Adding Q4_K - scalar, AVX2, CUDA Performance is the same or perhaps very slightly better than Q4_0 on the CPU. On the GPU, single token prediction is ~10% better than Q4_0, batch mode (perplexity is about the same). * Adding Q6_K - scalar, AVX2, CUDA Performance is ~40% lower compared to Q4_K on the CPU. This is to be expected, considering that we are memory bound on the CPU and the 6-bit model is ~44% larger than the 4-bit. On the GPU, single token prediction is ~6% lower than Q4_0, batch mode (perplexity) is even closer (but still slower). * Adding Q5_K - scalar, AVX2, CUDA Performance is ~20% lower compared to Q4_K on the CPU. This is to be expected, considering that we are memory bound on the CPU and the 5-bit model is ~22% larger than the 4-bit. On the GPU, single token prediction is about the same as Q4_0 for both, single token and batch prediction. * Per convention, all QX_K quantizations use Q5_K for output.weight * Adding quantization mixes * Quantization mixes: didn't quite get what I wanted in the last commit * Q4_K dot product for ARM_NEON * Q6_K dot product for ARM_NEON * Q5_K dot product for ARM_NEON * Adding Q3_K dot for ARM_NEON It is 22% slower than Q4_K, despite the smaller model size. On x86_64, where we are memory bound, the Q3_K model is quite a bit faster than Q4_K. * A very slightly faster ARM_NEON Q3_K dot * Adding Q2_K - just CUDA for now Token prediction is pretty good - about 15.5 ms on a RTX 4080. Perplexity is about the same as Q4_K. * Adding scalar and AVX2 Q2_K dot * Adding ARM_NEON Q2_K dot About the same performance as Q4_K. * A slightly faster ARM_NEON Q2_K dot Single token prediction is now ~36 ms on M2 Max. The code is much simpler too. * Fixed bug in Q2_K CUDA dot product kernel Stranegly enough, for the few prompts I tried with the 7B model the responses looked perfectly reasonable. Only realized something is not quite right when I tried the larger models and started getting nonse back. In any case, Q2_K single token evaluation time on an RTX 4080 in a Ryzen7950X box iusing CUDA and model fully loaded on the GPU are ~15.5 ms for 7B, ~25.4 ms for 13B, and ~55.8 ms for 30B. The max number of layers that fit in VRAM for The 65B is 32. With that, we get ~330 ms per token, which is not that much faster than just running on the CPU (~470 ms per token). * Don't print zeros/NaNs when no count histogram has been collected * A 10% faster CUDA vector dot kernel for Q3_K Q3_K is now running at ~18.5 ms / token on CUDA, so the gap to Q4_0 is only 10%. It seems memory acccess pattern is more important for performance than the amount of computation the kernel does. * A slightly daster Q4_K AVX2 dot product For perplexity, where we are less memory bound, time per pass drops by ~5%. Barely measurable difference for single token prediction. * A slightly faster ARM_NEON A4_K dot product * Minor * Fix quantization error test We cannot possibly be expecting rmse < 0.002 for 2- and 3-bit quantization variants. * Fix docker build I have been sloppy with vector reinterpret casts on ARM_NEON. It seems clang is very forgiving in that regard. * Added forgotten ggml.o dependence on k_quants.h to the Makefile * Had unintentionally committed the Makefile with -Ofast enabled * ggml : rename k_quants -> ggml-quants-k, use lowercase in code --------- Co-authored-by: Iwan Kawrakow <iwan.kawrakow@gmail.com> Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
2023-06-05 19:56:18 +00:00
for (int i = 0; i < nb; i++) {
ggml : quantization refactoring (#3833) * ggml : factor all quantization code in ggml-quants ggml-ci * ggml-quants : fix Zig and Swift builds + quantize tool ggml-ci * quantize : --pure option for disabling k-quant mixtures --------- Co-authored-by: cebtenzzre <cebtenzzre@gmail.com>
2023-10-29 16:32:28 +00:00
memcpy(&qh, x[i].qh, sizeof(uint32_t));
ggml : add SOTA 2,3,4,5,6 bit k-quantizations (#1684) * Starting to add k-quantization to ggml I think it is better to have quantization separate from ggml. For now just adding the k-quants there, but it would be better to also factor out the existing ggml quantizations. * Adding Q3_K and Q8_K (de)-quantization * Q3_K now working on CUDA and AVX2/scalar CUDA is not ideal - ~50% slower than Q4_0 for single token prediction, about the same in batch mode (perplexity). CPU single token is ~55 ms (on Ryzen 7950X). * Some improvement for Q3_K on CUDA It is now ~22.5 ms/token on my GPU, so ~30% slower than Q4_0. * Some more CUDA optimizations for Q3_K Single token is now 20.5 ms/token (~20% slower than Q4_0). Perplexity is on par with Q4_0. * Adding Q4_K - scalar, AVX2, CUDA Performance is the same or perhaps very slightly better than Q4_0 on the CPU. On the GPU, single token prediction is ~10% better than Q4_0, batch mode (perplexity is about the same). * Adding Q6_K - scalar, AVX2, CUDA Performance is ~40% lower compared to Q4_K on the CPU. This is to be expected, considering that we are memory bound on the CPU and the 6-bit model is ~44% larger than the 4-bit. On the GPU, single token prediction is ~6% lower than Q4_0, batch mode (perplexity) is even closer (but still slower). * Adding Q5_K - scalar, AVX2, CUDA Performance is ~20% lower compared to Q4_K on the CPU. This is to be expected, considering that we are memory bound on the CPU and the 5-bit model is ~22% larger than the 4-bit. On the GPU, single token prediction is about the same as Q4_0 for both, single token and batch prediction. * Per convention, all QX_K quantizations use Q5_K for output.weight * Adding quantization mixes * Quantization mixes: didn't quite get what I wanted in the last commit * Q4_K dot product for ARM_NEON * Q6_K dot product for ARM_NEON * Q5_K dot product for ARM_NEON * Adding Q3_K dot for ARM_NEON It is 22% slower than Q4_K, despite the smaller model size. On x86_64, where we are memory bound, the Q3_K model is quite a bit faster than Q4_K. * A very slightly faster ARM_NEON Q3_K dot * Adding Q2_K - just CUDA for now Token prediction is pretty good - about 15.5 ms on a RTX 4080. Perplexity is about the same as Q4_K. * Adding scalar and AVX2 Q2_K dot * Adding ARM_NEON Q2_K dot About the same performance as Q4_K. * A slightly faster ARM_NEON Q2_K dot Single token prediction is now ~36 ms on M2 Max. The code is much simpler too. * Fixed bug in Q2_K CUDA dot product kernel Stranegly enough, for the few prompts I tried with the 7B model the responses looked perfectly reasonable. Only realized something is not quite right when I tried the larger models and started getting nonse back. In any case, Q2_K single token evaluation time on an RTX 4080 in a Ryzen7950X box iusing CUDA and model fully loaded on the GPU are ~15.5 ms for 7B, ~25.4 ms for 13B, and ~55.8 ms for 30B. The max number of layers that fit in VRAM for The 65B is 32. With that, we get ~330 ms per token, which is not that much faster than just running on the CPU (~470 ms per token). * Don't print zeros/NaNs when no count histogram has been collected * A 10% faster CUDA vector dot kernel for Q3_K Q3_K is now running at ~18.5 ms / token on CUDA, so the gap to Q4_0 is only 10%. It seems memory acccess pattern is more important for performance than the amount of computation the kernel does. * A slightly daster Q4_K AVX2 dot product For perplexity, where we are less memory bound, time per pass drops by ~5%. Barely measurable difference for single token prediction. * A slightly faster ARM_NEON A4_K dot product * Minor * Fix quantization error test We cannot possibly be expecting rmse < 0.002 for 2- and 3-bit quantization variants. * Fix docker build I have been sloppy with vector reinterpret casts on ARM_NEON. It seems clang is very forgiving in that regard. * Added forgotten ggml.o dependence on k_quants.h to the Makefile * Had unintentionally committed the Makefile with -Ofast enabled * ggml : rename k_quants -> ggml-quants-k, use lowercase in code --------- Co-authored-by: Iwan Kawrakow <iwan.kawrakow@gmail.com> Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
2023-06-05 19:56:18 +00:00
ggml : quantization refactoring (#3833) * ggml : factor all quantization code in ggml-quants ggml-ci * ggml-quants : fix Zig and Swift builds + quantize tool ggml-ci * quantize : --pure option for disabling k-quant mixtures --------- Co-authored-by: cebtenzzre <cebtenzzre@gmail.com>
2023-10-29 16:32:28 +00:00
// load qh
vuint32m2_t vqh = __riscv_vmv_v_x_u32m2(qh, vl);
ggml : add SOTA 2,3,4,5,6 bit k-quantizations (#1684) * Starting to add k-quantization to ggml I think it is better to have quantization separate from ggml. For now just adding the k-quants there, but it would be better to also factor out the existing ggml quantizations. * Adding Q3_K and Q8_K (de)-quantization * Q3_K now working on CUDA and AVX2/scalar CUDA is not ideal - ~50% slower than Q4_0 for single token prediction, about the same in batch mode (perplexity). CPU single token is ~55 ms (on Ryzen 7950X). * Some improvement for Q3_K on CUDA It is now ~22.5 ms/token on my GPU, so ~30% slower than Q4_0. * Some more CUDA optimizations for Q3_K Single token is now 20.5 ms/token (~20% slower than Q4_0). Perplexity is on par with Q4_0. * Adding Q4_K - scalar, AVX2, CUDA Performance is the same or perhaps very slightly better than Q4_0 on the CPU. On the GPU, single token prediction is ~10% better than Q4_0, batch mode (perplexity is about the same). * Adding Q6_K - scalar, AVX2, CUDA Performance is ~40% lower compared to Q4_K on the CPU. This is to be expected, considering that we are memory bound on the CPU and the 6-bit model is ~44% larger than the 4-bit. On the GPU, single token prediction is ~6% lower than Q4_0, batch mode (perplexity) is even closer (but still slower). * Adding Q5_K - scalar, AVX2, CUDA Performance is ~20% lower compared to Q4_K on the CPU. This is to be expected, considering that we are memory bound on the CPU and the 5-bit model is ~22% larger than the 4-bit. On the GPU, single token prediction is about the same as Q4_0 for both, single token and batch prediction. * Per convention, all QX_K quantizations use Q5_K for output.weight * Adding quantization mixes * Quantization mixes: didn't quite get what I wanted in the last commit * Q4_K dot product for ARM_NEON * Q6_K dot product for ARM_NEON * Q5_K dot product for ARM_NEON * Adding Q3_K dot for ARM_NEON It is 22% slower than Q4_K, despite the smaller model size. On x86_64, where we are memory bound, the Q3_K model is quite a bit faster than Q4_K. * A very slightly faster ARM_NEON Q3_K dot * Adding Q2_K - just CUDA for now Token prediction is pretty good - about 15.5 ms on a RTX 4080. Perplexity is about the same as Q4_K. * Adding scalar and AVX2 Q2_K dot * Adding ARM_NEON Q2_K dot About the same performance as Q4_K. * A slightly faster ARM_NEON Q2_K dot Single token prediction is now ~36 ms on M2 Max. The code is much simpler too. * Fixed bug in Q2_K CUDA dot product kernel Stranegly enough, for the few prompts I tried with the 7B model the responses looked perfectly reasonable. Only realized something is not quite right when I tried the larger models and started getting nonse back. In any case, Q2_K single token evaluation time on an RTX 4080 in a Ryzen7950X box iusing CUDA and model fully loaded on the GPU are ~15.5 ms for 7B, ~25.4 ms for 13B, and ~55.8 ms for 30B. The max number of layers that fit in VRAM for The 65B is 32. With that, we get ~330 ms per token, which is not that much faster than just running on the CPU (~470 ms per token). * Don't print zeros/NaNs when no count histogram has been collected * A 10% faster CUDA vector dot kernel for Q3_K Q3_K is now running at ~18.5 ms / token on CUDA, so the gap to Q4_0 is only 10%. It seems memory acccess pattern is more important for performance than the amount of computation the kernel does. * A slightly daster Q4_K AVX2 dot product For perplexity, where we are less memory bound, time per pass drops by ~5%. Barely measurable difference for single token prediction. * A slightly faster ARM_NEON A4_K dot product * Minor * Fix quantization error test We cannot possibly be expecting rmse < 0.002 for 2- and 3-bit quantization variants. * Fix docker build I have been sloppy with vector reinterpret casts on ARM_NEON. It seems clang is very forgiving in that regard. * Added forgotten ggml.o dependence on k_quants.h to the Makefile * Had unintentionally committed the Makefile with -Ofast enabled * ggml : rename k_quants -> ggml-quants-k, use lowercase in code --------- Co-authored-by: Iwan Kawrakow <iwan.kawrakow@gmail.com> Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
2023-06-05 19:56:18 +00:00
ggml : quantization refactoring (#3833) * ggml : factor all quantization code in ggml-quants ggml-ci * ggml-quants : fix Zig and Swift builds + quantize tool ggml-ci * quantize : --pure option for disabling k-quant mixtures --------- Co-authored-by: cebtenzzre <cebtenzzre@gmail.com>
2023-10-29 16:32:28 +00:00
// ((qh >> (j + 0)) << 4) & 0x10;
vuint32m2_t xhr_0 = __riscv_vsrl_vv_u32m2(vqh, vt_1, vl);
vuint32m2_t xhl_0 = __riscv_vsll_vx_u32m2(xhr_0, 4, vl);
vuint32m2_t xha_0 = __riscv_vand_vx_u32m2(xhl_0, 0x10, vl);
ggml : add SOTA 2,3,4,5,6 bit k-quantizations (#1684) * Starting to add k-quantization to ggml I think it is better to have quantization separate from ggml. For now just adding the k-quants there, but it would be better to also factor out the existing ggml quantizations. * Adding Q3_K and Q8_K (de)-quantization * Q3_K now working on CUDA and AVX2/scalar CUDA is not ideal - ~50% slower than Q4_0 for single token prediction, about the same in batch mode (perplexity). CPU single token is ~55 ms (on Ryzen 7950X). * Some improvement for Q3_K on CUDA It is now ~22.5 ms/token on my GPU, so ~30% slower than Q4_0. * Some more CUDA optimizations for Q3_K Single token is now 20.5 ms/token (~20% slower than Q4_0). Perplexity is on par with Q4_0. * Adding Q4_K - scalar, AVX2, CUDA Performance is the same or perhaps very slightly better than Q4_0 on the CPU. On the GPU, single token prediction is ~10% better than Q4_0, batch mode (perplexity is about the same). * Adding Q6_K - scalar, AVX2, CUDA Performance is ~40% lower compared to Q4_K on the CPU. This is to be expected, considering that we are memory bound on the CPU and the 6-bit model is ~44% larger than the 4-bit. On the GPU, single token prediction is ~6% lower than Q4_0, batch mode (perplexity) is even closer (but still slower). * Adding Q5_K - scalar, AVX2, CUDA Performance is ~20% lower compared to Q4_K on the CPU. This is to be expected, considering that we are memory bound on the CPU and the 5-bit model is ~22% larger than the 4-bit. On the GPU, single token prediction is about the same as Q4_0 for both, single token and batch prediction. * Per convention, all QX_K quantizations use Q5_K for output.weight * Adding quantization mixes * Quantization mixes: didn't quite get what I wanted in the last commit * Q4_K dot product for ARM_NEON * Q6_K dot product for ARM_NEON * Q5_K dot product for ARM_NEON * Adding Q3_K dot for ARM_NEON It is 22% slower than Q4_K, despite the smaller model size. On x86_64, where we are memory bound, the Q3_K model is quite a bit faster than Q4_K. * A very slightly faster ARM_NEON Q3_K dot * Adding Q2_K - just CUDA for now Token prediction is pretty good - about 15.5 ms on a RTX 4080. Perplexity is about the same as Q4_K. * Adding scalar and AVX2 Q2_K dot * Adding ARM_NEON Q2_K dot About the same performance as Q4_K. * A slightly faster ARM_NEON Q2_K dot Single token prediction is now ~36 ms on M2 Max. The code is much simpler too. * Fixed bug in Q2_K CUDA dot product kernel Stranegly enough, for the few prompts I tried with the 7B model the responses looked perfectly reasonable. Only realized something is not quite right when I tried the larger models and started getting nonse back. In any case, Q2_K single token evaluation time on an RTX 4080 in a Ryzen7950X box iusing CUDA and model fully loaded on the GPU are ~15.5 ms for 7B, ~25.4 ms for 13B, and ~55.8 ms for 30B. The max number of layers that fit in VRAM for The 65B is 32. With that, we get ~330 ms per token, which is not that much faster than just running on the CPU (~470 ms per token). * Don't print zeros/NaNs when no count histogram has been collected * A 10% faster CUDA vector dot kernel for Q3_K Q3_K is now running at ~18.5 ms / token on CUDA, so the gap to Q4_0 is only 10%. It seems memory acccess pattern is more important for performance than the amount of computation the kernel does. * A slightly daster Q4_K AVX2 dot product For perplexity, where we are less memory bound, time per pass drops by ~5%. Barely measurable difference for single token prediction. * A slightly faster ARM_NEON A4_K dot product * Minor * Fix quantization error test We cannot possibly be expecting rmse < 0.002 for 2- and 3-bit quantization variants. * Fix docker build I have been sloppy with vector reinterpret casts on ARM_NEON. It seems clang is very forgiving in that regard. * Added forgotten ggml.o dependence on k_quants.h to the Makefile * Had unintentionally committed the Makefile with -Ofast enabled * ggml : rename k_quants -> ggml-quants-k, use lowercase in code --------- Co-authored-by: Iwan Kawrakow <iwan.kawrakow@gmail.com> Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
2023-06-05 19:56:18 +00:00
ggml : quantization refactoring (#3833) * ggml : factor all quantization code in ggml-quants ggml-ci * ggml-quants : fix Zig and Swift builds + quantize tool ggml-ci * quantize : --pure option for disabling k-quant mixtures --------- Co-authored-by: cebtenzzre <cebtenzzre@gmail.com>
2023-10-29 16:32:28 +00:00
// ((qh >> (j + 12)) ) & 0x10;
vuint32m2_t xhr_1 = __riscv_vsrl_vv_u32m2(vqh, vt_2, vl);
vuint32m2_t xha_1 = __riscv_vand_vx_u32m2(xhr_1, 0x10, vl);
ggml : add SOTA 2,3,4,5,6 bit k-quantizations (#1684) * Starting to add k-quantization to ggml I think it is better to have quantization separate from ggml. For now just adding the k-quants there, but it would be better to also factor out the existing ggml quantizations. * Adding Q3_K and Q8_K (de)-quantization * Q3_K now working on CUDA and AVX2/scalar CUDA is not ideal - ~50% slower than Q4_0 for single token prediction, about the same in batch mode (perplexity). CPU single token is ~55 ms (on Ryzen 7950X). * Some improvement for Q3_K on CUDA It is now ~22.5 ms/token on my GPU, so ~30% slower than Q4_0. * Some more CUDA optimizations for Q3_K Single token is now 20.5 ms/token (~20% slower than Q4_0). Perplexity is on par with Q4_0. * Adding Q4_K - scalar, AVX2, CUDA Performance is the same or perhaps very slightly better than Q4_0 on the CPU. On the GPU, single token prediction is ~10% better than Q4_0, batch mode (perplexity is about the same). * Adding Q6_K - scalar, AVX2, CUDA Performance is ~40% lower compared to Q4_K on the CPU. This is to be expected, considering that we are memory bound on the CPU and the 6-bit model is ~44% larger than the 4-bit. On the GPU, single token prediction is ~6% lower than Q4_0, batch mode (perplexity) is even closer (but still slower). * Adding Q5_K - scalar, AVX2, CUDA Performance is ~20% lower compared to Q4_K on the CPU. This is to be expected, considering that we are memory bound on the CPU and the 5-bit model is ~22% larger than the 4-bit. On the GPU, single token prediction is about the same as Q4_0 for both, single token and batch prediction. * Per convention, all QX_K quantizations use Q5_K for output.weight * Adding quantization mixes * Quantization mixes: didn't quite get what I wanted in the last commit * Q4_K dot product for ARM_NEON * Q6_K dot product for ARM_NEON * Q5_K dot product for ARM_NEON * Adding Q3_K dot for ARM_NEON It is 22% slower than Q4_K, despite the smaller model size. On x86_64, where we are memory bound, the Q3_K model is quite a bit faster than Q4_K. * A very slightly faster ARM_NEON Q3_K dot * Adding Q2_K - just CUDA for now Token prediction is pretty good - about 15.5 ms on a RTX 4080. Perplexity is about the same as Q4_K. * Adding scalar and AVX2 Q2_K dot * Adding ARM_NEON Q2_K dot About the same performance as Q4_K. * A slightly faster ARM_NEON Q2_K dot Single token prediction is now ~36 ms on M2 Max. The code is much simpler too. * Fixed bug in Q2_K CUDA dot product kernel Stranegly enough, for the few prompts I tried with the 7B model the responses looked perfectly reasonable. Only realized something is not quite right when I tried the larger models and started getting nonse back. In any case, Q2_K single token evaluation time on an RTX 4080 in a Ryzen7950X box iusing CUDA and model fully loaded on the GPU are ~15.5 ms for 7B, ~25.4 ms for 13B, and ~55.8 ms for 30B. The max number of layers that fit in VRAM for The 65B is 32. With that, we get ~330 ms per token, which is not that much faster than just running on the CPU (~470 ms per token). * Don't print zeros/NaNs when no count histogram has been collected * A 10% faster CUDA vector dot kernel for Q3_K Q3_K is now running at ~18.5 ms / token on CUDA, so the gap to Q4_0 is only 10%. It seems memory acccess pattern is more important for performance than the amount of computation the kernel does. * A slightly daster Q4_K AVX2 dot product For perplexity, where we are less memory bound, time per pass drops by ~5%. Barely measurable difference for single token prediction. * A slightly faster ARM_NEON A4_K dot product * Minor * Fix quantization error test We cannot possibly be expecting rmse < 0.002 for 2- and 3-bit quantization variants. * Fix docker build I have been sloppy with vector reinterpret casts on ARM_NEON. It seems clang is very forgiving in that regard. * Added forgotten ggml.o dependence on k_quants.h to the Makefile * Had unintentionally committed the Makefile with -Ofast enabled * ggml : rename k_quants -> ggml-quants-k, use lowercase in code --------- Co-authored-by: Iwan Kawrakow <iwan.kawrakow@gmail.com> Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
2023-06-05 19:56:18 +00:00
ggml : quantization refactoring (#3833) * ggml : factor all quantization code in ggml-quants ggml-ci * ggml-quants : fix Zig and Swift builds + quantize tool ggml-ci * quantize : --pure option for disabling k-quant mixtures --------- Co-authored-by: cebtenzzre <cebtenzzre@gmail.com>
2023-10-29 16:32:28 +00:00
// narrowing
vuint16m1_t xhc_0 = __riscv_vncvt_x_x_w_u16m1(xha_0, vl);
vuint8mf2_t xh_0 = __riscv_vncvt_x_x_w_u8mf2(xhc_0, vl);
ggml : add SOTA 2,3,4,5,6 bit k-quantizations (#1684) * Starting to add k-quantization to ggml I think it is better to have quantization separate from ggml. For now just adding the k-quants there, but it would be better to also factor out the existing ggml quantizations. * Adding Q3_K and Q8_K (de)-quantization * Q3_K now working on CUDA and AVX2/scalar CUDA is not ideal - ~50% slower than Q4_0 for single token prediction, about the same in batch mode (perplexity). CPU single token is ~55 ms (on Ryzen 7950X). * Some improvement for Q3_K on CUDA It is now ~22.5 ms/token on my GPU, so ~30% slower than Q4_0. * Some more CUDA optimizations for Q3_K Single token is now 20.5 ms/token (~20% slower than Q4_0). Perplexity is on par with Q4_0. * Adding Q4_K - scalar, AVX2, CUDA Performance is the same or perhaps very slightly better than Q4_0 on the CPU. On the GPU, single token prediction is ~10% better than Q4_0, batch mode (perplexity is about the same). * Adding Q6_K - scalar, AVX2, CUDA Performance is ~40% lower compared to Q4_K on the CPU. This is to be expected, considering that we are memory bound on the CPU and the 6-bit model is ~44% larger than the 4-bit. On the GPU, single token prediction is ~6% lower than Q4_0, batch mode (perplexity) is even closer (but still slower). * Adding Q5_K - scalar, AVX2, CUDA Performance is ~20% lower compared to Q4_K on the CPU. This is to be expected, considering that we are memory bound on the CPU and the 5-bit model is ~22% larger than the 4-bit. On the GPU, single token prediction is about the same as Q4_0 for both, single token and batch prediction. * Per convention, all QX_K quantizations use Q5_K for output.weight * Adding quantization mixes * Quantization mixes: didn't quite get what I wanted in the last commit * Q4_K dot product for ARM_NEON * Q6_K dot product for ARM_NEON * Q5_K dot product for ARM_NEON * Adding Q3_K dot for ARM_NEON It is 22% slower than Q4_K, despite the smaller model size. On x86_64, where we are memory bound, the Q3_K model is quite a bit faster than Q4_K. * A very slightly faster ARM_NEON Q3_K dot * Adding Q2_K - just CUDA for now Token prediction is pretty good - about 15.5 ms on a RTX 4080. Perplexity is about the same as Q4_K. * Adding scalar and AVX2 Q2_K dot * Adding ARM_NEON Q2_K dot About the same performance as Q4_K. * A slightly faster ARM_NEON Q2_K dot Single token prediction is now ~36 ms on M2 Max. The code is much simpler too. * Fixed bug in Q2_K CUDA dot product kernel Stranegly enough, for the few prompts I tried with the 7B model the responses looked perfectly reasonable. Only realized something is not quite right when I tried the larger models and started getting nonse back. In any case, Q2_K single token evaluation time on an RTX 4080 in a Ryzen7950X box iusing CUDA and model fully loaded on the GPU are ~15.5 ms for 7B, ~25.4 ms for 13B, and ~55.8 ms for 30B. The max number of layers that fit in VRAM for The 65B is 32. With that, we get ~330 ms per token, which is not that much faster than just running on the CPU (~470 ms per token). * Don't print zeros/NaNs when no count histogram has been collected * A 10% faster CUDA vector dot kernel for Q3_K Q3_K is now running at ~18.5 ms / token on CUDA, so the gap to Q4_0 is only 10%. It seems memory acccess pattern is more important for performance than the amount of computation the kernel does. * A slightly daster Q4_K AVX2 dot product For perplexity, where we are less memory bound, time per pass drops by ~5%. Barely measurable difference for single token prediction. * A slightly faster ARM_NEON A4_K dot product * Minor * Fix quantization error test We cannot possibly be expecting rmse < 0.002 for 2- and 3-bit quantization variants. * Fix docker build I have been sloppy with vector reinterpret casts on ARM_NEON. It seems clang is very forgiving in that regard. * Added forgotten ggml.o dependence on k_quants.h to the Makefile * Had unintentionally committed the Makefile with -Ofast enabled * ggml : rename k_quants -> ggml-quants-k, use lowercase in code --------- Co-authored-by: Iwan Kawrakow <iwan.kawrakow@gmail.com> Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
2023-06-05 19:56:18 +00:00
ggml : quantization refactoring (#3833) * ggml : factor all quantization code in ggml-quants ggml-ci * ggml-quants : fix Zig and Swift builds + quantize tool ggml-ci * quantize : --pure option for disabling k-quant mixtures --------- Co-authored-by: cebtenzzre <cebtenzzre@gmail.com>
2023-10-29 16:32:28 +00:00
vuint16m1_t xhc_1 = __riscv_vncvt_x_x_w_u16m1(xha_1, vl);
vuint8mf2_t xh_1 = __riscv_vncvt_x_x_w_u8mf2(xhc_1, vl);
ggml : add SOTA 2,3,4,5,6 bit k-quantizations (#1684) * Starting to add k-quantization to ggml I think it is better to have quantization separate from ggml. For now just adding the k-quants there, but it would be better to also factor out the existing ggml quantizations. * Adding Q3_K and Q8_K (de)-quantization * Q3_K now working on CUDA and AVX2/scalar CUDA is not ideal - ~50% slower than Q4_0 for single token prediction, about the same in batch mode (perplexity). CPU single token is ~55 ms (on Ryzen 7950X). * Some improvement for Q3_K on CUDA It is now ~22.5 ms/token on my GPU, so ~30% slower than Q4_0. * Some more CUDA optimizations for Q3_K Single token is now 20.5 ms/token (~20% slower than Q4_0). Perplexity is on par with Q4_0. * Adding Q4_K - scalar, AVX2, CUDA Performance is the same or perhaps very slightly better than Q4_0 on the CPU. On the GPU, single token prediction is ~10% better than Q4_0, batch mode (perplexity is about the same). * Adding Q6_K - scalar, AVX2, CUDA Performance is ~40% lower compared to Q4_K on the CPU. This is to be expected, considering that we are memory bound on the CPU and the 6-bit model is ~44% larger than the 4-bit. On the GPU, single token prediction is ~6% lower than Q4_0, batch mode (perplexity) is even closer (but still slower). * Adding Q5_K - scalar, AVX2, CUDA Performance is ~20% lower compared to Q4_K on the CPU. This is to be expected, considering that we are memory bound on the CPU and the 5-bit model is ~22% larger than the 4-bit. On the GPU, single token prediction is about the same as Q4_0 for both, single token and batch prediction. * Per convention, all QX_K quantizations use Q5_K for output.weight * Adding quantization mixes * Quantization mixes: didn't quite get what I wanted in the last commit * Q4_K dot product for ARM_NEON * Q6_K dot product for ARM_NEON * Q5_K dot product for ARM_NEON * Adding Q3_K dot for ARM_NEON It is 22% slower than Q4_K, despite the smaller model size. On x86_64, where we are memory bound, the Q3_K model is quite a bit faster than Q4_K. * A very slightly faster ARM_NEON Q3_K dot * Adding Q2_K - just CUDA for now Token prediction is pretty good - about 15.5 ms on a RTX 4080. Perplexity is about the same as Q4_K. * Adding scalar and AVX2 Q2_K dot * Adding ARM_NEON Q2_K dot About the same performance as Q4_K. * A slightly faster ARM_NEON Q2_K dot Single token prediction is now ~36 ms on M2 Max. The code is much simpler too. * Fixed bug in Q2_K CUDA dot product kernel Stranegly enough, for the few prompts I tried with the 7B model the responses looked perfectly reasonable. Only realized something is not quite right when I tried the larger models and started getting nonse back. In any case, Q2_K single token evaluation time on an RTX 4080 in a Ryzen7950X box iusing CUDA and model fully loaded on the GPU are ~15.5 ms for 7B, ~25.4 ms for 13B, and ~55.8 ms for 30B. The max number of layers that fit in VRAM for The 65B is 32. With that, we get ~330 ms per token, which is not that much faster than just running on the CPU (~470 ms per token). * Don't print zeros/NaNs when no count histogram has been collected * A 10% faster CUDA vector dot kernel for Q3_K Q3_K is now running at ~18.5 ms / token on CUDA, so the gap to Q4_0 is only 10%. It seems memory acccess pattern is more important for performance than the amount of computation the kernel does. * A slightly daster Q4_K AVX2 dot product For perplexity, where we are less memory bound, time per pass drops by ~5%. Barely measurable difference for single token prediction. * A slightly faster ARM_NEON A4_K dot product * Minor * Fix quantization error test We cannot possibly be expecting rmse < 0.002 for 2- and 3-bit quantization variants. * Fix docker build I have been sloppy with vector reinterpret casts on ARM_NEON. It seems clang is very forgiving in that regard. * Added forgotten ggml.o dependence on k_quants.h to the Makefile * Had unintentionally committed the Makefile with -Ofast enabled * ggml : rename k_quants -> ggml-quants-k, use lowercase in code --------- Co-authored-by: Iwan Kawrakow <iwan.kawrakow@gmail.com> Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
2023-06-05 19:56:18 +00:00
ggml : quantization refactoring (#3833) * ggml : factor all quantization code in ggml-quants ggml-ci * ggml-quants : fix Zig and Swift builds + quantize tool ggml-ci * quantize : --pure option for disabling k-quant mixtures --------- Co-authored-by: cebtenzzre <cebtenzzre@gmail.com>
2023-10-29 16:32:28 +00:00
// load
vuint8mf2_t tx = __riscv_vle8_v_u8mf2(x[i].qs, vl);
Guard against all weights in a super-block being zero (#3010) * Guard against all weights in a super-block being zero * Also guard against extremely small weights Closes #2982 --------- Co-authored-by: Iwan Kawrakow <iwan.kawrakow@gmail.com>
2023-09-05 07:55:33 +00:00
ggml : quantization refactoring (#3833) * ggml : factor all quantization code in ggml-quants ggml-ci * ggml-quants : fix Zig and Swift builds + quantize tool ggml-ci * quantize : --pure option for disabling k-quant mixtures --------- Co-authored-by: cebtenzzre <cebtenzzre@gmail.com>
2023-10-29 16:32:28 +00:00
vint8mf2_t y0 = __riscv_vle8_v_i8mf2(y[i].qs, vl);
vint8mf2_t y1 = __riscv_vle8_v_i8mf2(y[i].qs+16, vl);
ggml : add SOTA 2,3,4,5,6 bit k-quantizations (#1684) * Starting to add k-quantization to ggml I think it is better to have quantization separate from ggml. For now just adding the k-quants there, but it would be better to also factor out the existing ggml quantizations. * Adding Q3_K and Q8_K (de)-quantization * Q3_K now working on CUDA and AVX2/scalar CUDA is not ideal - ~50% slower than Q4_0 for single token prediction, about the same in batch mode (perplexity). CPU single token is ~55 ms (on Ryzen 7950X). * Some improvement for Q3_K on CUDA It is now ~22.5 ms/token on my GPU, so ~30% slower than Q4_0. * Some more CUDA optimizations for Q3_K Single token is now 20.5 ms/token (~20% slower than Q4_0). Perplexity is on par with Q4_0. * Adding Q4_K - scalar, AVX2, CUDA Performance is the same or perhaps very slightly better than Q4_0 on the CPU. On the GPU, single token prediction is ~10% better than Q4_0, batch mode (perplexity is about the same). * Adding Q6_K - scalar, AVX2, CUDA Performance is ~40% lower compared to Q4_K on the CPU. This is to be expected, considering that we are memory bound on the CPU and the 6-bit model is ~44% larger than the 4-bit. On the GPU, single token prediction is ~6% lower than Q4_0, batch mode (perplexity) is even closer (but still slower). * Adding Q5_K - scalar, AVX2, CUDA Performance is ~20% lower compared to Q4_K on the CPU. This is to be expected, considering that we are memory bound on the CPU and the 5-bit model is ~22% larger than the 4-bit. On the GPU, single token prediction is about the same as Q4_0 for both, single token and batch prediction. * Per convention, all QX_K quantizations use Q5_K for output.weight * Adding quantization mixes * Quantization mixes: didn't quite get what I wanted in the last commit * Q4_K dot product for ARM_NEON * Q6_K dot product for ARM_NEON * Q5_K dot product for ARM_NEON * Adding Q3_K dot for ARM_NEON It is 22% slower than Q4_K, despite the smaller model size. On x86_64, where we are memory bound, the Q3_K model is quite a bit faster than Q4_K. * A very slightly faster ARM_NEON Q3_K dot * Adding Q2_K - just CUDA for now Token prediction is pretty good - about 15.5 ms on a RTX 4080. Perplexity is about the same as Q4_K. * Adding scalar and AVX2 Q2_K dot * Adding ARM_NEON Q2_K dot About the same performance as Q4_K. * A slightly faster ARM_NEON Q2_K dot Single token prediction is now ~36 ms on M2 Max. The code is much simpler too. * Fixed bug in Q2_K CUDA dot product kernel Stranegly enough, for the few prompts I tried with the 7B model the responses looked perfectly reasonable. Only realized something is not quite right when I tried the larger models and started getting nonse back. In any case, Q2_K single token evaluation time on an RTX 4080 in a Ryzen7950X box iusing CUDA and model fully loaded on the GPU are ~15.5 ms for 7B, ~25.4 ms for 13B, and ~55.8 ms for 30B. The max number of layers that fit in VRAM for The 65B is 32. With that, we get ~330 ms per token, which is not that much faster than just running on the CPU (~470 ms per token). * Don't print zeros/NaNs when no count histogram has been collected * A 10% faster CUDA vector dot kernel for Q3_K Q3_K is now running at ~18.5 ms / token on CUDA, so the gap to Q4_0 is only 10%. It seems memory acccess pattern is more important for performance than the amount of computation the kernel does. * A slightly daster Q4_K AVX2 dot product For perplexity, where we are less memory bound, time per pass drops by ~5%. Barely measurable difference for single token prediction. * A slightly faster ARM_NEON A4_K dot product * Minor * Fix quantization error test We cannot possibly be expecting rmse < 0.002 for 2- and 3-bit quantization variants. * Fix docker build I have been sloppy with vector reinterpret casts on ARM_NEON. It seems clang is very forgiving in that regard. * Added forgotten ggml.o dependence on k_quants.h to the Makefile * Had unintentionally committed the Makefile with -Ofast enabled * ggml : rename k_quants -> ggml-quants-k, use lowercase in code --------- Co-authored-by: Iwan Kawrakow <iwan.kawrakow@gmail.com> Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
2023-06-05 19:56:18 +00:00
ggml : quantization refactoring (#3833) * ggml : factor all quantization code in ggml-quants ggml-ci * ggml-quants : fix Zig and Swift builds + quantize tool ggml-ci * quantize : --pure option for disabling k-quant mixtures --------- Co-authored-by: cebtenzzre <cebtenzzre@gmail.com>
2023-10-29 16:32:28 +00:00
vuint8mf2_t x_at = __riscv_vand_vx_u8mf2(tx, 0x0F, vl);
vuint8mf2_t x_lt = __riscv_vsrl_vx_u8mf2(tx, 0x04, vl);
ggml : add SOTA 2,3,4,5,6 bit k-quantizations (#1684) * Starting to add k-quantization to ggml I think it is better to have quantization separate from ggml. For now just adding the k-quants there, but it would be better to also factor out the existing ggml quantizations. * Adding Q3_K and Q8_K (de)-quantization * Q3_K now working on CUDA and AVX2/scalar CUDA is not ideal - ~50% slower than Q4_0 for single token prediction, about the same in batch mode (perplexity). CPU single token is ~55 ms (on Ryzen 7950X). * Some improvement for Q3_K on CUDA It is now ~22.5 ms/token on my GPU, so ~30% slower than Q4_0. * Some more CUDA optimizations for Q3_K Single token is now 20.5 ms/token (~20% slower than Q4_0). Perplexity is on par with Q4_0. * Adding Q4_K - scalar, AVX2, CUDA Performance is the same or perhaps very slightly better than Q4_0 on the CPU. On the GPU, single token prediction is ~10% better than Q4_0, batch mode (perplexity is about the same). * Adding Q6_K - scalar, AVX2, CUDA Performance is ~40% lower compared to Q4_K on the CPU. This is to be expected, considering that we are memory bound on the CPU and the 6-bit model is ~44% larger than the 4-bit. On the GPU, single token prediction is ~6% lower than Q4_0, batch mode (perplexity) is even closer (but still slower). * Adding Q5_K - scalar, AVX2, CUDA Performance is ~20% lower compared to Q4_K on the CPU. This is to be expected, considering that we are memory bound on the CPU and the 5-bit model is ~22% larger than the 4-bit. On the GPU, single token prediction is about the same as Q4_0 for both, single token and batch prediction. * Per convention, all QX_K quantizations use Q5_K for output.weight * Adding quantization mixes * Quantization mixes: didn't quite get what I wanted in the last commit * Q4_K dot product for ARM_NEON * Q6_K dot product for ARM_NEON * Q5_K dot product for ARM_NEON * Adding Q3_K dot for ARM_NEON It is 22% slower than Q4_K, despite the smaller model size. On x86_64, where we are memory bound, the Q3_K model is quite a bit faster than Q4_K. * A very slightly faster ARM_NEON Q3_K dot * Adding Q2_K - just CUDA for now Token prediction is pretty good - about 15.5 ms on a RTX 4080. Perplexity is about the same as Q4_K. * Adding scalar and AVX2 Q2_K dot * Adding ARM_NEON Q2_K dot About the same performance as Q4_K. * A slightly faster ARM_NEON Q2_K dot Single token prediction is now ~36 ms on M2 Max. The code is much simpler too. * Fixed bug in Q2_K CUDA dot product kernel Stranegly enough, for the few prompts I tried with the 7B model the responses looked perfectly reasonable. Only realized something is not quite right when I tried the larger models and started getting nonse back. In any case, Q2_K single token evaluation time on an RTX 4080 in a Ryzen7950X box iusing CUDA and model fully loaded on the GPU are ~15.5 ms for 7B, ~25.4 ms for 13B, and ~55.8 ms for 30B. The max number of layers that fit in VRAM for The 65B is 32. With that, we get ~330 ms per token, which is not that much faster than just running on the CPU (~470 ms per token). * Don't print zeros/NaNs when no count histogram has been collected * A 10% faster CUDA vector dot kernel for Q3_K Q3_K is now running at ~18.5 ms / token on CUDA, so the gap to Q4_0 is only 10%. It seems memory acccess pattern is more important for performance than the amount of computation the kernel does. * A slightly daster Q4_K AVX2 dot product For perplexity, where we are less memory bound, time per pass drops by ~5%. Barely measurable difference for single token prediction. * A slightly faster ARM_NEON A4_K dot product * Minor * Fix quantization error test We cannot possibly be expecting rmse < 0.002 for 2- and 3-bit quantization variants. * Fix docker build I have been sloppy with vector reinterpret casts on ARM_NEON. It seems clang is very forgiving in that regard. * Added forgotten ggml.o dependence on k_quants.h to the Makefile * Had unintentionally committed the Makefile with -Ofast enabled * ggml : rename k_quants -> ggml-quants-k, use lowercase in code --------- Co-authored-by: Iwan Kawrakow <iwan.kawrakow@gmail.com> Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
2023-06-05 19:56:18 +00:00
ggml : quantization refactoring (#3833) * ggml : factor all quantization code in ggml-quants ggml-ci * ggml-quants : fix Zig and Swift builds + quantize tool ggml-ci * quantize : --pure option for disabling k-quant mixtures --------- Co-authored-by: cebtenzzre <cebtenzzre@gmail.com>
2023-10-29 16:32:28 +00:00
vuint8mf2_t x_a = __riscv_vor_vv_u8mf2(x_at, xh_0, vl);
vuint8mf2_t x_l = __riscv_vor_vv_u8mf2(x_lt, xh_1, vl);
ggml : add SOTA 2,3,4,5,6 bit k-quantizations (#1684) * Starting to add k-quantization to ggml I think it is better to have quantization separate from ggml. For now just adding the k-quants there, but it would be better to also factor out the existing ggml quantizations. * Adding Q3_K and Q8_K (de)-quantization * Q3_K now working on CUDA and AVX2/scalar CUDA is not ideal - ~50% slower than Q4_0 for single token prediction, about the same in batch mode (perplexity). CPU single token is ~55 ms (on Ryzen 7950X). * Some improvement for Q3_K on CUDA It is now ~22.5 ms/token on my GPU, so ~30% slower than Q4_0. * Some more CUDA optimizations for Q3_K Single token is now 20.5 ms/token (~20% slower than Q4_0). Perplexity is on par with Q4_0. * Adding Q4_K - scalar, AVX2, CUDA Performance is the same or perhaps very slightly better than Q4_0 on the CPU. On the GPU, single token prediction is ~10% better than Q4_0, batch mode (perplexity is about the same). * Adding Q6_K - scalar, AVX2, CUDA Performance is ~40% lower compared to Q4_K on the CPU. This is to be expected, considering that we are memory bound on the CPU and the 6-bit model is ~44% larger than the 4-bit. On the GPU, single token prediction is ~6% lower than Q4_0, batch mode (perplexity) is even closer (but still slower). * Adding Q5_K - scalar, AVX2, CUDA Performance is ~20% lower compared to Q4_K on the CPU. This is to be expected, considering that we are memory bound on the CPU and the 5-bit model is ~22% larger than the 4-bit. On the GPU, single token prediction is about the same as Q4_0 for both, single token and batch prediction. * Per convention, all QX_K quantizations use Q5_K for output.weight * Adding quantization mixes * Quantization mixes: didn't quite get what I wanted in the last commit * Q4_K dot product for ARM_NEON * Q6_K dot product for ARM_NEON * Q5_K dot product for ARM_NEON * Adding Q3_K dot for ARM_NEON It is 22% slower than Q4_K, despite the smaller model size. On x86_64, where we are memory bound, the Q3_K model is quite a bit faster than Q4_K. * A very slightly faster ARM_NEON Q3_K dot * Adding Q2_K - just CUDA for now Token prediction is pretty good - about 15.5 ms on a RTX 4080. Perplexity is about the same as Q4_K. * Adding scalar and AVX2 Q2_K dot * Adding ARM_NEON Q2_K dot About the same performance as Q4_K. * A slightly faster ARM_NEON Q2_K dot Single token prediction is now ~36 ms on M2 Max. The code is much simpler too. * Fixed bug in Q2_K CUDA dot product kernel Stranegly enough, for the few prompts I tried with the 7B model the responses looked perfectly reasonable. Only realized something is not quite right when I tried the larger models and started getting nonse back. In any case, Q2_K single token evaluation time on an RTX 4080 in a Ryzen7950X box iusing CUDA and model fully loaded on the GPU are ~15.5 ms for 7B, ~25.4 ms for 13B, and ~55.8 ms for 30B. The max number of layers that fit in VRAM for The 65B is 32. With that, we get ~330 ms per token, which is not that much faster than just running on the CPU (~470 ms per token). * Don't print zeros/NaNs when no count histogram has been collected * A 10% faster CUDA vector dot kernel for Q3_K Q3_K is now running at ~18.5 ms / token on CUDA, so the gap to Q4_0 is only 10%. It seems memory acccess pattern is more important for performance than the amount of computation the kernel does. * A slightly daster Q4_K AVX2 dot product For perplexity, where we are less memory bound, time per pass drops by ~5%. Barely measurable difference for single token prediction. * A slightly faster ARM_NEON A4_K dot product * Minor * Fix quantization error test We cannot possibly be expecting rmse < 0.002 for 2- and 3-bit quantization variants. * Fix docker build I have been sloppy with vector reinterpret casts on ARM_NEON. It seems clang is very forgiving in that regard. * Added forgotten ggml.o dependence on k_quants.h to the Makefile * Had unintentionally committed the Makefile with -Ofast enabled * ggml : rename k_quants -> ggml-quants-k, use lowercase in code --------- Co-authored-by: Iwan Kawrakow <iwan.kawrakow@gmail.com> Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
2023-06-05 19:56:18 +00:00
ggml : quantization refactoring (#3833) * ggml : factor all quantization code in ggml-quants ggml-ci * ggml-quants : fix Zig and Swift builds + quantize tool ggml-ci * quantize : --pure option for disabling k-quant mixtures --------- Co-authored-by: cebtenzzre <cebtenzzre@gmail.com>
2023-10-29 16:32:28 +00:00
vint8mf2_t v0 = __riscv_vreinterpret_v_u8mf2_i8mf2(x_a);
vint8mf2_t v1 = __riscv_vreinterpret_v_u8mf2_i8mf2(x_l);
vint16m1_t vec_mul1 = __riscv_vwmul_vv_i16m1(v0, y0, vl);
vint16m1_t vec_mul2 = __riscv_vwmul_vv_i16m1(v1, y1, vl);
vint32m1_t vec_zero = __riscv_vmv_v_x_i32m1(0, vl);
vint32m1_t vs1 = __riscv_vwredsum_vs_i16m1_i32m1(vec_mul1, vec_zero, vl);
vint32m1_t vs2 = __riscv_vwredsum_vs_i16m1_i32m1(vec_mul2, vs1, vl);
ggml : add SOTA 2,3,4,5,6 bit k-quantizations (#1684) * Starting to add k-quantization to ggml I think it is better to have quantization separate from ggml. For now just adding the k-quants there, but it would be better to also factor out the existing ggml quantizations. * Adding Q3_K and Q8_K (de)-quantization * Q3_K now working on CUDA and AVX2/scalar CUDA is not ideal - ~50% slower than Q4_0 for single token prediction, about the same in batch mode (perplexity). CPU single token is ~55 ms (on Ryzen 7950X). * Some improvement for Q3_K on CUDA It is now ~22.5 ms/token on my GPU, so ~30% slower than Q4_0. * Some more CUDA optimizations for Q3_K Single token is now 20.5 ms/token (~20% slower than Q4_0). Perplexity is on par with Q4_0. * Adding Q4_K - scalar, AVX2, CUDA Performance is the same or perhaps very slightly better than Q4_0 on the CPU. On the GPU, single token prediction is ~10% better than Q4_0, batch mode (perplexity is about the same). * Adding Q6_K - scalar, AVX2, CUDA Performance is ~40% lower compared to Q4_K on the CPU. This is to be expected, considering that we are memory bound on the CPU and the 6-bit model is ~44% larger than the 4-bit. On the GPU, single token prediction is ~6% lower than Q4_0, batch mode (perplexity) is even closer (but still slower). * Adding Q5_K - scalar, AVX2, CUDA Performance is ~20% lower compared to Q4_K on the CPU. This is to be expected, considering that we are memory bound on the CPU and the 5-bit model is ~22% larger than the 4-bit. On the GPU, single token prediction is about the same as Q4_0 for both, single token and batch prediction. * Per convention, all QX_K quantizations use Q5_K for output.weight * Adding quantization mixes * Quantization mixes: didn't quite get what I wanted in the last commit * Q4_K dot product for ARM_NEON * Q6_K dot product for ARM_NEON * Q5_K dot product for ARM_NEON * Adding Q3_K dot for ARM_NEON It is 22% slower than Q4_K, despite the smaller model size. On x86_64, where we are memory bound, the Q3_K model is quite a bit faster than Q4_K. * A very slightly faster ARM_NEON Q3_K dot * Adding Q2_K - just CUDA for now Token prediction is pretty good - about 15.5 ms on a RTX 4080. Perplexity is about the same as Q4_K. * Adding scalar and AVX2 Q2_K dot * Adding ARM_NEON Q2_K dot About the same performance as Q4_K. * A slightly faster ARM_NEON Q2_K dot Single token prediction is now ~36 ms on M2 Max. The code is much simpler too. * Fixed bug in Q2_K CUDA dot product kernel Stranegly enough, for the few prompts I tried with the 7B model the responses looked perfectly reasonable. Only realized something is not quite right when I tried the larger models and started getting nonse back. In any case, Q2_K single token evaluation time on an RTX 4080 in a Ryzen7950X box iusing CUDA and model fully loaded on the GPU are ~15.5 ms for 7B, ~25.4 ms for 13B, and ~55.8 ms for 30B. The max number of layers that fit in VRAM for The 65B is 32. With that, we get ~330 ms per token, which is not that much faster than just running on the CPU (~470 ms per token). * Don't print zeros/NaNs when no count histogram has been collected * A 10% faster CUDA vector dot kernel for Q3_K Q3_K is now running at ~18.5 ms / token on CUDA, so the gap to Q4_0 is only 10%. It seems memory acccess pattern is more important for performance than the amount of computation the kernel does. * A slightly daster Q4_K AVX2 dot product For perplexity, where we are less memory bound, time per pass drops by ~5%. Barely measurable difference for single token prediction. * A slightly faster ARM_NEON A4_K dot product * Minor * Fix quantization error test We cannot possibly be expecting rmse < 0.002 for 2- and 3-bit quantization variants. * Fix docker build I have been sloppy with vector reinterpret casts on ARM_NEON. It seems clang is very forgiving in that regard. * Added forgotten ggml.o dependence on k_quants.h to the Makefile * Had unintentionally committed the Makefile with -Ofast enabled * ggml : rename k_quants -> ggml-quants-k, use lowercase in code --------- Co-authored-by: Iwan Kawrakow <iwan.kawrakow@gmail.com> Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
2023-06-05 19:56:18 +00:00
ggml : quantization refactoring (#3833) * ggml : factor all quantization code in ggml-quants ggml-ci * ggml-quants : fix Zig and Swift builds + quantize tool ggml-ci * quantize : --pure option for disabling k-quant mixtures --------- Co-authored-by: cebtenzzre <cebtenzzre@gmail.com>
2023-10-29 16:32:28 +00:00
int sumi = __riscv_vmv_x_s_i32m1_i32(vs2);
ggml : move FP16 <-> FP32 code to ggml-impl.h (#3861) * ggml : move FP16 <-> FP32 stuff to ggml-impl.h ggml-ci * tests : fix ARM build * ggml : explicitly initialize deprecated type traits * ggml : add math.h to ggml-impl.h * ggml : remove duplicate static assert macros * ggml : prefix lookup tables with ggml_ ggml-ci * ggml-impl : move extern "C" to start of file
2023-10-30 17:19:15 +00:00
sumf += (GGML_FP16_TO_FP32(x[i].d)*y[i].d)*sumi + GGML_FP16_TO_FP32(x[i].m)*y[i].s;
ggml : add SOTA 2,3,4,5,6 bit k-quantizations (#1684) * Starting to add k-quantization to ggml I think it is better to have quantization separate from ggml. For now just adding the k-quants there, but it would be better to also factor out the existing ggml quantizations. * Adding Q3_K and Q8_K (de)-quantization * Q3_K now working on CUDA and AVX2/scalar CUDA is not ideal - ~50% slower than Q4_0 for single token prediction, about the same in batch mode (perplexity). CPU single token is ~55 ms (on Ryzen 7950X). * Some improvement for Q3_K on CUDA It is now ~22.5 ms/token on my GPU, so ~30% slower than Q4_0. * Some more CUDA optimizations for Q3_K Single token is now 20.5 ms/token (~20% slower than Q4_0). Perplexity is on par with Q4_0. * Adding Q4_K - scalar, AVX2, CUDA Performance is the same or perhaps very slightly better than Q4_0 on the CPU. On the GPU, single token prediction is ~10% better than Q4_0, batch mode (perplexity is about the same). * Adding Q6_K - scalar, AVX2, CUDA Performance is ~40% lower compared to Q4_K on the CPU. This is to be expected, considering that we are memory bound on the CPU and the 6-bit model is ~44% larger than the 4-bit. On the GPU, single token prediction is ~6% lower than Q4_0, batch mode (perplexity) is even closer (but still slower). * Adding Q5_K - scalar, AVX2, CUDA Performance is ~20% lower compared to Q4_K on the CPU. This is to be expected, considering that we are memory bound on the CPU and the 5-bit model is ~22% larger than the 4-bit. On the GPU, single token prediction is about the same as Q4_0 for both, single token and batch prediction. * Per convention, all QX_K quantizations use Q5_K for output.weight * Adding quantization mixes * Quantization mixes: didn't quite get what I wanted in the last commit * Q4_K dot product for ARM_NEON * Q6_K dot product for ARM_NEON * Q5_K dot product for ARM_NEON * Adding Q3_K dot for ARM_NEON It is 22% slower than Q4_K, despite the smaller model size. On x86_64, where we are memory bound, the Q3_K model is quite a bit faster than Q4_K. * A very slightly faster ARM_NEON Q3_K dot * Adding Q2_K - just CUDA for now Token prediction is pretty good - about 15.5 ms on a RTX 4080. Perplexity is about the same as Q4_K. * Adding scalar and AVX2 Q2_K dot * Adding ARM_NEON Q2_K dot About the same performance as Q4_K. * A slightly faster ARM_NEON Q2_K dot Single token prediction is now ~36 ms on M2 Max. The code is much simpler too. * Fixed bug in Q2_K CUDA dot product kernel Stranegly enough, for the few prompts I tried with the 7B model the responses looked perfectly reasonable. Only realized something is not quite right when I tried the larger models and started getting nonse back. In any case, Q2_K single token evaluation time on an RTX 4080 in a Ryzen7950X box iusing CUDA and model fully loaded on the GPU are ~15.5 ms for 7B, ~25.4 ms for 13B, and ~55.8 ms for 30B. The max number of layers that fit in VRAM for The 65B is 32. With that, we get ~330 ms per token, which is not that much faster than just running on the CPU (~470 ms per token). * Don't print zeros/NaNs when no count histogram has been collected * A 10% faster CUDA vector dot kernel for Q3_K Q3_K is now running at ~18.5 ms / token on CUDA, so the gap to Q4_0 is only 10%. It seems memory acccess pattern is more important for performance than the amount of computation the kernel does. * A slightly daster Q4_K AVX2 dot product For perplexity, where we are less memory bound, time per pass drops by ~5%. Barely measurable difference for single token prediction. * A slightly faster ARM_NEON A4_K dot product * Minor * Fix quantization error test We cannot possibly be expecting rmse < 0.002 for 2- and 3-bit quantization variants. * Fix docker build I have been sloppy with vector reinterpret casts on ARM_NEON. It seems clang is very forgiving in that regard. * Added forgotten ggml.o dependence on k_quants.h to the Makefile * Had unintentionally committed the Makefile with -Ofast enabled * ggml : rename k_quants -> ggml-quants-k, use lowercase in code --------- Co-authored-by: Iwan Kawrakow <iwan.kawrakow@gmail.com> Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
2023-06-05 19:56:18 +00:00
}
ggml : quantization refactoring (#3833) * ggml : factor all quantization code in ggml-quants ggml-ci * ggml-quants : fix Zig and Swift builds + quantize tool ggml-ci * quantize : --pure option for disabling k-quant mixtures --------- Co-authored-by: cebtenzzre <cebtenzzre@gmail.com>
2023-10-29 16:32:28 +00:00
*s = sumf;
#else
// scalar
float sumf = 0.0;
ggml : add SOTA 2,3,4,5,6 bit k-quantizations (#1684) * Starting to add k-quantization to ggml I think it is better to have quantization separate from ggml. For now just adding the k-quants there, but it would be better to also factor out the existing ggml quantizations. * Adding Q3_K and Q8_K (de)-quantization * Q3_K now working on CUDA and AVX2/scalar CUDA is not ideal - ~50% slower than Q4_0 for single token prediction, about the same in batch mode (perplexity). CPU single token is ~55 ms (on Ryzen 7950X). * Some improvement for Q3_K on CUDA It is now ~22.5 ms/token on my GPU, so ~30% slower than Q4_0. * Some more CUDA optimizations for Q3_K Single token is now 20.5 ms/token (~20% slower than Q4_0). Perplexity is on par with Q4_0. * Adding Q4_K - scalar, AVX2, CUDA Performance is the same or perhaps very slightly better than Q4_0 on the CPU. On the GPU, single token prediction is ~10% better than Q4_0, batch mode (perplexity is about the same). * Adding Q6_K - scalar, AVX2, CUDA Performance is ~40% lower compared to Q4_K on the CPU. This is to be expected, considering that we are memory bound on the CPU and the 6-bit model is ~44% larger than the 4-bit. On the GPU, single token prediction is ~6% lower than Q4_0, batch mode (perplexity) is even closer (but still slower). * Adding Q5_K - scalar, AVX2, CUDA Performance is ~20% lower compared to Q4_K on the CPU. This is to be expected, considering that we are memory bound on the CPU and the 5-bit model is ~22% larger than the 4-bit. On the GPU, single token prediction is about the same as Q4_0 for both, single token and batch prediction. * Per convention, all QX_K quantizations use Q5_K for output.weight * Adding quantization mixes * Quantization mixes: didn't quite get what I wanted in the last commit * Q4_K dot product for ARM_NEON * Q6_K dot product for ARM_NEON * Q5_K dot product for ARM_NEON * Adding Q3_K dot for ARM_NEON It is 22% slower than Q4_K, despite the smaller model size. On x86_64, where we are memory bound, the Q3_K model is quite a bit faster than Q4_K. * A very slightly faster ARM_NEON Q3_K dot * Adding Q2_K - just CUDA for now Token prediction is pretty good - about 15.5 ms on a RTX 4080. Perplexity is about the same as Q4_K. * Adding scalar and AVX2 Q2_K dot * Adding ARM_NEON Q2_K dot About the same performance as Q4_K. * A slightly faster ARM_NEON Q2_K dot Single token prediction is now ~36 ms on M2 Max. The code is much simpler too. * Fixed bug in Q2_K CUDA dot product kernel Stranegly enough, for the few prompts I tried with the 7B model the responses looked perfectly reasonable. Only realized something is not quite right when I tried the larger models and started getting nonse back. In any case, Q2_K single token evaluation time on an RTX 4080 in a Ryzen7950X box iusing CUDA and model fully loaded on the GPU are ~15.5 ms for 7B, ~25.4 ms for 13B, and ~55.8 ms for 30B. The max number of layers that fit in VRAM for The 65B is 32. With that, we get ~330 ms per token, which is not that much faster than just running on the CPU (~470 ms per token). * Don't print zeros/NaNs when no count histogram has been collected * A 10% faster CUDA vector dot kernel for Q3_K Q3_K is now running at ~18.5 ms / token on CUDA, so the gap to Q4_0 is only 10%. It seems memory acccess pattern is more important for performance than the amount of computation the kernel does. * A slightly daster Q4_K AVX2 dot product For perplexity, where we are less memory bound, time per pass drops by ~5%. Barely measurable difference for single token prediction. * A slightly faster ARM_NEON A4_K dot product * Minor * Fix quantization error test We cannot possibly be expecting rmse < 0.002 for 2- and 3-bit quantization variants. * Fix docker build I have been sloppy with vector reinterpret casts on ARM_NEON. It seems clang is very forgiving in that regard. * Added forgotten ggml.o dependence on k_quants.h to the Makefile * Had unintentionally committed the Makefile with -Ofast enabled * ggml : rename k_quants -> ggml-quants-k, use lowercase in code --------- Co-authored-by: Iwan Kawrakow <iwan.kawrakow@gmail.com> Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
2023-06-05 19:56:18 +00:00
for (int i = 0; i < nb; i++) {
ggml : quantization refactoring (#3833) * ggml : factor all quantization code in ggml-quants ggml-ci * ggml-quants : fix Zig and Swift builds + quantize tool ggml-ci * quantize : --pure option for disabling k-quant mixtures --------- Co-authored-by: cebtenzzre <cebtenzzre@gmail.com>
2023-10-29 16:32:28 +00:00
uint32_t qh;
memcpy(&qh, x[i].qh, sizeof(qh));
ggml : add SOTA 2,3,4,5,6 bit k-quantizations (#1684) * Starting to add k-quantization to ggml I think it is better to have quantization separate from ggml. For now just adding the k-quants there, but it would be better to also factor out the existing ggml quantizations. * Adding Q3_K and Q8_K (de)-quantization * Q3_K now working on CUDA and AVX2/scalar CUDA is not ideal - ~50% slower than Q4_0 for single token prediction, about the same in batch mode (perplexity). CPU single token is ~55 ms (on Ryzen 7950X). * Some improvement for Q3_K on CUDA It is now ~22.5 ms/token on my GPU, so ~30% slower than Q4_0. * Some more CUDA optimizations for Q3_K Single token is now 20.5 ms/token (~20% slower than Q4_0). Perplexity is on par with Q4_0. * Adding Q4_K - scalar, AVX2, CUDA Performance is the same or perhaps very slightly better than Q4_0 on the CPU. On the GPU, single token prediction is ~10% better than Q4_0, batch mode (perplexity is about the same). * Adding Q6_K - scalar, AVX2, CUDA Performance is ~40% lower compared to Q4_K on the CPU. This is to be expected, considering that we are memory bound on the CPU and the 6-bit model is ~44% larger than the 4-bit. On the GPU, single token prediction is ~6% lower than Q4_0, batch mode (perplexity) is even closer (but still slower). * Adding Q5_K - scalar, AVX2, CUDA Performance is ~20% lower compared to Q4_K on the CPU. This is to be expected, considering that we are memory bound on the CPU and the 5-bit model is ~22% larger than the 4-bit. On the GPU, single token prediction is about the same as Q4_0 for both, single token and batch prediction. * Per convention, all QX_K quantizations use Q5_K for output.weight * Adding quantization mixes * Quantization mixes: didn't quite get what I wanted in the last commit * Q4_K dot product for ARM_NEON * Q6_K dot product for ARM_NEON * Q5_K dot product for ARM_NEON * Adding Q3_K dot for ARM_NEON It is 22% slower than Q4_K, despite the smaller model size. On x86_64, where we are memory bound, the Q3_K model is quite a bit faster than Q4_K. * A very slightly faster ARM_NEON Q3_K dot * Adding Q2_K - just CUDA for now Token prediction is pretty good - about 15.5 ms on a RTX 4080. Perplexity is about the same as Q4_K. * Adding scalar and AVX2 Q2_K dot * Adding ARM_NEON Q2_K dot About the same performance as Q4_K. * A slightly faster ARM_NEON Q2_K dot Single token prediction is now ~36 ms on M2 Max. The code is much simpler too. * Fixed bug in Q2_K CUDA dot product kernel Stranegly enough, for the few prompts I tried with the 7B model the responses looked perfectly reasonable. Only realized something is not quite right when I tried the larger models and started getting nonse back. In any case, Q2_K single token evaluation time on an RTX 4080 in a Ryzen7950X box iusing CUDA and model fully loaded on the GPU are ~15.5 ms for 7B, ~25.4 ms for 13B, and ~55.8 ms for 30B. The max number of layers that fit in VRAM for The 65B is 32. With that, we get ~330 ms per token, which is not that much faster than just running on the CPU (~470 ms per token). * Don't print zeros/NaNs when no count histogram has been collected * A 10% faster CUDA vector dot kernel for Q3_K Q3_K is now running at ~18.5 ms / token on CUDA, so the gap to Q4_0 is only 10%. It seems memory acccess pattern is more important for performance than the amount of computation the kernel does. * A slightly daster Q4_K AVX2 dot product For perplexity, where we are less memory bound, time per pass drops by ~5%. Barely measurable difference for single token prediction. * A slightly faster ARM_NEON A4_K dot product * Minor * Fix quantization error test We cannot possibly be expecting rmse < 0.002 for 2- and 3-bit quantization variants. * Fix docker build I have been sloppy with vector reinterpret casts on ARM_NEON. It seems clang is very forgiving in that regard. * Added forgotten ggml.o dependence on k_quants.h to the Makefile * Had unintentionally committed the Makefile with -Ofast enabled * ggml : rename k_quants -> ggml-quants-k, use lowercase in code --------- Co-authored-by: Iwan Kawrakow <iwan.kawrakow@gmail.com> Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
2023-06-05 19:56:18 +00:00
ggml : quantization refactoring (#3833) * ggml : factor all quantization code in ggml-quants ggml-ci * ggml-quants : fix Zig and Swift builds + quantize tool ggml-ci * quantize : --pure option for disabling k-quant mixtures --------- Co-authored-by: cebtenzzre <cebtenzzre@gmail.com>
2023-10-29 16:32:28 +00:00
int sumi = 0;
ggml : add SOTA 2,3,4,5,6 bit k-quantizations (#1684) * Starting to add k-quantization to ggml I think it is better to have quantization separate from ggml. For now just adding the k-quants there, but it would be better to also factor out the existing ggml quantizations. * Adding Q3_K and Q8_K (de)-quantization * Q3_K now working on CUDA and AVX2/scalar CUDA is not ideal - ~50% slower than Q4_0 for single token prediction, about the same in batch mode (perplexity). CPU single token is ~55 ms (on Ryzen 7950X). * Some improvement for Q3_K on CUDA It is now ~22.5 ms/token on my GPU, so ~30% slower than Q4_0. * Some more CUDA optimizations for Q3_K Single token is now 20.5 ms/token (~20% slower than Q4_0). Perplexity is on par with Q4_0. * Adding Q4_K - scalar, AVX2, CUDA Performance is the same or perhaps very slightly better than Q4_0 on the CPU. On the GPU, single token prediction is ~10% better than Q4_0, batch mode (perplexity is about the same). * Adding Q6_K - scalar, AVX2, CUDA Performance is ~40% lower compared to Q4_K on the CPU. This is to be expected, considering that we are memory bound on the CPU and the 6-bit model is ~44% larger than the 4-bit. On the GPU, single token prediction is ~6% lower than Q4_0, batch mode (perplexity) is even closer (but still slower). * Adding Q5_K - scalar, AVX2, CUDA Performance is ~20% lower compared to Q4_K on the CPU. This is to be expected, considering that we are memory bound on the CPU and the 5-bit model is ~22% larger than the 4-bit. On the GPU, single token prediction is about the same as Q4_0 for both, single token and batch prediction. * Per convention, all QX_K quantizations use Q5_K for output.weight * Adding quantization mixes * Quantization mixes: didn't quite get what I wanted in the last commit * Q4_K dot product for ARM_NEON * Q6_K dot product for ARM_NEON * Q5_K dot product for ARM_NEON * Adding Q3_K dot for ARM_NEON It is 22% slower than Q4_K, despite the smaller model size. On x86_64, where we are memory bound, the Q3_K model is quite a bit faster than Q4_K. * A very slightly faster ARM_NEON Q3_K dot * Adding Q2_K - just CUDA for now Token prediction is pretty good - about 15.5 ms on a RTX 4080. Perplexity is about the same as Q4_K. * Adding scalar and AVX2 Q2_K dot * Adding ARM_NEON Q2_K dot About the same performance as Q4_K. * A slightly faster ARM_NEON Q2_K dot Single token prediction is now ~36 ms on M2 Max. The code is much simpler too. * Fixed bug in Q2_K CUDA dot product kernel Stranegly enough, for the few prompts I tried with the 7B model the responses looked perfectly reasonable. Only realized something is not quite right when I tried the larger models and started getting nonse back. In any case, Q2_K single token evaluation time on an RTX 4080 in a Ryzen7950X box iusing CUDA and model fully loaded on the GPU are ~15.5 ms for 7B, ~25.4 ms for 13B, and ~55.8 ms for 30B. The max number of layers that fit in VRAM for The 65B is 32. With that, we get ~330 ms per token, which is not that much faster than just running on the CPU (~470 ms per token). * Don't print zeros/NaNs when no count histogram has been collected * A 10% faster CUDA vector dot kernel for Q3_K Q3_K is now running at ~18.5 ms / token on CUDA, so the gap to Q4_0 is only 10%. It seems memory acccess pattern is more important for performance than the amount of computation the kernel does. * A slightly daster Q4_K AVX2 dot product For perplexity, where we are less memory bound, time per pass drops by ~5%. Barely measurable difference for single token prediction. * A slightly faster ARM_NEON A4_K dot product * Minor * Fix quantization error test We cannot possibly be expecting rmse < 0.002 for 2- and 3-bit quantization variants. * Fix docker build I have been sloppy with vector reinterpret casts on ARM_NEON. It seems clang is very forgiving in that regard. * Added forgotten ggml.o dependence on k_quants.h to the Makefile * Had unintentionally committed the Makefile with -Ofast enabled * ggml : rename k_quants -> ggml-quants-k, use lowercase in code --------- Co-authored-by: Iwan Kawrakow <iwan.kawrakow@gmail.com> Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
2023-06-05 19:56:18 +00:00
ggml : quantization refactoring (#3833) * ggml : factor all quantization code in ggml-quants ggml-ci * ggml-quants : fix Zig and Swift builds + quantize tool ggml-ci * quantize : --pure option for disabling k-quant mixtures --------- Co-authored-by: cebtenzzre <cebtenzzre@gmail.com>
2023-10-29 16:32:28 +00:00
for (int j = 0; j < qk/2; ++j) {
const uint8_t xh_0 = ((qh >> (j + 0)) << 4) & 0x10;
const uint8_t xh_1 = ((qh >> (j + 12)) ) & 0x10;
ggml : add SOTA 2,3,4,5,6 bit k-quantizations (#1684) * Starting to add k-quantization to ggml I think it is better to have quantization separate from ggml. For now just adding the k-quants there, but it would be better to also factor out the existing ggml quantizations. * Adding Q3_K and Q8_K (de)-quantization * Q3_K now working on CUDA and AVX2/scalar CUDA is not ideal - ~50% slower than Q4_0 for single token prediction, about the same in batch mode (perplexity). CPU single token is ~55 ms (on Ryzen 7950X). * Some improvement for Q3_K on CUDA It is now ~22.5 ms/token on my GPU, so ~30% slower than Q4_0. * Some more CUDA optimizations for Q3_K Single token is now 20.5 ms/token (~20% slower than Q4_0). Perplexity is on par with Q4_0. * Adding Q4_K - scalar, AVX2, CUDA Performance is the same or perhaps very slightly better than Q4_0 on the CPU. On the GPU, single token prediction is ~10% better than Q4_0, batch mode (perplexity is about the same). * Adding Q6_K - scalar, AVX2, CUDA Performance is ~40% lower compared to Q4_K on the CPU. This is to be expected, considering that we are memory bound on the CPU and the 6-bit model is ~44% larger than the 4-bit. On the GPU, single token prediction is ~6% lower than Q4_0, batch mode (perplexity) is even closer (but still slower). * Adding Q5_K - scalar, AVX2, CUDA Performance is ~20% lower compared to Q4_K on the CPU. This is to be expected, considering that we are memory bound on the CPU and the 5-bit model is ~22% larger than the 4-bit. On the GPU, single token prediction is about the same as Q4_0 for both, single token and batch prediction. * Per convention, all QX_K quantizations use Q5_K for output.weight * Adding quantization mixes * Quantization mixes: didn't quite get what I wanted in the last commit * Q4_K dot product for ARM_NEON * Q6_K dot product for ARM_NEON * Q5_K dot product for ARM_NEON * Adding Q3_K dot for ARM_NEON It is 22% slower than Q4_K, despite the smaller model size. On x86_64, where we are memory bound, the Q3_K model is quite a bit faster than Q4_K. * A very slightly faster ARM_NEON Q3_K dot * Adding Q2_K - just CUDA for now Token prediction is pretty good - about 15.5 ms on a RTX 4080. Perplexity is about the same as Q4_K. * Adding scalar and AVX2 Q2_K dot * Adding ARM_NEON Q2_K dot About the same performance as Q4_K. * A slightly faster ARM_NEON Q2_K dot Single token prediction is now ~36 ms on M2 Max. The code is much simpler too. * Fixed bug in Q2_K CUDA dot product kernel Stranegly enough, for the few prompts I tried with the 7B model the responses looked perfectly reasonable. Only realized something is not quite right when I tried the larger models and started getting nonse back. In any case, Q2_K single token evaluation time on an RTX 4080 in a Ryzen7950X box iusing CUDA and model fully loaded on the GPU are ~15.5 ms for 7B, ~25.4 ms for 13B, and ~55.8 ms for 30B. The max number of layers that fit in VRAM for The 65B is 32. With that, we get ~330 ms per token, which is not that much faster than just running on the CPU (~470 ms per token). * Don't print zeros/NaNs when no count histogram has been collected * A 10% faster CUDA vector dot kernel for Q3_K Q3_K is now running at ~18.5 ms / token on CUDA, so the gap to Q4_0 is only 10%. It seems memory acccess pattern is more important for performance than the amount of computation the kernel does. * A slightly daster Q4_K AVX2 dot product For perplexity, where we are less memory bound, time per pass drops by ~5%. Barely measurable difference for single token prediction. * A slightly faster ARM_NEON A4_K dot product * Minor * Fix quantization error test We cannot possibly be expecting rmse < 0.002 for 2- and 3-bit quantization variants. * Fix docker build I have been sloppy with vector reinterpret casts on ARM_NEON. It seems clang is very forgiving in that regard. * Added forgotten ggml.o dependence on k_quants.h to the Makefile * Had unintentionally committed the Makefile with -Ofast enabled * ggml : rename k_quants -> ggml-quants-k, use lowercase in code --------- Co-authored-by: Iwan Kawrakow <iwan.kawrakow@gmail.com> Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
2023-06-05 19:56:18 +00:00
ggml : quantization refactoring (#3833) * ggml : factor all quantization code in ggml-quants ggml-ci * ggml-quants : fix Zig and Swift builds + quantize tool ggml-ci * quantize : --pure option for disabling k-quant mixtures --------- Co-authored-by: cebtenzzre <cebtenzzre@gmail.com>
2023-10-29 16:32:28 +00:00
const int32_t x0 = (x[i].qs[j] & 0xF) | xh_0;
const int32_t x1 = (x[i].qs[j] >> 4) | xh_1;
sumi += (x0 * y[i].qs[j]) + (x1 * y[i].qs[j + qk/2]);
k-quants : support for super-block size of 64 (#2001) * k_quants: WIP super-blocks with 64 weights * k_quants: WIP super-blocks with 64 weights Q6_K scalar and AVX2 works * k_quants: WIP super-blocks with 64 weights Q4_K scalar and AVX2 works * k_quants: WIP super-blocks with 64 weights Q2_K scalar and AVX2 works. Q2_K is way too slow (it is actually slower than the scalar implementation) * k_quants: WIP super-blocks with 64 weights Q3_K scalar and AVX2 works. * k_quants: WIP super-blocks with 64 weights Q5_K scalar and AVX2 works, and with that all k_quants are done on AVX2 and scalar * k_quants: WIP super-blocks with 64 weights Q6_K working on CUDA. Cannot make it run quite as gast as with super-blocks with 256 weigths: 8% slower on 4080, 20% slower on the 1660 (but there we fit 1 less layer on the GPU because pf the larger model size), so some fraction of these 20% is due to that, * k_quants: WIP super-blocks with 64 weights Q4_K working on CUDA. ~10% slower on GTX-1660, 16% slower on 4080. * k_quants: WIP super-blocks with 64 weights Q2_K working on CUDA. ~3% slower on GTX-1660, 10% slower on 4080. * k_quants: WIP super-blocks with 64 weights Q3_K working on CUDA. * k_quants: WIP super-blocks with 64 weights Q5_K working on CUDA, and with this CUDA is done. * k_quants: WIP super-blocks with 64 weights Q6_K working on ARM_NEON * k_quants: WIP super-blocks with 64 weights Q4_K working on ARM_NEON, but quite a bit slower than 256 weights * k_quants: WIP super-blocks with 64 weights Q2_K working on ARM_NEON, but quite a bit slower than 256 weights * k_quants: WIP super-blocks with 64 weights Q3_K working on ARM_NEON, but quite a bit slower than 256 weights. * k_quants: WIP super-blocks with 64 weights Q5_K working on ARM_NEON, but quite a bit slower than 256 weights. With that, we have full support for ARM_NEON, although performance is not quite there. * k_quants: WIP super-blocks with 64 weights Slightly more efficient Q3_K and Q5_K * k_quants: WIP super-blocks with 64 weights Another small improvement for Q3_K and Q5_K on ARM_NEON * k_quants: WIP super-blocks with 64 weights Yet another speedup for Q5_K on ARM_NEON. We are now within 10% of the QK_K = 256 version. * k_quants: WIP super-blocks with 64 weights * We are able to pass preprocessor macros to the Metal compiler * Q6_K works and is actually slightly more efficient than the QK_K = 256 version (25.2 ms vs 25.8 ms) * k_quants: WIP super-blocks with 64 weights Q4_K works on Metal and is actually slightly faster than QK_K = 256 (21.95 ms vs 24.0 ms). * k_quants: WIP super-blocks with 64 weights Q2_K works on Metal and is very slightly faster than QK_K = 256 (23.8 ms vs 24.2 ms). * k_quants: WIP super-blocks with 64 weights Q3_K works on Metal and is slightly faster than QK_K = 256 (26.6 ms vs 28.3 ms). * k_quants: WIP super-blocks with 64 weights Q5_K works on Metal and is slightly faster than QK_K = 256 (23.7 ms vs 26.3 ms). * k_quants: call them _K, not _k, also on Metal * k_quants: correctly define QK_K in llama.cpp * Fixed bug in q4_K quantization added with the 64-block addition * Simplify via lambda * k_quants: swicth Q3_K to 4-bit scales when QK_K = 64 Otherwise there isn't much benefit from this quantization type. There is some very slight loss in accuracy, but we reduce size by ~7%. E.g., for OpenLLaMA-3B, Q3_K_S perplexity is 8.6131 with 8-bit scales and 8.6352 with 4-bit, while file size decreases from 1.53G to 1.44G. * k_quants: switch Q4_K to 4-bit scales when QK_K = 64 Here the loss in accuracy is greater than for Q3_K, but the Q4_K points still move further to the left on the perplexity vs size curve. * k_quants: forgot to add the Metal changes in last commit * k_quants: change Q5_K to be type 0 when QK_K = 64 Still needs AVX2 implementation * k_quants: AVX2 implementation for new 64-weight Q5_K * k_quants: 10% faster ARM_NEON Q5_K dot product * k_quants: fixed issue caused by merging with master --------- Co-authored-by: Iwan Kawrakow <iwan.kawrakow@gmail.com>
2023-06-26 16:43:07 +00:00
}
ggml : add SOTA 2,3,4,5,6 bit k-quantizations (#1684) * Starting to add k-quantization to ggml I think it is better to have quantization separate from ggml. For now just adding the k-quants there, but it would be better to also factor out the existing ggml quantizations. * Adding Q3_K and Q8_K (de)-quantization * Q3_K now working on CUDA and AVX2/scalar CUDA is not ideal - ~50% slower than Q4_0 for single token prediction, about the same in batch mode (perplexity). CPU single token is ~55 ms (on Ryzen 7950X). * Some improvement for Q3_K on CUDA It is now ~22.5 ms/token on my GPU, so ~30% slower than Q4_0. * Some more CUDA optimizations for Q3_K Single token is now 20.5 ms/token (~20% slower than Q4_0). Perplexity is on par with Q4_0. * Adding Q4_K - scalar, AVX2, CUDA Performance is the same or perhaps very slightly better than Q4_0 on the CPU. On the GPU, single token prediction is ~10% better than Q4_0, batch mode (perplexity is about the same). * Adding Q6_K - scalar, AVX2, CUDA Performance is ~40% lower compared to Q4_K on the CPU. This is to be expected, considering that we are memory bound on the CPU and the 6-bit model is ~44% larger than the 4-bit. On the GPU, single token prediction is ~6% lower than Q4_0, batch mode (perplexity) is even closer (but still slower). * Adding Q5_K - scalar, AVX2, CUDA Performance is ~20% lower compared to Q4_K on the CPU. This is to be expected, considering that we are memory bound on the CPU and the 5-bit model is ~22% larger than the 4-bit. On the GPU, single token prediction is about the same as Q4_0 for both, single token and batch prediction. * Per convention, all QX_K quantizations use Q5_K for output.weight * Adding quantization mixes * Quantization mixes: didn't quite get what I wanted in the last commit * Q4_K dot product for ARM_NEON * Q6_K dot product for ARM_NEON * Q5_K dot product for ARM_NEON * Adding Q3_K dot for ARM_NEON It is 22% slower than Q4_K, despite the smaller model size. On x86_64, where we are memory bound, the Q3_K model is quite a bit faster than Q4_K. * A very slightly faster ARM_NEON Q3_K dot * Adding Q2_K - just CUDA for now Token prediction is pretty good - about 15.5 ms on a RTX 4080. Perplexity is about the same as Q4_K. * Adding scalar and AVX2 Q2_K dot * Adding ARM_NEON Q2_K dot About the same performance as Q4_K. * A slightly faster ARM_NEON Q2_K dot Single token prediction is now ~36 ms on M2 Max. The code is much simpler too. * Fixed bug in Q2_K CUDA dot product kernel Stranegly enough, for the few prompts I tried with the 7B model the responses looked perfectly reasonable. Only realized something is not quite right when I tried the larger models and started getting nonse back. In any case, Q2_K single token evaluation time on an RTX 4080 in a Ryzen7950X box iusing CUDA and model fully loaded on the GPU are ~15.5 ms for 7B, ~25.4 ms for 13B, and ~55.8 ms for 30B. The max number of layers that fit in VRAM for The 65B is 32. With that, we get ~330 ms per token, which is not that much faster than just running on the CPU (~470 ms per token). * Don't print zeros/NaNs when no count histogram has been collected * A 10% faster CUDA vector dot kernel for Q3_K Q3_K is now running at ~18.5 ms / token on CUDA, so the gap to Q4_0 is only 10%. It seems memory acccess pattern is more important for performance than the amount of computation the kernel does. * A slightly daster Q4_K AVX2 dot product For perplexity, where we are less memory bound, time per pass drops by ~5%. Barely measurable difference for single token prediction. * A slightly faster ARM_NEON A4_K dot product * Minor * Fix quantization error test We cannot possibly be expecting rmse < 0.002 for 2- and 3-bit quantization variants. * Fix docker build I have been sloppy with vector reinterpret casts on ARM_NEON. It seems clang is very forgiving in that regard. * Added forgotten ggml.o dependence on k_quants.h to the Makefile * Had unintentionally committed the Makefile with -Ofast enabled * ggml : rename k_quants -> ggml-quants-k, use lowercase in code --------- Co-authored-by: Iwan Kawrakow <iwan.kawrakow@gmail.com> Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
2023-06-05 19:56:18 +00:00
ggml : move FP16 <-> FP32 code to ggml-impl.h (#3861) * ggml : move FP16 <-> FP32 stuff to ggml-impl.h ggml-ci * tests : fix ARM build * ggml : explicitly initialize deprecated type traits * ggml : add math.h to ggml-impl.h * ggml : remove duplicate static assert macros * ggml : prefix lookup tables with ggml_ ggml-ci * ggml-impl : move extern "C" to start of file
2023-10-30 17:19:15 +00:00
sumf += (GGML_FP16_TO_FP32(x[i].d)*y[i].d)*sumi + GGML_FP16_TO_FP32(x[i].m)*y[i].s;
ggml : add SOTA 2,3,4,5,6 bit k-quantizations (#1684) * Starting to add k-quantization to ggml I think it is better to have quantization separate from ggml. For now just adding the k-quants there, but it would be better to also factor out the existing ggml quantizations. * Adding Q3_K and Q8_K (de)-quantization * Q3_K now working on CUDA and AVX2/scalar CUDA is not ideal - ~50% slower than Q4_0 for single token prediction, about the same in batch mode (perplexity). CPU single token is ~55 ms (on Ryzen 7950X). * Some improvement for Q3_K on CUDA It is now ~22.5 ms/token on my GPU, so ~30% slower than Q4_0. * Some more CUDA optimizations for Q3_K Single token is now 20.5 ms/token (~20% slower than Q4_0). Perplexity is on par with Q4_0. * Adding Q4_K - scalar, AVX2, CUDA Performance is the same or perhaps very slightly better than Q4_0 on the CPU. On the GPU, single token prediction is ~10% better than Q4_0, batch mode (perplexity is about the same). * Adding Q6_K - scalar, AVX2, CUDA Performance is ~40% lower compared to Q4_K on the CPU. This is to be expected, considering that we are memory bound on the CPU and the 6-bit model is ~44% larger than the 4-bit. On the GPU, single token prediction is ~6% lower than Q4_0, batch mode (perplexity) is even closer (but still slower). * Adding Q5_K - scalar, AVX2, CUDA Performance is ~20% lower compared to Q4_K on the CPU. This is to be expected, considering that we are memory bound on the CPU and the 5-bit model is ~22% larger than the 4-bit. On the GPU, single token prediction is about the same as Q4_0 for both, single token and batch prediction. * Per convention, all QX_K quantizations use Q5_K for output.weight * Adding quantization mixes * Quantization mixes: didn't quite get what I wanted in the last commit * Q4_K dot product for ARM_NEON * Q6_K dot product for ARM_NEON * Q5_K dot product for ARM_NEON * Adding Q3_K dot for ARM_NEON It is 22% slower than Q4_K, despite the smaller model size. On x86_64, where we are memory bound, the Q3_K model is quite a bit faster than Q4_K. * A very slightly faster ARM_NEON Q3_K dot * Adding Q2_K - just CUDA for now Token prediction is pretty good - about 15.5 ms on a RTX 4080. Perplexity is about the same as Q4_K. * Adding scalar and AVX2 Q2_K dot * Adding ARM_NEON Q2_K dot About the same performance as Q4_K. * A slightly faster ARM_NEON Q2_K dot Single token prediction is now ~36 ms on M2 Max. The code is much simpler too. * Fixed bug in Q2_K CUDA dot product kernel Stranegly enough, for the few prompts I tried with the 7B model the responses looked perfectly reasonable. Only realized something is not quite right when I tried the larger models and started getting nonse back. In any case, Q2_K single token evaluation time on an RTX 4080 in a Ryzen7950X box iusing CUDA and model fully loaded on the GPU are ~15.5 ms for 7B, ~25.4 ms for 13B, and ~55.8 ms for 30B. The max number of layers that fit in VRAM for The 65B is 32. With that, we get ~330 ms per token, which is not that much faster than just running on the CPU (~470 ms per token). * Don't print zeros/NaNs when no count histogram has been collected * A 10% faster CUDA vector dot kernel for Q3_K Q3_K is now running at ~18.5 ms / token on CUDA, so the gap to Q4_0 is only 10%. It seems memory acccess pattern is more important for performance than the amount of computation the kernel does. * A slightly daster Q4_K AVX2 dot product For perplexity, where we are less memory bound, time per pass drops by ~5%. Barely measurable difference for single token prediction. * A slightly faster ARM_NEON A4_K dot product * Minor * Fix quantization error test We cannot possibly be expecting rmse < 0.002 for 2- and 3-bit quantization variants. * Fix docker build I have been sloppy with vector reinterpret casts on ARM_NEON. It seems clang is very forgiving in that regard. * Added forgotten ggml.o dependence on k_quants.h to the Makefile * Had unintentionally committed the Makefile with -Ofast enabled * ggml : rename k_quants -> ggml-quants-k, use lowercase in code --------- Co-authored-by: Iwan Kawrakow <iwan.kawrakow@gmail.com> Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
2023-06-05 19:56:18 +00:00
}
ggml : quantization refactoring (#3833) * ggml : factor all quantization code in ggml-quants ggml-ci * ggml-quants : fix Zig and Swift builds + quantize tool ggml-ci * quantize : --pure option for disabling k-quant mixtures --------- Co-authored-by: cebtenzzre <cebtenzzre@gmail.com>
2023-10-29 16:32:28 +00:00
*s = sumf;
#endif
ggml : add SOTA 2,3,4,5,6 bit k-quantizations (#1684) * Starting to add k-quantization to ggml I think it is better to have quantization separate from ggml. For now just adding the k-quants there, but it would be better to also factor out the existing ggml quantizations. * Adding Q3_K and Q8_K (de)-quantization * Q3_K now working on CUDA and AVX2/scalar CUDA is not ideal - ~50% slower than Q4_0 for single token prediction, about the same in batch mode (perplexity). CPU single token is ~55 ms (on Ryzen 7950X). * Some improvement for Q3_K on CUDA It is now ~22.5 ms/token on my GPU, so ~30% slower than Q4_0. * Some more CUDA optimizations for Q3_K Single token is now 20.5 ms/token (~20% slower than Q4_0). Perplexity is on par with Q4_0. * Adding Q4_K - scalar, AVX2, CUDA Performance is the same or perhaps very slightly better than Q4_0 on the CPU. On the GPU, single token prediction is ~10% better than Q4_0, batch mode (perplexity is about the same). * Adding Q6_K - scalar, AVX2, CUDA Performance is ~40% lower compared to Q4_K on the CPU. This is to be expected, considering that we are memory bound on the CPU and the 6-bit model is ~44% larger than the 4-bit. On the GPU, single token prediction is ~6% lower than Q4_0, batch mode (perplexity) is even closer (but still slower). * Adding Q5_K - scalar, AVX2, CUDA Performance is ~20% lower compared to Q4_K on the CPU. This is to be expected, considering that we are memory bound on the CPU and the 5-bit model is ~22% larger than the 4-bit. On the GPU, single token prediction is about the same as Q4_0 for both, single token and batch prediction. * Per convention, all QX_K quantizations use Q5_K for output.weight * Adding quantization mixes * Quantization mixes: didn't quite get what I wanted in the last commit * Q4_K dot product for ARM_NEON * Q6_K dot product for ARM_NEON * Q5_K dot product for ARM_NEON * Adding Q3_K dot for ARM_NEON It is 22% slower than Q4_K, despite the smaller model size. On x86_64, where we are memory bound, the Q3_K model is quite a bit faster than Q4_K. * A very slightly faster ARM_NEON Q3_K dot * Adding Q2_K - just CUDA for now Token prediction is pretty good - about 15.5 ms on a RTX 4080. Perplexity is about the same as Q4_K. * Adding scalar and AVX2 Q2_K dot * Adding ARM_NEON Q2_K dot About the same performance as Q4_K. * A slightly faster ARM_NEON Q2_K dot Single token prediction is now ~36 ms on M2 Max. The code is much simpler too. * Fixed bug in Q2_K CUDA dot product kernel Stranegly enough, for the few prompts I tried with the 7B model the responses looked perfectly reasonable. Only realized something is not quite right when I tried the larger models and started getting nonse back. In any case, Q2_K single token evaluation time on an RTX 4080 in a Ryzen7950X box iusing CUDA and model fully loaded on the GPU are ~15.5 ms for 7B, ~25.4 ms for 13B, and ~55.8 ms for 30B. The max number of layers that fit in VRAM for The 65B is 32. With that, we get ~330 ms per token, which is not that much faster than just running on the CPU (~470 ms per token). * Don't print zeros/NaNs when no count histogram has been collected * A 10% faster CUDA vector dot kernel for Q3_K Q3_K is now running at ~18.5 ms / token on CUDA, so the gap to Q4_0 is only 10%. It seems memory acccess pattern is more important for performance than the amount of computation the kernel does. * A slightly daster Q4_K AVX2 dot product For perplexity, where we are less memory bound, time per pass drops by ~5%. Barely measurable difference for single token prediction. * A slightly faster ARM_NEON A4_K dot product * Minor * Fix quantization error test We cannot possibly be expecting rmse < 0.002 for 2- and 3-bit quantization variants. * Fix docker build I have been sloppy with vector reinterpret casts on ARM_NEON. It seems clang is very forgiving in that regard. * Added forgotten ggml.o dependence on k_quants.h to the Makefile * Had unintentionally committed the Makefile with -Ofast enabled * ggml : rename k_quants -> ggml-quants-k, use lowercase in code --------- Co-authored-by: Iwan Kawrakow <iwan.kawrakow@gmail.com> Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
2023-06-05 19:56:18 +00:00
}
ggml : quantization refactoring (#3833) * ggml : factor all quantization code in ggml-quants ggml-ci * ggml-quants : fix Zig and Swift builds + quantize tool ggml-ci * quantize : --pure option for disabling k-quant mixtures --------- Co-authored-by: cebtenzzre <cebtenzzre@gmail.com>
2023-10-29 16:32:28 +00:00
void ggml_vec_dot_q8_0_q8_0(const int n, float * restrict s, const void * restrict vx, const void * restrict vy) {
const int qk = QK8_0;
const int nb = n / qk;
ggml : add SOTA 2,3,4,5,6 bit k-quantizations (#1684) * Starting to add k-quantization to ggml I think it is better to have quantization separate from ggml. For now just adding the k-quants there, but it would be better to also factor out the existing ggml quantizations. * Adding Q3_K and Q8_K (de)-quantization * Q3_K now working on CUDA and AVX2/scalar CUDA is not ideal - ~50% slower than Q4_0 for single token prediction, about the same in batch mode (perplexity). CPU single token is ~55 ms (on Ryzen 7950X). * Some improvement for Q3_K on CUDA It is now ~22.5 ms/token on my GPU, so ~30% slower than Q4_0. * Some more CUDA optimizations for Q3_K Single token is now 20.5 ms/token (~20% slower than Q4_0). Perplexity is on par with Q4_0. * Adding Q4_K - scalar, AVX2, CUDA Performance is the same or perhaps very slightly better than Q4_0 on the CPU. On the GPU, single token prediction is ~10% better than Q4_0, batch mode (perplexity is about the same). * Adding Q6_K - scalar, AVX2, CUDA Performance is ~40% lower compared to Q4_K on the CPU. This is to be expected, considering that we are memory bound on the CPU and the 6-bit model is ~44% larger than the 4-bit. On the GPU, single token prediction is ~6% lower than Q4_0, batch mode (perplexity) is even closer (but still slower). * Adding Q5_K - scalar, AVX2, CUDA Performance is ~20% lower compared to Q4_K on the CPU. This is to be expected, considering that we are memory bound on the CPU and the 5-bit model is ~22% larger than the 4-bit. On the GPU, single token prediction is about the same as Q4_0 for both, single token and batch prediction. * Per convention, all QX_K quantizations use Q5_K for output.weight * Adding quantization mixes * Quantization mixes: didn't quite get what I wanted in the last commit * Q4_K dot product for ARM_NEON * Q6_K dot product for ARM_NEON * Q5_K dot product for ARM_NEON * Adding Q3_K dot for ARM_NEON It is 22% slower than Q4_K, despite the smaller model size. On x86_64, where we are memory bound, the Q3_K model is quite a bit faster than Q4_K. * A very slightly faster ARM_NEON Q3_K dot * Adding Q2_K - just CUDA for now Token prediction is pretty good - about 15.5 ms on a RTX 4080. Perplexity is about the same as Q4_K. * Adding scalar and AVX2 Q2_K dot * Adding ARM_NEON Q2_K dot About the same performance as Q4_K. * A slightly faster ARM_NEON Q2_K dot Single token prediction is now ~36 ms on M2 Max. The code is much simpler too. * Fixed bug in Q2_K CUDA dot product kernel Stranegly enough, for the few prompts I tried with the 7B model the responses looked perfectly reasonable. Only realized something is not quite right when I tried the larger models and started getting nonse back. In any case, Q2_K single token evaluation time on an RTX 4080 in a Ryzen7950X box iusing CUDA and model fully loaded on the GPU are ~15.5 ms for 7B, ~25.4 ms for 13B, and ~55.8 ms for 30B. The max number of layers that fit in VRAM for The 65B is 32. With that, we get ~330 ms per token, which is not that much faster than just running on the CPU (~470 ms per token). * Don't print zeros/NaNs when no count histogram has been collected * A 10% faster CUDA vector dot kernel for Q3_K Q3_K is now running at ~18.5 ms / token on CUDA, so the gap to Q4_0 is only 10%. It seems memory acccess pattern is more important for performance than the amount of computation the kernel does. * A slightly daster Q4_K AVX2 dot product For perplexity, where we are less memory bound, time per pass drops by ~5%. Barely measurable difference for single token prediction. * A slightly faster ARM_NEON A4_K dot product * Minor * Fix quantization error test We cannot possibly be expecting rmse < 0.002 for 2- and 3-bit quantization variants. * Fix docker build I have been sloppy with vector reinterpret casts on ARM_NEON. It seems clang is very forgiving in that regard. * Added forgotten ggml.o dependence on k_quants.h to the Makefile * Had unintentionally committed the Makefile with -Ofast enabled * ggml : rename k_quants -> ggml-quants-k, use lowercase in code --------- Co-authored-by: Iwan Kawrakow <iwan.kawrakow@gmail.com> Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
2023-06-05 19:56:18 +00:00
ggml : quantization refactoring (#3833) * ggml : factor all quantization code in ggml-quants ggml-ci * ggml-quants : fix Zig and Swift builds + quantize tool ggml-ci * quantize : --pure option for disabling k-quant mixtures --------- Co-authored-by: cebtenzzre <cebtenzzre@gmail.com>
2023-10-29 16:32:28 +00:00
assert(n % qk == 0);
const block_q8_0 * restrict x = vx;
const block_q8_0 * restrict y = vy;
#if defined(__ARM_NEON)
float32x4_t sumv0 = vdupq_n_f32(0.0f);
float32x4_t sumv1 = vdupq_n_f32(0.0f);
assert(nb % 2 == 0); // TODO: handle odd nb
for (int i = 0; i < nb; i += 2) {
const block_q8_0 * restrict x0 = &x[i + 0];
const block_q8_0 * restrict x1 = &x[i + 1];
const block_q8_0 * restrict y0 = &y[i + 0];
const block_q8_0 * restrict y1 = &y[i + 1];
const int8x16_t x0_0 = vld1q_s8(x0->qs);
const int8x16_t x0_1 = vld1q_s8(x0->qs + 16);
const int8x16_t x1_0 = vld1q_s8(x1->qs);
const int8x16_t x1_1 = vld1q_s8(x1->qs + 16);
// load y
const int8x16_t y0_0 = vld1q_s8(y0->qs);
const int8x16_t y0_1 = vld1q_s8(y0->qs + 16);
const int8x16_t y1_0 = vld1q_s8(y1->qs);
const int8x16_t y1_1 = vld1q_s8(y1->qs + 16);
sumv0 = vmlaq_n_f32(sumv0, vcvtq_f32_s32(vaddq_s32(
ggml : add ggml_vdotq_s32 alias (#4715) ggml-ci
2023-12-31 09:43:31 +00:00
ggml_vdotq_s32(vdupq_n_s32(0), x0_0, y0_0),
ggml_vdotq_s32(vdupq_n_s32(0), x0_1, y0_1))), GGML_FP16_TO_FP32(x0->d)*GGML_FP16_TO_FP32(y0->d));
ggml : quantization refactoring (#3833) * ggml : factor all quantization code in ggml-quants ggml-ci * ggml-quants : fix Zig and Swift builds + quantize tool ggml-ci * quantize : --pure option for disabling k-quant mixtures --------- Co-authored-by: cebtenzzre <cebtenzzre@gmail.com>
2023-10-29 16:32:28 +00:00
sumv1 = vmlaq_n_f32(sumv1, vcvtq_f32_s32(vaddq_s32(
ggml : add ggml_vdotq_s32 alias (#4715) ggml-ci
2023-12-31 09:43:31 +00:00
ggml_vdotq_s32(vdupq_n_s32(0), x1_0, y1_0),
ggml_vdotq_s32(vdupq_n_s32(0), x1_1, y1_1))), GGML_FP16_TO_FP32(x1->d)*GGML_FP16_TO_FP32(y1->d));
ggml : add SOTA 2,3,4,5,6 bit k-quantizations (#1684) * Starting to add k-quantization to ggml I think it is better to have quantization separate from ggml. For now just adding the k-quants there, but it would be better to also factor out the existing ggml quantizations. * Adding Q3_K and Q8_K (de)-quantization * Q3_K now working on CUDA and AVX2/scalar CUDA is not ideal - ~50% slower than Q4_0 for single token prediction, about the same in batch mode (perplexity). CPU single token is ~55 ms (on Ryzen 7950X). * Some improvement for Q3_K on CUDA It is now ~22.5 ms/token on my GPU, so ~30% slower than Q4_0. * Some more CUDA optimizations for Q3_K Single token is now 20.5 ms/token (~20% slower than Q4_0). Perplexity is on par with Q4_0. * Adding Q4_K - scalar, AVX2, CUDA Performance is the same or perhaps very slightly better than Q4_0 on the CPU. On the GPU, single token prediction is ~10% better than Q4_0, batch mode (perplexity is about the same). * Adding Q6_K - scalar, AVX2, CUDA Performance is ~40% lower compared to Q4_K on the CPU. This is to be expected, considering that we are memory bound on the CPU and the 6-bit model is ~44% larger than the 4-bit. On the GPU, single token prediction is ~6% lower than Q4_0, batch mode (perplexity) is even closer (but still slower). * Adding Q5_K - scalar, AVX2, CUDA Performance is ~20% lower compared to Q4_K on the CPU. This is to be expected, considering that we are memory bound on the CPU and the 5-bit model is ~22% larger than the 4-bit. On the GPU, single token prediction is about the same as Q4_0 for both, single token and batch prediction. * Per convention, all QX_K quantizations use Q5_K for output.weight * Adding quantization mixes * Quantization mixes: didn't quite get what I wanted in the last commit * Q4_K dot product for ARM_NEON * Q6_K dot product for ARM_NEON * Q5_K dot product for ARM_NEON * Adding Q3_K dot for ARM_NEON It is 22% slower than Q4_K, despite the smaller model size. On x86_64, where we are memory bound, the Q3_K model is quite a bit faster than Q4_K. * A very slightly faster ARM_NEON Q3_K dot * Adding Q2_K - just CUDA for now Token prediction is pretty good - about 15.5 ms on a RTX 4080. Perplexity is about the same as Q4_K. * Adding scalar and AVX2 Q2_K dot * Adding ARM_NEON Q2_K dot About the same performance as Q4_K. * A slightly faster ARM_NEON Q2_K dot Single token prediction is now ~36 ms on M2 Max. The code is much simpler too. * Fixed bug in Q2_K CUDA dot product kernel Stranegly enough, for the few prompts I tried with the 7B model the responses looked perfectly reasonable. Only realized something is not quite right when I tried the larger models and started getting nonse back. In any case, Q2_K single token evaluation time on an RTX 4080 in a Ryzen7950X box iusing CUDA and model fully loaded on the GPU are ~15.5 ms for 7B, ~25.4 ms for 13B, and ~55.8 ms for 30B. The max number of layers that fit in VRAM for The 65B is 32. With that, we get ~330 ms per token, which is not that much faster than just running on the CPU (~470 ms per token). * Don't print zeros/NaNs when no count histogram has been collected * A 10% faster CUDA vector dot kernel for Q3_K Q3_K is now running at ~18.5 ms / token on CUDA, so the gap to Q4_0 is only 10%. It seems memory acccess pattern is more important for performance than the amount of computation the kernel does. * A slightly daster Q4_K AVX2 dot product For perplexity, where we are less memory bound, time per pass drops by ~5%. Barely measurable difference for single token prediction. * A slightly faster ARM_NEON A4_K dot product * Minor * Fix quantization error test We cannot possibly be expecting rmse < 0.002 for 2- and 3-bit quantization variants. * Fix docker build I have been sloppy with vector reinterpret casts on ARM_NEON. It seems clang is very forgiving in that regard. * Added forgotten ggml.o dependence on k_quants.h to the Makefile * Had unintentionally committed the Makefile with -Ofast enabled * ggml : rename k_quants -> ggml-quants-k, use lowercase in code --------- Co-authored-by: Iwan Kawrakow <iwan.kawrakow@gmail.com> Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
2023-06-05 19:56:18 +00:00
}
ggml : quantization refactoring (#3833) * ggml : factor all quantization code in ggml-quants ggml-ci * ggml-quants : fix Zig and Swift builds + quantize tool ggml-ci * quantize : --pure option for disabling k-quant mixtures --------- Co-authored-by: cebtenzzre <cebtenzzre@gmail.com>
2023-10-29 16:32:28 +00:00
*s = vaddvq_f32(sumv0) + vaddvq_f32(sumv1);
#elif defined(__AVX2__) || defined(__AVX__)
// Initialize accumulator with zeros
__m256 acc = _mm256_setzero_ps();
ggml : add SOTA 2,3,4,5,6 bit k-quantizations (#1684) * Starting to add k-quantization to ggml I think it is better to have quantization separate from ggml. For now just adding the k-quants there, but it would be better to also factor out the existing ggml quantizations. * Adding Q3_K and Q8_K (de)-quantization * Q3_K now working on CUDA and AVX2/scalar CUDA is not ideal - ~50% slower than Q4_0 for single token prediction, about the same in batch mode (perplexity). CPU single token is ~55 ms (on Ryzen 7950X). * Some improvement for Q3_K on CUDA It is now ~22.5 ms/token on my GPU, so ~30% slower than Q4_0. * Some more CUDA optimizations for Q3_K Single token is now 20.5 ms/token (~20% slower than Q4_0). Perplexity is on par with Q4_0. * Adding Q4_K - scalar, AVX2, CUDA Performance is the same or perhaps very slightly better than Q4_0 on the CPU. On the GPU, single token prediction is ~10% better than Q4_0, batch mode (perplexity is about the same). * Adding Q6_K - scalar, AVX2, CUDA Performance is ~40% lower compared to Q4_K on the CPU. This is to be expected, considering that we are memory bound on the CPU and the 6-bit model is ~44% larger than the 4-bit. On the GPU, single token prediction is ~6% lower than Q4_0, batch mode (perplexity) is even closer (but still slower). * Adding Q5_K - scalar, AVX2, CUDA Performance is ~20% lower compared to Q4_K on the CPU. This is to be expected, considering that we are memory bound on the CPU and the 5-bit model is ~22% larger than the 4-bit. On the GPU, single token prediction is about the same as Q4_0 for both, single token and batch prediction. * Per convention, all QX_K quantizations use Q5_K for output.weight * Adding quantization mixes * Quantization mixes: didn't quite get what I wanted in the last commit * Q4_K dot product for ARM_NEON * Q6_K dot product for ARM_NEON * Q5_K dot product for ARM_NEON * Adding Q3_K dot for ARM_NEON It is 22% slower than Q4_K, despite the smaller model size. On x86_64, where we are memory bound, the Q3_K model is quite a bit faster than Q4_K. * A very slightly faster ARM_NEON Q3_K dot * Adding Q2_K - just CUDA for now Token prediction is pretty good - about 15.5 ms on a RTX 4080. Perplexity is about the same as Q4_K. * Adding scalar and AVX2 Q2_K dot * Adding ARM_NEON Q2_K dot About the same performance as Q4_K. * A slightly faster ARM_NEON Q2_K dot Single token prediction is now ~36 ms on M2 Max. The code is much simpler too. * Fixed bug in Q2_K CUDA dot product kernel Stranegly enough, for the few prompts I tried with the 7B model the responses looked perfectly reasonable. Only realized something is not quite right when I tried the larger models and started getting nonse back. In any case, Q2_K single token evaluation time on an RTX 4080 in a Ryzen7950X box iusing CUDA and model fully loaded on the GPU are ~15.5 ms for 7B, ~25.4 ms for 13B, and ~55.8 ms for 30B. The max number of layers that fit in VRAM for The 65B is 32. With that, we get ~330 ms per token, which is not that much faster than just running on the CPU (~470 ms per token). * Don't print zeros/NaNs when no count histogram has been collected * A 10% faster CUDA vector dot kernel for Q3_K Q3_K is now running at ~18.5 ms / token on CUDA, so the gap to Q4_0 is only 10%. It seems memory acccess pattern is more important for performance than the amount of computation the kernel does. * A slightly daster Q4_K AVX2 dot product For perplexity, where we are less memory bound, time per pass drops by ~5%. Barely measurable difference for single token prediction. * A slightly faster ARM_NEON A4_K dot product * Minor * Fix quantization error test We cannot possibly be expecting rmse < 0.002 for 2- and 3-bit quantization variants. * Fix docker build I have been sloppy with vector reinterpret casts on ARM_NEON. It seems clang is very forgiving in that regard. * Added forgotten ggml.o dependence on k_quants.h to the Makefile * Had unintentionally committed the Makefile with -Ofast enabled * ggml : rename k_quants -> ggml-quants-k, use lowercase in code --------- Co-authored-by: Iwan Kawrakow <iwan.kawrakow@gmail.com> Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
2023-06-05 19:56:18 +00:00
ggml : quantization refactoring (#3833) * ggml : factor all quantization code in ggml-quants ggml-ci * ggml-quants : fix Zig and Swift builds + quantize tool ggml-ci * quantize : --pure option for disabling k-quant mixtures --------- Co-authored-by: cebtenzzre <cebtenzzre@gmail.com>
2023-10-29 16:32:28 +00:00
// Main loop
for (int i = 0; i < nb; ++i) {
// Compute combined scale for the block
ggml : move FP16 <-> FP32 code to ggml-impl.h (#3861) * ggml : move FP16 <-> FP32 stuff to ggml-impl.h ggml-ci * tests : fix ARM build * ggml : explicitly initialize deprecated type traits * ggml : add math.h to ggml-impl.h * ggml : remove duplicate static assert macros * ggml : prefix lookup tables with ggml_ ggml-ci * ggml-impl : move extern "C" to start of file
2023-10-30 17:19:15 +00:00
const __m256 d = _mm256_set1_ps(GGML_FP16_TO_FP32(x[i].d) * GGML_FP16_TO_FP32(y[i].d));
ggml : quantization refactoring (#3833) * ggml : factor all quantization code in ggml-quants ggml-ci * ggml-quants : fix Zig and Swift builds + quantize tool ggml-ci * quantize : --pure option for disabling k-quant mixtures --------- Co-authored-by: cebtenzzre <cebtenzzre@gmail.com>
2023-10-29 16:32:28 +00:00
__m256i bx = _mm256_loadu_si256((const __m256i *)x[i].qs);
__m256i by = _mm256_loadu_si256((const __m256i *)y[i].qs);
ggml : add SOTA 2,3,4,5,6 bit k-quantizations (#1684) * Starting to add k-quantization to ggml I think it is better to have quantization separate from ggml. For now just adding the k-quants there, but it would be better to also factor out the existing ggml quantizations. * Adding Q3_K and Q8_K (de)-quantization * Q3_K now working on CUDA and AVX2/scalar CUDA is not ideal - ~50% slower than Q4_0 for single token prediction, about the same in batch mode (perplexity). CPU single token is ~55 ms (on Ryzen 7950X). * Some improvement for Q3_K on CUDA It is now ~22.5 ms/token on my GPU, so ~30% slower than Q4_0. * Some more CUDA optimizations for Q3_K Single token is now 20.5 ms/token (~20% slower than Q4_0). Perplexity is on par with Q4_0. * Adding Q4_K - scalar, AVX2, CUDA Performance is the same or perhaps very slightly better than Q4_0 on the CPU. On the GPU, single token prediction is ~10% better than Q4_0, batch mode (perplexity is about the same). * Adding Q6_K - scalar, AVX2, CUDA Performance is ~40% lower compared to Q4_K on the CPU. This is to be expected, considering that we are memory bound on the CPU and the 6-bit model is ~44% larger than the 4-bit. On the GPU, single token prediction is ~6% lower than Q4_0, batch mode (perplexity) is even closer (but still slower). * Adding Q5_K - scalar, AVX2, CUDA Performance is ~20% lower compared to Q4_K on the CPU. This is to be expected, considering that we are memory bound on the CPU and the 5-bit model is ~22% larger than the 4-bit. On the GPU, single token prediction is about the same as Q4_0 for both, single token and batch prediction. * Per convention, all QX_K quantizations use Q5_K for output.weight * Adding quantization mixes * Quantization mixes: didn't quite get what I wanted in the last commit * Q4_K dot product for ARM_NEON * Q6_K dot product for ARM_NEON * Q5_K dot product for ARM_NEON * Adding Q3_K dot for ARM_NEON It is 22% slower than Q4_K, despite the smaller model size. On x86_64, where we are memory bound, the Q3_K model is quite a bit faster than Q4_K. * A very slightly faster ARM_NEON Q3_K dot * Adding Q2_K - just CUDA for now Token prediction is pretty good - about 15.5 ms on a RTX 4080. Perplexity is about the same as Q4_K. * Adding scalar and AVX2 Q2_K dot * Adding ARM_NEON Q2_K dot About the same performance as Q4_K. * A slightly faster ARM_NEON Q2_K dot Single token prediction is now ~36 ms on M2 Max. The code is much simpler too. * Fixed bug in Q2_K CUDA dot product kernel Stranegly enough, for the few prompts I tried with the 7B model the responses looked perfectly reasonable. Only realized something is not quite right when I tried the larger models and started getting nonse back. In any case, Q2_K single token evaluation time on an RTX 4080 in a Ryzen7950X box iusing CUDA and model fully loaded on the GPU are ~15.5 ms for 7B, ~25.4 ms for 13B, and ~55.8 ms for 30B. The max number of layers that fit in VRAM for The 65B is 32. With that, we get ~330 ms per token, which is not that much faster than just running on the CPU (~470 ms per token). * Don't print zeros/NaNs when no count histogram has been collected * A 10% faster CUDA vector dot kernel for Q3_K Q3_K is now running at ~18.5 ms / token on CUDA, so the gap to Q4_0 is only 10%. It seems memory acccess pattern is more important for performance than the amount of computation the kernel does. * A slightly daster Q4_K AVX2 dot product For perplexity, where we are less memory bound, time per pass drops by ~5%. Barely measurable difference for single token prediction. * A slightly faster ARM_NEON A4_K dot product * Minor * Fix quantization error test We cannot possibly be expecting rmse < 0.002 for 2- and 3-bit quantization variants. * Fix docker build I have been sloppy with vector reinterpret casts on ARM_NEON. It seems clang is very forgiving in that regard. * Added forgotten ggml.o dependence on k_quants.h to the Makefile * Had unintentionally committed the Makefile with -Ofast enabled * ggml : rename k_quants -> ggml-quants-k, use lowercase in code --------- Co-authored-by: Iwan Kawrakow <iwan.kawrakow@gmail.com> Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
2023-06-05 19:56:18 +00:00
ggml : quantization refactoring (#3833) * ggml : factor all quantization code in ggml-quants ggml-ci * ggml-quants : fix Zig and Swift builds + quantize tool ggml-ci * quantize : --pure option for disabling k-quant mixtures --------- Co-authored-by: cebtenzzre <cebtenzzre@gmail.com>
2023-10-29 16:32:28 +00:00
const __m256 q = mul_sum_i8_pairs_float(bx, by);
ggml : add SOTA 2,3,4,5,6 bit k-quantizations (#1684) * Starting to add k-quantization to ggml I think it is better to have quantization separate from ggml. For now just adding the k-quants there, but it would be better to also factor out the existing ggml quantizations. * Adding Q3_K and Q8_K (de)-quantization * Q3_K now working on CUDA and AVX2/scalar CUDA is not ideal - ~50% slower than Q4_0 for single token prediction, about the same in batch mode (perplexity). CPU single token is ~55 ms (on Ryzen 7950X). * Some improvement for Q3_K on CUDA It is now ~22.5 ms/token on my GPU, so ~30% slower than Q4_0. * Some more CUDA optimizations for Q3_K Single token is now 20.5 ms/token (~20% slower than Q4_0). Perplexity is on par with Q4_0. * Adding Q4_K - scalar, AVX2, CUDA Performance is the same or perhaps very slightly better than Q4_0 on the CPU. On the GPU, single token prediction is ~10% better than Q4_0, batch mode (perplexity is about the same). * Adding Q6_K - scalar, AVX2, CUDA Performance is ~40% lower compared to Q4_K on the CPU. This is to be expected, considering that we are memory bound on the CPU and the 6-bit model is ~44% larger than the 4-bit. On the GPU, single token prediction is ~6% lower than Q4_0, batch mode (perplexity) is even closer (but still slower). * Adding Q5_K - scalar, AVX2, CUDA Performance is ~20% lower compared to Q4_K on the CPU. This is to be expected, considering that we are memory bound on the CPU and the 5-bit model is ~22% larger than the 4-bit. On the GPU, single token prediction is about the same as Q4_0 for both, single token and batch prediction. * Per convention, all QX_K quantizations use Q5_K for output.weight * Adding quantization mixes * Quantization mixes: didn't quite get what I wanted in the last commit * Q4_K dot product for ARM_NEON * Q6_K dot product for ARM_NEON * Q5_K dot product for ARM_NEON * Adding Q3_K dot for ARM_NEON It is 22% slower than Q4_K, despite the smaller model size. On x86_64, where we are memory bound, the Q3_K model is quite a bit faster than Q4_K. * A very slightly faster ARM_NEON Q3_K dot * Adding Q2_K - just CUDA for now Token prediction is pretty good - about 15.5 ms on a RTX 4080. Perplexity is about the same as Q4_K. * Adding scalar and AVX2 Q2_K dot * Adding ARM_NEON Q2_K dot About the same performance as Q4_K. * A slightly faster ARM_NEON Q2_K dot Single token prediction is now ~36 ms on M2 Max. The code is much simpler too. * Fixed bug in Q2_K CUDA dot product kernel Stranegly enough, for the few prompts I tried with the 7B model the responses looked perfectly reasonable. Only realized something is not quite right when I tried the larger models and started getting nonse back. In any case, Q2_K single token evaluation time on an RTX 4080 in a Ryzen7950X box iusing CUDA and model fully loaded on the GPU are ~15.5 ms for 7B, ~25.4 ms for 13B, and ~55.8 ms for 30B. The max number of layers that fit in VRAM for The 65B is 32. With that, we get ~330 ms per token, which is not that much faster than just running on the CPU (~470 ms per token). * Don't print zeros/NaNs when no count histogram has been collected * A 10% faster CUDA vector dot kernel for Q3_K Q3_K is now running at ~18.5 ms / token on CUDA, so the gap to Q4_0 is only 10%. It seems memory acccess pattern is more important for performance than the amount of computation the kernel does. * A slightly daster Q4_K AVX2 dot product For perplexity, where we are less memory bound, time per pass drops by ~5%. Barely measurable difference for single token prediction. * A slightly faster ARM_NEON A4_K dot product * Minor * Fix quantization error test We cannot possibly be expecting rmse < 0.002 for 2- and 3-bit quantization variants. * Fix docker build I have been sloppy with vector reinterpret casts on ARM_NEON. It seems clang is very forgiving in that regard. * Added forgotten ggml.o dependence on k_quants.h to the Makefile * Had unintentionally committed the Makefile with -Ofast enabled * ggml : rename k_quants -> ggml-quants-k, use lowercase in code --------- Co-authored-by: Iwan Kawrakow <iwan.kawrakow@gmail.com> Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
2023-06-05 19:56:18 +00:00
ggml : quantization refactoring (#3833) * ggml : factor all quantization code in ggml-quants ggml-ci * ggml-quants : fix Zig and Swift builds + quantize tool ggml-ci * quantize : --pure option for disabling k-quant mixtures --------- Co-authored-by: cebtenzzre <cebtenzzre@gmail.com>
2023-10-29 16:32:28 +00:00
// Multiply q with scale and accumulate
#if defined(__AVX2__)
acc = _mm256_fmadd_ps( d, q, acc );
#else
acc = _mm256_add_ps( _mm256_mul_ps( d, q ), acc );
#endif
ggml : add SOTA 2,3,4,5,6 bit k-quantizations (#1684) * Starting to add k-quantization to ggml I think it is better to have quantization separate from ggml. For now just adding the k-quants there, but it would be better to also factor out the existing ggml quantizations. * Adding Q3_K and Q8_K (de)-quantization * Q3_K now working on CUDA and AVX2/scalar CUDA is not ideal - ~50% slower than Q4_0 for single token prediction, about the same in batch mode (perplexity). CPU single token is ~55 ms (on Ryzen 7950X). * Some improvement for Q3_K on CUDA It is now ~22.5 ms/token on my GPU, so ~30% slower than Q4_0. * Some more CUDA optimizations for Q3_K Single token is now 20.5 ms/token (~20% slower than Q4_0). Perplexity is on par with Q4_0. * Adding Q4_K - scalar, AVX2, CUDA Performance is the same or perhaps very slightly better than Q4_0 on the CPU. On the GPU, single token prediction is ~10% better than Q4_0, batch mode (perplexity is about the same). * Adding Q6_K - scalar, AVX2, CUDA Performance is ~40% lower compared to Q4_K on the CPU. This is to be expected, considering that we are memory bound on the CPU and the 6-bit model is ~44% larger than the 4-bit. On the GPU, single token prediction is ~6% lower than Q4_0, batch mode (perplexity) is even closer (but still slower). * Adding Q5_K - scalar, AVX2, CUDA Performance is ~20% lower compared to Q4_K on the CPU. This is to be expected, considering that we are memory bound on the CPU and the 5-bit model is ~22% larger than the 4-bit. On the GPU, single token prediction is about the same as Q4_0 for both, single token and batch prediction. * Per convention, all QX_K quantizations use Q5_K for output.weight * Adding quantization mixes * Quantization mixes: didn't quite get what I wanted in the last commit * Q4_K dot product for ARM_NEON * Q6_K dot product for ARM_NEON * Q5_K dot product for ARM_NEON * Adding Q3_K dot for ARM_NEON It is 22% slower than Q4_K, despite the smaller model size. On x86_64, where we are memory bound, the Q3_K model is quite a bit faster than Q4_K. * A very slightly faster ARM_NEON Q3_K dot * Adding Q2_K - just CUDA for now Token prediction is pretty good - about 15.5 ms on a RTX 4080. Perplexity is about the same as Q4_K. * Adding scalar and AVX2 Q2_K dot * Adding ARM_NEON Q2_K dot About the same performance as Q4_K. * A slightly faster ARM_NEON Q2_K dot Single token prediction is now ~36 ms on M2 Max. The code is much simpler too. * Fixed bug in Q2_K CUDA dot product kernel Stranegly enough, for the few prompts I tried with the 7B model the responses looked perfectly reasonable. Only realized something is not quite right when I tried the larger models and started getting nonse back. In any case, Q2_K single token evaluation time on an RTX 4080 in a Ryzen7950X box iusing CUDA and model fully loaded on the GPU are ~15.5 ms for 7B, ~25.4 ms for 13B, and ~55.8 ms for 30B. The max number of layers that fit in VRAM for The 65B is 32. With that, we get ~330 ms per token, which is not that much faster than just running on the CPU (~470 ms per token). * Don't print zeros/NaNs when no count histogram has been collected * A 10% faster CUDA vector dot kernel for Q3_K Q3_K is now running at ~18.5 ms / token on CUDA, so the gap to Q4_0 is only 10%. It seems memory acccess pattern is more important for performance than the amount of computation the kernel does. * A slightly daster Q4_K AVX2 dot product For perplexity, where we are less memory bound, time per pass drops by ~5%. Barely measurable difference for single token prediction. * A slightly faster ARM_NEON A4_K dot product * Minor * Fix quantization error test We cannot possibly be expecting rmse < 0.002 for 2- and 3-bit quantization variants. * Fix docker build I have been sloppy with vector reinterpret casts on ARM_NEON. It seems clang is very forgiving in that regard. * Added forgotten ggml.o dependence on k_quants.h to the Makefile * Had unintentionally committed the Makefile with -Ofast enabled * ggml : rename k_quants -> ggml-quants-k, use lowercase in code --------- Co-authored-by: Iwan Kawrakow <iwan.kawrakow@gmail.com> Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
2023-06-05 19:56:18 +00:00
}
ggml : quantization refactoring (#3833) * ggml : factor all quantization code in ggml-quants ggml-ci * ggml-quants : fix Zig and Swift builds + quantize tool ggml-ci * quantize : --pure option for disabling k-quant mixtures --------- Co-authored-by: cebtenzzre <cebtenzzre@gmail.com>
2023-10-29 16:32:28 +00:00
*s = hsum_float_8(acc);
#elif defined(__riscv_v_intrinsic)
float sumf = 0.0;
size_t vl = __riscv_vsetvl_e8m1(qk);
ggml : add SOTA 2,3,4,5,6 bit k-quantizations (#1684) * Starting to add k-quantization to ggml I think it is better to have quantization separate from ggml. For now just adding the k-quants there, but it would be better to also factor out the existing ggml quantizations. * Adding Q3_K and Q8_K (de)-quantization * Q3_K now working on CUDA and AVX2/scalar CUDA is not ideal - ~50% slower than Q4_0 for single token prediction, about the same in batch mode (perplexity). CPU single token is ~55 ms (on Ryzen 7950X). * Some improvement for Q3_K on CUDA It is now ~22.5 ms/token on my GPU, so ~30% slower than Q4_0. * Some more CUDA optimizations for Q3_K Single token is now 20.5 ms/token (~20% slower than Q4_0). Perplexity is on par with Q4_0. * Adding Q4_K - scalar, AVX2, CUDA Performance is the same or perhaps very slightly better than Q4_0 on the CPU. On the GPU, single token prediction is ~10% better than Q4_0, batch mode (perplexity is about the same). * Adding Q6_K - scalar, AVX2, CUDA Performance is ~40% lower compared to Q4_K on the CPU. This is to be expected, considering that we are memory bound on the CPU and the 6-bit model is ~44% larger than the 4-bit. On the GPU, single token prediction is ~6% lower than Q4_0, batch mode (perplexity) is even closer (but still slower). * Adding Q5_K - scalar, AVX2, CUDA Performance is ~20% lower compared to Q4_K on the CPU. This is to be expected, considering that we are memory bound on the CPU and the 5-bit model is ~22% larger than the 4-bit. On the GPU, single token prediction is about the same as Q4_0 for both, single token and batch prediction. * Per convention, all QX_K quantizations use Q5_K for output.weight * Adding quantization mixes * Quantization mixes: didn't quite get what I wanted in the last commit * Q4_K dot product for ARM_NEON * Q6_K dot product for ARM_NEON * Q5_K dot product for ARM_NEON * Adding Q3_K dot for ARM_NEON It is 22% slower than Q4_K, despite the smaller model size. On x86_64, where we are memory bound, the Q3_K model is quite a bit faster than Q4_K. * A very slightly faster ARM_NEON Q3_K dot * Adding Q2_K - just CUDA for now Token prediction is pretty good - about 15.5 ms on a RTX 4080. Perplexity is about the same as Q4_K. * Adding scalar and AVX2 Q2_K dot * Adding ARM_NEON Q2_K dot About the same performance as Q4_K. * A slightly faster ARM_NEON Q2_K dot Single token prediction is now ~36 ms on M2 Max. The code is much simpler too. * Fixed bug in Q2_K CUDA dot product kernel Stranegly enough, for the few prompts I tried with the 7B model the responses looked perfectly reasonable. Only realized something is not quite right when I tried the larger models and started getting nonse back. In any case, Q2_K single token evaluation time on an RTX 4080 in a Ryzen7950X box iusing CUDA and model fully loaded on the GPU are ~15.5 ms for 7B, ~25.4 ms for 13B, and ~55.8 ms for 30B. The max number of layers that fit in VRAM for The 65B is 32. With that, we get ~330 ms per token, which is not that much faster than just running on the CPU (~470 ms per token). * Don't print zeros/NaNs when no count histogram has been collected * A 10% faster CUDA vector dot kernel for Q3_K Q3_K is now running at ~18.5 ms / token on CUDA, so the gap to Q4_0 is only 10%. It seems memory acccess pattern is more important for performance than the amount of computation the kernel does. * A slightly daster Q4_K AVX2 dot product For perplexity, where we are less memory bound, time per pass drops by ~5%. Barely measurable difference for single token prediction. * A slightly faster ARM_NEON A4_K dot product * Minor * Fix quantization error test We cannot possibly be expecting rmse < 0.002 for 2- and 3-bit quantization variants. * Fix docker build I have been sloppy with vector reinterpret casts on ARM_NEON. It seems clang is very forgiving in that regard. * Added forgotten ggml.o dependence on k_quants.h to the Makefile * Had unintentionally committed the Makefile with -Ofast enabled * ggml : rename k_quants -> ggml-quants-k, use lowercase in code --------- Co-authored-by: Iwan Kawrakow <iwan.kawrakow@gmail.com> Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
2023-06-05 19:56:18 +00:00
for (int i = 0; i < nb; i++) {
ggml : quantization refactoring (#3833) * ggml : factor all quantization code in ggml-quants ggml-ci * ggml-quants : fix Zig and Swift builds + quantize tool ggml-ci * quantize : --pure option for disabling k-quant mixtures --------- Co-authored-by: cebtenzzre <cebtenzzre@gmail.com>
2023-10-29 16:32:28 +00:00
// load elements
vint8m1_t bx = __riscv_vle8_v_i8m1(x[i].qs, vl);
vint8m1_t by = __riscv_vle8_v_i8m1(y[i].qs, vl);
vint16m2_t vw_mul = __riscv_vwmul_vv_i16m2(bx, by, vl);
vint32m1_t v_zero = __riscv_vmv_v_x_i32m1(0, vl);
vint32m1_t v_sum = __riscv_vwredsum_vs_i16m2_i32m1(vw_mul, v_zero, vl);
int sumi = __riscv_vmv_x_s_i32m1_i32(v_sum);
ggml : move FP16 <-> FP32 code to ggml-impl.h (#3861) * ggml : move FP16 <-> FP32 stuff to ggml-impl.h ggml-ci * tests : fix ARM build * ggml : explicitly initialize deprecated type traits * ggml : add math.h to ggml-impl.h * ggml : remove duplicate static assert macros * ggml : prefix lookup tables with ggml_ ggml-ci * ggml-impl : move extern "C" to start of file
2023-10-30 17:19:15 +00:00
sumf += sumi*(GGML_FP16_TO_FP32(x[i].d)*GGML_FP16_TO_FP32(y[i].d));
ggml : add SOTA 2,3,4,5,6 bit k-quantizations (#1684) * Starting to add k-quantization to ggml I think it is better to have quantization separate from ggml. For now just adding the k-quants there, but it would be better to also factor out the existing ggml quantizations. * Adding Q3_K and Q8_K (de)-quantization * Q3_K now working on CUDA and AVX2/scalar CUDA is not ideal - ~50% slower than Q4_0 for single token prediction, about the same in batch mode (perplexity). CPU single token is ~55 ms (on Ryzen 7950X). * Some improvement for Q3_K on CUDA It is now ~22.5 ms/token on my GPU, so ~30% slower than Q4_0. * Some more CUDA optimizations for Q3_K Single token is now 20.5 ms/token (~20% slower than Q4_0). Perplexity is on par with Q4_0. * Adding Q4_K - scalar, AVX2, CUDA Performance is the same or perhaps very slightly better than Q4_0 on the CPU. On the GPU, single token prediction is ~10% better than Q4_0, batch mode (perplexity is about the same). * Adding Q6_K - scalar, AVX2, CUDA Performance is ~40% lower compared to Q4_K on the CPU. This is to be expected, considering that we are memory bound on the CPU and the 6-bit model is ~44% larger than the 4-bit. On the GPU, single token prediction is ~6% lower than Q4_0, batch mode (perplexity) is even closer (but still slower). * Adding Q5_K - scalar, AVX2, CUDA Performance is ~20% lower compared to Q4_K on the CPU. This is to be expected, considering that we are memory bound on the CPU and the 5-bit model is ~22% larger than the 4-bit. On the GPU, single token prediction is about the same as Q4_0 for both, single token and batch prediction. * Per convention, all QX_K quantizations use Q5_K for output.weight * Adding quantization mixes * Quantization mixes: didn't quite get what I wanted in the last commit * Q4_K dot product for ARM_NEON * Q6_K dot product for ARM_NEON * Q5_K dot product for ARM_NEON * Adding Q3_K dot for ARM_NEON It is 22% slower than Q4_K, despite the smaller model size. On x86_64, where we are memory bound, the Q3_K model is quite a bit faster than Q4_K. * A very slightly faster ARM_NEON Q3_K dot * Adding Q2_K - just CUDA for now Token prediction is pretty good - about 15.5 ms on a RTX 4080. Perplexity is about the same as Q4_K. * Adding scalar and AVX2 Q2_K dot * Adding ARM_NEON Q2_K dot About the same performance as Q4_K. * A slightly faster ARM_NEON Q2_K dot Single token prediction is now ~36 ms on M2 Max. The code is much simpler too. * Fixed bug in Q2_K CUDA dot product kernel Stranegly enough, for the few prompts I tried with the 7B model the responses looked perfectly reasonable. Only realized something is not quite right when I tried the larger models and started getting nonse back. In any case, Q2_K single token evaluation time on an RTX 4080 in a Ryzen7950X box iusing CUDA and model fully loaded on the GPU are ~15.5 ms for 7B, ~25.4 ms for 13B, and ~55.8 ms for 30B. The max number of layers that fit in VRAM for The 65B is 32. With that, we get ~330 ms per token, which is not that much faster than just running on the CPU (~470 ms per token). * Don't print zeros/NaNs when no count histogram has been collected * A 10% faster CUDA vector dot kernel for Q3_K Q3_K is now running at ~18.5 ms / token on CUDA, so the gap to Q4_0 is only 10%. It seems memory acccess pattern is more important for performance than the amount of computation the kernel does. * A slightly daster Q4_K AVX2 dot product For perplexity, where we are less memory bound, time per pass drops by ~5%. Barely measurable difference for single token prediction. * A slightly faster ARM_NEON A4_K dot product * Minor * Fix quantization error test We cannot possibly be expecting rmse < 0.002 for 2- and 3-bit quantization variants. * Fix docker build I have been sloppy with vector reinterpret casts on ARM_NEON. It seems clang is very forgiving in that regard. * Added forgotten ggml.o dependence on k_quants.h to the Makefile * Had unintentionally committed the Makefile with -Ofast enabled * ggml : rename k_quants -> ggml-quants-k, use lowercase in code --------- Co-authored-by: Iwan Kawrakow <iwan.kawrakow@gmail.com> Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
2023-06-05 19:56:18 +00:00
}
ggml : quantization refactoring (#3833) * ggml : factor all quantization code in ggml-quants ggml-ci * ggml-quants : fix Zig and Swift builds + quantize tool ggml-ci * quantize : --pure option for disabling k-quant mixtures --------- Co-authored-by: cebtenzzre <cebtenzzre@gmail.com>
2023-10-29 16:32:28 +00:00
*s = sumf;
#else
// scalar
float sumf = 0.0;
ggml : add SOTA 2,3,4,5,6 bit k-quantizations (#1684) * Starting to add k-quantization to ggml I think it is better to have quantization separate from ggml. For now just adding the k-quants there, but it would be better to also factor out the existing ggml quantizations. * Adding Q3_K and Q8_K (de)-quantization * Q3_K now working on CUDA and AVX2/scalar CUDA is not ideal - ~50% slower than Q4_0 for single token prediction, about the same in batch mode (perplexity). CPU single token is ~55 ms (on Ryzen 7950X). * Some improvement for Q3_K on CUDA It is now ~22.5 ms/token on my GPU, so ~30% slower than Q4_0. * Some more CUDA optimizations for Q3_K Single token is now 20.5 ms/token (~20% slower than Q4_0). Perplexity is on par with Q4_0. * Adding Q4_K - scalar, AVX2, CUDA Performance is the same or perhaps very slightly better than Q4_0 on the CPU. On the GPU, single token prediction is ~10% better than Q4_0, batch mode (perplexity is about the same). * Adding Q6_K - scalar, AVX2, CUDA Performance is ~40% lower compared to Q4_K on the CPU. This is to be expected, considering that we are memory bound on the CPU and the 6-bit model is ~44% larger than the 4-bit. On the GPU, single token prediction is ~6% lower than Q4_0, batch mode (perplexity) is even closer (but still slower). * Adding Q5_K - scalar, AVX2, CUDA Performance is ~20% lower compared to Q4_K on the CPU. This is to be expected, considering that we are memory bound on the CPU and the 5-bit model is ~22% larger than the 4-bit. On the GPU, single token prediction is about the same as Q4_0 for both, single token and batch prediction. * Per convention, all QX_K quantizations use Q5_K for output.weight * Adding quantization mixes * Quantization mixes: didn't quite get what I wanted in the last commit * Q4_K dot product for ARM_NEON * Q6_K dot product for ARM_NEON * Q5_K dot product for ARM_NEON * Adding Q3_K dot for ARM_NEON It is 22% slower than Q4_K, despite the smaller model size. On x86_64, where we are memory bound, the Q3_K model is quite a bit faster than Q4_K. * A very slightly faster ARM_NEON Q3_K dot * Adding Q2_K - just CUDA for now Token prediction is pretty good - about 15.5 ms on a RTX 4080. Perplexity is about the same as Q4_K. * Adding scalar and AVX2 Q2_K dot * Adding ARM_NEON Q2_K dot About the same performance as Q4_K. * A slightly faster ARM_NEON Q2_K dot Single token prediction is now ~36 ms on M2 Max. The code is much simpler too. * Fixed bug in Q2_K CUDA dot product kernel Stranegly enough, for the few prompts I tried with the 7B model the responses looked perfectly reasonable. Only realized something is not quite right when I tried the larger models and started getting nonse back. In any case, Q2_K single token evaluation time on an RTX 4080 in a Ryzen7950X box iusing CUDA and model fully loaded on the GPU are ~15.5 ms for 7B, ~25.4 ms for 13B, and ~55.8 ms for 30B. The max number of layers that fit in VRAM for The 65B is 32. With that, we get ~330 ms per token, which is not that much faster than just running on the CPU (~470 ms per token). * Don't print zeros/NaNs when no count histogram has been collected * A 10% faster CUDA vector dot kernel for Q3_K Q3_K is now running at ~18.5 ms / token on CUDA, so the gap to Q4_0 is only 10%. It seems memory acccess pattern is more important for performance than the amount of computation the kernel does. * A slightly daster Q4_K AVX2 dot product For perplexity, where we are less memory bound, time per pass drops by ~5%. Barely measurable difference for single token prediction. * A slightly faster ARM_NEON A4_K dot product * Minor * Fix quantization error test We cannot possibly be expecting rmse < 0.002 for 2- and 3-bit quantization variants. * Fix docker build I have been sloppy with vector reinterpret casts on ARM_NEON. It seems clang is very forgiving in that regard. * Added forgotten ggml.o dependence on k_quants.h to the Makefile * Had unintentionally committed the Makefile with -Ofast enabled * ggml : rename k_quants -> ggml-quants-k, use lowercase in code --------- Co-authored-by: Iwan Kawrakow <iwan.kawrakow@gmail.com> Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
2023-06-05 19:56:18 +00:00
ggml : quantization refactoring (#3833) * ggml : factor all quantization code in ggml-quants ggml-ci * ggml-quants : fix Zig and Swift builds + quantize tool ggml-ci * quantize : --pure option for disabling k-quant mixtures --------- Co-authored-by: cebtenzzre <cebtenzzre@gmail.com>
2023-10-29 16:32:28 +00:00
for (int i = 0; i < nb; i++) {
int sumi = 0;
ggml : add SOTA 2,3,4,5,6 bit k-quantizations (#1684) * Starting to add k-quantization to ggml I think it is better to have quantization separate from ggml. For now just adding the k-quants there, but it would be better to also factor out the existing ggml quantizations. * Adding Q3_K and Q8_K (de)-quantization * Q3_K now working on CUDA and AVX2/scalar CUDA is not ideal - ~50% slower than Q4_0 for single token prediction, about the same in batch mode (perplexity). CPU single token is ~55 ms (on Ryzen 7950X). * Some improvement for Q3_K on CUDA It is now ~22.5 ms/token on my GPU, so ~30% slower than Q4_0. * Some more CUDA optimizations for Q3_K Single token is now 20.5 ms/token (~20% slower than Q4_0). Perplexity is on par with Q4_0. * Adding Q4_K - scalar, AVX2, CUDA Performance is the same or perhaps very slightly better than Q4_0 on the CPU. On the GPU, single token prediction is ~10% better than Q4_0, batch mode (perplexity is about the same). * Adding Q6_K - scalar, AVX2, CUDA Performance is ~40% lower compared to Q4_K on the CPU. This is to be expected, considering that we are memory bound on the CPU and the 6-bit model is ~44% larger than the 4-bit. On the GPU, single token prediction is ~6% lower than Q4_0, batch mode (perplexity) is even closer (but still slower). * Adding Q5_K - scalar, AVX2, CUDA Performance is ~20% lower compared to Q4_K on the CPU. This is to be expected, considering that we are memory bound on the CPU and the 5-bit model is ~22% larger than the 4-bit. On the GPU, single token prediction is about the same as Q4_0 for both, single token and batch prediction. * Per convention, all QX_K quantizations use Q5_K for output.weight * Adding quantization mixes * Quantization mixes: didn't quite get what I wanted in the last commit * Q4_K dot product for ARM_NEON * Q6_K dot product for ARM_NEON * Q5_K dot product for ARM_NEON * Adding Q3_K dot for ARM_NEON It is 22% slower than Q4_K, despite the smaller model size. On x86_64, where we are memory bound, the Q3_K model is quite a bit faster than Q4_K. * A very slightly faster ARM_NEON Q3_K dot * Adding Q2_K - just CUDA for now Token prediction is pretty good - about 15.5 ms on a RTX 4080. Perplexity is about the same as Q4_K. * Adding scalar and AVX2 Q2_K dot * Adding ARM_NEON Q2_K dot About the same performance as Q4_K. * A slightly faster ARM_NEON Q2_K dot Single token prediction is now ~36 ms on M2 Max. The code is much simpler too. * Fixed bug in Q2_K CUDA dot product kernel Stranegly enough, for the few prompts I tried with the 7B model the responses looked perfectly reasonable. Only realized something is not quite right when I tried the larger models and started getting nonse back. In any case, Q2_K single token evaluation time on an RTX 4080 in a Ryzen7950X box iusing CUDA and model fully loaded on the GPU are ~15.5 ms for 7B, ~25.4 ms for 13B, and ~55.8 ms for 30B. The max number of layers that fit in VRAM for The 65B is 32. With that, we get ~330 ms per token, which is not that much faster than just running on the CPU (~470 ms per token). * Don't print zeros/NaNs when no count histogram has been collected * A 10% faster CUDA vector dot kernel for Q3_K Q3_K is now running at ~18.5 ms / token on CUDA, so the gap to Q4_0 is only 10%. It seems memory acccess pattern is more important for performance than the amount of computation the kernel does. * A slightly daster Q4_K AVX2 dot product For perplexity, where we are less memory bound, time per pass drops by ~5%. Barely measurable difference for single token prediction. * A slightly faster ARM_NEON A4_K dot product * Minor * Fix quantization error test We cannot possibly be expecting rmse < 0.002 for 2- and 3-bit quantization variants. * Fix docker build I have been sloppy with vector reinterpret casts on ARM_NEON. It seems clang is very forgiving in that regard. * Added forgotten ggml.o dependence on k_quants.h to the Makefile * Had unintentionally committed the Makefile with -Ofast enabled * ggml : rename k_quants -> ggml-quants-k, use lowercase in code --------- Co-authored-by: Iwan Kawrakow <iwan.kawrakow@gmail.com> Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
2023-06-05 19:56:18 +00:00
ggml : quantization refactoring (#3833) * ggml : factor all quantization code in ggml-quants ggml-ci * ggml-quants : fix Zig and Swift builds + quantize tool ggml-ci * quantize : --pure option for disabling k-quant mixtures --------- Co-authored-by: cebtenzzre <cebtenzzre@gmail.com>
2023-10-29 16:32:28 +00:00
for (int j = 0; j < qk; j++) {
sumi += x[i].qs[j]*y[i].qs[j];
}
ggml : add SOTA 2,3,4,5,6 bit k-quantizations (#1684) * Starting to add k-quantization to ggml I think it is better to have quantization separate from ggml. For now just adding the k-quants there, but it would be better to also factor out the existing ggml quantizations. * Adding Q3_K and Q8_K (de)-quantization * Q3_K now working on CUDA and AVX2/scalar CUDA is not ideal - ~50% slower than Q4_0 for single token prediction, about the same in batch mode (perplexity). CPU single token is ~55 ms (on Ryzen 7950X). * Some improvement for Q3_K on CUDA It is now ~22.5 ms/token on my GPU, so ~30% slower than Q4_0. * Some more CUDA optimizations for Q3_K Single token is now 20.5 ms/token (~20% slower than Q4_0). Perplexity is on par with Q4_0. * Adding Q4_K - scalar, AVX2, CUDA Performance is the same or perhaps very slightly better than Q4_0 on the CPU. On the GPU, single token prediction is ~10% better than Q4_0, batch mode (perplexity is about the same). * Adding Q6_K - scalar, AVX2, CUDA Performance is ~40% lower compared to Q4_K on the CPU. This is to be expected, considering that we are memory bound on the CPU and the 6-bit model is ~44% larger than the 4-bit. On the GPU, single token prediction is ~6% lower than Q4_0, batch mode (perplexity) is even closer (but still slower). * Adding Q5_K - scalar, AVX2, CUDA Performance is ~20% lower compared to Q4_K on the CPU. This is to be expected, considering that we are memory bound on the CPU and the 5-bit model is ~22% larger than the 4-bit. On the GPU, single token prediction is about the same as Q4_0 for both, single token and batch prediction. * Per convention, all QX_K quantizations use Q5_K for output.weight * Adding quantization mixes * Quantization mixes: didn't quite get what I wanted in the last commit * Q4_K dot product for ARM_NEON * Q6_K dot product for ARM_NEON * Q5_K dot product for ARM_NEON * Adding Q3_K dot for ARM_NEON It is 22% slower than Q4_K, despite the smaller model size. On x86_64, where we are memory bound, the Q3_K model is quite a bit faster than Q4_K. * A very slightly faster ARM_NEON Q3_K dot * Adding Q2_K - just CUDA for now Token prediction is pretty good - about 15.5 ms on a RTX 4080. Perplexity is about the same as Q4_K. * Adding scalar and AVX2 Q2_K dot * Adding ARM_NEON Q2_K dot About the same performance as Q4_K. * A slightly faster ARM_NEON Q2_K dot Single token prediction is now ~36 ms on M2 Max. The code is much simpler too. * Fixed bug in Q2_K CUDA dot product kernel Stranegly enough, for the few prompts I tried with the 7B model the responses looked perfectly reasonable. Only realized something is not quite right when I tried the larger models and started getting nonse back. In any case, Q2_K single token evaluation time on an RTX 4080 in a Ryzen7950X box iusing CUDA and model fully loaded on the GPU are ~15.5 ms for 7B, ~25.4 ms for 13B, and ~55.8 ms for 30B. The max number of layers that fit in VRAM for The 65B is 32. With that, we get ~330 ms per token, which is not that much faster than just running on the CPU (~470 ms per token). * Don't print zeros/NaNs when no count histogram has been collected * A 10% faster CUDA vector dot kernel for Q3_K Q3_K is now running at ~18.5 ms / token on CUDA, so the gap to Q4_0 is only 10%. It seems memory acccess pattern is more important for performance than the amount of computation the kernel does. * A slightly daster Q4_K AVX2 dot product For perplexity, where we are less memory bound, time per pass drops by ~5%. Barely measurable difference for single token prediction. * A slightly faster ARM_NEON A4_K dot product * Minor * Fix quantization error test We cannot possibly be expecting rmse < 0.002 for 2- and 3-bit quantization variants. * Fix docker build I have been sloppy with vector reinterpret casts on ARM_NEON. It seems clang is very forgiving in that regard. * Added forgotten ggml.o dependence on k_quants.h to the Makefile * Had unintentionally committed the Makefile with -Ofast enabled * ggml : rename k_quants -> ggml-quants-k, use lowercase in code --------- Co-authored-by: Iwan Kawrakow <iwan.kawrakow@gmail.com> Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
2023-06-05 19:56:18 +00:00
ggml : move FP16 <-> FP32 code to ggml-impl.h (#3861) * ggml : move FP16 <-> FP32 stuff to ggml-impl.h ggml-ci * tests : fix ARM build * ggml : explicitly initialize deprecated type traits * ggml : add math.h to ggml-impl.h * ggml : remove duplicate static assert macros * ggml : prefix lookup tables with ggml_ ggml-ci * ggml-impl : move extern "C" to start of file
2023-10-30 17:19:15 +00:00
sumf += sumi*(GGML_FP16_TO_FP32(x[i].d)*GGML_FP16_TO_FP32(y[i].d));
ggml : quantization refactoring (#3833) * ggml : factor all quantization code in ggml-quants ggml-ci * ggml-quants : fix Zig and Swift builds + quantize tool ggml-ci * quantize : --pure option for disabling k-quant mixtures --------- Co-authored-by: cebtenzzre <cebtenzzre@gmail.com>
2023-10-29 16:32:28 +00:00
}
ggml : add SOTA 2,3,4,5,6 bit k-quantizations (#1684) * Starting to add k-quantization to ggml I think it is better to have quantization separate from ggml. For now just adding the k-quants there, but it would be better to also factor out the existing ggml quantizations. * Adding Q3_K and Q8_K (de)-quantization * Q3_K now working on CUDA and AVX2/scalar CUDA is not ideal - ~50% slower than Q4_0 for single token prediction, about the same in batch mode (perplexity). CPU single token is ~55 ms (on Ryzen 7950X). * Some improvement for Q3_K on CUDA It is now ~22.5 ms/token on my GPU, so ~30% slower than Q4_0. * Some more CUDA optimizations for Q3_K Single token is now 20.5 ms/token (~20% slower than Q4_0). Perplexity is on par with Q4_0. * Adding Q4_K - scalar, AVX2, CUDA Performance is the same or perhaps very slightly better than Q4_0 on the CPU. On the GPU, single token prediction is ~10% better than Q4_0, batch mode (perplexity is about the same). * Adding Q6_K - scalar, AVX2, CUDA Performance is ~40% lower compared to Q4_K on the CPU. This is to be expected, considering that we are memory bound on the CPU and the 6-bit model is ~44% larger than the 4-bit. On the GPU, single token prediction is ~6% lower than Q4_0, batch mode (perplexity) is even closer (but still slower). * Adding Q5_K - scalar, AVX2, CUDA Performance is ~20% lower compared to Q4_K on the CPU. This is to be expected, considering that we are memory bound on the CPU and the 5-bit model is ~22% larger than the 4-bit. On the GPU, single token prediction is about the same as Q4_0 for both, single token and batch prediction. * Per convention, all QX_K quantizations use Q5_K for output.weight * Adding quantization mixes * Quantization mixes: didn't quite get what I wanted in the last commit * Q4_K dot product for ARM_NEON * Q6_K dot product for ARM_NEON * Q5_K dot product for ARM_NEON * Adding Q3_K dot for ARM_NEON It is 22% slower than Q4_K, despite the smaller model size. On x86_64, where we are memory bound, the Q3_K model is quite a bit faster than Q4_K. * A very slightly faster ARM_NEON Q3_K dot * Adding Q2_K - just CUDA for now Token prediction is pretty good - about 15.5 ms on a RTX 4080. Perplexity is about the same as Q4_K. * Adding scalar and AVX2 Q2_K dot * Adding ARM_NEON Q2_K dot About the same performance as Q4_K. * A slightly faster ARM_NEON Q2_K dot Single token prediction is now ~36 ms on M2 Max. The code is much simpler too. * Fixed bug in Q2_K CUDA dot product kernel Stranegly enough, for the few prompts I tried with the 7B model the responses looked perfectly reasonable. Only realized something is not quite right when I tried the larger models and started getting nonse back. In any case, Q2_K single token evaluation time on an RTX 4080 in a Ryzen7950X box iusing CUDA and model fully loaded on the GPU are ~15.5 ms for 7B, ~25.4 ms for 13B, and ~55.8 ms for 30B. The max number of layers that fit in VRAM for The 65B is 32. With that, we get ~330 ms per token, which is not that much faster than just running on the CPU (~470 ms per token). * Don't print zeros/NaNs when no count histogram has been collected * A 10% faster CUDA vector dot kernel for Q3_K Q3_K is now running at ~18.5 ms / token on CUDA, so the gap to Q4_0 is only 10%. It seems memory acccess pattern is more important for performance than the amount of computation the kernel does. * A slightly daster Q4_K AVX2 dot product For perplexity, where we are less memory bound, time per pass drops by ~5%. Barely measurable difference for single token prediction. * A slightly faster ARM_NEON A4_K dot product * Minor * Fix quantization error test We cannot possibly be expecting rmse < 0.002 for 2- and 3-bit quantization variants. * Fix docker build I have been sloppy with vector reinterpret casts on ARM_NEON. It seems clang is very forgiving in that regard. * Added forgotten ggml.o dependence on k_quants.h to the Makefile * Had unintentionally committed the Makefile with -Ofast enabled * ggml : rename k_quants -> ggml-quants-k, use lowercase in code --------- Co-authored-by: Iwan Kawrakow <iwan.kawrakow@gmail.com> Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
2023-06-05 19:56:18 +00:00
ggml : quantization refactoring (#3833) * ggml : factor all quantization code in ggml-quants ggml-ci * ggml-quants : fix Zig and Swift builds + quantize tool ggml-ci * quantize : --pure option for disabling k-quant mixtures --------- Co-authored-by: cebtenzzre <cebtenzzre@gmail.com>
2023-10-29 16:32:28 +00:00
*s = sumf;
ggml : add SOTA 2,3,4,5,6 bit k-quantizations (#1684) * Starting to add k-quantization to ggml I think it is better to have quantization separate from ggml. For now just adding the k-quants there, but it would be better to also factor out the existing ggml quantizations. * Adding Q3_K and Q8_K (de)-quantization * Q3_K now working on CUDA and AVX2/scalar CUDA is not ideal - ~50% slower than Q4_0 for single token prediction, about the same in batch mode (perplexity). CPU single token is ~55 ms (on Ryzen 7950X). * Some improvement for Q3_K on CUDA It is now ~22.5 ms/token on my GPU, so ~30% slower than Q4_0. * Some more CUDA optimizations for Q3_K Single token is now 20.5 ms/token (~20% slower than Q4_0). Perplexity is on par with Q4_0. * Adding Q4_K - scalar, AVX2, CUDA Performance is the same or perhaps very slightly better than Q4_0 on the CPU. On the GPU, single token prediction is ~10% better than Q4_0, batch mode (perplexity is about the same). * Adding Q6_K - scalar, AVX2, CUDA Performance is ~40% lower compared to Q4_K on the CPU. This is to be expected, considering that we are memory bound on the CPU and the 6-bit model is ~44% larger than the 4-bit. On the GPU, single token prediction is ~6% lower than Q4_0, batch mode (perplexity) is even closer (but still slower). * Adding Q5_K - scalar, AVX2, CUDA Performance is ~20% lower compared to Q4_K on the CPU. This is to be expected, considering that we are memory bound on the CPU and the 5-bit model is ~22% larger than the 4-bit. On the GPU, single token prediction is about the same as Q4_0 for both, single token and batch prediction. * Per convention, all QX_K quantizations use Q5_K for output.weight * Adding quantization mixes * Quantization mixes: didn't quite get what I wanted in the last commit * Q4_K dot product for ARM_NEON * Q6_K dot product for ARM_NEON * Q5_K dot product for ARM_NEON * Adding Q3_K dot for ARM_NEON It is 22% slower than Q4_K, despite the smaller model size. On x86_64, where we are memory bound, the Q3_K model is quite a bit faster than Q4_K. * A very slightly faster ARM_NEON Q3_K dot * Adding Q2_K - just CUDA for now Token prediction is pretty good - about 15.5 ms on a RTX 4080. Perplexity is about the same as Q4_K. * Adding scalar and AVX2 Q2_K dot * Adding ARM_NEON Q2_K dot About the same performance as Q4_K. * A slightly faster ARM_NEON Q2_K dot Single token prediction is now ~36 ms on M2 Max. The code is much simpler too. * Fixed bug in Q2_K CUDA dot product kernel Stranegly enough, for the few prompts I tried with the 7B model the responses looked perfectly reasonable. Only realized something is not quite right when I tried the larger models and started getting nonse back. In any case, Q2_K single token evaluation time on an RTX 4080 in a Ryzen7950X box iusing CUDA and model fully loaded on the GPU are ~15.5 ms for 7B, ~25.4 ms for 13B, and ~55.8 ms for 30B. The max number of layers that fit in VRAM for The 65B is 32. With that, we get ~330 ms per token, which is not that much faster than just running on the CPU (~470 ms per token). * Don't print zeros/NaNs when no count histogram has been collected * A 10% faster CUDA vector dot kernel for Q3_K Q3_K is now running at ~18.5 ms / token on CUDA, so the gap to Q4_0 is only 10%. It seems memory acccess pattern is more important for performance than the amount of computation the kernel does. * A slightly daster Q4_K AVX2 dot product For perplexity, where we are less memory bound, time per pass drops by ~5%. Barely measurable difference for single token prediction. * A slightly faster ARM_NEON A4_K dot product * Minor * Fix quantization error test We cannot possibly be expecting rmse < 0.002 for 2- and 3-bit quantization variants. * Fix docker build I have been sloppy with vector reinterpret casts on ARM_NEON. It seems clang is very forgiving in that regard. * Added forgotten ggml.o dependence on k_quants.h to the Makefile * Had unintentionally committed the Makefile with -Ofast enabled * ggml : rename k_quants -> ggml-quants-k, use lowercase in code --------- Co-authored-by: Iwan Kawrakow <iwan.kawrakow@gmail.com> Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
2023-06-05 19:56:18 +00:00
#endif
ggml : quantization refactoring (#3833) * ggml : factor all quantization code in ggml-quants ggml-ci * ggml-quants : fix Zig and Swift builds + quantize tool ggml-ci * quantize : --pure option for disabling k-quant mixtures --------- Co-authored-by: cebtenzzre <cebtenzzre@gmail.com>
2023-10-29 16:32:28 +00:00
}
ggml : add SOTA 2,3,4,5,6 bit k-quantizations (#1684) * Starting to add k-quantization to ggml I think it is better to have quantization separate from ggml. For now just adding the k-quants there, but it would be better to also factor out the existing ggml quantizations. * Adding Q3_K and Q8_K (de)-quantization * Q3_K now working on CUDA and AVX2/scalar CUDA is not ideal - ~50% slower than Q4_0 for single token prediction, about the same in batch mode (perplexity). CPU single token is ~55 ms (on Ryzen 7950X). * Some improvement for Q3_K on CUDA It is now ~22.5 ms/token on my GPU, so ~30% slower than Q4_0. * Some more CUDA optimizations for Q3_K Single token is now 20.5 ms/token (~20% slower than Q4_0). Perplexity is on par with Q4_0. * Adding Q4_K - scalar, AVX2, CUDA Performance is the same or perhaps very slightly better than Q4_0 on the CPU. On the GPU, single token prediction is ~10% better than Q4_0, batch mode (perplexity is about the same). * Adding Q6_K - scalar, AVX2, CUDA Performance is ~40% lower compared to Q4_K on the CPU. This is to be expected, considering that we are memory bound on the CPU and the 6-bit model is ~44% larger than the 4-bit. On the GPU, single token prediction is ~6% lower than Q4_0, batch mode (perplexity) is even closer (but still slower). * Adding Q5_K - scalar, AVX2, CUDA Performance is ~20% lower compared to Q4_K on the CPU. This is to be expected, considering that we are memory bound on the CPU and the 5-bit model is ~22% larger than the 4-bit. On the GPU, single token prediction is about the same as Q4_0 for both, single token and batch prediction. * Per convention, all QX_K quantizations use Q5_K for output.weight * Adding quantization mixes * Quantization mixes: didn't quite get what I wanted in the last commit * Q4_K dot product for ARM_NEON * Q6_K dot product for ARM_NEON * Q5_K dot product for ARM_NEON * Adding Q3_K dot for ARM_NEON It is 22% slower than Q4_K, despite the smaller model size. On x86_64, where we are memory bound, the Q3_K model is quite a bit faster than Q4_K. * A very slightly faster ARM_NEON Q3_K dot * Adding Q2_K - just CUDA for now Token prediction is pretty good - about 15.5 ms on a RTX 4080. Perplexity is about the same as Q4_K. * Adding scalar and AVX2 Q2_K dot * Adding ARM_NEON Q2_K dot About the same performance as Q4_K. * A slightly faster ARM_NEON Q2_K dot Single token prediction is now ~36 ms on M2 Max. The code is much simpler too. * Fixed bug in Q2_K CUDA dot product kernel Stranegly enough, for the few prompts I tried with the 7B model the responses looked perfectly reasonable. Only realized something is not quite right when I tried the larger models and started getting nonse back. In any case, Q2_K single token evaluation time on an RTX 4080 in a Ryzen7950X box iusing CUDA and model fully loaded on the GPU are ~15.5 ms for 7B, ~25.4 ms for 13B, and ~55.8 ms for 30B. The max number of layers that fit in VRAM for The 65B is 32. With that, we get ~330 ms per token, which is not that much faster than just running on the CPU (~470 ms per token). * Don't print zeros/NaNs when no count histogram has been collected * A 10% faster CUDA vector dot kernel for Q3_K Q3_K is now running at ~18.5 ms / token on CUDA, so the gap to Q4_0 is only 10%. It seems memory acccess pattern is more important for performance than the amount of computation the kernel does. * A slightly daster Q4_K AVX2 dot product For perplexity, where we are less memory bound, time per pass drops by ~5%. Barely measurable difference for single token prediction. * A slightly faster ARM_NEON A4_K dot product * Minor * Fix quantization error test We cannot possibly be expecting rmse < 0.002 for 2- and 3-bit quantization variants. * Fix docker build I have been sloppy with vector reinterpret casts on ARM_NEON. It seems clang is very forgiving in that regard. * Added forgotten ggml.o dependence on k_quants.h to the Makefile * Had unintentionally committed the Makefile with -Ofast enabled * ggml : rename k_quants -> ggml-quants-k, use lowercase in code --------- Co-authored-by: Iwan Kawrakow <iwan.kawrakow@gmail.com> Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
2023-06-05 19:56:18 +00:00
k-quants : support for super-block size of 64 (#2001) * k_quants: WIP super-blocks with 64 weights * k_quants: WIP super-blocks with 64 weights Q6_K scalar and AVX2 works * k_quants: WIP super-blocks with 64 weights Q4_K scalar and AVX2 works * k_quants: WIP super-blocks with 64 weights Q2_K scalar and AVX2 works. Q2_K is way too slow (it is actually slower than the scalar implementation) * k_quants: WIP super-blocks with 64 weights Q3_K scalar and AVX2 works. * k_quants: WIP super-blocks with 64 weights Q5_K scalar and AVX2 works, and with that all k_quants are done on AVX2 and scalar * k_quants: WIP super-blocks with 64 weights Q6_K working on CUDA. Cannot make it run quite as gast as with super-blocks with 256 weigths: 8% slower on 4080, 20% slower on the 1660 (but there we fit 1 less layer on the GPU because pf the larger model size), so some fraction of these 20% is due to that, * k_quants: WIP super-blocks with 64 weights Q4_K working on CUDA. ~10% slower on GTX-1660, 16% slower on 4080. * k_quants: WIP super-blocks with 64 weights Q2_K working on CUDA. ~3% slower on GTX-1660, 10% slower on 4080. * k_quants: WIP super-blocks with 64 weights Q3_K working on CUDA. * k_quants: WIP super-blocks with 64 weights Q5_K working on CUDA, and with this CUDA is done. * k_quants: WIP super-blocks with 64 weights Q6_K working on ARM_NEON * k_quants: WIP super-blocks with 64 weights Q4_K working on ARM_NEON, but quite a bit slower than 256 weights * k_quants: WIP super-blocks with 64 weights Q2_K working on ARM_NEON, but quite a bit slower than 256 weights * k_quants: WIP super-blocks with 64 weights Q3_K working on ARM_NEON, but quite a bit slower than 256 weights. * k_quants: WIP super-blocks with 64 weights Q5_K working on ARM_NEON, but quite a bit slower than 256 weights. With that, we have full support for ARM_NEON, although performance is not quite there. * k_quants: WIP super-blocks with 64 weights Slightly more efficient Q3_K and Q5_K * k_quants: WIP super-blocks with 64 weights Another small improvement for Q3_K and Q5_K on ARM_NEON * k_quants: WIP super-blocks with 64 weights Yet another speedup for Q5_K on ARM_NEON. We are now within 10% of the QK_K = 256 version. * k_quants: WIP super-blocks with 64 weights * We are able to pass preprocessor macros to the Metal compiler * Q6_K works and is actually slightly more efficient than the QK_K = 256 version (25.2 ms vs 25.8 ms) * k_quants: WIP super-blocks with 64 weights Q4_K works on Metal and is actually slightly faster than QK_K = 256 (21.95 ms vs 24.0 ms). * k_quants: WIP super-blocks with 64 weights Q2_K works on Metal and is very slightly faster than QK_K = 256 (23.8 ms vs 24.2 ms). * k_quants: WIP super-blocks with 64 weights Q3_K works on Metal and is slightly faster than QK_K = 256 (26.6 ms vs 28.3 ms). * k_quants: WIP super-blocks with 64 weights Q5_K works on Metal and is slightly faster than QK_K = 256 (23.7 ms vs 26.3 ms). * k_quants: call them _K, not _k, also on Metal * k_quants: correctly define QK_K in llama.cpp * Fixed bug in q4_K quantization added with the 64-block addition * Simplify via lambda * k_quants: swicth Q3_K to 4-bit scales when QK_K = 64 Otherwise there isn't much benefit from this quantization type. There is some very slight loss in accuracy, but we reduce size by ~7%. E.g., for OpenLLaMA-3B, Q3_K_S perplexity is 8.6131 with 8-bit scales and 8.6352 with 4-bit, while file size decreases from 1.53G to 1.44G. * k_quants: switch Q4_K to 4-bit scales when QK_K = 64 Here the loss in accuracy is greater than for Q3_K, but the Q4_K points still move further to the left on the perplexity vs size curve. * k_quants: forgot to add the Metal changes in last commit * k_quants: change Q5_K to be type 0 when QK_K = 64 Still needs AVX2 implementation * k_quants: AVX2 implementation for new 64-weight Q5_K * k_quants: 10% faster ARM_NEON Q5_K dot product * k_quants: fixed issue caused by merging with master --------- Co-authored-by: Iwan Kawrakow <iwan.kawrakow@gmail.com>
2023-06-26 16:43:07 +00:00
#if QK_K == 256
k-quants : allow to optionally disable at compile time (#1734) * k-quants : put behind optional compile flag LLAMA_K_QUANTS * build : enable k-quants by default
2023-06-07 07:59:52 +00:00
void ggml_vec_dot_q2_K_q8_K(const int n, float * restrict s, const void * restrict vx, const void * restrict vy) {
ggml : add SOTA 2,3,4,5,6 bit k-quantizations (#1684) * Starting to add k-quantization to ggml I think it is better to have quantization separate from ggml. For now just adding the k-quants there, but it would be better to also factor out the existing ggml quantizations. * Adding Q3_K and Q8_K (de)-quantization * Q3_K now working on CUDA and AVX2/scalar CUDA is not ideal - ~50% slower than Q4_0 for single token prediction, about the same in batch mode (perplexity). CPU single token is ~55 ms (on Ryzen 7950X). * Some improvement for Q3_K on CUDA It is now ~22.5 ms/token on my GPU, so ~30% slower than Q4_0. * Some more CUDA optimizations for Q3_K Single token is now 20.5 ms/token (~20% slower than Q4_0). Perplexity is on par with Q4_0. * Adding Q4_K - scalar, AVX2, CUDA Performance is the same or perhaps very slightly better than Q4_0 on the CPU. On the GPU, single token prediction is ~10% better than Q4_0, batch mode (perplexity is about the same). * Adding Q6_K - scalar, AVX2, CUDA Performance is ~40% lower compared to Q4_K on the CPU. This is to be expected, considering that we are memory bound on the CPU and the 6-bit model is ~44% larger than the 4-bit. On the GPU, single token prediction is ~6% lower than Q4_0, batch mode (perplexity) is even closer (but still slower). * Adding Q5_K - scalar, AVX2, CUDA Performance is ~20% lower compared to Q4_K on the CPU. This is to be expected, considering that we are memory bound on the CPU and the 5-bit model is ~22% larger than the 4-bit. On the GPU, single token prediction is about the same as Q4_0 for both, single token and batch prediction. * Per convention, all QX_K quantizations use Q5_K for output.weight * Adding quantization mixes * Quantization mixes: didn't quite get what I wanted in the last commit * Q4_K dot product for ARM_NEON * Q6_K dot product for ARM_NEON * Q5_K dot product for ARM_NEON * Adding Q3_K dot for ARM_NEON It is 22% slower than Q4_K, despite the smaller model size. On x86_64, where we are memory bound, the Q3_K model is quite a bit faster than Q4_K. * A very slightly faster ARM_NEON Q3_K dot * Adding Q2_K - just CUDA for now Token prediction is pretty good - about 15.5 ms on a RTX 4080. Perplexity is about the same as Q4_K. * Adding scalar and AVX2 Q2_K dot * Adding ARM_NEON Q2_K dot About the same performance as Q4_K. * A slightly faster ARM_NEON Q2_K dot Single token prediction is now ~36 ms on M2 Max. The code is much simpler too. * Fixed bug in Q2_K CUDA dot product kernel Stranegly enough, for the few prompts I tried with the 7B model the responses looked perfectly reasonable. Only realized something is not quite right when I tried the larger models and started getting nonse back. In any case, Q2_K single token evaluation time on an RTX 4080 in a Ryzen7950X box iusing CUDA and model fully loaded on the GPU are ~15.5 ms for 7B, ~25.4 ms for 13B, and ~55.8 ms for 30B. The max number of layers that fit in VRAM for The 65B is 32. With that, we get ~330 ms per token, which is not that much faster than just running on the CPU (~470 ms per token). * Don't print zeros/NaNs when no count histogram has been collected * A 10% faster CUDA vector dot kernel for Q3_K Q3_K is now running at ~18.5 ms / token on CUDA, so the gap to Q4_0 is only 10%. It seems memory acccess pattern is more important for performance than the amount of computation the kernel does. * A slightly daster Q4_K AVX2 dot product For perplexity, where we are less memory bound, time per pass drops by ~5%. Barely measurable difference for single token prediction. * A slightly faster ARM_NEON A4_K dot product * Minor * Fix quantization error test We cannot possibly be expecting rmse < 0.002 for 2- and 3-bit quantization variants. * Fix docker build I have been sloppy with vector reinterpret casts on ARM_NEON. It seems clang is very forgiving in that regard. * Added forgotten ggml.o dependence on k_quants.h to the Makefile * Had unintentionally committed the Makefile with -Ofast enabled * ggml : rename k_quants -> ggml-quants-k, use lowercase in code --------- Co-authored-by: Iwan Kawrakow <iwan.kawrakow@gmail.com> Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
2023-06-05 19:56:18 +00:00
k-quants : allow to optionally disable at compile time (#1734) * k-quants : put behind optional compile flag LLAMA_K_QUANTS * build : enable k-quants by default
2023-06-07 07:59:52 +00:00
const block_q2_K * restrict x = vx;
const block_q8_K * restrict y = vy;
ggml : add SOTA 2,3,4,5,6 bit k-quantizations (#1684) * Starting to add k-quantization to ggml I think it is better to have quantization separate from ggml. For now just adding the k-quants there, but it would be better to also factor out the existing ggml quantizations. * Adding Q3_K and Q8_K (de)-quantization * Q3_K now working on CUDA and AVX2/scalar CUDA is not ideal - ~50% slower than Q4_0 for single token prediction, about the same in batch mode (perplexity). CPU single token is ~55 ms (on Ryzen 7950X). * Some improvement for Q3_K on CUDA It is now ~22.5 ms/token on my GPU, so ~30% slower than Q4_0. * Some more CUDA optimizations for Q3_K Single token is now 20.5 ms/token (~20% slower than Q4_0). Perplexity is on par with Q4_0. * Adding Q4_K - scalar, AVX2, CUDA Performance is the same or perhaps very slightly better than Q4_0 on the CPU. On the GPU, single token prediction is ~10% better than Q4_0, batch mode (perplexity is about the same). * Adding Q6_K - scalar, AVX2, CUDA Performance is ~40% lower compared to Q4_K on the CPU. This is to be expected, considering that we are memory bound on the CPU and the 6-bit model is ~44% larger than the 4-bit. On the GPU, single token prediction is ~6% lower than Q4_0, batch mode (perplexity) is even closer (but still slower). * Adding Q5_K - scalar, AVX2, CUDA Performance is ~20% lower compared to Q4_K on the CPU. This is to be expected, considering that we are memory bound on the CPU and the 5-bit model is ~22% larger than the 4-bit. On the GPU, single token prediction is about the same as Q4_0 for both, single token and batch prediction. * Per convention, all QX_K quantizations use Q5_K for output.weight * Adding quantization mixes * Quantization mixes: didn't quite get what I wanted in the last commit * Q4_K dot product for ARM_NEON * Q6_K dot product for ARM_NEON * Q5_K dot product for ARM_NEON * Adding Q3_K dot for ARM_NEON It is 22% slower than Q4_K, despite the smaller model size. On x86_64, where we are memory bound, the Q3_K model is quite a bit faster than Q4_K. * A very slightly faster ARM_NEON Q3_K dot * Adding Q2_K - just CUDA for now Token prediction is pretty good - about 15.5 ms on a RTX 4080. Perplexity is about the same as Q4_K. * Adding scalar and AVX2 Q2_K dot * Adding ARM_NEON Q2_K dot About the same performance as Q4_K. * A slightly faster ARM_NEON Q2_K dot Single token prediction is now ~36 ms on M2 Max. The code is much simpler too. * Fixed bug in Q2_K CUDA dot product kernel Stranegly enough, for the few prompts I tried with the 7B model the responses looked perfectly reasonable. Only realized something is not quite right when I tried the larger models and started getting nonse back. In any case, Q2_K single token evaluation time on an RTX 4080 in a Ryzen7950X box iusing CUDA and model fully loaded on the GPU are ~15.5 ms for 7B, ~25.4 ms for 13B, and ~55.8 ms for 30B. The max number of layers that fit in VRAM for The 65B is 32. With that, we get ~330 ms per token, which is not that much faster than just running on the CPU (~470 ms per token). * Don't print zeros/NaNs when no count histogram has been collected * A 10% faster CUDA vector dot kernel for Q3_K Q3_K is now running at ~18.5 ms / token on CUDA, so the gap to Q4_0 is only 10%. It seems memory acccess pattern is more important for performance than the amount of computation the kernel does. * A slightly daster Q4_K AVX2 dot product For perplexity, where we are less memory bound, time per pass drops by ~5%. Barely measurable difference for single token prediction. * A slightly faster ARM_NEON A4_K dot product * Minor * Fix quantization error test We cannot possibly be expecting rmse < 0.002 for 2- and 3-bit quantization variants. * Fix docker build I have been sloppy with vector reinterpret casts on ARM_NEON. It seems clang is very forgiving in that regard. * Added forgotten ggml.o dependence on k_quants.h to the Makefile * Had unintentionally committed the Makefile with -Ofast enabled * ggml : rename k_quants -> ggml-quants-k, use lowercase in code --------- Co-authored-by: Iwan Kawrakow <iwan.kawrakow@gmail.com> Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
2023-06-05 19:56:18 +00:00
const int nb = n / QK_K;
#ifdef __ARM_NEON
const uint8x16_t m3 = vdupq_n_u8(0x3);
const uint8x16_t m4 = vdupq_n_u8(0xF);
ggml : fix dot product for ARM (#4630) ggml-ci
2023-12-27 09:02:13 +00:00
const int32x4_t vzero = vdupq_n_s32(0);
ggml : add SOTA 2,3,4,5,6 bit k-quantizations (#1684) * Starting to add k-quantization to ggml I think it is better to have quantization separate from ggml. For now just adding the k-quants there, but it would be better to also factor out the existing ggml quantizations. * Adding Q3_K and Q8_K (de)-quantization * Q3_K now working on CUDA and AVX2/scalar CUDA is not ideal - ~50% slower than Q4_0 for single token prediction, about the same in batch mode (perplexity). CPU single token is ~55 ms (on Ryzen 7950X). * Some improvement for Q3_K on CUDA It is now ~22.5 ms/token on my GPU, so ~30% slower than Q4_0. * Some more CUDA optimizations for Q3_K Single token is now 20.5 ms/token (~20% slower than Q4_0). Perplexity is on par with Q4_0. * Adding Q4_K - scalar, AVX2, CUDA Performance is the same or perhaps very slightly better than Q4_0 on the CPU. On the GPU, single token prediction is ~10% better than Q4_0, batch mode (perplexity is about the same). * Adding Q6_K - scalar, AVX2, CUDA Performance is ~40% lower compared to Q4_K on the CPU. This is to be expected, considering that we are memory bound on the CPU and the 6-bit model is ~44% larger than the 4-bit. On the GPU, single token prediction is ~6% lower than Q4_0, batch mode (perplexity) is even closer (but still slower). * Adding Q5_K - scalar, AVX2, CUDA Performance is ~20% lower compared to Q4_K on the CPU. This is to be expected, considering that we are memory bound on the CPU and the 5-bit model is ~22% larger than the 4-bit. On the GPU, single token prediction is about the same as Q4_0 for both, single token and batch prediction. * Per convention, all QX_K quantizations use Q5_K for output.weight * Adding quantization mixes * Quantization mixes: didn't quite get what I wanted in the last commit * Q4_K dot product for ARM_NEON * Q6_K dot product for ARM_NEON * Q5_K dot product for ARM_NEON * Adding Q3_K dot for ARM_NEON It is 22% slower than Q4_K, despite the smaller model size. On x86_64, where we are memory bound, the Q3_K model is quite a bit faster than Q4_K. * A very slightly faster ARM_NEON Q3_K dot * Adding Q2_K - just CUDA for now Token prediction is pretty good - about 15.5 ms on a RTX 4080. Perplexity is about the same as Q4_K. * Adding scalar and AVX2 Q2_K dot * Adding ARM_NEON Q2_K dot About the same performance as Q4_K. * A slightly faster ARM_NEON Q2_K dot Single token prediction is now ~36 ms on M2 Max. The code is much simpler too. * Fixed bug in Q2_K CUDA dot product kernel Stranegly enough, for the few prompts I tried with the 7B model the responses looked perfectly reasonable. Only realized something is not quite right when I tried the larger models and started getting nonse back. In any case, Q2_K single token evaluation time on an RTX 4080 in a Ryzen7950X box iusing CUDA and model fully loaded on the GPU are ~15.5 ms for 7B, ~25.4 ms for 13B, and ~55.8 ms for 30B. The max number of layers that fit in VRAM for The 65B is 32. With that, we get ~330 ms per token, which is not that much faster than just running on the CPU (~470 ms per token). * Don't print zeros/NaNs when no count histogram has been collected * A 10% faster CUDA vector dot kernel for Q3_K Q3_K is now running at ~18.5 ms / token on CUDA, so the gap to Q4_0 is only 10%. It seems memory acccess pattern is more important for performance than the amount of computation the kernel does. * A slightly daster Q4_K AVX2 dot product For perplexity, where we are less memory bound, time per pass drops by ~5%. Barely measurable difference for single token prediction. * A slightly faster ARM_NEON A4_K dot product * Minor * Fix quantization error test We cannot possibly be expecting rmse < 0.002 for 2- and 3-bit quantization variants. * Fix docker build I have been sloppy with vector reinterpret casts on ARM_NEON. It seems clang is very forgiving in that regard. * Added forgotten ggml.o dependence on k_quants.h to the Makefile * Had unintentionally committed the Makefile with -Ofast enabled * ggml : rename k_quants -> ggml-quants-k, use lowercase in code --------- Co-authored-by: Iwan Kawrakow <iwan.kawrakow@gmail.com> Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
2023-06-05 19:56:18 +00:00
ggml : sync (im2col, GPU conv, 32-bit arm compat) (#4060) ggml-ci
2023-11-13 14:55:52 +00:00
ggml_int8x16x2_t q2bytes;
ggml : add SOTA 2,3,4,5,6 bit k-quantizations (#1684) * Starting to add k-quantization to ggml I think it is better to have quantization separate from ggml. For now just adding the k-quants there, but it would be better to also factor out the existing ggml quantizations. * Adding Q3_K and Q8_K (de)-quantization * Q3_K now working on CUDA and AVX2/scalar CUDA is not ideal - ~50% slower than Q4_0 for single token prediction, about the same in batch mode (perplexity). CPU single token is ~55 ms (on Ryzen 7950X). * Some improvement for Q3_K on CUDA It is now ~22.5 ms/token on my GPU, so ~30% slower than Q4_0. * Some more CUDA optimizations for Q3_K Single token is now 20.5 ms/token (~20% slower than Q4_0). Perplexity is on par with Q4_0. * Adding Q4_K - scalar, AVX2, CUDA Performance is the same or perhaps very slightly better than Q4_0 on the CPU. On the GPU, single token prediction is ~10% better than Q4_0, batch mode (perplexity is about the same). * Adding Q6_K - scalar, AVX2, CUDA Performance is ~40% lower compared to Q4_K on the CPU. This is to be expected, considering that we are memory bound on the CPU and the 6-bit model is ~44% larger than the 4-bit. On the GPU, single token prediction is ~6% lower than Q4_0, batch mode (perplexity) is even closer (but still slower). * Adding Q5_K - scalar, AVX2, CUDA Performance is ~20% lower compared to Q4_K on the CPU. This is to be expected, considering that we are memory bound on the CPU and the 5-bit model is ~22% larger than the 4-bit. On the GPU, single token prediction is about the same as Q4_0 for both, single token and batch prediction. * Per convention, all QX_K quantizations use Q5_K for output.weight * Adding quantization mixes * Quantization mixes: didn't quite get what I wanted in the last commit * Q4_K dot product for ARM_NEON * Q6_K dot product for ARM_NEON * Q5_K dot product for ARM_NEON * Adding Q3_K dot for ARM_NEON It is 22% slower than Q4_K, despite the smaller model size. On x86_64, where we are memory bound, the Q3_K model is quite a bit faster than Q4_K. * A very slightly faster ARM_NEON Q3_K dot * Adding Q2_K - just CUDA for now Token prediction is pretty good - about 15.5 ms on a RTX 4080. Perplexity is about the same as Q4_K. * Adding scalar and AVX2 Q2_K dot * Adding ARM_NEON Q2_K dot About the same performance as Q4_K. * A slightly faster ARM_NEON Q2_K dot Single token prediction is now ~36 ms on M2 Max. The code is much simpler too. * Fixed bug in Q2_K CUDA dot product kernel Stranegly enough, for the few prompts I tried with the 7B model the responses looked perfectly reasonable. Only realized something is not quite right when I tried the larger models and started getting nonse back. In any case, Q2_K single token evaluation time on an RTX 4080 in a Ryzen7950X box iusing CUDA and model fully loaded on the GPU are ~15.5 ms for 7B, ~25.4 ms for 13B, and ~55.8 ms for 30B. The max number of layers that fit in VRAM for The 65B is 32. With that, we get ~330 ms per token, which is not that much faster than just running on the CPU (~470 ms per token). * Don't print zeros/NaNs when no count histogram has been collected * A 10% faster CUDA vector dot kernel for Q3_K Q3_K is now running at ~18.5 ms / token on CUDA, so the gap to Q4_0 is only 10%. It seems memory acccess pattern is more important for performance than the amount of computation the kernel does. * A slightly daster Q4_K AVX2 dot product For perplexity, where we are less memory bound, time per pass drops by ~5%. Barely measurable difference for single token prediction. * A slightly faster ARM_NEON A4_K dot product * Minor * Fix quantization error test We cannot possibly be expecting rmse < 0.002 for 2- and 3-bit quantization variants. * Fix docker build I have been sloppy with vector reinterpret casts on ARM_NEON. It seems clang is very forgiving in that regard. * Added forgotten ggml.o dependence on k_quants.h to the Makefile * Had unintentionally committed the Makefile with -Ofast enabled * ggml : rename k_quants -> ggml-quants-k, use lowercase in code --------- Co-authored-by: Iwan Kawrakow <iwan.kawrakow@gmail.com> Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
2023-06-05 19:56:18 +00:00
uint8_t aux[16];
float sum = 0;
for (int i = 0; i < nb; ++i) {
ggml : move FP16 <-> FP32 code to ggml-impl.h (#3861) * ggml : move FP16 <-> FP32 stuff to ggml-impl.h ggml-ci * tests : fix ARM build * ggml : explicitly initialize deprecated type traits * ggml : add math.h to ggml-impl.h * ggml : remove duplicate static assert macros * ggml : prefix lookup tables with ggml_ ggml-ci * ggml-impl : move extern "C" to start of file
2023-10-30 17:19:15 +00:00
const float d = y[i].d * GGML_FP16_TO_FP32(x[i].d);
const float dmin = -y[i].d * GGML_FP16_TO_FP32(x[i].dmin);
ggml : add SOTA 2,3,4,5,6 bit k-quantizations (#1684) * Starting to add k-quantization to ggml I think it is better to have quantization separate from ggml. For now just adding the k-quants there, but it would be better to also factor out the existing ggml quantizations. * Adding Q3_K and Q8_K (de)-quantization * Q3_K now working on CUDA and AVX2/scalar CUDA is not ideal - ~50% slower than Q4_0 for single token prediction, about the same in batch mode (perplexity). CPU single token is ~55 ms (on Ryzen 7950X). * Some improvement for Q3_K on CUDA It is now ~22.5 ms/token on my GPU, so ~30% slower than Q4_0. * Some more CUDA optimizations for Q3_K Single token is now 20.5 ms/token (~20% slower than Q4_0). Perplexity is on par with Q4_0. * Adding Q4_K - scalar, AVX2, CUDA Performance is the same or perhaps very slightly better than Q4_0 on the CPU. On the GPU, single token prediction is ~10% better than Q4_0, batch mode (perplexity is about the same). * Adding Q6_K - scalar, AVX2, CUDA Performance is ~40% lower compared to Q4_K on the CPU. This is to be expected, considering that we are memory bound on the CPU and the 6-bit model is ~44% larger than the 4-bit. On the GPU, single token prediction is ~6% lower than Q4_0, batch mode (perplexity) is even closer (but still slower). * Adding Q5_K - scalar, AVX2, CUDA Performance is ~20% lower compared to Q4_K on the CPU. This is to be expected, considering that we are memory bound on the CPU and the 5-bit model is ~22% larger than the 4-bit. On the GPU, single token prediction is about the same as Q4_0 for both, single token and batch prediction. * Per convention, all QX_K quantizations use Q5_K for output.weight * Adding quantization mixes * Quantization mixes: didn't quite get what I wanted in the last commit * Q4_K dot product for ARM_NEON * Q6_K dot product for ARM_NEON * Q5_K dot product for ARM_NEON * Adding Q3_K dot for ARM_NEON It is 22% slower than Q4_K, despite the smaller model size. On x86_64, where we are memory bound, the Q3_K model is quite a bit faster than Q4_K. * A very slightly faster ARM_NEON Q3_K dot * Adding Q2_K - just CUDA for now Token prediction is pretty good - about 15.5 ms on a RTX 4080. Perplexity is about the same as Q4_K. * Adding scalar and AVX2 Q2_K dot * Adding ARM_NEON Q2_K dot About the same performance as Q4_K. * A slightly faster ARM_NEON Q2_K dot Single token prediction is now ~36 ms on M2 Max. The code is much simpler too. * Fixed bug in Q2_K CUDA dot product kernel Stranegly enough, for the few prompts I tried with the 7B model the responses looked perfectly reasonable. Only realized something is not quite right when I tried the larger models and started getting nonse back. In any case, Q2_K single token evaluation time on an RTX 4080 in a Ryzen7950X box iusing CUDA and model fully loaded on the GPU are ~15.5 ms for 7B, ~25.4 ms for 13B, and ~55.8 ms for 30B. The max number of layers that fit in VRAM for The 65B is 32. With that, we get ~330 ms per token, which is not that much faster than just running on the CPU (~470 ms per token). * Don't print zeros/NaNs when no count histogram has been collected * A 10% faster CUDA vector dot kernel for Q3_K Q3_K is now running at ~18.5 ms / token on CUDA, so the gap to Q4_0 is only 10%. It seems memory acccess pattern is more important for performance than the amount of computation the kernel does. * A slightly daster Q4_K AVX2 dot product For perplexity, where we are less memory bound, time per pass drops by ~5%. Barely measurable difference for single token prediction. * A slightly faster ARM_NEON A4_K dot product * Minor * Fix quantization error test We cannot possibly be expecting rmse < 0.002 for 2- and 3-bit quantization variants. * Fix docker build I have been sloppy with vector reinterpret casts on ARM_NEON. It seems clang is very forgiving in that regard. * Added forgotten ggml.o dependence on k_quants.h to the Makefile * Had unintentionally committed the Makefile with -Ofast enabled * ggml : rename k_quants -> ggml-quants-k, use lowercase in code --------- Co-authored-by: Iwan Kawrakow <iwan.kawrakow@gmail.com> Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
2023-06-05 19:56:18 +00:00
const uint8_t * restrict q2 = x[i].qs;
const int8_t * restrict q8 = y[i].qs;
const uint8_t * restrict sc = x[i].scales;
const uint8x16_t mins_and_scales = vld1q_u8(sc);
const uint8x16_t scales = vandq_u8(mins_and_scales, m4);
vst1q_u8(aux, scales);
const uint8x16_t mins = vshrq_n_u8(mins_and_scales, 4);
ggml : sync (im2col, GPU conv, 32-bit arm compat) (#4060) ggml-ci
2023-11-13 14:55:52 +00:00
const ggml_int16x8x2_t q8sums = ggml_vld1q_s16_x2(y[i].bsums);
cuda : fix jetson compile error (#4560) * fix old jetson compile error * Update Makefile * update jetson detect and cuda version detect * update cuda marco define * update makefile and cuda,fix some issue * Update README.md Co-authored-by: Georgi Gerganov <ggerganov@gmail.com> * Update Makefile * Update README.md --------- Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
2023-12-22 15:11:12 +00:00
const ggml_int16x8x2_t mins16 = {{vreinterpretq_s16_u16(vmovl_u8(vget_low_u8(mins))), vreinterpretq_s16_u16(vmovl_u8(vget_high_u8(mins)))}};
ggml : add SOTA 2,3,4,5,6 bit k-quantizations (#1684) * Starting to add k-quantization to ggml I think it is better to have quantization separate from ggml. For now just adding the k-quants there, but it would be better to also factor out the existing ggml quantizations. * Adding Q3_K and Q8_K (de)-quantization * Q3_K now working on CUDA and AVX2/scalar CUDA is not ideal - ~50% slower than Q4_0 for single token prediction, about the same in batch mode (perplexity). CPU single token is ~55 ms (on Ryzen 7950X). * Some improvement for Q3_K on CUDA It is now ~22.5 ms/token on my GPU, so ~30% slower than Q4_0. * Some more CUDA optimizations for Q3_K Single token is now 20.5 ms/token (~20% slower than Q4_0). Perplexity is on par with Q4_0. * Adding Q4_K - scalar, AVX2, CUDA Performance is the same or perhaps very slightly better than Q4_0 on the CPU. On the GPU, single token prediction is ~10% better than Q4_0, batch mode (perplexity is about the same). * Adding Q6_K - scalar, AVX2, CUDA Performance is ~40% lower compared to Q4_K on the CPU. This is to be expected, considering that we are memory bound on the CPU and the 6-bit model is ~44% larger than the 4-bit. On the GPU, single token prediction is ~6% lower than Q4_0, batch mode (perplexity) is even closer (but still slower). * Adding Q5_K - scalar, AVX2, CUDA Performance is ~20% lower compared to Q4_K on the CPU. This is to be expected, considering that we are memory bound on the CPU and the 5-bit model is ~22% larger than the 4-bit. On the GPU, single token prediction is about the same as Q4_0 for both, single token and batch prediction. * Per convention, all QX_K quantizations use Q5_K for output.weight * Adding quantization mixes * Quantization mixes: didn't quite get what I wanted in the last commit * Q4_K dot product for ARM_NEON * Q6_K dot product for ARM_NEON * Q5_K dot product for ARM_NEON * Adding Q3_K dot for ARM_NEON It is 22% slower than Q4_K, despite the smaller model size. On x86_64, where we are memory bound, the Q3_K model is quite a bit faster than Q4_K. * A very slightly faster ARM_NEON Q3_K dot * Adding Q2_K - just CUDA for now Token prediction is pretty good - about 15.5 ms on a RTX 4080. Perplexity is about the same as Q4_K. * Adding scalar and AVX2 Q2_K dot * Adding ARM_NEON Q2_K dot About the same performance as Q4_K. * A slightly faster ARM_NEON Q2_K dot Single token prediction is now ~36 ms on M2 Max. The code is much simpler too. * Fixed bug in Q2_K CUDA dot product kernel Stranegly enough, for the few prompts I tried with the 7B model the responses looked perfectly reasonable. Only realized something is not quite right when I tried the larger models and started getting nonse back. In any case, Q2_K single token evaluation time on an RTX 4080 in a Ryzen7950X box iusing CUDA and model fully loaded on the GPU are ~15.5 ms for 7B, ~25.4 ms for 13B, and ~55.8 ms for 30B. The max number of layers that fit in VRAM for The 65B is 32. With that, we get ~330 ms per token, which is not that much faster than just running on the CPU (~470 ms per token). * Don't print zeros/NaNs when no count histogram has been collected * A 10% faster CUDA vector dot kernel for Q3_K Q3_K is now running at ~18.5 ms / token on CUDA, so the gap to Q4_0 is only 10%. It seems memory acccess pattern is more important for performance than the amount of computation the kernel does. * A slightly daster Q4_K AVX2 dot product For perplexity, where we are less memory bound, time per pass drops by ~5%. Barely measurable difference for single token prediction. * A slightly faster ARM_NEON A4_K dot product * Minor * Fix quantization error test We cannot possibly be expecting rmse < 0.002 for 2- and 3-bit quantization variants. * Fix docker build I have been sloppy with vector reinterpret casts on ARM_NEON. It seems clang is very forgiving in that regard. * Added forgotten ggml.o dependence on k_quants.h to the Makefile * Had unintentionally committed the Makefile with -Ofast enabled * ggml : rename k_quants -> ggml-quants-k, use lowercase in code --------- Co-authored-by: Iwan Kawrakow <iwan.kawrakow@gmail.com> Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
2023-06-05 19:56:18 +00:00
const int32x4_t s0 = vaddq_s32(vmull_s16(vget_low_s16 (mins16.val[0]), vget_low_s16 (q8sums.val[0])),
vmull_s16(vget_high_s16(mins16.val[0]), vget_high_s16(q8sums.val[0])));
const int32x4_t s1 = vaddq_s32(vmull_s16(vget_low_s16 (mins16.val[1]), vget_low_s16 (q8sums.val[1])),
vmull_s16(vget_high_s16(mins16.val[1]), vget_high_s16(q8sums.val[1])));
sum += dmin * vaddvq_s32(vaddq_s32(s0, s1));
int isum = 0;
int is = 0;
// We use this macro instead of a function call because for some reason
// the code runs 2-3% slower, even if the function is declared inline
#define MULTIPLY_ACCUM_WITH_SCALE(index)\
ggml : add ggml_vdotq_s32 alias (#4715) ggml-ci
2023-12-31 09:43:31 +00:00
isum += vaddvq_s32(ggml_vdotq_s32(vzero, q2bytes.val[0], q8bytes.val[0])) * aux[is+(index)];\
isum += vaddvq_s32(ggml_vdotq_s32(vzero, q2bytes.val[1], q8bytes.val[1])) * aux[is+1+(index)];
ggml : add SOTA 2,3,4,5,6 bit k-quantizations (#1684) * Starting to add k-quantization to ggml I think it is better to have quantization separate from ggml. For now just adding the k-quants there, but it would be better to also factor out the existing ggml quantizations. * Adding Q3_K and Q8_K (de)-quantization * Q3_K now working on CUDA and AVX2/scalar CUDA is not ideal - ~50% slower than Q4_0 for single token prediction, about the same in batch mode (perplexity). CPU single token is ~55 ms (on Ryzen 7950X). * Some improvement for Q3_K on CUDA It is now ~22.5 ms/token on my GPU, so ~30% slower than Q4_0. * Some more CUDA optimizations for Q3_K Single token is now 20.5 ms/token (~20% slower than Q4_0). Perplexity is on par with Q4_0. * Adding Q4_K - scalar, AVX2, CUDA Performance is the same or perhaps very slightly better than Q4_0 on the CPU. On the GPU, single token prediction is ~10% better than Q4_0, batch mode (perplexity is about the same). * Adding Q6_K - scalar, AVX2, CUDA Performance is ~40% lower compared to Q4_K on the CPU. This is to be expected, considering that we are memory bound on the CPU and the 6-bit model is ~44% larger than the 4-bit. On the GPU, single token prediction is ~6% lower than Q4_0, batch mode (perplexity) is even closer (but still slower). * Adding Q5_K - scalar, AVX2, CUDA Performance is ~20% lower compared to Q4_K on the CPU. This is to be expected, considering that we are memory bound on the CPU and the 5-bit model is ~22% larger than the 4-bit. On the GPU, single token prediction is about the same as Q4_0 for both, single token and batch prediction. * Per convention, all QX_K quantizations use Q5_K for output.weight * Adding quantization mixes * Quantization mixes: didn't quite get what I wanted in the last commit * Q4_K dot product for ARM_NEON * Q6_K dot product for ARM_NEON * Q5_K dot product for ARM_NEON * Adding Q3_K dot for ARM_NEON It is 22% slower than Q4_K, despite the smaller model size. On x86_64, where we are memory bound, the Q3_K model is quite a bit faster than Q4_K. * A very slightly faster ARM_NEON Q3_K dot * Adding Q2_K - just CUDA for now Token prediction is pretty good - about 15.5 ms on a RTX 4080. Perplexity is about the same as Q4_K. * Adding scalar and AVX2 Q2_K dot * Adding ARM_NEON Q2_K dot About the same performance as Q4_K. * A slightly faster ARM_NEON Q2_K dot Single token prediction is now ~36 ms on M2 Max. The code is much simpler too. * Fixed bug in Q2_K CUDA dot product kernel Stranegly enough, for the few prompts I tried with the 7B model the responses looked perfectly reasonable. Only realized something is not quite right when I tried the larger models and started getting nonse back. In any case, Q2_K single token evaluation time on an RTX 4080 in a Ryzen7950X box iusing CUDA and model fully loaded on the GPU are ~15.5 ms for 7B, ~25.4 ms for 13B, and ~55.8 ms for 30B. The max number of layers that fit in VRAM for The 65B is 32. With that, we get ~330 ms per token, which is not that much faster than just running on the CPU (~470 ms per token). * Don't print zeros/NaNs when no count histogram has been collected * A 10% faster CUDA vector dot kernel for Q3_K Q3_K is now running at ~18.5 ms / token on CUDA, so the gap to Q4_0 is only 10%. It seems memory acccess pattern is more important for performance than the amount of computation the kernel does. * A slightly daster Q4_K AVX2 dot product For perplexity, where we are less memory bound, time per pass drops by ~5%. Barely measurable difference for single token prediction. * A slightly faster ARM_NEON A4_K dot product * Minor * Fix quantization error test We cannot possibly be expecting rmse < 0.002 for 2- and 3-bit quantization variants. * Fix docker build I have been sloppy with vector reinterpret casts on ARM_NEON. It seems clang is very forgiving in that regard. * Added forgotten ggml.o dependence on k_quants.h to the Makefile * Had unintentionally committed the Makefile with -Ofast enabled * ggml : rename k_quants -> ggml-quants-k, use lowercase in code --------- Co-authored-by: Iwan Kawrakow <iwan.kawrakow@gmail.com> Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
2023-06-05 19:56:18 +00:00
#define SHIFT_MULTIPLY_ACCUM_WITH_SCALE(shift, index)\
ggml : sync (im2col, GPU conv, 32-bit arm compat) (#4060) ggml-ci
2023-11-13 14:55:52 +00:00
q8bytes = ggml_vld1q_s8_x2(q8); q8 += 32;\
ggml : add SOTA 2,3,4,5,6 bit k-quantizations (#1684) * Starting to add k-quantization to ggml I think it is better to have quantization separate from ggml. For now just adding the k-quants there, but it would be better to also factor out the existing ggml quantizations. * Adding Q3_K and Q8_K (de)-quantization * Q3_K now working on CUDA and AVX2/scalar CUDA is not ideal - ~50% slower than Q4_0 for single token prediction, about the same in batch mode (perplexity). CPU single token is ~55 ms (on Ryzen 7950X). * Some improvement for Q3_K on CUDA It is now ~22.5 ms/token on my GPU, so ~30% slower than Q4_0. * Some more CUDA optimizations for Q3_K Single token is now 20.5 ms/token (~20% slower than Q4_0). Perplexity is on par with Q4_0. * Adding Q4_K - scalar, AVX2, CUDA Performance is the same or perhaps very slightly better than Q4_0 on the CPU. On the GPU, single token prediction is ~10% better than Q4_0, batch mode (perplexity is about the same). * Adding Q6_K - scalar, AVX2, CUDA Performance is ~40% lower compared to Q4_K on the CPU. This is to be expected, considering that we are memory bound on the CPU and the 6-bit model is ~44% larger than the 4-bit. On the GPU, single token prediction is ~6% lower than Q4_0, batch mode (perplexity) is even closer (but still slower). * Adding Q5_K - scalar, AVX2, CUDA Performance is ~20% lower compared to Q4_K on the CPU. This is to be expected, considering that we are memory bound on the CPU and the 5-bit model is ~22% larger than the 4-bit. On the GPU, single token prediction is about the same as Q4_0 for both, single token and batch prediction. * Per convention, all QX_K quantizations use Q5_K for output.weight * Adding quantization mixes * Quantization mixes: didn't quite get what I wanted in the last commit * Q4_K dot product for ARM_NEON * Q6_K dot product for ARM_NEON * Q5_K dot product for ARM_NEON * Adding Q3_K dot for ARM_NEON It is 22% slower than Q4_K, despite the smaller model size. On x86_64, where we are memory bound, the Q3_K model is quite a bit faster than Q4_K. * A very slightly faster ARM_NEON Q3_K dot * Adding Q2_K - just CUDA for now Token prediction is pretty good - about 15.5 ms on a RTX 4080. Perplexity is about the same as Q4_K. * Adding scalar and AVX2 Q2_K dot * Adding ARM_NEON Q2_K dot About the same performance as Q4_K. * A slightly faster ARM_NEON Q2_K dot Single token prediction is now ~36 ms on M2 Max. The code is much simpler too. * Fixed bug in Q2_K CUDA dot product kernel Stranegly enough, for the few prompts I tried with the 7B model the responses looked perfectly reasonable. Only realized something is not quite right when I tried the larger models and started getting nonse back. In any case, Q2_K single token evaluation time on an RTX 4080 in a Ryzen7950X box iusing CUDA and model fully loaded on the GPU are ~15.5 ms for 7B, ~25.4 ms for 13B, and ~55.8 ms for 30B. The max number of layers that fit in VRAM for The 65B is 32. With that, we get ~330 ms per token, which is not that much faster than just running on the CPU (~470 ms per token). * Don't print zeros/NaNs when no count histogram has been collected * A 10% faster CUDA vector dot kernel for Q3_K Q3_K is now running at ~18.5 ms / token on CUDA, so the gap to Q4_0 is only 10%. It seems memory acccess pattern is more important for performance than the amount of computation the kernel does. * A slightly daster Q4_K AVX2 dot product For perplexity, where we are less memory bound, time per pass drops by ~5%. Barely measurable difference for single token prediction. * A slightly faster ARM_NEON A4_K dot product * Minor * Fix quantization error test We cannot possibly be expecting rmse < 0.002 for 2- and 3-bit quantization variants. * Fix docker build I have been sloppy with vector reinterpret casts on ARM_NEON. It seems clang is very forgiving in that regard. * Added forgotten ggml.o dependence on k_quants.h to the Makefile * Had unintentionally committed the Makefile with -Ofast enabled * ggml : rename k_quants -> ggml-quants-k, use lowercase in code --------- Co-authored-by: Iwan Kawrakow <iwan.kawrakow@gmail.com> Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
2023-06-05 19:56:18 +00:00
q2bytes.val[0] = vreinterpretq_s8_u8(vandq_u8(vshrq_n_u8(q2bits.val[0], (shift)), m3));\
q2bytes.val[1] = vreinterpretq_s8_u8(vandq_u8(vshrq_n_u8(q2bits.val[1], (shift)), m3));\
MULTIPLY_ACCUM_WITH_SCALE((index));
for (int j = 0; j < QK_K/128; ++j) {
ggml : sync (im2col, GPU conv, 32-bit arm compat) (#4060) ggml-ci
2023-11-13 14:55:52 +00:00
const ggml_uint8x16x2_t q2bits = ggml_vld1q_u8_x2(q2); q2 += 32;
ggml : add SOTA 2,3,4,5,6 bit k-quantizations (#1684) * Starting to add k-quantization to ggml I think it is better to have quantization separate from ggml. For now just adding the k-quants there, but it would be better to also factor out the existing ggml quantizations. * Adding Q3_K and Q8_K (de)-quantization * Q3_K now working on CUDA and AVX2/scalar CUDA is not ideal - ~50% slower than Q4_0 for single token prediction, about the same in batch mode (perplexity). CPU single token is ~55 ms (on Ryzen 7950X). * Some improvement for Q3_K on CUDA It is now ~22.5 ms/token on my GPU, so ~30% slower than Q4_0. * Some more CUDA optimizations for Q3_K Single token is now 20.5 ms/token (~20% slower than Q4_0). Perplexity is on par with Q4_0. * Adding Q4_K - scalar, AVX2, CUDA Performance is the same or perhaps very slightly better than Q4_0 on the CPU. On the GPU, single token prediction is ~10% better than Q4_0, batch mode (perplexity is about the same). * Adding Q6_K - scalar, AVX2, CUDA Performance is ~40% lower compared to Q4_K on the CPU. This is to be expected, considering that we are memory bound on the CPU and the 6-bit model is ~44% larger than the 4-bit. On the GPU, single token prediction is ~6% lower than Q4_0, batch mode (perplexity) is even closer (but still slower). * Adding Q5_K - scalar, AVX2, CUDA Performance is ~20% lower compared to Q4_K on the CPU. This is to be expected, considering that we are memory bound on the CPU and the 5-bit model is ~22% larger than the 4-bit. On the GPU, single token prediction is about the same as Q4_0 for both, single token and batch prediction. * Per convention, all QX_K quantizations use Q5_K for output.weight * Adding quantization mixes * Quantization mixes: didn't quite get what I wanted in the last commit * Q4_K dot product for ARM_NEON * Q6_K dot product for ARM_NEON * Q5_K dot product for ARM_NEON * Adding Q3_K dot for ARM_NEON It is 22% slower than Q4_K, despite the smaller model size. On x86_64, where we are memory bound, the Q3_K model is quite a bit faster than Q4_K. * A very slightly faster ARM_NEON Q3_K dot * Adding Q2_K - just CUDA for now Token prediction is pretty good - about 15.5 ms on a RTX 4080. Perplexity is about the same as Q4_K. * Adding scalar and AVX2 Q2_K dot * Adding ARM_NEON Q2_K dot About the same performance as Q4_K. * A slightly faster ARM_NEON Q2_K dot Single token prediction is now ~36 ms on M2 Max. The code is much simpler too. * Fixed bug in Q2_K CUDA dot product kernel Stranegly enough, for the few prompts I tried with the 7B model the responses looked perfectly reasonable. Only realized something is not quite right when I tried the larger models and started getting nonse back. In any case, Q2_K single token evaluation time on an RTX 4080 in a Ryzen7950X box iusing CUDA and model fully loaded on the GPU are ~15.5 ms for 7B, ~25.4 ms for 13B, and ~55.8 ms for 30B. The max number of layers that fit in VRAM for The 65B is 32. With that, we get ~330 ms per token, which is not that much faster than just running on the CPU (~470 ms per token). * Don't print zeros/NaNs when no count histogram has been collected * A 10% faster CUDA vector dot kernel for Q3_K Q3_K is now running at ~18.5 ms / token on CUDA, so the gap to Q4_0 is only 10%. It seems memory acccess pattern is more important for performance than the amount of computation the kernel does. * A slightly daster Q4_K AVX2 dot product For perplexity, where we are less memory bound, time per pass drops by ~5%. Barely measurable difference for single token prediction. * A slightly faster ARM_NEON A4_K dot product * Minor * Fix quantization error test We cannot possibly be expecting rmse < 0.002 for 2- and 3-bit quantization variants. * Fix docker build I have been sloppy with vector reinterpret casts on ARM_NEON. It seems clang is very forgiving in that regard. * Added forgotten ggml.o dependence on k_quants.h to the Makefile * Had unintentionally committed the Makefile with -Ofast enabled * ggml : rename k_quants -> ggml-quants-k, use lowercase in code --------- Co-authored-by: Iwan Kawrakow <iwan.kawrakow@gmail.com> Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
2023-06-05 19:56:18 +00:00
ggml : sync (im2col, GPU conv, 32-bit arm compat) (#4060) ggml-ci
2023-11-13 14:55:52 +00:00
ggml_int8x16x2_t q8bytes = ggml_vld1q_s8_x2(q8); q8 += 32;
ggml : add SOTA 2,3,4,5,6 bit k-quantizations (#1684) * Starting to add k-quantization to ggml I think it is better to have quantization separate from ggml. For now just adding the k-quants there, but it would be better to also factor out the existing ggml quantizations. * Adding Q3_K and Q8_K (de)-quantization * Q3_K now working on CUDA and AVX2/scalar CUDA is not ideal - ~50% slower than Q4_0 for single token prediction, about the same in batch mode (perplexity). CPU single token is ~55 ms (on Ryzen 7950X). * Some improvement for Q3_K on CUDA It is now ~22.5 ms/token on my GPU, so ~30% slower than Q4_0. * Some more CUDA optimizations for Q3_K Single token is now 20.5 ms/token (~20% slower than Q4_0). Perplexity is on par with Q4_0. * Adding Q4_K - scalar, AVX2, CUDA Performance is the same or perhaps very slightly better than Q4_0 on the CPU. On the GPU, single token prediction is ~10% better than Q4_0, batch mode (perplexity is about the same). * Adding Q6_K - scalar, AVX2, CUDA Performance is ~40% lower compared to Q4_K on the CPU. This is to be expected, considering that we are memory bound on the CPU and the 6-bit model is ~44% larger than the 4-bit. On the GPU, single token prediction is ~6% lower than Q4_0, batch mode (perplexity) is even closer (but still slower). * Adding Q5_K - scalar, AVX2, CUDA Performance is ~20% lower compared to Q4_K on the CPU. This is to be expected, considering that we are memory bound on the CPU and the 5-bit model is ~22% larger than the 4-bit. On the GPU, single token prediction is about the same as Q4_0 for both, single token and batch prediction. * Per convention, all QX_K quantizations use Q5_K for output.weight * Adding quantization mixes * Quantization mixes: didn't quite get what I wanted in the last commit * Q4_K dot product for ARM_NEON * Q6_K dot product for ARM_NEON * Q5_K dot product for ARM_NEON * Adding Q3_K dot for ARM_NEON It is 22% slower than Q4_K, despite the smaller model size. On x86_64, where we are memory bound, the Q3_K model is quite a bit faster than Q4_K. * A very slightly faster ARM_NEON Q3_K dot * Adding Q2_K - just CUDA for now Token prediction is pretty good - about 15.5 ms on a RTX 4080. Perplexity is about the same as Q4_K. * Adding scalar and AVX2 Q2_K dot * Adding ARM_NEON Q2_K dot About the same performance as Q4_K. * A slightly faster ARM_NEON Q2_K dot Single token prediction is now ~36 ms on M2 Max. The code is much simpler too. * Fixed bug in Q2_K CUDA dot product kernel Stranegly enough, for the few prompts I tried with the 7B model the responses looked perfectly reasonable. Only realized something is not quite right when I tried the larger models and started getting nonse back. In any case, Q2_K single token evaluation time on an RTX 4080 in a Ryzen7950X box iusing CUDA and model fully loaded on the GPU are ~15.5 ms for 7B, ~25.4 ms for 13B, and ~55.8 ms for 30B. The max number of layers that fit in VRAM for The 65B is 32. With that, we get ~330 ms per token, which is not that much faster than just running on the CPU (~470 ms per token). * Don't print zeros/NaNs when no count histogram has been collected * A 10% faster CUDA vector dot kernel for Q3_K Q3_K is now running at ~18.5 ms / token on CUDA, so the gap to Q4_0 is only 10%. It seems memory acccess pattern is more important for performance than the amount of computation the kernel does. * A slightly daster Q4_K AVX2 dot product For perplexity, where we are less memory bound, time per pass drops by ~5%. Barely measurable difference for single token prediction. * A slightly faster ARM_NEON A4_K dot product * Minor * Fix quantization error test We cannot possibly be expecting rmse < 0.002 for 2- and 3-bit quantization variants. * Fix docker build I have been sloppy with vector reinterpret casts on ARM_NEON. It seems clang is very forgiving in that regard. * Added forgotten ggml.o dependence on k_quants.h to the Makefile * Had unintentionally committed the Makefile with -Ofast enabled * ggml : rename k_quants -> ggml-quants-k, use lowercase in code --------- Co-authored-by: Iwan Kawrakow <iwan.kawrakow@gmail.com> Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
2023-06-05 19:56:18 +00:00
q2bytes.val[0] = vreinterpretq_s8_u8(vandq_u8(q2bits.val[0], m3));
q2bytes.val[1] = vreinterpretq_s8_u8(vandq_u8(q2bits.val[1], m3));
ggml : fix dot product for ARM (#4630) ggml-ci
2023-12-27 09:02:13 +00:00
ggml : add SOTA 2,3,4,5,6 bit k-quantizations (#1684) * Starting to add k-quantization to ggml I think it is better to have quantization separate from ggml. For now just adding the k-quants there, but it would be better to also factor out the existing ggml quantizations. * Adding Q3_K and Q8_K (de)-quantization * Q3_K now working on CUDA and AVX2/scalar CUDA is not ideal - ~50% slower than Q4_0 for single token prediction, about the same in batch mode (perplexity). CPU single token is ~55 ms (on Ryzen 7950X). * Some improvement for Q3_K on CUDA It is now ~22.5 ms/token on my GPU, so ~30% slower than Q4_0. * Some more CUDA optimizations for Q3_K Single token is now 20.5 ms/token (~20% slower than Q4_0). Perplexity is on par with Q4_0. * Adding Q4_K - scalar, AVX2, CUDA Performance is the same or perhaps very slightly better than Q4_0 on the CPU. On the GPU, single token prediction is ~10% better than Q4_0, batch mode (perplexity is about the same). * Adding Q6_K - scalar, AVX2, CUDA Performance is ~40% lower compared to Q4_K on the CPU. This is to be expected, considering that we are memory bound on the CPU and the 6-bit model is ~44% larger than the 4-bit. On the GPU, single token prediction is ~6% lower than Q4_0, batch mode (perplexity) is even closer (but still slower). * Adding Q5_K - scalar, AVX2, CUDA Performance is ~20% lower compared to Q4_K on the CPU. This is to be expected, considering that we are memory bound on the CPU and the 5-bit model is ~22% larger than the 4-bit. On the GPU, single token prediction is about the same as Q4_0 for both, single token and batch prediction. * Per convention, all QX_K quantizations use Q5_K for output.weight * Adding quantization mixes * Quantization mixes: didn't quite get what I wanted in the last commit * Q4_K dot product for ARM_NEON * Q6_K dot product for ARM_NEON * Q5_K dot product for ARM_NEON * Adding Q3_K dot for ARM_NEON It is 22% slower than Q4_K, despite the smaller model size. On x86_64, where we are memory bound, the Q3_K model is quite a bit faster than Q4_K. * A very slightly faster ARM_NEON Q3_K dot * Adding Q2_K - just CUDA for now Token prediction is pretty good - about 15.5 ms on a RTX 4080. Perplexity is about the same as Q4_K. * Adding scalar and AVX2 Q2_K dot * Adding ARM_NEON Q2_K dot About the same performance as Q4_K. * A slightly faster ARM_NEON Q2_K dot Single token prediction is now ~36 ms on M2 Max. The code is much simpler too. * Fixed bug in Q2_K CUDA dot product kernel Stranegly enough, for the few prompts I tried with the 7B model the responses looked perfectly reasonable. Only realized something is not quite right when I tried the larger models and started getting nonse back. In any case, Q2_K single token evaluation time on an RTX 4080 in a Ryzen7950X box iusing CUDA and model fully loaded on the GPU are ~15.5 ms for 7B, ~25.4 ms for 13B, and ~55.8 ms for 30B. The max number of layers that fit in VRAM for The 65B is 32. With that, we get ~330 ms per token, which is not that much faster than just running on the CPU (~470 ms per token). * Don't print zeros/NaNs when no count histogram has been collected * A 10% faster CUDA vector dot kernel for Q3_K Q3_K is now running at ~18.5 ms / token on CUDA, so the gap to Q4_0 is only 10%. It seems memory acccess pattern is more important for performance than the amount of computation the kernel does. * A slightly daster Q4_K AVX2 dot product For perplexity, where we are less memory bound, time per pass drops by ~5%. Barely measurable difference for single token prediction. * A slightly faster ARM_NEON A4_K dot product * Minor * Fix quantization error test We cannot possibly be expecting rmse < 0.002 for 2- and 3-bit quantization variants. * Fix docker build I have been sloppy with vector reinterpret casts on ARM_NEON. It seems clang is very forgiving in that regard. * Added forgotten ggml.o dependence on k_quants.h to the Makefile * Had unintentionally committed the Makefile with -Ofast enabled * ggml : rename k_quants -> ggml-quants-k, use lowercase in code --------- Co-authored-by: Iwan Kawrakow <iwan.kawrakow@gmail.com> Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
2023-06-05 19:56:18 +00:00
MULTIPLY_ACCUM_WITH_SCALE(0);
SHIFT_MULTIPLY_ACCUM_WITH_SCALE(2, 2);
SHIFT_MULTIPLY_ACCUM_WITH_SCALE(4, 4);
SHIFT_MULTIPLY_ACCUM_WITH_SCALE(6, 6);
is += 8;
}
ggml : fix dot product for ARM (#4630) ggml-ci
2023-12-27 09:02:13 +00:00
sum += d * isum;
ggml : add SOTA 2,3,4,5,6 bit k-quantizations (#1684) * Starting to add k-quantization to ggml I think it is better to have quantization separate from ggml. For now just adding the k-quants there, but it would be better to also factor out the existing ggml quantizations. * Adding Q3_K and Q8_K (de)-quantization * Q3_K now working on CUDA and AVX2/scalar CUDA is not ideal - ~50% slower than Q4_0 for single token prediction, about the same in batch mode (perplexity). CPU single token is ~55 ms (on Ryzen 7950X). * Some improvement for Q3_K on CUDA It is now ~22.5 ms/token on my GPU, so ~30% slower than Q4_0. * Some more CUDA optimizations for Q3_K Single token is now 20.5 ms/token (~20% slower than Q4_0). Perplexity is on par with Q4_0. * Adding Q4_K - scalar, AVX2, CUDA Performance is the same or perhaps very slightly better than Q4_0 on the CPU. On the GPU, single token prediction is ~10% better than Q4_0, batch mode (perplexity is about the same). * Adding Q6_K - scalar, AVX2, CUDA Performance is ~40% lower compared to Q4_K on the CPU. This is to be expected, considering that we are memory bound on the CPU and the 6-bit model is ~44% larger than the 4-bit. On the GPU, single token prediction is ~6% lower than Q4_0, batch mode (perplexity) is even closer (but still slower). * Adding Q5_K - scalar, AVX2, CUDA Performance is ~20% lower compared to Q4_K on the CPU. This is to be expected, considering that we are memory bound on the CPU and the 5-bit model is ~22% larger than the 4-bit. On the GPU, single token prediction is about the same as Q4_0 for both, single token and batch prediction. * Per convention, all QX_K quantizations use Q5_K for output.weight * Adding quantization mixes * Quantization mixes: didn't quite get what I wanted in the last commit * Q4_K dot product for ARM_NEON * Q6_K dot product for ARM_NEON * Q5_K dot product for ARM_NEON * Adding Q3_K dot for ARM_NEON It is 22% slower than Q4_K, despite the smaller model size. On x86_64, where we are memory bound, the Q3_K model is quite a bit faster than Q4_K. * A very slightly faster ARM_NEON Q3_K dot * Adding Q2_K - just CUDA for now Token prediction is pretty good - about 15.5 ms on a RTX 4080. Perplexity is about the same as Q4_K. * Adding scalar and AVX2 Q2_K dot * Adding ARM_NEON Q2_K dot About the same performance as Q4_K. * A slightly faster ARM_NEON Q2_K dot Single token prediction is now ~36 ms on M2 Max. The code is much simpler too. * Fixed bug in Q2_K CUDA dot product kernel Stranegly enough, for the few prompts I tried with the 7B model the responses looked perfectly reasonable. Only realized something is not quite right when I tried the larger models and started getting nonse back. In any case, Q2_K single token evaluation time on an RTX 4080 in a Ryzen7950X box iusing CUDA and model fully loaded on the GPU are ~15.5 ms for 7B, ~25.4 ms for 13B, and ~55.8 ms for 30B. The max number of layers that fit in VRAM for The 65B is 32. With that, we get ~330 ms per token, which is not that much faster than just running on the CPU (~470 ms per token). * Don't print zeros/NaNs when no count histogram has been collected * A 10% faster CUDA vector dot kernel for Q3_K Q3_K is now running at ~18.5 ms / token on CUDA, so the gap to Q4_0 is only 10%. It seems memory acccess pattern is more important for performance than the amount of computation the kernel does. * A slightly daster Q4_K AVX2 dot product For perplexity, where we are less memory bound, time per pass drops by ~5%. Barely measurable difference for single token prediction. * A slightly faster ARM_NEON A4_K dot product * Minor * Fix quantization error test We cannot possibly be expecting rmse < 0.002 for 2- and 3-bit quantization variants. * Fix docker build I have been sloppy with vector reinterpret casts on ARM_NEON. It seems clang is very forgiving in that regard. * Added forgotten ggml.o dependence on k_quants.h to the Makefile * Had unintentionally committed the Makefile with -Ofast enabled * ggml : rename k_quants -> ggml-quants-k, use lowercase in code --------- Co-authored-by: Iwan Kawrakow <iwan.kawrakow@gmail.com> Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
2023-06-05 19:56:18 +00:00
}
*s = sum;
#elif defined __AVX2__
const __m256i m3 = _mm256_set1_epi8(3);
const __m128i m4 = _mm_set1_epi8(0xF);
__m256 acc = _mm256_setzero_ps();
for (int i = 0; i < nb; ++i) {
ggml : move FP16 <-> FP32 code to ggml-impl.h (#3861) * ggml : move FP16 <-> FP32 stuff to ggml-impl.h ggml-ci * tests : fix ARM build * ggml : explicitly initialize deprecated type traits * ggml : add math.h to ggml-impl.h * ggml : remove duplicate static assert macros * ggml : prefix lookup tables with ggml_ ggml-ci * ggml-impl : move extern "C" to start of file
2023-10-30 17:19:15 +00:00
const float d = y[i].d * GGML_FP16_TO_FP32(x[i].d);
const float dmin = -y[i].d * GGML_FP16_TO_FP32(x[i].dmin);
ggml : add SOTA 2,3,4,5,6 bit k-quantizations (#1684) * Starting to add k-quantization to ggml I think it is better to have quantization separate from ggml. For now just adding the k-quants there, but it would be better to also factor out the existing ggml quantizations. * Adding Q3_K and Q8_K (de)-quantization * Q3_K now working on CUDA and AVX2/scalar CUDA is not ideal - ~50% slower than Q4_0 for single token prediction, about the same in batch mode (perplexity). CPU single token is ~55 ms (on Ryzen 7950X). * Some improvement for Q3_K on CUDA It is now ~22.5 ms/token on my GPU, so ~30% slower than Q4_0. * Some more CUDA optimizations for Q3_K Single token is now 20.5 ms/token (~20% slower than Q4_0). Perplexity is on par with Q4_0. * Adding Q4_K - scalar, AVX2, CUDA Performance is the same or perhaps very slightly better than Q4_0 on the CPU. On the GPU, single token prediction is ~10% better than Q4_0, batch mode (perplexity is about the same). * Adding Q6_K - scalar, AVX2, CUDA Performance is ~40% lower compared to Q4_K on the CPU. This is to be expected, considering that we are memory bound on the CPU and the 6-bit model is ~44% larger than the 4-bit. On the GPU, single token prediction is ~6% lower than Q4_0, batch mode (perplexity) is even closer (but still slower). * Adding Q5_K - scalar, AVX2, CUDA Performance is ~20% lower compared to Q4_K on the CPU. This is to be expected, considering that we are memory bound on the CPU and the 5-bit model is ~22% larger than the 4-bit. On the GPU, single token prediction is about the same as Q4_0 for both, single token and batch prediction. * Per convention, all QX_K quantizations use Q5_K for output.weight * Adding quantization mixes * Quantization mixes: didn't quite get what I wanted in the last commit * Q4_K dot product for ARM_NEON * Q6_K dot product for ARM_NEON * Q5_K dot product for ARM_NEON * Adding Q3_K dot for ARM_NEON It is 22% slower than Q4_K, despite the smaller model size. On x86_64, where we are memory bound, the Q3_K model is quite a bit faster than Q4_K. * A very slightly faster ARM_NEON Q3_K dot * Adding Q2_K - just CUDA for now Token prediction is pretty good - about 15.5 ms on a RTX 4080. Perplexity is about the same as Q4_K. * Adding scalar and AVX2 Q2_K dot * Adding ARM_NEON Q2_K dot About the same performance as Q4_K. * A slightly faster ARM_NEON Q2_K dot Single token prediction is now ~36 ms on M2 Max. The code is much simpler too. * Fixed bug in Q2_K CUDA dot product kernel Stranegly enough, for the few prompts I tried with the 7B model the responses looked perfectly reasonable. Only realized something is not quite right when I tried the larger models and started getting nonse back. In any case, Q2_K single token evaluation time on an RTX 4080 in a Ryzen7950X box iusing CUDA and model fully loaded on the GPU are ~15.5 ms for 7B, ~25.4 ms for 13B, and ~55.8 ms for 30B. The max number of layers that fit in VRAM for The 65B is 32. With that, we get ~330 ms per token, which is not that much faster than just running on the CPU (~470 ms per token). * Don't print zeros/NaNs when no count histogram has been collected * A 10% faster CUDA vector dot kernel for Q3_K Q3_K is now running at ~18.5 ms / token on CUDA, so the gap to Q4_0 is only 10%. It seems memory acccess pattern is more important for performance than the amount of computation the kernel does. * A slightly daster Q4_K AVX2 dot product For perplexity, where we are less memory bound, time per pass drops by ~5%. Barely measurable difference for single token prediction. * A slightly faster ARM_NEON A4_K dot product * Minor * Fix quantization error test We cannot possibly be expecting rmse < 0.002 for 2- and 3-bit quantization variants. * Fix docker build I have been sloppy with vector reinterpret casts on ARM_NEON. It seems clang is very forgiving in that regard. * Added forgotten ggml.o dependence on k_quants.h to the Makefile * Had unintentionally committed the Makefile with -Ofast enabled * ggml : rename k_quants -> ggml-quants-k, use lowercase in code --------- Co-authored-by: Iwan Kawrakow <iwan.kawrakow@gmail.com> Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
2023-06-05 19:56:18 +00:00
const uint8_t * restrict q2 = x[i].qs;
const int8_t * restrict q8 = y[i].qs;
const __m128i mins_and_scales = _mm_loadu_si128((const __m128i*)x[i].scales);
const __m128i scales8 = _mm_and_si128(mins_and_scales, m4);
const __m128i mins8 = _mm_and_si128(_mm_srli_epi16(mins_and_scales, 4), m4);
const __m256i mins = _mm256_cvtepi8_epi16(mins8);
const __m256i prod = _mm256_madd_epi16(mins, _mm256_loadu_si256((const __m256i*)y[i].bsums));
acc = _mm256_fmadd_ps(_mm256_broadcast_ss(&dmin), _mm256_cvtepi32_ps(prod), acc);
const __m256i all_scales = _mm256_cvtepi8_epi16(scales8);
const __m128i l_scales = _mm256_extracti128_si256(all_scales, 0);
const __m128i h_scales = _mm256_extracti128_si256(all_scales, 1);
ggml : workaround for missing _mm256_setr_m128i in GCC < 8 in k_quants.c (#2405)
2023-07-28 18:17:45 +00:00
const __m256i scales[2] = {MM256_SET_M128I(l_scales, l_scales), MM256_SET_M128I(h_scales, h_scales)};
ggml : add SOTA 2,3,4,5,6 bit k-quantizations (#1684) * Starting to add k-quantization to ggml I think it is better to have quantization separate from ggml. For now just adding the k-quants there, but it would be better to also factor out the existing ggml quantizations. * Adding Q3_K and Q8_K (de)-quantization * Q3_K now working on CUDA and AVX2/scalar CUDA is not ideal - ~50% slower than Q4_0 for single token prediction, about the same in batch mode (perplexity). CPU single token is ~55 ms (on Ryzen 7950X). * Some improvement for Q3_K on CUDA It is now ~22.5 ms/token on my GPU, so ~30% slower than Q4_0. * Some more CUDA optimizations for Q3_K Single token is now 20.5 ms/token (~20% slower than Q4_0). Perplexity is on par with Q4_0. * Adding Q4_K - scalar, AVX2, CUDA Performance is the same or perhaps very slightly better than Q4_0 on the CPU. On the GPU, single token prediction is ~10% better than Q4_0, batch mode (perplexity is about the same). * Adding Q6_K - scalar, AVX2, CUDA Performance is ~40% lower compared to Q4_K on the CPU. This is to be expected, considering that we are memory bound on the CPU and the 6-bit model is ~44% larger than the 4-bit. On the GPU, single token prediction is ~6% lower than Q4_0, batch mode (perplexity) is even closer (but still slower). * Adding Q5_K - scalar, AVX2, CUDA Performance is ~20% lower compared to Q4_K on the CPU. This is to be expected, considering that we are memory bound on the CPU and the 5-bit model is ~22% larger than the 4-bit. On the GPU, single token prediction is about the same as Q4_0 for both, single token and batch prediction. * Per convention, all QX_K quantizations use Q5_K for output.weight * Adding quantization mixes * Quantization mixes: didn't quite get what I wanted in the last commit * Q4_K dot product for ARM_NEON * Q6_K dot product for ARM_NEON * Q5_K dot product for ARM_NEON * Adding Q3_K dot for ARM_NEON It is 22% slower than Q4_K, despite the smaller model size. On x86_64, where we are memory bound, the Q3_K model is quite a bit faster than Q4_K. * A very slightly faster ARM_NEON Q3_K dot * Adding Q2_K - just CUDA for now Token prediction is pretty good - about 15.5 ms on a RTX 4080. Perplexity is about the same as Q4_K. * Adding scalar and AVX2 Q2_K dot * Adding ARM_NEON Q2_K dot About the same performance as Q4_K. * A slightly faster ARM_NEON Q2_K dot Single token prediction is now ~36 ms on M2 Max. The code is much simpler too. * Fixed bug in Q2_K CUDA dot product kernel Stranegly enough, for the few prompts I tried with the 7B model the responses looked perfectly reasonable. Only realized something is not quite right when I tried the larger models and started getting nonse back. In any case, Q2_K single token evaluation time on an RTX 4080 in a Ryzen7950X box iusing CUDA and model fully loaded on the GPU are ~15.5 ms for 7B, ~25.4 ms for 13B, and ~55.8 ms for 30B. The max number of layers that fit in VRAM for The 65B is 32. With that, we get ~330 ms per token, which is not that much faster than just running on the CPU (~470 ms per token). * Don't print zeros/NaNs when no count histogram has been collected * A 10% faster CUDA vector dot kernel for Q3_K Q3_K is now running at ~18.5 ms / token on CUDA, so the gap to Q4_0 is only 10%. It seems memory acccess pattern is more important for performance than the amount of computation the kernel does. * A slightly daster Q4_K AVX2 dot product For perplexity, where we are less memory bound, time per pass drops by ~5%. Barely measurable difference for single token prediction. * A slightly faster ARM_NEON A4_K dot product * Minor * Fix quantization error test We cannot possibly be expecting rmse < 0.002 for 2- and 3-bit quantization variants. * Fix docker build I have been sloppy with vector reinterpret casts on ARM_NEON. It seems clang is very forgiving in that regard. * Added forgotten ggml.o dependence on k_quants.h to the Makefile * Had unintentionally committed the Makefile with -Ofast enabled * ggml : rename k_quants -> ggml-quants-k, use lowercase in code --------- Co-authored-by: Iwan Kawrakow <iwan.kawrakow@gmail.com> Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
2023-06-05 19:56:18 +00:00
__m256i sumi = _mm256_setzero_si256();
for (int j = 0; j < QK_K/128; ++j) {
const __m256i q2bits = _mm256_loadu_si256((const __m256i*)q2); q2 += 32;
const __m256i q8_0 = _mm256_loadu_si256((const __m256i*)q8); q8 += 32;
const __m256i q8_1 = _mm256_loadu_si256((const __m256i*)q8); q8 += 32;
const __m256i q8_2 = _mm256_loadu_si256((const __m256i*)q8); q8 += 32;
const __m256i q8_3 = _mm256_loadu_si256((const __m256i*)q8); q8 += 32;
const __m256i q2_0 = _mm256_and_si256(q2bits, m3);
const __m256i q2_1 = _mm256_and_si256(_mm256_srli_epi16(q2bits, 2), m3);
const __m256i q2_2 = _mm256_and_si256(_mm256_srli_epi16(q2bits, 4), m3);
const __m256i q2_3 = _mm256_and_si256(_mm256_srli_epi16(q2bits, 6), m3);
__m256i p0 = _mm256_maddubs_epi16(q2_0, q8_0);
__m256i p1 = _mm256_maddubs_epi16(q2_1, q8_1);
__m256i p2 = _mm256_maddubs_epi16(q2_2, q8_2);
__m256i p3 = _mm256_maddubs_epi16(q2_3, q8_3);
p0 = _mm256_madd_epi16(_mm256_shuffle_epi8(scales[j], get_scale_shuffle_q3k(0)), p0);
p1 = _mm256_madd_epi16(_mm256_shuffle_epi8(scales[j], get_scale_shuffle_q3k(1)), p1);
p2 = _mm256_madd_epi16(_mm256_shuffle_epi8(scales[j], get_scale_shuffle_q3k(2)), p2);
p3 = _mm256_madd_epi16(_mm256_shuffle_epi8(scales[j], get_scale_shuffle_q3k(3)), p3);
p0 = _mm256_add_epi32(p0, p1);
p2 = _mm256_add_epi32(p2, p3);
sumi = _mm256_add_epi32(sumi, _mm256_add_epi32(p0, p2));
}
acc = _mm256_fmadd_ps(_mm256_broadcast_ss(&d), _mm256_cvtepi32_ps(sumi), acc);
}
*s = hsum_float_8(acc);
k-quants : add AVX support to dot functions (#1916) * k_quants : add AVX support * k_quants : apply review comments
2023-06-26 16:46:07 +00:00
#elif defined __AVX__
const __m128i m3 = _mm_set1_epi8(0x3);
const __m128i m4 = _mm_set1_epi8(0xF);
const __m128i m2 = _mm_set1_epi8(0x2);
__m256 acc = _mm256_setzero_ps();
for (int i = 0; i < nb; ++i) {
ggml : move FP16 <-> FP32 code to ggml-impl.h (#3861) * ggml : move FP16 <-> FP32 stuff to ggml-impl.h ggml-ci * tests : fix ARM build * ggml : explicitly initialize deprecated type traits * ggml : add math.h to ggml-impl.h * ggml : remove duplicate static assert macros * ggml : prefix lookup tables with ggml_ ggml-ci * ggml-impl : move extern "C" to start of file
2023-10-30 17:19:15 +00:00
const float dall = y[i].d * GGML_FP16_TO_FP32(x[i].d);
const float dmin = -y[i].d * GGML_FP16_TO_FP32(x[i].dmin);
k-quants : add AVX support to dot functions (#1916) * k_quants : add AVX support * k_quants : apply review comments
2023-06-26 16:46:07 +00:00
const uint8_t * restrict q2 = x[i].qs;
const int8_t * restrict q8 = y[i].qs;
// load mins and scales from block_q2_K.scales[QK_K/16]
const __m128i mins_and_scales = _mm_loadu_si128((const __m128i*)x[i].scales);
const __m128i scales16 = _mm_and_si128(mins_and_scales, m4);
const __m128i mins16 = _mm_and_si128(_mm_srli_epi16(mins_and_scales, 4), m4);
const __m128i mins_0 = _mm_cvtepi8_epi16(mins16);
const __m128i mins_1 = _mm_cvtepi8_epi16(_mm_unpackhi_epi64(mins16, mins16));
// summs = y[i].bsums * (x[i].scales >> 4) in 16bits*8*2 to 32bits*4*2
const __m128i summs_0 = _mm_madd_epi16(mins_0, _mm_loadu_si128((const __m128i*)&y[i].bsums[0]));
const __m128i summs_1 = _mm_madd_epi16(mins_1, _mm_loadu_si128((const __m128i*)&y[i].bsums[8]));
// sumf += -dmin * summs in 32bits*8
ggml : workaround for missing _mm256_setr_m128i in GCC < 8 in k_quants.c (#2405)
2023-07-28 18:17:45 +00:00
acc = _mm256_add_ps(_mm256_mul_ps(_mm256_broadcast_ss(&dmin), _mm256_cvtepi32_ps(MM256_SET_M128I(summs_1, summs_0))), acc);
k-quants : add AVX support to dot functions (#1916) * k_quants : add AVX support * k_quants : apply review comments
2023-06-26 16:46:07 +00:00
const __m128i scales_0 = _mm_cvtepi8_epi16(scales16);
const __m128i scales_1 = _mm_cvtepi8_epi16(_mm_unpackhi_epi64(scales16, scales16));
const __m128i scales[2] = { scales_0, scales_1 };
__m128i sumi_0 = _mm_setzero_si128();
__m128i sumi_1 = _mm_setzero_si128();
for (int j = 0; j < QK_K/128; ++j) {
// load Q8 quants int8*16*8 from block_q8_K.qs[QK_K]
const __m128i q8_0 = _mm_loadu_si128((const __m128i*)q8); q8 += 16;
const __m128i q8_1 = _mm_loadu_si128((const __m128i*)q8); q8 += 16;
const __m128i q8_2 = _mm_loadu_si128((const __m128i*)q8); q8 += 16;
const __m128i q8_3 = _mm_loadu_si128((const __m128i*)q8); q8 += 16;
const __m128i q8_4 = _mm_loadu_si128((const __m128i*)q8); q8 += 16;
const __m128i q8_5 = _mm_loadu_si128((const __m128i*)q8); q8 += 16;
const __m128i q8_6 = _mm_loadu_si128((const __m128i*)q8); q8 += 16;
const __m128i q8_7 = _mm_loadu_si128((const __m128i*)q8); q8 += 16;
// load 2bits*16*8 from block_q2_K.qs[QK_K/4]
__m128i q2bits = _mm_loadu_si128((const __m128i*)q2); q2 += 16;
const __m128i q2_0 = _mm_and_si128(q2bits, m3);
const __m128i q2_2 = _mm_and_si128(_mm_srli_epi16(q2bits, 2), m3);
const __m128i q2_4 = _mm_and_si128(_mm_srli_epi16(q2bits, 4), m3);
const __m128i q2_6 = _mm_and_si128(_mm_srli_epi16(q2bits, 6), m3);
q2bits = _mm_loadu_si128((const __m128i*)q2); q2 += 16;
const __m128i q2_1 = _mm_and_si128(q2bits, m3);
const __m128i q2_3 = _mm_and_si128(_mm_srli_epi16(q2bits, 2), m3);
const __m128i q2_5 = _mm_and_si128(_mm_srli_epi16(q2bits, 4), m3);
const __m128i q2_7 = _mm_and_si128(_mm_srli_epi16(q2bits, 6), m3);
// isuml = q8[l] * ((q2[l] >> shift) & 3) in 8bits*16*8 to 16bits*8*8
__m128i p0 = _mm_maddubs_epi16(q2_0, q8_0);
__m128i p1 = _mm_maddubs_epi16(q2_1, q8_1);
__m128i p2 = _mm_maddubs_epi16(q2_2, q8_2);
__m128i p3 = _mm_maddubs_epi16(q2_3, q8_3);
__m128i p4 = _mm_maddubs_epi16(q2_4, q8_4);
__m128i p5 = _mm_maddubs_epi16(q2_5, q8_5);
__m128i p6 = _mm_maddubs_epi16(q2_6, q8_6);
__m128i p7 = _mm_maddubs_epi16(q2_7, q8_7);
// isum += (x[i].scales[is++] & 0xF) * isuml in 16bits*8*8 to 32bits*4*8
__m128i shuffle = _mm_set1_epi16(0x0100);
p0 = _mm_madd_epi16(_mm_shuffle_epi8(scales[j], shuffle), p0);
shuffle = _mm_add_epi16(shuffle, m2);
p1 = _mm_madd_epi16(_mm_shuffle_epi8(scales[j], shuffle), p1);
shuffle = _mm_add_epi16(shuffle, m2);
p2 = _mm_madd_epi16(_mm_shuffle_epi8(scales[j], shuffle), p2);
shuffle = _mm_add_epi16(shuffle, m2);
p3 = _mm_madd_epi16(_mm_shuffle_epi8(scales[j], shuffle), p3);
shuffle = _mm_add_epi16(shuffle, m2);
p4 = _mm_madd_epi16(_mm_shuffle_epi8(scales[j], shuffle), p4);
shuffle = _mm_add_epi16(shuffle, m2);
p5 = _mm_madd_epi16(_mm_shuffle_epi8(scales[j], shuffle), p5);
shuffle = _mm_add_epi16(shuffle, m2);
p6 = _mm_madd_epi16(_mm_shuffle_epi8(scales[j], shuffle), p6);
shuffle = _mm_add_epi16(shuffle, m2);
p7 = _mm_madd_epi16(_mm_shuffle_epi8(scales[j], shuffle), p7);
p0 = _mm_add_epi32(p0, p1);
p2 = _mm_add_epi32(p2, p3);
p4 = _mm_add_epi32(p4, p5);
p6 = _mm_add_epi32(p6, p7);
// isum in 32bits*4*2
sumi_0 = _mm_add_epi32(sumi_0, _mm_add_epi32(p0, p2));
sumi_1 = _mm_add_epi32(sumi_1, _mm_add_epi32(p4, p6));
}
// sumf += dall * isum - dmin * summs in 32bits
ggml : workaround for missing _mm256_setr_m128i in GCC < 8 in k_quants.c (#2405)
2023-07-28 18:17:45 +00:00
__m256i sumi = MM256_SET_M128I(sumi_1, sumi_0);
k-quants : add AVX support to dot functions (#1916) * k_quants : add AVX support * k_quants : apply review comments
2023-06-26 16:46:07 +00:00
acc = _mm256_add_ps(_mm256_mul_ps(_mm256_broadcast_ss(&dall), _mm256_cvtepi32_ps(sumi)), acc);
}
*s = hsum_float_8(acc);
ggml : add RISC-V Vector Support for K-Quants and improved the existing intrinsics (#3453) * Added RVV intrinsics support for Q8 quantize row and also improved the existing dot product function for risc-v. The RVV intrinsics is added for the following quantize row functions quantize_row_q8_0 quantize_row_q8_1 The following dot product functions have also been optimized by using LMUL = 1/2 instead of LMUL = 1 ggml_vec_dot_q4_0_q8_0 ggml_vec_dot_q4_1_q8_1 ggml_vec_dot_q5_0_q8_0 ggml_vec_dot_q5_1_q8_1 And vector initialization in Q5 by temporary array is also replaced by the vid intrinsics Signed-off-by: Ahmad Tameem <ahmad.tameem@10xengineers.ai> * Added RVV intrinsics support for k_quants This adds RISC-V Vector intrinsics support for the following K_quants functions for both QKK = 256 and QKK = 64 ggml_vec_dot_q2_K_q8_K ggml_vec_dot_q3_K_q8_K ggml_vec_dot_q4_K_q8_K ggml_vec_dot_q5_K_q8_K ggml_vec_dot_q6_K_q8_K Signed-off-by: Ahmad Tameem <ahmad.tameem@10xengineers.ai> --------- Signed-off-by: Ahmad Tameem <ahmad.tameem@10xengineers.ai>
2023-10-03 18:38:19 +00:00
#elif defined __riscv_v_intrinsic
float sumf = 0;
uint8_t temp_01[32] = {0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1};
for (int i = 0; i < nb; ++i) {
const uint8_t * q2 = x[i].qs;
const int8_t * q8 = y[i].qs;
const uint8_t * sc = x[i].scales;
ggml : move FP16 <-> FP32 code to ggml-impl.h (#3861) * ggml : move FP16 <-> FP32 stuff to ggml-impl.h ggml-ci * tests : fix ARM build * ggml : explicitly initialize deprecated type traits * ggml : add math.h to ggml-impl.h * ggml : remove duplicate static assert macros * ggml : prefix lookup tables with ggml_ ggml-ci * ggml-impl : move extern "C" to start of file
2023-10-30 17:19:15 +00:00
const float dall = y[i].d * GGML_FP16_TO_FP32(x[i].d);
const float dmin = -y[i].d * GGML_FP16_TO_FP32(x[i].dmin);
ggml : add RISC-V Vector Support for K-Quants and improved the existing intrinsics (#3453) * Added RVV intrinsics support for Q8 quantize row and also improved the existing dot product function for risc-v. The RVV intrinsics is added for the following quantize row functions quantize_row_q8_0 quantize_row_q8_1 The following dot product functions have also been optimized by using LMUL = 1/2 instead of LMUL = 1 ggml_vec_dot_q4_0_q8_0 ggml_vec_dot_q4_1_q8_1 ggml_vec_dot_q5_0_q8_0 ggml_vec_dot_q5_1_q8_1 And vector initialization in Q5 by temporary array is also replaced by the vid intrinsics Signed-off-by: Ahmad Tameem <ahmad.tameem@10xengineers.ai> * Added RVV intrinsics support for k_quants This adds RISC-V Vector intrinsics support for the following K_quants functions for both QKK = 256 and QKK = 64 ggml_vec_dot_q2_K_q8_K ggml_vec_dot_q3_K_q8_K ggml_vec_dot_q4_K_q8_K ggml_vec_dot_q5_K_q8_K ggml_vec_dot_q6_K_q8_K Signed-off-by: Ahmad Tameem <ahmad.tameem@10xengineers.ai> --------- Signed-off-by: Ahmad Tameem <ahmad.tameem@10xengineers.ai>
2023-10-03 18:38:19 +00:00
size_t vl = 16;
vuint8m1_t scales = __riscv_vle8_v_u8m1(sc, vl);
vuint8m1_t aux = __riscv_vand_vx_u8m1(scales, 0x0F, vl);
vint16m1_t q8sums = __riscv_vle16_v_i16m1(y[i].bsums, vl);
vuint8mf2_t scales_2 = __riscv_vle8_v_u8mf2(sc, vl);
vuint8mf2_t mins8 = __riscv_vsrl_vx_u8mf2(scales_2, 0x4, vl);
vint16m1_t mins = __riscv_vreinterpret_v_u16m1_i16m1(__riscv_vzext_vf2_u16m1(mins8, vl));
vint32m2_t prod = __riscv_vwmul_vv_i32m2(q8sums, mins, vl);
vint32m1_t vsums = __riscv_vredsum_vs_i32m2_i32m1(prod, __riscv_vmv_v_x_i32m1(0, 1), vl);
sumf += dmin * __riscv_vmv_x_s_i32m1_i32(vsums);
vl = 32;
vint32m1_t vzero = __riscv_vmv_v_x_i32m1(0, 1);
vuint8m1_t v_b = __riscv_vle8_v_u8m1(temp_01, vl);
uint8_t is=0;
int isum=0;
for (int j = 0; j < QK_K/128; ++j) {
// load Q2
vuint8m1_t q2_x = __riscv_vle8_v_u8m1(q2, vl);
vuint8m1_t q2_0 = __riscv_vand_vx_u8m1(q2_x, 0x03, vl);
vuint8m1_t q2_1 = __riscv_vand_vx_u8m1(__riscv_vsrl_vx_u8m1(q2_x, 0x2, vl), 0x03 , vl);
vuint8m1_t q2_2 = __riscv_vand_vx_u8m1(__riscv_vsrl_vx_u8m1(q2_x, 0x4, vl), 0x03 , vl);
vuint8m1_t q2_3 = __riscv_vand_vx_u8m1(__riscv_vsrl_vx_u8m1(q2_x, 0x6, vl), 0x03 , vl);
// duplicate scale elements for product
vuint8m1_t sc0 = __riscv_vrgather_vv_u8m1(aux, __riscv_vadd_vx_u8m1(v_b, 0+is, vl), vl);
vuint8m1_t sc1 = __riscv_vrgather_vv_u8m1(aux, __riscv_vadd_vx_u8m1(v_b, 2+is, vl), vl);
vuint8m1_t sc2 = __riscv_vrgather_vv_u8m1(aux, __riscv_vadd_vx_u8m1(v_b, 4+is, vl), vl);
vuint8m1_t sc3 = __riscv_vrgather_vv_u8m1(aux, __riscv_vadd_vx_u8m1(v_b, 6+is, vl), vl);
vint16m2_t p0 = __riscv_vreinterpret_v_u16m2_i16m2(__riscv_vwmulu_vv_u16m2(q2_0, sc0, vl));
vint16m2_t p1 = __riscv_vreinterpret_v_u16m2_i16m2(__riscv_vwmulu_vv_u16m2(q2_1, sc1, vl));
vint16m2_t p2 = __riscv_vreinterpret_v_u16m2_i16m2(__riscv_vwmulu_vv_u16m2(q2_2, sc2, vl));
vint16m2_t p3 = __riscv_vreinterpret_v_u16m2_i16m2(__riscv_vwmulu_vv_u16m2(q2_3, sc3, vl));
// load Q8
vint8m1_t q8_0 = __riscv_vle8_v_i8m1(q8, vl);
vint8m1_t q8_1 = __riscv_vle8_v_i8m1(q8+32, vl);
vint8m1_t q8_2 = __riscv_vle8_v_i8m1(q8+64, vl);
vint8m1_t q8_3 = __riscv_vle8_v_i8m1(q8+96, vl);
vint32m4_t s0 = __riscv_vwmul_vv_i32m4(p0, __riscv_vwcvt_x_x_v_i16m2(q8_0, vl), vl);
vint32m4_t s1 = __riscv_vwmul_vv_i32m4(p1, __riscv_vwcvt_x_x_v_i16m2(q8_1, vl), vl);
vint32m4_t s2 = __riscv_vwmul_vv_i32m4(p2, __riscv_vwcvt_x_x_v_i16m2(q8_2, vl), vl);
vint32m4_t s3 = __riscv_vwmul_vv_i32m4(p3, __riscv_vwcvt_x_x_v_i16m2(q8_3, vl), vl);
vint32m1_t isum0 = __riscv_vredsum_vs_i32m4_i32m1(__riscv_vadd_vv_i32m4(s0, s1, vl), vzero, vl);
vint32m1_t isum1 = __riscv_vredsum_vs_i32m4_i32m1(__riscv_vadd_vv_i32m4(s2, s3, vl), isum0, vl);
isum += __riscv_vmv_x_s_i32m1_i32(isum1);
q2+=32; q8+=128; is=8;
}
sumf += dall * isum;
}
*s = sumf;
ggml : add SOTA 2,3,4,5,6 bit k-quantizations (#1684) * Starting to add k-quantization to ggml I think it is better to have quantization separate from ggml. For now just adding the k-quants there, but it would be better to also factor out the existing ggml quantizations. * Adding Q3_K and Q8_K (de)-quantization * Q3_K now working on CUDA and AVX2/scalar CUDA is not ideal - ~50% slower than Q4_0 for single token prediction, about the same in batch mode (perplexity). CPU single token is ~55 ms (on Ryzen 7950X). * Some improvement for Q3_K on CUDA It is now ~22.5 ms/token on my GPU, so ~30% slower than Q4_0. * Some more CUDA optimizations for Q3_K Single token is now 20.5 ms/token (~20% slower than Q4_0). Perplexity is on par with Q4_0. * Adding Q4_K - scalar, AVX2, CUDA Performance is the same or perhaps very slightly better than Q4_0 on the CPU. On the GPU, single token prediction is ~10% better than Q4_0, batch mode (perplexity is about the same). * Adding Q6_K - scalar, AVX2, CUDA Performance is ~40% lower compared to Q4_K on the CPU. This is to be expected, considering that we are memory bound on the CPU and the 6-bit model is ~44% larger than the 4-bit. On the GPU, single token prediction is ~6% lower than Q4_0, batch mode (perplexity) is even closer (but still slower). * Adding Q5_K - scalar, AVX2, CUDA Performance is ~20% lower compared to Q4_K on the CPU. This is to be expected, considering that we are memory bound on the CPU and the 5-bit model is ~22% larger than the 4-bit. On the GPU, single token prediction is about the same as Q4_0 for both, single token and batch prediction. * Per convention, all QX_K quantizations use Q5_K for output.weight * Adding quantization mixes * Quantization mixes: didn't quite get what I wanted in the last commit * Q4_K dot product for ARM_NEON * Q6_K dot product for ARM_NEON * Q5_K dot product for ARM_NEON * Adding Q3_K dot for ARM_NEON It is 22% slower than Q4_K, despite the smaller model size. On x86_64, where we are memory bound, the Q3_K model is quite a bit faster than Q4_K. * A very slightly faster ARM_NEON Q3_K dot * Adding Q2_K - just CUDA for now Token prediction is pretty good - about 15.5 ms on a RTX 4080. Perplexity is about the same as Q4_K. * Adding scalar and AVX2 Q2_K dot * Adding ARM_NEON Q2_K dot About the same performance as Q4_K. * A slightly faster ARM_NEON Q2_K dot Single token prediction is now ~36 ms on M2 Max. The code is much simpler too. * Fixed bug in Q2_K CUDA dot product kernel Stranegly enough, for the few prompts I tried with the 7B model the responses looked perfectly reasonable. Only realized something is not quite right when I tried the larger models and started getting nonse back. In any case, Q2_K single token evaluation time on an RTX 4080 in a Ryzen7950X box iusing CUDA and model fully loaded on the GPU are ~15.5 ms for 7B, ~25.4 ms for 13B, and ~55.8 ms for 30B. The max number of layers that fit in VRAM for The 65B is 32. With that, we get ~330 ms per token, which is not that much faster than just running on the CPU (~470 ms per token). * Don't print zeros/NaNs when no count histogram has been collected * A 10% faster CUDA vector dot kernel for Q3_K Q3_K is now running at ~18.5 ms / token on CUDA, so the gap to Q4_0 is only 10%. It seems memory acccess pattern is more important for performance than the amount of computation the kernel does. * A slightly daster Q4_K AVX2 dot product For perplexity, where we are less memory bound, time per pass drops by ~5%. Barely measurable difference for single token prediction. * A slightly faster ARM_NEON A4_K dot product * Minor * Fix quantization error test We cannot possibly be expecting rmse < 0.002 for 2- and 3-bit quantization variants. * Fix docker build I have been sloppy with vector reinterpret casts on ARM_NEON. It seems clang is very forgiving in that regard. * Added forgotten ggml.o dependence on k_quants.h to the Makefile * Had unintentionally committed the Makefile with -Ofast enabled * ggml : rename k_quants -> ggml-quants-k, use lowercase in code --------- Co-authored-by: Iwan Kawrakow <iwan.kawrakow@gmail.com> Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
2023-06-05 19:56:18 +00:00
#else
float sumf = 0;
for (int i = 0; i < nb; ++i) {
const uint8_t * q2 = x[i].qs;
const int8_t * q8 = y[i].qs;
const uint8_t * sc = x[i].scales;
int summs = 0;
for (int j = 0; j < 16; ++j) {
summs += y[i].bsums[j] * (sc[j] >> 4);
}
ggml : move FP16 <-> FP32 code to ggml-impl.h (#3861) * ggml : move FP16 <-> FP32 stuff to ggml-impl.h ggml-ci * tests : fix ARM build * ggml : explicitly initialize deprecated type traits * ggml : add math.h to ggml-impl.h * ggml : remove duplicate static assert macros * ggml : prefix lookup tables with ggml_ ggml-ci * ggml-impl : move extern "C" to start of file
2023-10-30 17:19:15 +00:00
const float dall = y[i].d * GGML_FP16_TO_FP32(x[i].d);
const float dmin = y[i].d * GGML_FP16_TO_FP32(x[i].dmin);
ggml : add SOTA 2,3,4,5,6 bit k-quantizations (#1684) * Starting to add k-quantization to ggml I think it is better to have quantization separate from ggml. For now just adding the k-quants there, but it would be better to also factor out the existing ggml quantizations. * Adding Q3_K and Q8_K (de)-quantization * Q3_K now working on CUDA and AVX2/scalar CUDA is not ideal - ~50% slower than Q4_0 for single token prediction, about the same in batch mode (perplexity). CPU single token is ~55 ms (on Ryzen 7950X). * Some improvement for Q3_K on CUDA It is now ~22.5 ms/token on my GPU, so ~30% slower than Q4_0. * Some more CUDA optimizations for Q3_K Single token is now 20.5 ms/token (~20% slower than Q4_0). Perplexity is on par with Q4_0. * Adding Q4_K - scalar, AVX2, CUDA Performance is the same or perhaps very slightly better than Q4_0 on the CPU. On the GPU, single token prediction is ~10% better than Q4_0, batch mode (perplexity is about the same). * Adding Q6_K - scalar, AVX2, CUDA Performance is ~40% lower compared to Q4_K on the CPU. This is to be expected, considering that we are memory bound on the CPU and the 6-bit model is ~44% larger than the 4-bit. On the GPU, single token prediction is ~6% lower than Q4_0, batch mode (perplexity) is even closer (but still slower). * Adding Q5_K - scalar, AVX2, CUDA Performance is ~20% lower compared to Q4_K on the CPU. This is to be expected, considering that we are memory bound on the CPU and the 5-bit model is ~22% larger than the 4-bit. On the GPU, single token prediction is about the same as Q4_0 for both, single token and batch prediction. * Per convention, all QX_K quantizations use Q5_K for output.weight * Adding quantization mixes * Quantization mixes: didn't quite get what I wanted in the last commit * Q4_K dot product for ARM_NEON * Q6_K dot product for ARM_NEON * Q5_K dot product for ARM_NEON * Adding Q3_K dot for ARM_NEON It is 22% slower than Q4_K, despite the smaller model size. On x86_64, where we are memory bound, the Q3_K model is quite a bit faster than Q4_K. * A very slightly faster ARM_NEON Q3_K dot * Adding Q2_K - just CUDA for now Token prediction is pretty good - about 15.5 ms on a RTX 4080. Perplexity is about the same as Q4_K. * Adding scalar and AVX2 Q2_K dot * Adding ARM_NEON Q2_K dot About the same performance as Q4_K. * A slightly faster ARM_NEON Q2_K dot Single token prediction is now ~36 ms on M2 Max. The code is much simpler too. * Fixed bug in Q2_K CUDA dot product kernel Stranegly enough, for the few prompts I tried with the 7B model the responses looked perfectly reasonable. Only realized something is not quite right when I tried the larger models and started getting nonse back. In any case, Q2_K single token evaluation time on an RTX 4080 in a Ryzen7950X box iusing CUDA and model fully loaded on the GPU are ~15.5 ms for 7B, ~25.4 ms for 13B, and ~55.8 ms for 30B. The max number of layers that fit in VRAM for The 65B is 32. With that, we get ~330 ms per token, which is not that much faster than just running on the CPU (~470 ms per token). * Don't print zeros/NaNs when no count histogram has been collected * A 10% faster CUDA vector dot kernel for Q3_K Q3_K is now running at ~18.5 ms / token on CUDA, so the gap to Q4_0 is only 10%. It seems memory acccess pattern is more important for performance than the amount of computation the kernel does. * A slightly daster Q4_K AVX2 dot product For perplexity, where we are less memory bound, time per pass drops by ~5%. Barely measurable difference for single token prediction. * A slightly faster ARM_NEON A4_K dot product * Minor * Fix quantization error test We cannot possibly be expecting rmse < 0.002 for 2- and 3-bit quantization variants. * Fix docker build I have been sloppy with vector reinterpret casts on ARM_NEON. It seems clang is very forgiving in that regard. * Added forgotten ggml.o dependence on k_quants.h to the Makefile * Had unintentionally committed the Makefile with -Ofast enabled * ggml : rename k_quants -> ggml-quants-k, use lowercase in code --------- Co-authored-by: Iwan Kawrakow <iwan.kawrakow@gmail.com> Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
2023-06-05 19:56:18 +00:00
int isum = 0;
int is = 0;
int d;
for (int k = 0; k < QK_K/128; ++k) {
int shift = 0;
for (int j = 0; j < 4; ++j) {
d = sc[is++] & 0xF;
int isuml = 0;
for (int l = 0; l < 16; ++l) isuml += q8[l] * ((q2[l] >> shift) & 3);
isum += d * isuml;
d = sc[is++] & 0xF;
isuml = 0;
for (int l = 16; l < 32; ++l) isuml += q8[l] * ((q2[l] >> shift) & 3);
isum += d * isuml;
shift += 2;
q8 += 32;
}
q2 += 32;
}
sumf += dall * isum - dmin * summs;
}
*s = sumf;
#endif
}
k-quants : support for super-block size of 64 (#2001) * k_quants: WIP super-blocks with 64 weights * k_quants: WIP super-blocks with 64 weights Q6_K scalar and AVX2 works * k_quants: WIP super-blocks with 64 weights Q4_K scalar and AVX2 works * k_quants: WIP super-blocks with 64 weights Q2_K scalar and AVX2 works. Q2_K is way too slow (it is actually slower than the scalar implementation) * k_quants: WIP super-blocks with 64 weights Q3_K scalar and AVX2 works. * k_quants: WIP super-blocks with 64 weights Q5_K scalar and AVX2 works, and with that all k_quants are done on AVX2 and scalar * k_quants: WIP super-blocks with 64 weights Q6_K working on CUDA. Cannot make it run quite as gast as with super-blocks with 256 weigths: 8% slower on 4080, 20% slower on the 1660 (but there we fit 1 less layer on the GPU because pf the larger model size), so some fraction of these 20% is due to that, * k_quants: WIP super-blocks with 64 weights Q4_K working on CUDA. ~10% slower on GTX-1660, 16% slower on 4080. * k_quants: WIP super-blocks with 64 weights Q2_K working on CUDA. ~3% slower on GTX-1660, 10% slower on 4080. * k_quants: WIP super-blocks with 64 weights Q3_K working on CUDA. * k_quants: WIP super-blocks with 64 weights Q5_K working on CUDA, and with this CUDA is done. * k_quants: WIP super-blocks with 64 weights Q6_K working on ARM_NEON * k_quants: WIP super-blocks with 64 weights Q4_K working on ARM_NEON, but quite a bit slower than 256 weights * k_quants: WIP super-blocks with 64 weights Q2_K working on ARM_NEON, but quite a bit slower than 256 weights * k_quants: WIP super-blocks with 64 weights Q3_K working on ARM_NEON, but quite a bit slower than 256 weights. * k_quants: WIP super-blocks with 64 weights Q5_K working on ARM_NEON, but quite a bit slower than 256 weights. With that, we have full support for ARM_NEON, although performance is not quite there. * k_quants: WIP super-blocks with 64 weights Slightly more efficient Q3_K and Q5_K * k_quants: WIP super-blocks with 64 weights Another small improvement for Q3_K and Q5_K on ARM_NEON * k_quants: WIP super-blocks with 64 weights Yet another speedup for Q5_K on ARM_NEON. We are now within 10% of the QK_K = 256 version. * k_quants: WIP super-blocks with 64 weights * We are able to pass preprocessor macros to the Metal compiler * Q6_K works and is actually slightly more efficient than the QK_K = 256 version (25.2 ms vs 25.8 ms) * k_quants: WIP super-blocks with 64 weights Q4_K works on Metal and is actually slightly faster than QK_K = 256 (21.95 ms vs 24.0 ms). * k_quants: WIP super-blocks with 64 weights Q2_K works on Metal and is very slightly faster than QK_K = 256 (23.8 ms vs 24.2 ms). * k_quants: WIP super-blocks with 64 weights Q3_K works on Metal and is slightly faster than QK_K = 256 (26.6 ms vs 28.3 ms). * k_quants: WIP super-blocks with 64 weights Q5_K works on Metal and is slightly faster than QK_K = 256 (23.7 ms vs 26.3 ms). * k_quants: call them _K, not _k, also on Metal * k_quants: correctly define QK_K in llama.cpp * Fixed bug in q4_K quantization added with the 64-block addition * Simplify via lambda * k_quants: swicth Q3_K to 4-bit scales when QK_K = 64 Otherwise there isn't much benefit from this quantization type. There is some very slight loss in accuracy, but we reduce size by ~7%. E.g., for OpenLLaMA-3B, Q3_K_S perplexity is 8.6131 with 8-bit scales and 8.6352 with 4-bit, while file size decreases from 1.53G to 1.44G. * k_quants: switch Q4_K to 4-bit scales when QK_K = 64 Here the loss in accuracy is greater than for Q3_K, but the Q4_K points still move further to the left on the perplexity vs size curve. * k_quants: forgot to add the Metal changes in last commit * k_quants: change Q5_K to be type 0 when QK_K = 64 Still needs AVX2 implementation * k_quants: AVX2 implementation for new 64-weight Q5_K * k_quants: 10% faster ARM_NEON Q5_K dot product * k_quants: fixed issue caused by merging with master --------- Co-authored-by: Iwan Kawrakow <iwan.kawrakow@gmail.com>
2023-06-26 16:43:07 +00:00
#else
void ggml_vec_dot_q2_K_q8_K(const int n, float * restrict s, const void * restrict vx, const void * restrict vy) {
const block_q2_K * restrict x = vx;
const block_q8_K * restrict y = vy;
const int nb = n / QK_K;
#ifdef __ARM_NEON
const uint8x16_t m3 = vdupq_n_u8(0x3);
ggml : fix dot product for ARM (#4630) ggml-ci
2023-12-27 09:02:13 +00:00
const int32x4_t vzero = vdupq_n_s32(0);
k-quants : support for super-block size of 64 (#2001) * k_quants: WIP super-blocks with 64 weights * k_quants: WIP super-blocks with 64 weights Q6_K scalar and AVX2 works * k_quants: WIP super-blocks with 64 weights Q4_K scalar and AVX2 works * k_quants: WIP super-blocks with 64 weights Q2_K scalar and AVX2 works. Q2_K is way too slow (it is actually slower than the scalar implementation) * k_quants: WIP super-blocks with 64 weights Q3_K scalar and AVX2 works. * k_quants: WIP super-blocks with 64 weights Q5_K scalar and AVX2 works, and with that all k_quants are done on AVX2 and scalar * k_quants: WIP super-blocks with 64 weights Q6_K working on CUDA. Cannot make it run quite as gast as with super-blocks with 256 weigths: 8% slower on 4080, 20% slower on the 1660 (but there we fit 1 less layer on the GPU because pf the larger model size), so some fraction of these 20% is due to that, * k_quants: WIP super-blocks with 64 weights Q4_K working on CUDA. ~10% slower on GTX-1660, 16% slower on 4080. * k_quants: WIP super-blocks with 64 weights Q2_K working on CUDA. ~3% slower on GTX-1660, 10% slower on 4080. * k_quants: WIP super-blocks with 64 weights Q3_K working on CUDA. * k_quants: WIP super-blocks with 64 weights Q5_K working on CUDA, and with this CUDA is done. * k_quants: WIP super-blocks with 64 weights Q6_K working on ARM_NEON * k_quants: WIP super-blocks with 64 weights Q4_K working on ARM_NEON, but quite a bit slower than 256 weights * k_quants: WIP super-blocks with 64 weights Q2_K working on ARM_NEON, but quite a bit slower than 256 weights * k_quants: WIP super-blocks with 64 weights Q3_K working on ARM_NEON, but quite a bit slower than 256 weights. * k_quants: WIP super-blocks with 64 weights Q5_K working on ARM_NEON, but quite a bit slower than 256 weights. With that, we have full support for ARM_NEON, although performance is not quite there. * k_quants: WIP super-blocks with 64 weights Slightly more efficient Q3_K and Q5_K * k_quants: WIP super-blocks with 64 weights Another small improvement for Q3_K and Q5_K on ARM_NEON * k_quants: WIP super-blocks with 64 weights Yet another speedup for Q5_K on ARM_NEON. We are now within 10% of the QK_K = 256 version. * k_quants: WIP super-blocks with 64 weights * We are able to pass preprocessor macros to the Metal compiler * Q6_K works and is actually slightly more efficient than the QK_K = 256 version (25.2 ms vs 25.8 ms) * k_quants: WIP super-blocks with 64 weights Q4_K works on Metal and is actually slightly faster than QK_K = 256 (21.95 ms vs 24.0 ms). * k_quants: WIP super-blocks with 64 weights Q2_K works on Metal and is very slightly faster than QK_K = 256 (23.8 ms vs 24.2 ms). * k_quants: WIP super-blocks with 64 weights Q3_K works on Metal and is slightly faster than QK_K = 256 (26.6 ms vs 28.3 ms). * k_quants: WIP super-blocks with 64 weights Q5_K works on Metal and is slightly faster than QK_K = 256 (23.7 ms vs 26.3 ms). * k_quants: call them _K, not _k, also on Metal * k_quants: correctly define QK_K in llama.cpp * Fixed bug in q4_K quantization added with the 64-block addition * Simplify via lambda * k_quants: swicth Q3_K to 4-bit scales when QK_K = 64 Otherwise there isn't much benefit from this quantization type. There is some very slight loss in accuracy, but we reduce size by ~7%. E.g., for OpenLLaMA-3B, Q3_K_S perplexity is 8.6131 with 8-bit scales and 8.6352 with 4-bit, while file size decreases from 1.53G to 1.44G. * k_quants: switch Q4_K to 4-bit scales when QK_K = 64 Here the loss in accuracy is greater than for Q3_K, but the Q4_K points still move further to the left on the perplexity vs size curve. * k_quants: forgot to add the Metal changes in last commit * k_quants: change Q5_K to be type 0 when QK_K = 64 Still needs AVX2 implementation * k_quants: AVX2 implementation for new 64-weight Q5_K * k_quants: 10% faster ARM_NEON Q5_K dot product * k_quants: fixed issue caused by merging with master --------- Co-authored-by: Iwan Kawrakow <iwan.kawrakow@gmail.com>
2023-06-26 16:43:07 +00:00
ggml : sync (im2col, GPU conv, 32-bit arm compat) (#4060) ggml-ci
2023-11-13 14:55:52 +00:00
ggml_int8x16x4_t q2bytes;
k-quants : support for super-block size of 64 (#2001) * k_quants: WIP super-blocks with 64 weights * k_quants: WIP super-blocks with 64 weights Q6_K scalar and AVX2 works * k_quants: WIP super-blocks with 64 weights Q4_K scalar and AVX2 works * k_quants: WIP super-blocks with 64 weights Q2_K scalar and AVX2 works. Q2_K is way too slow (it is actually slower than the scalar implementation) * k_quants: WIP super-blocks with 64 weights Q3_K scalar and AVX2 works. * k_quants: WIP super-blocks with 64 weights Q5_K scalar and AVX2 works, and with that all k_quants are done on AVX2 and scalar * k_quants: WIP super-blocks with 64 weights Q6_K working on CUDA. Cannot make it run quite as gast as with super-blocks with 256 weigths: 8% slower on 4080, 20% slower on the 1660 (but there we fit 1 less layer on the GPU because pf the larger model size), so some fraction of these 20% is due to that, * k_quants: WIP super-blocks with 64 weights Q4_K working on CUDA. ~10% slower on GTX-1660, 16% slower on 4080. * k_quants: WIP super-blocks with 64 weights Q2_K working on CUDA. ~3% slower on GTX-1660, 10% slower on 4080. * k_quants: WIP super-blocks with 64 weights Q3_K working on CUDA. * k_quants: WIP super-blocks with 64 weights Q5_K working on CUDA, and with this CUDA is done. * k_quants: WIP super-blocks with 64 weights Q6_K working on ARM_NEON * k_quants: WIP super-blocks with 64 weights Q4_K working on ARM_NEON, but quite a bit slower than 256 weights * k_quants: WIP super-blocks with 64 weights Q2_K working on ARM_NEON, but quite a bit slower than 256 weights * k_quants: WIP super-blocks with 64 weights Q3_K working on ARM_NEON, but quite a bit slower than 256 weights. * k_quants: WIP super-blocks with 64 weights Q5_K working on ARM_NEON, but quite a bit slower than 256 weights. With that, we have full support for ARM_NEON, although performance is not quite there. * k_quants: WIP super-blocks with 64 weights Slightly more efficient Q3_K and Q5_K * k_quants: WIP super-blocks with 64 weights Another small improvement for Q3_K and Q5_K on ARM_NEON * k_quants: WIP super-blocks with 64 weights Yet another speedup for Q5_K on ARM_NEON. We are now within 10% of the QK_K = 256 version. * k_quants: WIP super-blocks with 64 weights * We are able to pass preprocessor macros to the Metal compiler * Q6_K works and is actually slightly more efficient than the QK_K = 256 version (25.2 ms vs 25.8 ms) * k_quants: WIP super-blocks with 64 weights Q4_K works on Metal and is actually slightly faster than QK_K = 256 (21.95 ms vs 24.0 ms). * k_quants: WIP super-blocks with 64 weights Q2_K works on Metal and is very slightly faster than QK_K = 256 (23.8 ms vs 24.2 ms). * k_quants: WIP super-blocks with 64 weights Q3_K works on Metal and is slightly faster than QK_K = 256 (26.6 ms vs 28.3 ms). * k_quants: WIP super-blocks with 64 weights Q5_K works on Metal and is slightly faster than QK_K = 256 (23.7 ms vs 26.3 ms). * k_quants: call them _K, not _k, also on Metal * k_quants: correctly define QK_K in llama.cpp * Fixed bug in q4_K quantization added with the 64-block addition * Simplify via lambda * k_quants: swicth Q3_K to 4-bit scales when QK_K = 64 Otherwise there isn't much benefit from this quantization type. There is some very slight loss in accuracy, but we reduce size by ~7%. E.g., for OpenLLaMA-3B, Q3_K_S perplexity is 8.6131 with 8-bit scales and 8.6352 with 4-bit, while file size decreases from 1.53G to 1.44G. * k_quants: switch Q4_K to 4-bit scales when QK_K = 64 Here the loss in accuracy is greater than for Q3_K, but the Q4_K points still move further to the left on the perplexity vs size curve. * k_quants: forgot to add the Metal changes in last commit * k_quants: change Q5_K to be type 0 when QK_K = 64 Still needs AVX2 implementation * k_quants: AVX2 implementation for new 64-weight Q5_K * k_quants: 10% faster ARM_NEON Q5_K dot product * k_quants: fixed issue caused by merging with master --------- Co-authored-by: Iwan Kawrakow <iwan.kawrakow@gmail.com>
2023-06-26 16:43:07 +00:00
uint32_t aux32[2];
const uint8_t * scales = (const uint8_t *)aux32;
float sum = 0;
for (int i = 0; i < nb; ++i) {
const float d = y[i].d * (float)x[i].d;
const float dmin = -y[i].d * (float)x[i].dmin;
const uint8_t * restrict q2 = x[i].qs;
const int8_t * restrict q8 = y[i].qs;
const uint32_t * restrict sc = (const uint32_t *)x[i].scales;
aux32[0] = sc[0] & 0x0f0f0f0f;
aux32[1] = (sc[0] >> 4) & 0x0f0f0f0f;
sum += dmin * (scales[4] * y[i].bsums[0] + scales[5] * y[i].bsums[1] + scales[6] * y[i].bsums[2] + scales[7] * y[i].bsums[3]);
int isum1 = 0, isum2 = 0;
const uint8x16_t q2bits = vld1q_u8(q2);
ggml : sync (im2col, GPU conv, 32-bit arm compat) (#4060) ggml-ci
2023-11-13 14:55:52 +00:00
const ggml_int8x16x4_t q8bytes = ggml_vld1q_s8_x4(q8);
k-quants : support for super-block size of 64 (#2001) * k_quants: WIP super-blocks with 64 weights * k_quants: WIP super-blocks with 64 weights Q6_K scalar and AVX2 works * k_quants: WIP super-blocks with 64 weights Q4_K scalar and AVX2 works * k_quants: WIP super-blocks with 64 weights Q2_K scalar and AVX2 works. Q2_K is way too slow (it is actually slower than the scalar implementation) * k_quants: WIP super-blocks with 64 weights Q3_K scalar and AVX2 works. * k_quants: WIP super-blocks with 64 weights Q5_K scalar and AVX2 works, and with that all k_quants are done on AVX2 and scalar * k_quants: WIP super-blocks with 64 weights Q6_K working on CUDA. Cannot make it run quite as gast as with super-blocks with 256 weigths: 8% slower on 4080, 20% slower on the 1660 (but there we fit 1 less layer on the GPU because pf the larger model size), so some fraction of these 20% is due to that, * k_quants: WIP super-blocks with 64 weights Q4_K working on CUDA. ~10% slower on GTX-1660, 16% slower on 4080. * k_quants: WIP super-blocks with 64 weights Q2_K working on CUDA. ~3% slower on GTX-1660, 10% slower on 4080. * k_quants: WIP super-blocks with 64 weights Q3_K working on CUDA. * k_quants: WIP super-blocks with 64 weights Q5_K working on CUDA, and with this CUDA is done. * k_quants: WIP super-blocks with 64 weights Q6_K working on ARM_NEON * k_quants: WIP super-blocks with 64 weights Q4_K working on ARM_NEON, but quite a bit slower than 256 weights * k_quants: WIP super-blocks with 64 weights Q2_K working on ARM_NEON, but quite a bit slower than 256 weights * k_quants: WIP super-blocks with 64 weights Q3_K working on ARM_NEON, but quite a bit slower than 256 weights. * k_quants: WIP super-blocks with 64 weights Q5_K working on ARM_NEON, but quite a bit slower than 256 weights. With that, we have full support for ARM_NEON, although performance is not quite there. * k_quants: WIP super-blocks with 64 weights Slightly more efficient Q3_K and Q5_K * k_quants: WIP super-blocks with 64 weights Another small improvement for Q3_K and Q5_K on ARM_NEON * k_quants: WIP super-blocks with 64 weights Yet another speedup for Q5_K on ARM_NEON. We are now within 10% of the QK_K = 256 version. * k_quants: WIP super-blocks with 64 weights * We are able to pass preprocessor macros to the Metal compiler * Q6_K works and is actually slightly more efficient than the QK_K = 256 version (25.2 ms vs 25.8 ms) * k_quants: WIP super-blocks with 64 weights Q4_K works on Metal and is actually slightly faster than QK_K = 256 (21.95 ms vs 24.0 ms). * k_quants: WIP super-blocks with 64 weights Q2_K works on Metal and is very slightly faster than QK_K = 256 (23.8 ms vs 24.2 ms). * k_quants: WIP super-blocks with 64 weights Q3_K works on Metal and is slightly faster than QK_K = 256 (26.6 ms vs 28.3 ms). * k_quants: WIP super-blocks with 64 weights Q5_K works on Metal and is slightly faster than QK_K = 256 (23.7 ms vs 26.3 ms). * k_quants: call them _K, not _k, also on Metal * k_quants: correctly define QK_K in llama.cpp * Fixed bug in q4_K quantization added with the 64-block addition * Simplify via lambda * k_quants: swicth Q3_K to 4-bit scales when QK_K = 64 Otherwise there isn't much benefit from this quantization type. There is some very slight loss in accuracy, but we reduce size by ~7%. E.g., for OpenLLaMA-3B, Q3_K_S perplexity is 8.6131 with 8-bit scales and 8.6352 with 4-bit, while file size decreases from 1.53G to 1.44G. * k_quants: switch Q4_K to 4-bit scales when QK_K = 64 Here the loss in accuracy is greater than for Q3_K, but the Q4_K points still move further to the left on the perplexity vs size curve. * k_quants: forgot to add the Metal changes in last commit * k_quants: change Q5_K to be type 0 when QK_K = 64 Still needs AVX2 implementation * k_quants: AVX2 implementation for new 64-weight Q5_K * k_quants: 10% faster ARM_NEON Q5_K dot product * k_quants: fixed issue caused by merging with master --------- Co-authored-by: Iwan Kawrakow <iwan.kawrakow@gmail.com>
2023-06-26 16:43:07 +00:00
q2bytes.val[0] = vreinterpretq_s8_u8(vandq_u8(q2bits, m3));
q2bytes.val[1] = vreinterpretq_s8_u8(vandq_u8(vshrq_n_u8(q2bits, 2), m3));
q2bytes.val[2] = vreinterpretq_s8_u8(vandq_u8(vshrq_n_u8(q2bits, 4), m3));
q2bytes.val[3] = vreinterpretq_s8_u8(vandq_u8(vshrq_n_u8(q2bits, 6), m3));
ggml : add ggml_vdotq_s32 alias (#4715) ggml-ci
2023-12-31 09:43:31 +00:00
isum1 += vaddvq_s32(ggml_vdotq_s32(vzero, q2bytes.val[0], q8bytes.val[0])) * scales[0];
isum2 += vaddvq_s32(ggml_vdotq_s32(vzero, q2bytes.val[1], q8bytes.val[1])) * scales[1];
isum1 += vaddvq_s32(ggml_vdotq_s32(vzero, q2bytes.val[2], q8bytes.val[2])) * scales[2];
isum2 += vaddvq_s32(ggml_vdotq_s32(vzero, q2bytes.val[3], q8bytes.val[3])) * scales[3];
k-quants : support for super-block size of 64 (#2001) * k_quants: WIP super-blocks with 64 weights * k_quants: WIP super-blocks with 64 weights Q6_K scalar and AVX2 works * k_quants: WIP super-blocks with 64 weights Q4_K scalar and AVX2 works * k_quants: WIP super-blocks with 64 weights Q2_K scalar and AVX2 works. Q2_K is way too slow (it is actually slower than the scalar implementation) * k_quants: WIP super-blocks with 64 weights Q3_K scalar and AVX2 works. * k_quants: WIP super-blocks with 64 weights Q5_K scalar and AVX2 works, and with that all k_quants are done on AVX2 and scalar * k_quants: WIP super-blocks with 64 weights Q6_K working on CUDA. Cannot make it run quite as gast as with super-blocks with 256 weigths: 8% slower on 4080, 20% slower on the 1660 (but there we fit 1 less layer on the GPU because pf the larger model size), so some fraction of these 20% is due to that, * k_quants: WIP super-blocks with 64 weights Q4_K working on CUDA. ~10% slower on GTX-1660, 16% slower on 4080. * k_quants: WIP super-blocks with 64 weights Q2_K working on CUDA. ~3% slower on GTX-1660, 10% slower on 4080. * k_quants: WIP super-blocks with 64 weights Q3_K working on CUDA. * k_quants: WIP super-blocks with 64 weights Q5_K working on CUDA, and with this CUDA is done. * k_quants: WIP super-blocks with 64 weights Q6_K working on ARM_NEON * k_quants: WIP super-blocks with 64 weights Q4_K working on ARM_NEON, but quite a bit slower than 256 weights * k_quants: WIP super-blocks with 64 weights Q2_K working on ARM_NEON, but quite a bit slower than 256 weights * k_quants: WIP super-blocks with 64 weights Q3_K working on ARM_NEON, but quite a bit slower than 256 weights. * k_quants: WIP super-blocks with 64 weights Q5_K working on ARM_NEON, but quite a bit slower than 256 weights. With that, we have full support for ARM_NEON, although performance is not quite there. * k_quants: WIP super-blocks with 64 weights Slightly more efficient Q3_K and Q5_K * k_quants: WIP super-blocks with 64 weights Another small improvement for Q3_K and Q5_K on ARM_NEON * k_quants: WIP super-blocks with 64 weights Yet another speedup for Q5_K on ARM_NEON. We are now within 10% of the QK_K = 256 version. * k_quants: WIP super-blocks with 64 weights * We are able to pass preprocessor macros to the Metal compiler * Q6_K works and is actually slightly more efficient than the QK_K = 256 version (25.2 ms vs 25.8 ms) * k_quants: WIP super-blocks with 64 weights Q4_K works on Metal and is actually slightly faster than QK_K = 256 (21.95 ms vs 24.0 ms). * k_quants: WIP super-blocks with 64 weights Q2_K works on Metal and is very slightly faster than QK_K = 256 (23.8 ms vs 24.2 ms). * k_quants: WIP super-blocks with 64 weights Q3_K works on Metal and is slightly faster than QK_K = 256 (26.6 ms vs 28.3 ms). * k_quants: WIP super-blocks with 64 weights Q5_K works on Metal and is slightly faster than QK_K = 256 (23.7 ms vs 26.3 ms). * k_quants: call them _K, not _k, also on Metal * k_quants: correctly define QK_K in llama.cpp * Fixed bug in q4_K quantization added with the 64-block addition * Simplify via lambda * k_quants: swicth Q3_K to 4-bit scales when QK_K = 64 Otherwise there isn't much benefit from this quantization type. There is some very slight loss in accuracy, but we reduce size by ~7%. E.g., for OpenLLaMA-3B, Q3_K_S perplexity is 8.6131 with 8-bit scales and 8.6352 with 4-bit, while file size decreases from 1.53G to 1.44G. * k_quants: switch Q4_K to 4-bit scales when QK_K = 64 Here the loss in accuracy is greater than for Q3_K, but the Q4_K points still move further to the left on the perplexity vs size curve. * k_quants: forgot to add the Metal changes in last commit * k_quants: change Q5_K to be type 0 when QK_K = 64 Still needs AVX2 implementation * k_quants: AVX2 implementation for new 64-weight Q5_K * k_quants: 10% faster ARM_NEON Q5_K dot product * k_quants: fixed issue caused by merging with master --------- Co-authored-by: Iwan Kawrakow <iwan.kawrakow@gmail.com>
2023-06-26 16:43:07 +00:00
ggml : fix dot product for ARM (#4630) ggml-ci
2023-12-27 09:02:13 +00:00
sum += d * (isum1 + isum2);
k-quants : support for super-block size of 64 (#2001) * k_quants: WIP super-blocks with 64 weights * k_quants: WIP super-blocks with 64 weights Q6_K scalar and AVX2 works * k_quants: WIP super-blocks with 64 weights Q4_K scalar and AVX2 works * k_quants: WIP super-blocks with 64 weights Q2_K scalar and AVX2 works. Q2_K is way too slow (it is actually slower than the scalar implementation) * k_quants: WIP super-blocks with 64 weights Q3_K scalar and AVX2 works. * k_quants: WIP super-blocks with 64 weights Q5_K scalar and AVX2 works, and with that all k_quants are done on AVX2 and scalar * k_quants: WIP super-blocks with 64 weights Q6_K working on CUDA. Cannot make it run quite as gast as with super-blocks with 256 weigths: 8% slower on 4080, 20% slower on the 1660 (but there we fit 1 less layer on the GPU because pf the larger model size), so some fraction of these 20% is due to that, * k_quants: WIP super-blocks with 64 weights Q4_K working on CUDA. ~10% slower on GTX-1660, 16% slower on 4080. * k_quants: WIP super-blocks with 64 weights Q2_K working on CUDA. ~3% slower on GTX-1660, 10% slower on 4080. * k_quants: WIP super-blocks with 64 weights Q3_K working on CUDA. * k_quants: WIP super-blocks with 64 weights Q5_K working on CUDA, and with this CUDA is done. * k_quants: WIP super-blocks with 64 weights Q6_K working on ARM_NEON * k_quants: WIP super-blocks with 64 weights Q4_K working on ARM_NEON, but quite a bit slower than 256 weights * k_quants: WIP super-blocks with 64 weights Q2_K working on ARM_NEON, but quite a bit slower than 256 weights * k_quants: WIP super-blocks with 64 weights Q3_K working on ARM_NEON, but quite a bit slower than 256 weights. * k_quants: WIP super-blocks with 64 weights Q5_K working on ARM_NEON, but quite a bit slower than 256 weights. With that, we have full support for ARM_NEON, although performance is not quite there. * k_quants: WIP super-blocks with 64 weights Slightly more efficient Q3_K and Q5_K * k_quants: WIP super-blocks with 64 weights Another small improvement for Q3_K and Q5_K on ARM_NEON * k_quants: WIP super-blocks with 64 weights Yet another speedup for Q5_K on ARM_NEON. We are now within 10% of the QK_K = 256 version. * k_quants: WIP super-blocks with 64 weights * We are able to pass preprocessor macros to the Metal compiler * Q6_K works and is actually slightly more efficient than the QK_K = 256 version (25.2 ms vs 25.8 ms) * k_quants: WIP super-blocks with 64 weights Q4_K works on Metal and is actually slightly faster than QK_K = 256 (21.95 ms vs 24.0 ms). * k_quants: WIP super-blocks with 64 weights Q2_K works on Metal and is very slightly faster than QK_K = 256 (23.8 ms vs 24.2 ms). * k_quants: WIP super-blocks with 64 weights Q3_K works on Metal and is slightly faster than QK_K = 256 (26.6 ms vs 28.3 ms). * k_quants: WIP super-blocks with 64 weights Q5_K works on Metal and is slightly faster than QK_K = 256 (23.7 ms vs 26.3 ms). * k_quants: call them _K, not _k, also on Metal * k_quants: correctly define QK_K in llama.cpp * Fixed bug in q4_K quantization added with the 64-block addition * Simplify via lambda * k_quants: swicth Q3_K to 4-bit scales when QK_K = 64 Otherwise there isn't much benefit from this quantization type. There is some very slight loss in accuracy, but we reduce size by ~7%. E.g., for OpenLLaMA-3B, Q3_K_S perplexity is 8.6131 with 8-bit scales and 8.6352 with 4-bit, while file size decreases from 1.53G to 1.44G. * k_quants: switch Q4_K to 4-bit scales when QK_K = 64 Here the loss in accuracy is greater than for Q3_K, but the Q4_K points still move further to the left on the perplexity vs size curve. * k_quants: forgot to add the Metal changes in last commit * k_quants: change Q5_K to be type 0 when QK_K = 64 Still needs AVX2 implementation * k_quants: AVX2 implementation for new 64-weight Q5_K * k_quants: 10% faster ARM_NEON Q5_K dot product * k_quants: fixed issue caused by merging with master --------- Co-authored-by: Iwan Kawrakow <iwan.kawrakow@gmail.com>
2023-06-26 16:43:07 +00:00
}
*s = sum;
#elif defined __AVX2__
const __m256i m3 = _mm256_set1_epi8(3);
__m256 acc = _mm256_setzero_ps();
uint32_t ud, um;
const uint8_t * restrict db = (const uint8_t *)&ud;
const uint8_t * restrict mb = (const uint8_t *)&um;
float summs = 0;
// TODO: optimize this
for (int i = 0; i < nb; ++i) {
ggml : move FP16 <-> FP32 code to ggml-impl.h (#3861) * ggml : move FP16 <-> FP32 stuff to ggml-impl.h ggml-ci * tests : fix ARM build * ggml : explicitly initialize deprecated type traits * ggml : add math.h to ggml-impl.h * ggml : remove duplicate static assert macros * ggml : prefix lookup tables with ggml_ ggml-ci * ggml-impl : move extern "C" to start of file
2023-10-30 17:19:15 +00:00
const float d = y[i].d * GGML_FP16_TO_FP32(x[i].d);
const float dmin = -y[i].d * GGML_FP16_TO_FP32(x[i].dmin);
k-quants : support for super-block size of 64 (#2001) * k_quants: WIP super-blocks with 64 weights * k_quants: WIP super-blocks with 64 weights Q6_K scalar and AVX2 works * k_quants: WIP super-blocks with 64 weights Q4_K scalar and AVX2 works * k_quants: WIP super-blocks with 64 weights Q2_K scalar and AVX2 works. Q2_K is way too slow (it is actually slower than the scalar implementation) * k_quants: WIP super-blocks with 64 weights Q3_K scalar and AVX2 works. * k_quants: WIP super-blocks with 64 weights Q5_K scalar and AVX2 works, and with that all k_quants are done on AVX2 and scalar * k_quants: WIP super-blocks with 64 weights Q6_K working on CUDA. Cannot make it run quite as gast as with super-blocks with 256 weigths: 8% slower on 4080, 20% slower on the 1660 (but there we fit 1 less layer on the GPU because pf the larger model size), so some fraction of these 20% is due to that, * k_quants: WIP super-blocks with 64 weights Q4_K working on CUDA. ~10% slower on GTX-1660, 16% slower on 4080. * k_quants: WIP super-blocks with 64 weights Q2_K working on CUDA. ~3% slower on GTX-1660, 10% slower on 4080. * k_quants: WIP super-blocks with 64 weights Q3_K working on CUDA. * k_quants: WIP super-blocks with 64 weights Q5_K working on CUDA, and with this CUDA is done. * k_quants: WIP super-blocks with 64 weights Q6_K working on ARM_NEON * k_quants: WIP super-blocks with 64 weights Q4_K working on ARM_NEON, but quite a bit slower than 256 weights * k_quants: WIP super-blocks with 64 weights Q2_K working on ARM_NEON, but quite a bit slower than 256 weights * k_quants: WIP super-blocks with 64 weights Q3_K working on ARM_NEON, but quite a bit slower than 256 weights. * k_quants: WIP super-blocks with 64 weights Q5_K working on ARM_NEON, but quite a bit slower than 256 weights. With that, we have full support for ARM_NEON, although performance is not quite there. * k_quants: WIP super-blocks with 64 weights Slightly more efficient Q3_K and Q5_K * k_quants: WIP super-blocks with 64 weights Another small improvement for Q3_K and Q5_K on ARM_NEON * k_quants: WIP super-blocks with 64 weights Yet another speedup for Q5_K on ARM_NEON. We are now within 10% of the QK_K = 256 version. * k_quants: WIP super-blocks with 64 weights * We are able to pass preprocessor macros to the Metal compiler * Q6_K works and is actually slightly more efficient than the QK_K = 256 version (25.2 ms vs 25.8 ms) * k_quants: WIP super-blocks with 64 weights Q4_K works on Metal and is actually slightly faster than QK_K = 256 (21.95 ms vs 24.0 ms). * k_quants: WIP super-blocks with 64 weights Q2_K works on Metal and is very slightly faster than QK_K = 256 (23.8 ms vs 24.2 ms). * k_quants: WIP super-blocks with 64 weights Q3_K works on Metal and is slightly faster than QK_K = 256 (26.6 ms vs 28.3 ms). * k_quants: WIP super-blocks with 64 weights Q5_K works on Metal and is slightly faster than QK_K = 256 (23.7 ms vs 26.3 ms). * k_quants: call them _K, not _k, also on Metal * k_quants: correctly define QK_K in llama.cpp * Fixed bug in q4_K quantization added with the 64-block addition * Simplify via lambda * k_quants: swicth Q3_K to 4-bit scales when QK_K = 64 Otherwise there isn't much benefit from this quantization type. There is some very slight loss in accuracy, but we reduce size by ~7%. E.g., for OpenLLaMA-3B, Q3_K_S perplexity is 8.6131 with 8-bit scales and 8.6352 with 4-bit, while file size decreases from 1.53G to 1.44G. * k_quants: switch Q4_K to 4-bit scales when QK_K = 64 Here the loss in accuracy is greater than for Q3_K, but the Q4_K points still move further to the left on the perplexity vs size curve. * k_quants: forgot to add the Metal changes in last commit * k_quants: change Q5_K to be type 0 when QK_K = 64 Still needs AVX2 implementation * k_quants: AVX2 implementation for new 64-weight Q5_K * k_quants: 10% faster ARM_NEON Q5_K dot product * k_quants: fixed issue caused by merging with master --------- Co-authored-by: Iwan Kawrakow <iwan.kawrakow@gmail.com>
2023-06-26 16:43:07 +00:00
const uint8_t * restrict q2 = x[i].qs;
const int8_t * restrict q8 = y[i].qs;
const uint32_t * restrict sc = (const uint32_t *)x[i].scales;
ud = (sc[0] >> 0) & 0x0f0f0f0f;
um = (sc[0] >> 4) & 0x0f0f0f0f;
int32_t smin = mb[0] * y[i].bsums[0] + mb[1] * y[i].bsums[1] + mb[2] * y[i].bsums[2] + mb[3] * y[i].bsums[3];
summs += dmin * smin;
const __m128i q2bits = _mm_loadu_si128((const __m128i*)q2);
ggml : workaround for missing _mm256_setr_m128i in GCC < 8 in k_quants.c (#2405)
2023-07-28 18:17:45 +00:00
const __m256i q2_0 = _mm256_and_si256(MM256_SET_M128I(_mm_srli_epi16(q2bits, 2), q2bits), m3);
const __m256i q2_1 = _mm256_and_si256(MM256_SET_M128I(_mm_srli_epi16(q2bits, 6), _mm_srli_epi16(q2bits, 4)), m3);
k-quants : support for super-block size of 64 (#2001) * k_quants: WIP super-blocks with 64 weights * k_quants: WIP super-blocks with 64 weights Q6_K scalar and AVX2 works * k_quants: WIP super-blocks with 64 weights Q4_K scalar and AVX2 works * k_quants: WIP super-blocks with 64 weights Q2_K scalar and AVX2 works. Q2_K is way too slow (it is actually slower than the scalar implementation) * k_quants: WIP super-blocks with 64 weights Q3_K scalar and AVX2 works. * k_quants: WIP super-blocks with 64 weights Q5_K scalar and AVX2 works, and with that all k_quants are done on AVX2 and scalar * k_quants: WIP super-blocks with 64 weights Q6_K working on CUDA. Cannot make it run quite as gast as with super-blocks with 256 weigths: 8% slower on 4080, 20% slower on the 1660 (but there we fit 1 less layer on the GPU because pf the larger model size), so some fraction of these 20% is due to that, * k_quants: WIP super-blocks with 64 weights Q4_K working on CUDA. ~10% slower on GTX-1660, 16% slower on 4080. * k_quants: WIP super-blocks with 64 weights Q2_K working on CUDA. ~3% slower on GTX-1660, 10% slower on 4080. * k_quants: WIP super-blocks with 64 weights Q3_K working on CUDA. * k_quants: WIP super-blocks with 64 weights Q5_K working on CUDA, and with this CUDA is done. * k_quants: WIP super-blocks with 64 weights Q6_K working on ARM_NEON * k_quants: WIP super-blocks with 64 weights Q4_K working on ARM_NEON, but quite a bit slower than 256 weights * k_quants: WIP super-blocks with 64 weights Q2_K working on ARM_NEON, but quite a bit slower than 256 weights * k_quants: WIP super-blocks with 64 weights Q3_K working on ARM_NEON, but quite a bit slower than 256 weights. * k_quants: WIP super-blocks with 64 weights Q5_K working on ARM_NEON, but quite a bit slower than 256 weights. With that, we have full support for ARM_NEON, although performance is not quite there. * k_quants: WIP super-blocks with 64 weights Slightly more efficient Q3_K and Q5_K * k_quants: WIP super-blocks with 64 weights Another small improvement for Q3_K and Q5_K on ARM_NEON * k_quants: WIP super-blocks with 64 weights Yet another speedup for Q5_K on ARM_NEON. We are now within 10% of the QK_K = 256 version. * k_quants: WIP super-blocks with 64 weights * We are able to pass preprocessor macros to the Metal compiler * Q6_K works and is actually slightly more efficient than the QK_K = 256 version (25.2 ms vs 25.8 ms) * k_quants: WIP super-blocks with 64 weights Q4_K works on Metal and is actually slightly faster than QK_K = 256 (21.95 ms vs 24.0 ms). * k_quants: WIP super-blocks with 64 weights Q2_K works on Metal and is very slightly faster than QK_K = 256 (23.8 ms vs 24.2 ms). * k_quants: WIP super-blocks with 64 weights Q3_K works on Metal and is slightly faster than QK_K = 256 (26.6 ms vs 28.3 ms). * k_quants: WIP super-blocks with 64 weights Q5_K works on Metal and is slightly faster than QK_K = 256 (23.7 ms vs 26.3 ms). * k_quants: call them _K, not _k, also on Metal * k_quants: correctly define QK_K in llama.cpp * Fixed bug in q4_K quantization added with the 64-block addition * Simplify via lambda * k_quants: swicth Q3_K to 4-bit scales when QK_K = 64 Otherwise there isn't much benefit from this quantization type. There is some very slight loss in accuracy, but we reduce size by ~7%. E.g., for OpenLLaMA-3B, Q3_K_S perplexity is 8.6131 with 8-bit scales and 8.6352 with 4-bit, while file size decreases from 1.53G to 1.44G. * k_quants: switch Q4_K to 4-bit scales when QK_K = 64 Here the loss in accuracy is greater than for Q3_K, but the Q4_K points still move further to the left on the perplexity vs size curve. * k_quants: forgot to add the Metal changes in last commit * k_quants: change Q5_K to be type 0 when QK_K = 64 Still needs AVX2 implementation * k_quants: AVX2 implementation for new 64-weight Q5_K * k_quants: 10% faster ARM_NEON Q5_K dot product * k_quants: fixed issue caused by merging with master --------- Co-authored-by: Iwan Kawrakow <iwan.kawrakow@gmail.com>
2023-06-26 16:43:07 +00:00
const __m256i q8_0 = _mm256_loadu_si256((const __m256i*)(q8+ 0));
const __m256i q8_1 = _mm256_loadu_si256((const __m256i*)(q8+32));
const __m256i p0 = _mm256_maddubs_epi16(q2_0, q8_0);
const __m256i p1 = _mm256_maddubs_epi16(q2_1, q8_1);
const __m256i p_0 = _mm256_cvtepi16_epi32(_mm256_extracti128_si256(p0, 0));
const __m256i p_1 = _mm256_cvtepi16_epi32(_mm256_extracti128_si256(p0, 1));
const __m256i p_2 = _mm256_cvtepi16_epi32(_mm256_extracti128_si256(p1, 0));
const __m256i p_3 = _mm256_cvtepi16_epi32(_mm256_extracti128_si256(p1, 1));
acc = _mm256_fmadd_ps(_mm256_set1_ps(d * db[0]), _mm256_cvtepi32_ps(p_0), acc);
acc = _mm256_fmadd_ps(_mm256_set1_ps(d * db[1]), _mm256_cvtepi32_ps(p_1), acc);
acc = _mm256_fmadd_ps(_mm256_set1_ps(d * db[2]), _mm256_cvtepi32_ps(p_2), acc);
acc = _mm256_fmadd_ps(_mm256_set1_ps(d * db[3]), _mm256_cvtepi32_ps(p_3), acc);
}
*s = hsum_float_8(acc) + summs;
k_quants : add AVX support to dot functions with QK_K as 64 (#2339) * add AVX to ggml_vec_dot_q2_K_q8_K() * add AVX to ggml_vec_dot_q3_K_q8_K() * add AVX to ggml_vec_dot_q4_K_q8_K() * add AVX to ggml_vec_dot_q5_K_q8_K() * add AVX to ggml_vec_dot_q6_K_q8_K() * refactor AVX code in ggml_vec_dot_q6_K_q8_K()
2023-07-25 12:13:41 +00:00
#elif defined __AVX__
const __m128i m3 = _mm_set1_epi8(3);
__m256 acc = _mm256_setzero_ps();
uint32_t ud, um;
const uint8_t * restrict db = (const uint8_t *)&ud;
const uint8_t * restrict mb = (const uint8_t *)&um;
float summs = 0;
// TODO: optimize this
for (int i = 0; i < nb; ++i) {
ggml : move FP16 <-> FP32 code to ggml-impl.h (#3861) * ggml : move FP16 <-> FP32 stuff to ggml-impl.h ggml-ci * tests : fix ARM build * ggml : explicitly initialize deprecated type traits * ggml : add math.h to ggml-impl.h * ggml : remove duplicate static assert macros * ggml : prefix lookup tables with ggml_ ggml-ci * ggml-impl : move extern "C" to start of file
2023-10-30 17:19:15 +00:00
const float d = y[i].d * GGML_FP16_TO_FP32(x[i].d);
const float dmin = -y[i].d * GGML_FP16_TO_FP32(x[i].dmin);
k_quants : add AVX support to dot functions with QK_K as 64 (#2339) * add AVX to ggml_vec_dot_q2_K_q8_K() * add AVX to ggml_vec_dot_q3_K_q8_K() * add AVX to ggml_vec_dot_q4_K_q8_K() * add AVX to ggml_vec_dot_q5_K_q8_K() * add AVX to ggml_vec_dot_q6_K_q8_K() * refactor AVX code in ggml_vec_dot_q6_K_q8_K()
2023-07-25 12:13:41 +00:00
const uint8_t * restrict q2 = x[i].qs;
const int8_t * restrict q8 = y[i].qs;
const uint32_t * restrict sc = (const uint32_t *)x[i].scales;
ud = (sc[0] >> 0) & 0x0f0f0f0f;
um = (sc[0] >> 4) & 0x0f0f0f0f;
int32_t smin = mb[0] * y[i].bsums[0] + mb[1] * y[i].bsums[1] + mb[2] * y[i].bsums[2] + mb[3] * y[i].bsums[3];
summs += dmin * smin;
const __m128i q2bits = _mm_loadu_si128((const __m128i*)q2);
const __m128i q2_0 = _mm_and_si128(q2bits, m3);
const __m128i q2_1 = _mm_and_si128(_mm_srli_epi16(q2bits, 2), m3);
const __m128i q2_2 = _mm_and_si128(_mm_srli_epi16(q2bits, 4), m3);
const __m128i q2_3 = _mm_and_si128(_mm_srli_epi16(q2bits, 6), m3);
const __m256i q8_0 = _mm256_loadu_si256((const __m256i*)(q8+ 0));
const __m256i q8_1 = _mm256_loadu_si256((const __m256i*)(q8+32));
const __m128i p0 = _mm_maddubs_epi16(q2_0, _mm256_extractf128_si256(q8_0, 0));
const __m128i p1 = _mm_maddubs_epi16(q2_1, _mm256_extractf128_si256(q8_0, 1));
const __m128i p2 = _mm_maddubs_epi16(q2_2, _mm256_extractf128_si256(q8_1, 0));
const __m128i p3 = _mm_maddubs_epi16(q2_3, _mm256_extractf128_si256(q8_1, 1));
ggml : workaround for missing _mm256_setr_m128i in GCC < 8 in k_quants.c (#2405)
2023-07-28 18:17:45 +00:00
const __m256i p_0 = MM256_SET_M128I(_mm_cvtepi16_epi32(_mm_unpackhi_epi64(p0, p0)), _mm_cvtepi16_epi32(p0));
const __m256i p_1 = MM256_SET_M128I(_mm_cvtepi16_epi32(_mm_unpackhi_epi64(p1, p1)), _mm_cvtepi16_epi32(p1));
const __m256i p_2 = MM256_SET_M128I(_mm_cvtepi16_epi32(_mm_unpackhi_epi64(p2, p2)), _mm_cvtepi16_epi32(p2));
const __m256i p_3 = MM256_SET_M128I(_mm_cvtepi16_epi32(_mm_unpackhi_epi64(p3, p3)), _mm_cvtepi16_epi32(p3));
k_quants : add AVX support to dot functions with QK_K as 64 (#2339) * add AVX to ggml_vec_dot_q2_K_q8_K() * add AVX to ggml_vec_dot_q3_K_q8_K() * add AVX to ggml_vec_dot_q4_K_q8_K() * add AVX to ggml_vec_dot_q5_K_q8_K() * add AVX to ggml_vec_dot_q6_K_q8_K() * refactor AVX code in ggml_vec_dot_q6_K_q8_K()
2023-07-25 12:13:41 +00:00
acc = _mm256_add_ps(_mm256_mul_ps(_mm256_set1_ps(d * db[0]), _mm256_cvtepi32_ps(p_0)), acc);
acc = _mm256_add_ps(_mm256_mul_ps(_mm256_set1_ps(d * db[1]), _mm256_cvtepi32_ps(p_1)), acc);
acc = _mm256_add_ps(_mm256_mul_ps(_mm256_set1_ps(d * db[2]), _mm256_cvtepi32_ps(p_2)), acc);
acc = _mm256_add_ps(_mm256_mul_ps(_mm256_set1_ps(d * db[3]), _mm256_cvtepi32_ps(p_3)), acc);
}
*s = hsum_float_8(acc) + summs;
ggml : add RISC-V Vector Support for K-Quants and improved the existing intrinsics (#3453) * Added RVV intrinsics support for Q8 quantize row and also improved the existing dot product function for risc-v. The RVV intrinsics is added for the following quantize row functions quantize_row_q8_0 quantize_row_q8_1 The following dot product functions have also been optimized by using LMUL = 1/2 instead of LMUL = 1 ggml_vec_dot_q4_0_q8_0 ggml_vec_dot_q4_1_q8_1 ggml_vec_dot_q5_0_q8_0 ggml_vec_dot_q5_1_q8_1 And vector initialization in Q5 by temporary array is also replaced by the vid intrinsics Signed-off-by: Ahmad Tameem <ahmad.tameem@10xengineers.ai> * Added RVV intrinsics support for k_quants This adds RISC-V Vector intrinsics support for the following K_quants functions for both QKK = 256 and QKK = 64 ggml_vec_dot_q2_K_q8_K ggml_vec_dot_q3_K_q8_K ggml_vec_dot_q4_K_q8_K ggml_vec_dot_q5_K_q8_K ggml_vec_dot_q6_K_q8_K Signed-off-by: Ahmad Tameem <ahmad.tameem@10xengineers.ai> --------- Signed-off-by: Ahmad Tameem <ahmad.tameem@10xengineers.ai>
2023-10-03 18:38:19 +00:00
#elif defined __riscv_v_intrinsic
uint32_t aux32[2];
const uint8_t * scales = (const uint8_t *)aux32;
float sumf = 0;
for (int i = 0; i < nb; ++i) {
const float d = y[i].d * (float)x[i].d;
const float dmin = -y[i].d * (float)x[i].dmin;
const uint8_t * restrict q2 = x[i].qs;
const int8_t * restrict q8 = y[i].qs;
const uint32_t * restrict sc = (const uint32_t *)x[i].scales;
aux32[0] = sc[0] & 0x0f0f0f0f;
aux32[1] = (sc[0] >> 4) & 0x0f0f0f0f;
sumf += dmin * (scales[4] * y[i].bsums[0] + scales[5] * y[i].bsums[1] + scales[6] * y[i].bsums[2] + scales[7] * y[i].bsums[3]);
int isum1 = 0;
int isum2 = 0;
size_t vl = 16;
vint16m1_t vzero = __riscv_vmv_v_x_i16m1(0, 1);
// load Q2
vuint8mf2_t q2_x = __riscv_vle8_v_u8mf2(q2, vl);
vint8mf2_t q2_0 = __riscv_vreinterpret_v_u8mf2_i8mf2(__riscv_vand_vx_u8mf2(q2_x, 0x03, vl));
vint8mf2_t q2_1 = __riscv_vreinterpret_v_u8mf2_i8mf2(__riscv_vand_vx_u8mf2(__riscv_vsrl_vx_u8mf2(q2_x, 0x2, vl), 0x03 , vl));
vint8mf2_t q2_2 = __riscv_vreinterpret_v_u8mf2_i8mf2(__riscv_vand_vx_u8mf2(__riscv_vsrl_vx_u8mf2(q2_x, 0x4, vl), 0x03 , vl));
vint8mf2_t q2_3 = __riscv_vreinterpret_v_u8mf2_i8mf2(__riscv_vand_vx_u8mf2(__riscv_vsrl_vx_u8mf2(q2_x, 0x6, vl), 0x03 , vl));
// load Q8, and take product with Q2
vint16m1_t p0 = __riscv_vwmul_vv_i16m1(q2_0, __riscv_vle8_v_i8mf2(q8, vl), vl);
vint16m1_t p1 = __riscv_vwmul_vv_i16m1(q2_1, __riscv_vle8_v_i8mf2(q8+16, vl), vl);
vint16m1_t p2 = __riscv_vwmul_vv_i16m1(q2_2, __riscv_vle8_v_i8mf2(q8+32, vl), vl);
vint16m1_t p3 = __riscv_vwmul_vv_i16m1(q2_3, __riscv_vle8_v_i8mf2(q8+48, vl), vl);
vint16m1_t vs_0 = __riscv_vredsum_vs_i16m1_i16m1(p0, vzero, vl);
vint16m1_t vs_1 = __riscv_vredsum_vs_i16m1_i16m1(p1, vzero, vl);
vint16m1_t vs_2 = __riscv_vredsum_vs_i16m1_i16m1(p2, vzero, vl);
vint16m1_t vs_3 = __riscv_vredsum_vs_i16m1_i16m1(p3, vzero, vl);
isum1 += __riscv_vmv_x_s_i16m1_i16(vs_0) * scales[0];
isum2 += __riscv_vmv_x_s_i16m1_i16(vs_1) * scales[1];
isum1 += __riscv_vmv_x_s_i16m1_i16(vs_2) * scales[2];
isum2 += __riscv_vmv_x_s_i16m1_i16(vs_3) * scales[3];
sumf += d * (isum1 + isum2);
}
*s = sumf;
k-quants : support for super-block size of 64 (#2001) * k_quants: WIP super-blocks with 64 weights * k_quants: WIP super-blocks with 64 weights Q6_K scalar and AVX2 works * k_quants: WIP super-blocks with 64 weights Q4_K scalar and AVX2 works * k_quants: WIP super-blocks with 64 weights Q2_K scalar and AVX2 works. Q2_K is way too slow (it is actually slower than the scalar implementation) * k_quants: WIP super-blocks with 64 weights Q3_K scalar and AVX2 works. * k_quants: WIP super-blocks with 64 weights Q5_K scalar and AVX2 works, and with that all k_quants are done on AVX2 and scalar * k_quants: WIP super-blocks with 64 weights Q6_K working on CUDA. Cannot make it run quite as gast as with super-blocks with 256 weigths: 8% slower on 4080, 20% slower on the 1660 (but there we fit 1 less layer on the GPU because pf the larger model size), so some fraction of these 20% is due to that, * k_quants: WIP super-blocks with 64 weights Q4_K working on CUDA. ~10% slower on GTX-1660, 16% slower on 4080. * k_quants: WIP super-blocks with 64 weights Q2_K working on CUDA. ~3% slower on GTX-1660, 10% slower on 4080. * k_quants: WIP super-blocks with 64 weights Q3_K working on CUDA. * k_quants: WIP super-blocks with 64 weights Q5_K working on CUDA, and with this CUDA is done. * k_quants: WIP super-blocks with 64 weights Q6_K working on ARM_NEON * k_quants: WIP super-blocks with 64 weights Q4_K working on ARM_NEON, but quite a bit slower than 256 weights * k_quants: WIP super-blocks with 64 weights Q2_K working on ARM_NEON, but quite a bit slower than 256 weights * k_quants: WIP super-blocks with 64 weights Q3_K working on ARM_NEON, but quite a bit slower than 256 weights. * k_quants: WIP super-blocks with 64 weights Q5_K working on ARM_NEON, but quite a bit slower than 256 weights. With that, we have full support for ARM_NEON, although performance is not quite there. * k_quants: WIP super-blocks with 64 weights Slightly more efficient Q3_K and Q5_K * k_quants: WIP super-blocks with 64 weights Another small improvement for Q3_K and Q5_K on ARM_NEON * k_quants: WIP super-blocks with 64 weights Yet another speedup for Q5_K on ARM_NEON. We are now within 10% of the QK_K = 256 version. * k_quants: WIP super-blocks with 64 weights * We are able to pass preprocessor macros to the Metal compiler * Q6_K works and is actually slightly more efficient than the QK_K = 256 version (25.2 ms vs 25.8 ms) * k_quants: WIP super-blocks with 64 weights Q4_K works on Metal and is actually slightly faster than QK_K = 256 (21.95 ms vs 24.0 ms). * k_quants: WIP super-blocks with 64 weights Q2_K works on Metal and is very slightly faster than QK_K = 256 (23.8 ms vs 24.2 ms). * k_quants: WIP super-blocks with 64 weights Q3_K works on Metal and is slightly faster than QK_K = 256 (26.6 ms vs 28.3 ms). * k_quants: WIP super-blocks with 64 weights Q5_K works on Metal and is slightly faster than QK_K = 256 (23.7 ms vs 26.3 ms). * k_quants: call them _K, not _k, also on Metal * k_quants: correctly define QK_K in llama.cpp * Fixed bug in q4_K quantization added with the 64-block addition * Simplify via lambda * k_quants: swicth Q3_K to 4-bit scales when QK_K = 64 Otherwise there isn't much benefit from this quantization type. There is some very slight loss in accuracy, but we reduce size by ~7%. E.g., for OpenLLaMA-3B, Q3_K_S perplexity is 8.6131 with 8-bit scales and 8.6352 with 4-bit, while file size decreases from 1.53G to 1.44G. * k_quants: switch Q4_K to 4-bit scales when QK_K = 64 Here the loss in accuracy is greater than for Q3_K, but the Q4_K points still move further to the left on the perplexity vs size curve. * k_quants: forgot to add the Metal changes in last commit * k_quants: change Q5_K to be type 0 when QK_K = 64 Still needs AVX2 implementation * k_quants: AVX2 implementation for new 64-weight Q5_K * k_quants: 10% faster ARM_NEON Q5_K dot product * k_quants: fixed issue caused by merging with master --------- Co-authored-by: Iwan Kawrakow <iwan.kawrakow@gmail.com>
2023-06-26 16:43:07 +00:00
#else
float sumf = 0;
int isum[4];
for (int i = 0; i < nb; ++i) {
const uint8_t * q2 = x[i].qs;
const int8_t * q8 = y[i].qs;
const uint8_t * sc = x[i].scales;
int summs = 0;
for (int j = 0; j < QK_K/16; ++j) {
summs += y[i].bsums[j] * (sc[j] >> 4);
}
ggml : move FP16 <-> FP32 code to ggml-impl.h (#3861) * ggml : move FP16 <-> FP32 stuff to ggml-impl.h ggml-ci * tests : fix ARM build * ggml : explicitly initialize deprecated type traits * ggml : add math.h to ggml-impl.h * ggml : remove duplicate static assert macros * ggml : prefix lookup tables with ggml_ ggml-ci * ggml-impl : move extern "C" to start of file
2023-10-30 17:19:15 +00:00
const float dall = y[i].d * GGML_FP16_TO_FP32(x[i].d);
const float dmin = y[i].d * GGML_FP16_TO_FP32(x[i].dmin);
k-quants : support for super-block size of 64 (#2001) * k_quants: WIP super-blocks with 64 weights * k_quants: WIP super-blocks with 64 weights Q6_K scalar and AVX2 works * k_quants: WIP super-blocks with 64 weights Q4_K scalar and AVX2 works * k_quants: WIP super-blocks with 64 weights Q2_K scalar and AVX2 works. Q2_K is way too slow (it is actually slower than the scalar implementation) * k_quants: WIP super-blocks with 64 weights Q3_K scalar and AVX2 works. * k_quants: WIP super-blocks with 64 weights Q5_K scalar and AVX2 works, and with that all k_quants are done on AVX2 and scalar * k_quants: WIP super-blocks with 64 weights Q6_K working on CUDA. Cannot make it run quite as gast as with super-blocks with 256 weigths: 8% slower on 4080, 20% slower on the 1660 (but there we fit 1 less layer on the GPU because pf the larger model size), so some fraction of these 20% is due to that, * k_quants: WIP super-blocks with 64 weights Q4_K working on CUDA. ~10% slower on GTX-1660, 16% slower on 4080. * k_quants: WIP super-blocks with 64 weights Q2_K working on CUDA. ~3% slower on GTX-1660, 10% slower on 4080. * k_quants: WIP super-blocks with 64 weights Q3_K working on CUDA. * k_quants: WIP super-blocks with 64 weights Q5_K working on CUDA, and with this CUDA is done. * k_quants: WIP super-blocks with 64 weights Q6_K working on ARM_NEON * k_quants: WIP super-blocks with 64 weights Q4_K working on ARM_NEON, but quite a bit slower than 256 weights * k_quants: WIP super-blocks with 64 weights Q2_K working on ARM_NEON, but quite a bit slower than 256 weights * k_quants: WIP super-blocks with 64 weights Q3_K working on ARM_NEON, but quite a bit slower than 256 weights. * k_quants: WIP super-blocks with 64 weights Q5_K working on ARM_NEON, but quite a bit slower than 256 weights. With that, we have full support for ARM_NEON, although performance is not quite there. * k_quants: WIP super-blocks with 64 weights Slightly more efficient Q3_K and Q5_K * k_quants: WIP super-blocks with 64 weights Another small improvement for Q3_K and Q5_K on ARM_NEON * k_quants: WIP super-blocks with 64 weights Yet another speedup for Q5_K on ARM_NEON. We are now within 10% of the QK_K = 256 version. * k_quants: WIP super-blocks with 64 weights * We are able to pass preprocessor macros to the Metal compiler * Q6_K works and is actually slightly more efficient than the QK_K = 256 version (25.2 ms vs 25.8 ms) * k_quants: WIP super-blocks with 64 weights Q4_K works on Metal and is actually slightly faster than QK_K = 256 (21.95 ms vs 24.0 ms). * k_quants: WIP super-blocks with 64 weights Q2_K works on Metal and is very slightly faster than QK_K = 256 (23.8 ms vs 24.2 ms). * k_quants: WIP super-blocks with 64 weights Q3_K works on Metal and is slightly faster than QK_K = 256 (26.6 ms vs 28.3 ms). * k_quants: WIP super-blocks with 64 weights Q5_K works on Metal and is slightly faster than QK_K = 256 (23.7 ms vs 26.3 ms). * k_quants: call them _K, not _k, also on Metal * k_quants: correctly define QK_K in llama.cpp * Fixed bug in q4_K quantization added with the 64-block addition * Simplify via lambda * k_quants: swicth Q3_K to 4-bit scales when QK_K = 64 Otherwise there isn't much benefit from this quantization type. There is some very slight loss in accuracy, but we reduce size by ~7%. E.g., for OpenLLaMA-3B, Q3_K_S perplexity is 8.6131 with 8-bit scales and 8.6352 with 4-bit, while file size decreases from 1.53G to 1.44G. * k_quants: switch Q4_K to 4-bit scales when QK_K = 64 Here the loss in accuracy is greater than for Q3_K, but the Q4_K points still move further to the left on the perplexity vs size curve. * k_quants: forgot to add the Metal changes in last commit * k_quants: change Q5_K to be type 0 when QK_K = 64 Still needs AVX2 implementation * k_quants: AVX2 implementation for new 64-weight Q5_K * k_quants: 10% faster ARM_NEON Q5_K dot product * k_quants: fixed issue caused by merging with master --------- Co-authored-by: Iwan Kawrakow <iwan.kawrakow@gmail.com>
2023-06-26 16:43:07 +00:00
isum[0] = isum[1] = isum[2] = isum[3] = 0;
for (int l = 0; l < 16; ++l) {
isum[0] += q8[l+ 0] * ((q2[l] >> 0) & 3);
isum[1] += q8[l+16] * ((q2[l] >> 2) & 3);
isum[2] += q8[l+32] * ((q2[l] >> 4) & 3);
isum[3] += q8[l+48] * ((q2[l] >> 6) & 3);
}
for (int l = 0; l < 4; ++l) {
isum[l] *= (sc[l] & 0xF);
}
sumf += dall * (isum[0] + isum[1] + isum[2] + isum[3]) - dmin * summs;
}
*s = sumf;
#endif
}
#endif
#if QK_K == 256
k-quants : allow to optionally disable at compile time (#1734) * k-quants : put behind optional compile flag LLAMA_K_QUANTS * build : enable k-quants by default
2023-06-07 07:59:52 +00:00
void ggml_vec_dot_q3_K_q8_K(const int n, float * restrict s, const void * restrict vx, const void * restrict vy) {
ggml : add SOTA 2,3,4,5,6 bit k-quantizations (#1684) * Starting to add k-quantization to ggml I think it is better to have quantization separate from ggml. For now just adding the k-quants there, but it would be better to also factor out the existing ggml quantizations. * Adding Q3_K and Q8_K (de)-quantization * Q3_K now working on CUDA and AVX2/scalar CUDA is not ideal - ~50% slower than Q4_0 for single token prediction, about the same in batch mode (perplexity). CPU single token is ~55 ms (on Ryzen 7950X). * Some improvement for Q3_K on CUDA It is now ~22.5 ms/token on my GPU, so ~30% slower than Q4_0. * Some more CUDA optimizations for Q3_K Single token is now 20.5 ms/token (~20% slower than Q4_0). Perplexity is on par with Q4_0. * Adding Q4_K - scalar, AVX2, CUDA Performance is the same or perhaps very slightly better than Q4_0 on the CPU. On the GPU, single token prediction is ~10% better than Q4_0, batch mode (perplexity is about the same). * Adding Q6_K - scalar, AVX2, CUDA Performance is ~40% lower compared to Q4_K on the CPU. This is to be expected, considering that we are memory bound on the CPU and the 6-bit model is ~44% larger than the 4-bit. On the GPU, single token prediction is ~6% lower than Q4_0, batch mode (perplexity) is even closer (but still slower). * Adding Q5_K - scalar, AVX2, CUDA Performance is ~20% lower compared to Q4_K on the CPU. This is to be expected, considering that we are memory bound on the CPU and the 5-bit model is ~22% larger than the 4-bit. On the GPU, single token prediction is about the same as Q4_0 for both, single token and batch prediction. * Per convention, all QX_K quantizations use Q5_K for output.weight * Adding quantization mixes * Quantization mixes: didn't quite get what I wanted in the last commit * Q4_K dot product for ARM_NEON * Q6_K dot product for ARM_NEON * Q5_K dot product for ARM_NEON * Adding Q3_K dot for ARM_NEON It is 22% slower than Q4_K, despite the smaller model size. On x86_64, where we are memory bound, the Q3_K model is quite a bit faster than Q4_K. * A very slightly faster ARM_NEON Q3_K dot * Adding Q2_K - just CUDA for now Token prediction is pretty good - about 15.5 ms on a RTX 4080. Perplexity is about the same as Q4_K. * Adding scalar and AVX2 Q2_K dot * Adding ARM_NEON Q2_K dot About the same performance as Q4_K. * A slightly faster ARM_NEON Q2_K dot Single token prediction is now ~36 ms on M2 Max. The code is much simpler too. * Fixed bug in Q2_K CUDA dot product kernel Stranegly enough, for the few prompts I tried with the 7B model the responses looked perfectly reasonable. Only realized something is not quite right when I tried the larger models and started getting nonse back. In any case, Q2_K single token evaluation time on an RTX 4080 in a Ryzen7950X box iusing CUDA and model fully loaded on the GPU are ~15.5 ms for 7B, ~25.4 ms for 13B, and ~55.8 ms for 30B. The max number of layers that fit in VRAM for The 65B is 32. With that, we get ~330 ms per token, which is not that much faster than just running on the CPU (~470 ms per token). * Don't print zeros/NaNs when no count histogram has been collected * A 10% faster CUDA vector dot kernel for Q3_K Q3_K is now running at ~18.5 ms / token on CUDA, so the gap to Q4_0 is only 10%. It seems memory acccess pattern is more important for performance than the amount of computation the kernel does. * A slightly daster Q4_K AVX2 dot product For perplexity, where we are less memory bound, time per pass drops by ~5%. Barely measurable difference for single token prediction. * A slightly faster ARM_NEON A4_K dot product * Minor * Fix quantization error test We cannot possibly be expecting rmse < 0.002 for 2- and 3-bit quantization variants. * Fix docker build I have been sloppy with vector reinterpret casts on ARM_NEON. It seems clang is very forgiving in that regard. * Added forgotten ggml.o dependence on k_quants.h to the Makefile * Had unintentionally committed the Makefile with -Ofast enabled * ggml : rename k_quants -> ggml-quants-k, use lowercase in code --------- Co-authored-by: Iwan Kawrakow <iwan.kawrakow@gmail.com> Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
2023-06-05 19:56:18 +00:00
assert(n % QK_K == 0);
const uint32_t kmask1 = 0x03030303;
const uint32_t kmask2 = 0x0f0f0f0f;
k-quants : allow to optionally disable at compile time (#1734) * k-quants : put behind optional compile flag LLAMA_K_QUANTS * build : enable k-quants by default
2023-06-07 07:59:52 +00:00
const block_q3_K * restrict x = vx;
const block_q8_K * restrict y = vy;
ggml : add SOTA 2,3,4,5,6 bit k-quantizations (#1684) * Starting to add k-quantization to ggml I think it is better to have quantization separate from ggml. For now just adding the k-quants there, but it would be better to also factor out the existing ggml quantizations. * Adding Q3_K and Q8_K (de)-quantization * Q3_K now working on CUDA and AVX2/scalar CUDA is not ideal - ~50% slower than Q4_0 for single token prediction, about the same in batch mode (perplexity). CPU single token is ~55 ms (on Ryzen 7950X). * Some improvement for Q3_K on CUDA It is now ~22.5 ms/token on my GPU, so ~30% slower than Q4_0. * Some more CUDA optimizations for Q3_K Single token is now 20.5 ms/token (~20% slower than Q4_0). Perplexity is on par with Q4_0. * Adding Q4_K - scalar, AVX2, CUDA Performance is the same or perhaps very slightly better than Q4_0 on the CPU. On the GPU, single token prediction is ~10% better than Q4_0, batch mode (perplexity is about the same). * Adding Q6_K - scalar, AVX2, CUDA Performance is ~40% lower compared to Q4_K on the CPU. This is to be expected, considering that we are memory bound on the CPU and the 6-bit model is ~44% larger than the 4-bit. On the GPU, single token prediction is ~6% lower than Q4_0, batch mode (perplexity) is even closer (but still slower). * Adding Q5_K - scalar, AVX2, CUDA Performance is ~20% lower compared to Q4_K on the CPU. This is to be expected, considering that we are memory bound on the CPU and the 5-bit model is ~22% larger than the 4-bit. On the GPU, single token prediction is about the same as Q4_0 for both, single token and batch prediction. * Per convention, all QX_K quantizations use Q5_K for output.weight * Adding quantization mixes * Quantization mixes: didn't quite get what I wanted in the last commit * Q4_K dot product for ARM_NEON * Q6_K dot product for ARM_NEON * Q5_K dot product for ARM_NEON * Adding Q3_K dot for ARM_NEON It is 22% slower than Q4_K, despite the smaller model size. On x86_64, where we are memory bound, the Q3_K model is quite a bit faster than Q4_K. * A very slightly faster ARM_NEON Q3_K dot * Adding Q2_K - just CUDA for now Token prediction is pretty good - about 15.5 ms on a RTX 4080. Perplexity is about the same as Q4_K. * Adding scalar and AVX2 Q2_K dot * Adding ARM_NEON Q2_K dot About the same performance as Q4_K. * A slightly faster ARM_NEON Q2_K dot Single token prediction is now ~36 ms on M2 Max. The code is much simpler too. * Fixed bug in Q2_K CUDA dot product kernel Stranegly enough, for the few prompts I tried with the 7B model the responses looked perfectly reasonable. Only realized something is not quite right when I tried the larger models and started getting nonse back. In any case, Q2_K single token evaluation time on an RTX 4080 in a Ryzen7950X box iusing CUDA and model fully loaded on the GPU are ~15.5 ms for 7B, ~25.4 ms for 13B, and ~55.8 ms for 30B. The max number of layers that fit in VRAM for The 65B is 32. With that, we get ~330 ms per token, which is not that much faster than just running on the CPU (~470 ms per token). * Don't print zeros/NaNs when no count histogram has been collected * A 10% faster CUDA vector dot kernel for Q3_K Q3_K is now running at ~18.5 ms / token on CUDA, so the gap to Q4_0 is only 10%. It seems memory acccess pattern is more important for performance than the amount of computation the kernel does. * A slightly daster Q4_K AVX2 dot product For perplexity, where we are less memory bound, time per pass drops by ~5%. Barely measurable difference for single token prediction. * A slightly faster ARM_NEON A4_K dot product * Minor * Fix quantization error test We cannot possibly be expecting rmse < 0.002 for 2- and 3-bit quantization variants. * Fix docker build I have been sloppy with vector reinterpret casts on ARM_NEON. It seems clang is very forgiving in that regard. * Added forgotten ggml.o dependence on k_quants.h to the Makefile * Had unintentionally committed the Makefile with -Ofast enabled * ggml : rename k_quants -> ggml-quants-k, use lowercase in code --------- Co-authored-by: Iwan Kawrakow <iwan.kawrakow@gmail.com> Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
2023-06-05 19:56:18 +00:00
const int nb = n / QK_K;
#ifdef __ARM_NEON
uint32_t aux[3];
uint32_t utmp[4];
const uint8x16_t m3b = vdupq_n_u8(0x3);
const int32x4_t vzero = vdupq_n_s32(0);
const uint8x16_t m0 = vdupq_n_u8(1);
const uint8x16_t m1 = vshlq_n_u8(m0, 1);
const uint8x16_t m2 = vshlq_n_u8(m0, 2);
const uint8x16_t m3 = vshlq_n_u8(m0, 3);
const int8_t m32 = 32;
ggml : sync (im2col, GPU conv, 32-bit arm compat) (#4060) ggml-ci
2023-11-13 14:55:52 +00:00
ggml_int8x16x4_t q3bytes;
ggml : add SOTA 2,3,4,5,6 bit k-quantizations (#1684) * Starting to add k-quantization to ggml I think it is better to have quantization separate from ggml. For now just adding the k-quants there, but it would be better to also factor out the existing ggml quantizations. * Adding Q3_K and Q8_K (de)-quantization * Q3_K now working on CUDA and AVX2/scalar CUDA is not ideal - ~50% slower than Q4_0 for single token prediction, about the same in batch mode (perplexity). CPU single token is ~55 ms (on Ryzen 7950X). * Some improvement for Q3_K on CUDA It is now ~22.5 ms/token on my GPU, so ~30% slower than Q4_0. * Some more CUDA optimizations for Q3_K Single token is now 20.5 ms/token (~20% slower than Q4_0). Perplexity is on par with Q4_0. * Adding Q4_K - scalar, AVX2, CUDA Performance is the same or perhaps very slightly better than Q4_0 on the CPU. On the GPU, single token prediction is ~10% better than Q4_0, batch mode (perplexity is about the same). * Adding Q6_K - scalar, AVX2, CUDA Performance is ~40% lower compared to Q4_K on the CPU. This is to be expected, considering that we are memory bound on the CPU and the 6-bit model is ~44% larger than the 4-bit. On the GPU, single token prediction is ~6% lower than Q4_0, batch mode (perplexity) is even closer (but still slower). * Adding Q5_K - scalar, AVX2, CUDA Performance is ~20% lower compared to Q4_K on the CPU. This is to be expected, considering that we are memory bound on the CPU and the 5-bit model is ~22% larger than the 4-bit. On the GPU, single token prediction is about the same as Q4_0 for both, single token and batch prediction. * Per convention, all QX_K quantizations use Q5_K for output.weight * Adding quantization mixes * Quantization mixes: didn't quite get what I wanted in the last commit * Q4_K dot product for ARM_NEON * Q6_K dot product for ARM_NEON * Q5_K dot product for ARM_NEON * Adding Q3_K dot for ARM_NEON It is 22% slower than Q4_K, despite the smaller model size. On x86_64, where we are memory bound, the Q3_K model is quite a bit faster than Q4_K. * A very slightly faster ARM_NEON Q3_K dot * Adding Q2_K - just CUDA for now Token prediction is pretty good - about 15.5 ms on a RTX 4080. Perplexity is about the same as Q4_K. * Adding scalar and AVX2 Q2_K dot * Adding ARM_NEON Q2_K dot About the same performance as Q4_K. * A slightly faster ARM_NEON Q2_K dot Single token prediction is now ~36 ms on M2 Max. The code is much simpler too. * Fixed bug in Q2_K CUDA dot product kernel Stranegly enough, for the few prompts I tried with the 7B model the responses looked perfectly reasonable. Only realized something is not quite right when I tried the larger models and started getting nonse back. In any case, Q2_K single token evaluation time on an RTX 4080 in a Ryzen7950X box iusing CUDA and model fully loaded on the GPU are ~15.5 ms for 7B, ~25.4 ms for 13B, and ~55.8 ms for 30B. The max number of layers that fit in VRAM for The 65B is 32. With that, we get ~330 ms per token, which is not that much faster than just running on the CPU (~470 ms per token). * Don't print zeros/NaNs when no count histogram has been collected * A 10% faster CUDA vector dot kernel for Q3_K Q3_K is now running at ~18.5 ms / token on CUDA, so the gap to Q4_0 is only 10%. It seems memory acccess pattern is more important for performance than the amount of computation the kernel does. * A slightly daster Q4_K AVX2 dot product For perplexity, where we are less memory bound, time per pass drops by ~5%. Barely measurable difference for single token prediction. * A slightly faster ARM_NEON A4_K dot product * Minor * Fix quantization error test We cannot possibly be expecting rmse < 0.002 for 2- and 3-bit quantization variants. * Fix docker build I have been sloppy with vector reinterpret casts on ARM_NEON. It seems clang is very forgiving in that regard. * Added forgotten ggml.o dependence on k_quants.h to the Makefile * Had unintentionally committed the Makefile with -Ofast enabled * ggml : rename k_quants -> ggml-quants-k, use lowercase in code --------- Co-authored-by: Iwan Kawrakow <iwan.kawrakow@gmail.com> Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
2023-06-05 19:56:18 +00:00
float sum = 0;
for (int i = 0; i < nb; ++i) {
ggml : move FP16 <-> FP32 code to ggml-impl.h (#3861) * ggml : move FP16 <-> FP32 stuff to ggml-impl.h ggml-ci * tests : fix ARM build * ggml : explicitly initialize deprecated type traits * ggml : add math.h to ggml-impl.h * ggml : remove duplicate static assert macros * ggml : prefix lookup tables with ggml_ ggml-ci * ggml-impl : move extern "C" to start of file
2023-10-30 17:19:15 +00:00
const float d = y[i].d * GGML_FP16_TO_FP32(x[i].d);
ggml : add SOTA 2,3,4,5,6 bit k-quantizations (#1684) * Starting to add k-quantization to ggml I think it is better to have quantization separate from ggml. For now just adding the k-quants there, but it would be better to also factor out the existing ggml quantizations. * Adding Q3_K and Q8_K (de)-quantization * Q3_K now working on CUDA and AVX2/scalar CUDA is not ideal - ~50% slower than Q4_0 for single token prediction, about the same in batch mode (perplexity). CPU single token is ~55 ms (on Ryzen 7950X). * Some improvement for Q3_K on CUDA It is now ~22.5 ms/token on my GPU, so ~30% slower than Q4_0. * Some more CUDA optimizations for Q3_K Single token is now 20.5 ms/token (~20% slower than Q4_0). Perplexity is on par with Q4_0. * Adding Q4_K - scalar, AVX2, CUDA Performance is the same or perhaps very slightly better than Q4_0 on the CPU. On the GPU, single token prediction is ~10% better than Q4_0, batch mode (perplexity is about the same). * Adding Q6_K - scalar, AVX2, CUDA Performance is ~40% lower compared to Q4_K on the CPU. This is to be expected, considering that we are memory bound on the CPU and the 6-bit model is ~44% larger than the 4-bit. On the GPU, single token prediction is ~6% lower than Q4_0, batch mode (perplexity) is even closer (but still slower). * Adding Q5_K - scalar, AVX2, CUDA Performance is ~20% lower compared to Q4_K on the CPU. This is to be expected, considering that we are memory bound on the CPU and the 5-bit model is ~22% larger than the 4-bit. On the GPU, single token prediction is about the same as Q4_0 for both, single token and batch prediction. * Per convention, all QX_K quantizations use Q5_K for output.weight * Adding quantization mixes * Quantization mixes: didn't quite get what I wanted in the last commit * Q4_K dot product for ARM_NEON * Q6_K dot product for ARM_NEON * Q5_K dot product for ARM_NEON * Adding Q3_K dot for ARM_NEON It is 22% slower than Q4_K, despite the smaller model size. On x86_64, where we are memory bound, the Q3_K model is quite a bit faster than Q4_K. * A very slightly faster ARM_NEON Q3_K dot * Adding Q2_K - just CUDA for now Token prediction is pretty good - about 15.5 ms on a RTX 4080. Perplexity is about the same as Q4_K. * Adding scalar and AVX2 Q2_K dot * Adding ARM_NEON Q2_K dot About the same performance as Q4_K. * A slightly faster ARM_NEON Q2_K dot Single token prediction is now ~36 ms on M2 Max. The code is much simpler too. * Fixed bug in Q2_K CUDA dot product kernel Stranegly enough, for the few prompts I tried with the 7B model the responses looked perfectly reasonable. Only realized something is not quite right when I tried the larger models and started getting nonse back. In any case, Q2_K single token evaluation time on an RTX 4080 in a Ryzen7950X box iusing CUDA and model fully loaded on the GPU are ~15.5 ms for 7B, ~25.4 ms for 13B, and ~55.8 ms for 30B. The max number of layers that fit in VRAM for The 65B is 32. With that, we get ~330 ms per token, which is not that much faster than just running on the CPU (~470 ms per token). * Don't print zeros/NaNs when no count histogram has been collected * A 10% faster CUDA vector dot kernel for Q3_K Q3_K is now running at ~18.5 ms / token on CUDA, so the gap to Q4_0 is only 10%. It seems memory acccess pattern is more important for performance than the amount of computation the kernel does. * A slightly daster Q4_K AVX2 dot product For perplexity, where we are less memory bound, time per pass drops by ~5%. Barely measurable difference for single token prediction. * A slightly faster ARM_NEON A4_K dot product * Minor * Fix quantization error test We cannot possibly be expecting rmse < 0.002 for 2- and 3-bit quantization variants. * Fix docker build I have been sloppy with vector reinterpret casts on ARM_NEON. It seems clang is very forgiving in that regard. * Added forgotten ggml.o dependence on k_quants.h to the Makefile * Had unintentionally committed the Makefile with -Ofast enabled * ggml : rename k_quants -> ggml-quants-k, use lowercase in code --------- Co-authored-by: Iwan Kawrakow <iwan.kawrakow@gmail.com> Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
2023-06-05 19:56:18 +00:00
const uint8_t * restrict q3 = x[i].qs;
const uint8_t * restrict qh = x[i].hmask;
const int8_t * restrict q8 = y[i].qs;
ggml : sync (im2col, GPU conv, 32-bit arm compat) (#4060) ggml-ci
2023-11-13 14:55:52 +00:00
ggml_uint8x16x2_t qhbits = ggml_vld1q_u8_x2(qh);
ggml : add SOTA 2,3,4,5,6 bit k-quantizations (#1684) * Starting to add k-quantization to ggml I think it is better to have quantization separate from ggml. For now just adding the k-quants there, but it would be better to also factor out the existing ggml quantizations. * Adding Q3_K and Q8_K (de)-quantization * Q3_K now working on CUDA and AVX2/scalar CUDA is not ideal - ~50% slower than Q4_0 for single token prediction, about the same in batch mode (perplexity). CPU single token is ~55 ms (on Ryzen 7950X). * Some improvement for Q3_K on CUDA It is now ~22.5 ms/token on my GPU, so ~30% slower than Q4_0. * Some more CUDA optimizations for Q3_K Single token is now 20.5 ms/token (~20% slower than Q4_0). Perplexity is on par with Q4_0. * Adding Q4_K - scalar, AVX2, CUDA Performance is the same or perhaps very slightly better than Q4_0 on the CPU. On the GPU, single token prediction is ~10% better than Q4_0, batch mode (perplexity is about the same). * Adding Q6_K - scalar, AVX2, CUDA Performance is ~40% lower compared to Q4_K on the CPU. This is to be expected, considering that we are memory bound on the CPU and the 6-bit model is ~44% larger than the 4-bit. On the GPU, single token prediction is ~6% lower than Q4_0, batch mode (perplexity) is even closer (but still slower). * Adding Q5_K - scalar, AVX2, CUDA Performance is ~20% lower compared to Q4_K on the CPU. This is to be expected, considering that we are memory bound on the CPU and the 5-bit model is ~22% larger than the 4-bit. On the GPU, single token prediction is about the same as Q4_0 for both, single token and batch prediction. * Per convention, all QX_K quantizations use Q5_K for output.weight * Adding quantization mixes * Quantization mixes: didn't quite get what I wanted in the last commit * Q4_K dot product for ARM_NEON * Q6_K dot product for ARM_NEON * Q5_K dot product for ARM_NEON * Adding Q3_K dot for ARM_NEON It is 22% slower than Q4_K, despite the smaller model size. On x86_64, where we are memory bound, the Q3_K model is quite a bit faster than Q4_K. * A very slightly faster ARM_NEON Q3_K dot * Adding Q2_K - just CUDA for now Token prediction is pretty good - about 15.5 ms on a RTX 4080. Perplexity is about the same as Q4_K. * Adding scalar and AVX2 Q2_K dot * Adding ARM_NEON Q2_K dot About the same performance as Q4_K. * A slightly faster ARM_NEON Q2_K dot Single token prediction is now ~36 ms on M2 Max. The code is much simpler too. * Fixed bug in Q2_K CUDA dot product kernel Stranegly enough, for the few prompts I tried with the 7B model the responses looked perfectly reasonable. Only realized something is not quite right when I tried the larger models and started getting nonse back. In any case, Q2_K single token evaluation time on an RTX 4080 in a Ryzen7950X box iusing CUDA and model fully loaded on the GPU are ~15.5 ms for 7B, ~25.4 ms for 13B, and ~55.8 ms for 30B. The max number of layers that fit in VRAM for The 65B is 32. With that, we get ~330 ms per token, which is not that much faster than just running on the CPU (~470 ms per token). * Don't print zeros/NaNs when no count histogram has been collected * A 10% faster CUDA vector dot kernel for Q3_K Q3_K is now running at ~18.5 ms / token on CUDA, so the gap to Q4_0 is only 10%. It seems memory acccess pattern is more important for performance than the amount of computation the kernel does. * A slightly daster Q4_K AVX2 dot product For perplexity, where we are less memory bound, time per pass drops by ~5%. Barely measurable difference for single token prediction. * A slightly faster ARM_NEON A4_K dot product * Minor * Fix quantization error test We cannot possibly be expecting rmse < 0.002 for 2- and 3-bit quantization variants. * Fix docker build I have been sloppy with vector reinterpret casts on ARM_NEON. It seems clang is very forgiving in that regard. * Added forgotten ggml.o dependence on k_quants.h to the Makefile * Had unintentionally committed the Makefile with -Ofast enabled * ggml : rename k_quants -> ggml-quants-k, use lowercase in code --------- Co-authored-by: Iwan Kawrakow <iwan.kawrakow@gmail.com> Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
2023-06-05 19:56:18 +00:00
ggml : sync (im2col, GPU conv, 32-bit arm compat) (#4060) ggml-ci
2023-11-13 14:55:52 +00:00
ggml_uint8x16x4_t q3h;
ggml : add SOTA 2,3,4,5,6 bit k-quantizations (#1684) * Starting to add k-quantization to ggml I think it is better to have quantization separate from ggml. For now just adding the k-quants there, but it would be better to also factor out the existing ggml quantizations. * Adding Q3_K and Q8_K (de)-quantization * Q3_K now working on CUDA and AVX2/scalar CUDA is not ideal - ~50% slower than Q4_0 for single token prediction, about the same in batch mode (perplexity). CPU single token is ~55 ms (on Ryzen 7950X). * Some improvement for Q3_K on CUDA It is now ~22.5 ms/token on my GPU, so ~30% slower than Q4_0. * Some more CUDA optimizations for Q3_K Single token is now 20.5 ms/token (~20% slower than Q4_0). Perplexity is on par with Q4_0. * Adding Q4_K - scalar, AVX2, CUDA Performance is the same or perhaps very slightly better than Q4_0 on the CPU. On the GPU, single token prediction is ~10% better than Q4_0, batch mode (perplexity is about the same). * Adding Q6_K - scalar, AVX2, CUDA Performance is ~40% lower compared to Q4_K on the CPU. This is to be expected, considering that we are memory bound on the CPU and the 6-bit model is ~44% larger than the 4-bit. On the GPU, single token prediction is ~6% lower than Q4_0, batch mode (perplexity) is even closer (but still slower). * Adding Q5_K - scalar, AVX2, CUDA Performance is ~20% lower compared to Q4_K on the CPU. This is to be expected, considering that we are memory bound on the CPU and the 5-bit model is ~22% larger than the 4-bit. On the GPU, single token prediction is about the same as Q4_0 for both, single token and batch prediction. * Per convention, all QX_K quantizations use Q5_K for output.weight * Adding quantization mixes * Quantization mixes: didn't quite get what I wanted in the last commit * Q4_K dot product for ARM_NEON * Q6_K dot product for ARM_NEON * Q5_K dot product for ARM_NEON * Adding Q3_K dot for ARM_NEON It is 22% slower than Q4_K, despite the smaller model size. On x86_64, where we are memory bound, the Q3_K model is quite a bit faster than Q4_K. * A very slightly faster ARM_NEON Q3_K dot * Adding Q2_K - just CUDA for now Token prediction is pretty good - about 15.5 ms on a RTX 4080. Perplexity is about the same as Q4_K. * Adding scalar and AVX2 Q2_K dot * Adding ARM_NEON Q2_K dot About the same performance as Q4_K. * A slightly faster ARM_NEON Q2_K dot Single token prediction is now ~36 ms on M2 Max. The code is much simpler too. * Fixed bug in Q2_K CUDA dot product kernel Stranegly enough, for the few prompts I tried with the 7B model the responses looked perfectly reasonable. Only realized something is not quite right when I tried the larger models and started getting nonse back. In any case, Q2_K single token evaluation time on an RTX 4080 in a Ryzen7950X box iusing CUDA and model fully loaded on the GPU are ~15.5 ms for 7B, ~25.4 ms for 13B, and ~55.8 ms for 30B. The max number of layers that fit in VRAM for The 65B is 32. With that, we get ~330 ms per token, which is not that much faster than just running on the CPU (~470 ms per token). * Don't print zeros/NaNs when no count histogram has been collected * A 10% faster CUDA vector dot kernel for Q3_K Q3_K is now running at ~18.5 ms / token on CUDA, so the gap to Q4_0 is only 10%. It seems memory acccess pattern is more important for performance than the amount of computation the kernel does. * A slightly daster Q4_K AVX2 dot product For perplexity, where we are less memory bound, time per pass drops by ~5%. Barely measurable difference for single token prediction. * A slightly faster ARM_NEON A4_K dot product * Minor * Fix quantization error test We cannot possibly be expecting rmse < 0.002 for 2- and 3-bit quantization variants. * Fix docker build I have been sloppy with vector reinterpret casts on ARM_NEON. It seems clang is very forgiving in that regard. * Added forgotten ggml.o dependence on k_quants.h to the Makefile * Had unintentionally committed the Makefile with -Ofast enabled * ggml : rename k_quants -> ggml-quants-k, use lowercase in code --------- Co-authored-by: Iwan Kawrakow <iwan.kawrakow@gmail.com> Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
2023-06-05 19:56:18 +00:00
int32_t isum = 0;
// Set up scales
memcpy(aux, x[i].scales, 12);
utmp[3] = ((aux[1] >> 4) & kmask2) | (((aux[2] >> 6) & kmask1) << 4);
utmp[2] = ((aux[0] >> 4) & kmask2) | (((aux[2] >> 4) & kmask1) << 4);
utmp[1] = (aux[1] & kmask2) | (((aux[2] >> 2) & kmask1) << 4);
utmp[0] = (aux[0] & kmask2) | (((aux[2] >> 0) & kmask1) << 4);
int8_t * scale = (int8_t *)utmp;
for (int j = 0; j < 16; ++j) scale[j] -= m32;
for (int j = 0; j < QK_K/128; ++j) {
ggml : sync (im2col, GPU conv, 32-bit arm compat) (#4060) ggml-ci
2023-11-13 14:55:52 +00:00
const ggml_uint8x16x2_t q3bits = ggml_vld1q_u8_x2(q3); q3 += 32;
const ggml_int8x16x4_t q8bytes_1 = ggml_vld1q_s8_x4(q8); q8 += 64;
const ggml_int8x16x4_t q8bytes_2 = ggml_vld1q_s8_x4(q8); q8 += 64;
ggml : add SOTA 2,3,4,5,6 bit k-quantizations (#1684) * Starting to add k-quantization to ggml I think it is better to have quantization separate from ggml. For now just adding the k-quants there, but it would be better to also factor out the existing ggml quantizations. * Adding Q3_K and Q8_K (de)-quantization * Q3_K now working on CUDA and AVX2/scalar CUDA is not ideal - ~50% slower than Q4_0 for single token prediction, about the same in batch mode (perplexity). CPU single token is ~55 ms (on Ryzen 7950X). * Some improvement for Q3_K on CUDA It is now ~22.5 ms/token on my GPU, so ~30% slower than Q4_0. * Some more CUDA optimizations for Q3_K Single token is now 20.5 ms/token (~20% slower than Q4_0). Perplexity is on par with Q4_0. * Adding Q4_K - scalar, AVX2, CUDA Performance is the same or perhaps very slightly better than Q4_0 on the CPU. On the GPU, single token prediction is ~10% better than Q4_0, batch mode (perplexity is about the same). * Adding Q6_K - scalar, AVX2, CUDA Performance is ~40% lower compared to Q4_K on the CPU. This is to be expected, considering that we are memory bound on the CPU and the 6-bit model is ~44% larger than the 4-bit. On the GPU, single token prediction is ~6% lower than Q4_0, batch mode (perplexity) is even closer (but still slower). * Adding Q5_K - scalar, AVX2, CUDA Performance is ~20% lower compared to Q4_K on the CPU. This is to be expected, considering that we are memory bound on the CPU and the 5-bit model is ~22% larger than the 4-bit. On the GPU, single token prediction is about the same as Q4_0 for both, single token and batch prediction. * Per convention, all QX_K quantizations use Q5_K for output.weight * Adding quantization mixes * Quantization mixes: didn't quite get what I wanted in the last commit * Q4_K dot product for ARM_NEON * Q6_K dot product for ARM_NEON * Q5_K dot product for ARM_NEON * Adding Q3_K dot for ARM_NEON It is 22% slower than Q4_K, despite the smaller model size. On x86_64, where we are memory bound, the Q3_K model is quite a bit faster than Q4_K. * A very slightly faster ARM_NEON Q3_K dot * Adding Q2_K - just CUDA for now Token prediction is pretty good - about 15.5 ms on a RTX 4080. Perplexity is about the same as Q4_K. * Adding scalar and AVX2 Q2_K dot * Adding ARM_NEON Q2_K dot About the same performance as Q4_K. * A slightly faster ARM_NEON Q2_K dot Single token prediction is now ~36 ms on M2 Max. The code is much simpler too. * Fixed bug in Q2_K CUDA dot product kernel Stranegly enough, for the few prompts I tried with the 7B model the responses looked perfectly reasonable. Only realized something is not quite right when I tried the larger models and started getting nonse back. In any case, Q2_K single token evaluation time on an RTX 4080 in a Ryzen7950X box iusing CUDA and model fully loaded on the GPU are ~15.5 ms for 7B, ~25.4 ms for 13B, and ~55.8 ms for 30B. The max number of layers that fit in VRAM for The 65B is 32. With that, we get ~330 ms per token, which is not that much faster than just running on the CPU (~470 ms per token). * Don't print zeros/NaNs when no count histogram has been collected * A 10% faster CUDA vector dot kernel for Q3_K Q3_K is now running at ~18.5 ms / token on CUDA, so the gap to Q4_0 is only 10%. It seems memory acccess pattern is more important for performance than the amount of computation the kernel does. * A slightly daster Q4_K AVX2 dot product For perplexity, where we are less memory bound, time per pass drops by ~5%. Barely measurable difference for single token prediction. * A slightly faster ARM_NEON A4_K dot product * Minor * Fix quantization error test We cannot possibly be expecting rmse < 0.002 for 2- and 3-bit quantization variants. * Fix docker build I have been sloppy with vector reinterpret casts on ARM_NEON. It seems clang is very forgiving in that regard. * Added forgotten ggml.o dependence on k_quants.h to the Makefile * Had unintentionally committed the Makefile with -Ofast enabled * ggml : rename k_quants -> ggml-quants-k, use lowercase in code --------- Co-authored-by: Iwan Kawrakow <iwan.kawrakow@gmail.com> Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
2023-06-05 19:56:18 +00:00
q3h.val[0] = vshlq_n_u8(vbicq_u8(m0, qhbits.val[0]), 2);
q3h.val[1] = vshlq_n_u8(vbicq_u8(m0, qhbits.val[1]), 2);
q3h.val[2] = vshlq_n_u8(vbicq_u8(m1, qhbits.val[0]), 1);
q3h.val[3] = vshlq_n_u8(vbicq_u8(m1, qhbits.val[1]), 1);
q3bytes.val[0] = vsubq_s8(vreinterpretq_s8_u8(vandq_u8(q3bits.val[0], m3b)), vreinterpretq_s8_u8(q3h.val[0]));
q3bytes.val[1] = vsubq_s8(vreinterpretq_s8_u8(vandq_u8(q3bits.val[1], m3b)), vreinterpretq_s8_u8(q3h.val[1]));
q3bytes.val[2] = vsubq_s8(vreinterpretq_s8_u8(vandq_u8(vshrq_n_u8(q3bits.val[0], 2), m3b)), vreinterpretq_s8_u8(q3h.val[2]));
q3bytes.val[3] = vsubq_s8(vreinterpretq_s8_u8(vandq_u8(vshrq_n_u8(q3bits.val[1], 2), m3b)), vreinterpretq_s8_u8(q3h.val[3]));
ggml : add ggml_vdotq_s32 alias (#4715) ggml-ci
2023-12-31 09:43:31 +00:00
isum += vaddvq_s32(ggml_vdotq_s32(vzero, q3bytes.val[0], q8bytes_1.val[0])) * scale[0];
isum += vaddvq_s32(ggml_vdotq_s32(vzero, q3bytes.val[1], q8bytes_1.val[1])) * scale[1];
isum += vaddvq_s32(ggml_vdotq_s32(vzero, q3bytes.val[2], q8bytes_1.val[2])) * scale[2];
isum += vaddvq_s32(ggml_vdotq_s32(vzero, q3bytes.val[3], q8bytes_1.val[3])) * scale[3];
ggml : fix dot product for ARM (#4630) ggml-ci
2023-12-27 09:02:13 +00:00
ggml : add SOTA 2,3,4,5,6 bit k-quantizations (#1684) * Starting to add k-quantization to ggml I think it is better to have quantization separate from ggml. For now just adding the k-quants there, but it would be better to also factor out the existing ggml quantizations. * Adding Q3_K and Q8_K (de)-quantization * Q3_K now working on CUDA and AVX2/scalar CUDA is not ideal - ~50% slower than Q4_0 for single token prediction, about the same in batch mode (perplexity). CPU single token is ~55 ms (on Ryzen 7950X). * Some improvement for Q3_K on CUDA It is now ~22.5 ms/token on my GPU, so ~30% slower than Q4_0. * Some more CUDA optimizations for Q3_K Single token is now 20.5 ms/token (~20% slower than Q4_0). Perplexity is on par with Q4_0. * Adding Q4_K - scalar, AVX2, CUDA Performance is the same or perhaps very slightly better than Q4_0 on the CPU. On the GPU, single token prediction is ~10% better than Q4_0, batch mode (perplexity is about the same). * Adding Q6_K - scalar, AVX2, CUDA Performance is ~40% lower compared to Q4_K on the CPU. This is to be expected, considering that we are memory bound on the CPU and the 6-bit model is ~44% larger than the 4-bit. On the GPU, single token prediction is ~6% lower than Q4_0, batch mode (perplexity) is even closer (but still slower). * Adding Q5_K - scalar, AVX2, CUDA Performance is ~20% lower compared to Q4_K on the CPU. This is to be expected, considering that we are memory bound on the CPU and the 5-bit model is ~22% larger than the 4-bit. On the GPU, single token prediction is about the same as Q4_0 for both, single token and batch prediction. * Per convention, all QX_K quantizations use Q5_K for output.weight * Adding quantization mixes * Quantization mixes: didn't quite get what I wanted in the last commit * Q4_K dot product for ARM_NEON * Q6_K dot product for ARM_NEON * Q5_K dot product for ARM_NEON * Adding Q3_K dot for ARM_NEON It is 22% slower than Q4_K, despite the smaller model size. On x86_64, where we are memory bound, the Q3_K model is quite a bit faster than Q4_K. * A very slightly faster ARM_NEON Q3_K dot * Adding Q2_K - just CUDA for now Token prediction is pretty good - about 15.5 ms on a RTX 4080. Perplexity is about the same as Q4_K. * Adding scalar and AVX2 Q2_K dot * Adding ARM_NEON Q2_K dot About the same performance as Q4_K. * A slightly faster ARM_NEON Q2_K dot Single token prediction is now ~36 ms on M2 Max. The code is much simpler too. * Fixed bug in Q2_K CUDA dot product kernel Stranegly enough, for the few prompts I tried with the 7B model the responses looked perfectly reasonable. Only realized something is not quite right when I tried the larger models and started getting nonse back. In any case, Q2_K single token evaluation time on an RTX 4080 in a Ryzen7950X box iusing CUDA and model fully loaded on the GPU are ~15.5 ms for 7B, ~25.4 ms for 13B, and ~55.8 ms for 30B. The max number of layers that fit in VRAM for The 65B is 32. With that, we get ~330 ms per token, which is not that much faster than just running on the CPU (~470 ms per token). * Don't print zeros/NaNs when no count histogram has been collected * A 10% faster CUDA vector dot kernel for Q3_K Q3_K is now running at ~18.5 ms / token on CUDA, so the gap to Q4_0 is only 10%. It seems memory acccess pattern is more important for performance than the amount of computation the kernel does. * A slightly daster Q4_K AVX2 dot product For perplexity, where we are less memory bound, time per pass drops by ~5%. Barely measurable difference for single token prediction. * A slightly faster ARM_NEON A4_K dot product * Minor * Fix quantization error test We cannot possibly be expecting rmse < 0.002 for 2- and 3-bit quantization variants. * Fix docker build I have been sloppy with vector reinterpret casts on ARM_NEON. It seems clang is very forgiving in that regard. * Added forgotten ggml.o dependence on k_quants.h to the Makefile * Had unintentionally committed the Makefile with -Ofast enabled * ggml : rename k_quants -> ggml-quants-k, use lowercase in code --------- Co-authored-by: Iwan Kawrakow <iwan.kawrakow@gmail.com> Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
2023-06-05 19:56:18 +00:00
scale += 4;
q3h.val[0] = vbicq_u8(m2, qhbits.val[0]);
q3h.val[1] = vbicq_u8(m2, qhbits.val[1]);
q3h.val[2] = vshrq_n_u8(vbicq_u8(m3, qhbits.val[0]), 1);
q3h.val[3] = vshrq_n_u8(vbicq_u8(m3, qhbits.val[1]), 1);
q3bytes.val[0] = vsubq_s8(vreinterpretq_s8_u8(vandq_u8(vshrq_n_u8(q3bits.val[0], 4), m3b)), vreinterpretq_s8_u8(q3h.val[0]));
q3bytes.val[1] = vsubq_s8(vreinterpretq_s8_u8(vandq_u8(vshrq_n_u8(q3bits.val[1], 4), m3b)), vreinterpretq_s8_u8(q3h.val[1]));
q3bytes.val[2] = vsubq_s8(vreinterpretq_s8_u8(vandq_u8(vshrq_n_u8(q3bits.val[0], 6), m3b)), vreinterpretq_s8_u8(q3h.val[2]));
q3bytes.val[3] = vsubq_s8(vreinterpretq_s8_u8(vandq_u8(vshrq_n_u8(q3bits.val[1], 6), m3b)), vreinterpretq_s8_u8(q3h.val[3]));
ggml : add ggml_vdotq_s32 alias (#4715) ggml-ci
2023-12-31 09:43:31 +00:00
isum += vaddvq_s32(ggml_vdotq_s32(vzero, q3bytes.val[0], q8bytes_2.val[0])) * scale[0];
isum += vaddvq_s32(ggml_vdotq_s32(vzero, q3bytes.val[1], q8bytes_2.val[1])) * scale[1];
isum += vaddvq_s32(ggml_vdotq_s32(vzero, q3bytes.val[2], q8bytes_2.val[2])) * scale[2];
isum += vaddvq_s32(ggml_vdotq_s32(vzero, q3bytes.val[3], q8bytes_2.val[3])) * scale[3];
ggml : fix dot product for ARM (#4630) ggml-ci
2023-12-27 09:02:13 +00:00
ggml : add SOTA 2,3,4,5,6 bit k-quantizations (#1684) * Starting to add k-quantization to ggml I think it is better to have quantization separate from ggml. For now just adding the k-quants there, but it would be better to also factor out the existing ggml quantizations. * Adding Q3_K and Q8_K (de)-quantization * Q3_K now working on CUDA and AVX2/scalar CUDA is not ideal - ~50% slower than Q4_0 for single token prediction, about the same in batch mode (perplexity). CPU single token is ~55 ms (on Ryzen 7950X). * Some improvement for Q3_K on CUDA It is now ~22.5 ms/token on my GPU, so ~30% slower than Q4_0. * Some more CUDA optimizations for Q3_K Single token is now 20.5 ms/token (~20% slower than Q4_0). Perplexity is on par with Q4_0. * Adding Q4_K - scalar, AVX2, CUDA Performance is the same or perhaps very slightly better than Q4_0 on the CPU. On the GPU, single token prediction is ~10% better than Q4_0, batch mode (perplexity is about the same). * Adding Q6_K - scalar, AVX2, CUDA Performance is ~40% lower compared to Q4_K on the CPU. This is to be expected, considering that we are memory bound on the CPU and the 6-bit model is ~44% larger than the 4-bit. On the GPU, single token prediction is ~6% lower than Q4_0, batch mode (perplexity) is even closer (but still slower). * Adding Q5_K - scalar, AVX2, CUDA Performance is ~20% lower compared to Q4_K on the CPU. This is to be expected, considering that we are memory bound on the CPU and the 5-bit model is ~22% larger than the 4-bit. On the GPU, single token prediction is about the same as Q4_0 for both, single token and batch prediction. * Per convention, all QX_K quantizations use Q5_K for output.weight * Adding quantization mixes * Quantization mixes: didn't quite get what I wanted in the last commit * Q4_K dot product for ARM_NEON * Q6_K dot product for ARM_NEON * Q5_K dot product for ARM_NEON * Adding Q3_K dot for ARM_NEON It is 22% slower than Q4_K, despite the smaller model size. On x86_64, where we are memory bound, the Q3_K model is quite a bit faster than Q4_K. * A very slightly faster ARM_NEON Q3_K dot * Adding Q2_K - just CUDA for now Token prediction is pretty good - about 15.5 ms on a RTX 4080. Perplexity is about the same as Q4_K. * Adding scalar and AVX2 Q2_K dot * Adding ARM_NEON Q2_K dot About the same performance as Q4_K. * A slightly faster ARM_NEON Q2_K dot Single token prediction is now ~36 ms on M2 Max. The code is much simpler too. * Fixed bug in Q2_K CUDA dot product kernel Stranegly enough, for the few prompts I tried with the 7B model the responses looked perfectly reasonable. Only realized something is not quite right when I tried the larger models and started getting nonse back. In any case, Q2_K single token evaluation time on an RTX 4080 in a Ryzen7950X box iusing CUDA and model fully loaded on the GPU are ~15.5 ms for 7B, ~25.4 ms for 13B, and ~55.8 ms for 30B. The max number of layers that fit in VRAM for The 65B is 32. With that, we get ~330 ms per token, which is not that much faster than just running on the CPU (~470 ms per token). * Don't print zeros/NaNs when no count histogram has been collected * A 10% faster CUDA vector dot kernel for Q3_K Q3_K is now running at ~18.5 ms / token on CUDA, so the gap to Q4_0 is only 10%. It seems memory acccess pattern is more important for performance than the amount of computation the kernel does. * A slightly daster Q4_K AVX2 dot product For perplexity, where we are less memory bound, time per pass drops by ~5%. Barely measurable difference for single token prediction. * A slightly faster ARM_NEON A4_K dot product * Minor * Fix quantization error test We cannot possibly be expecting rmse < 0.002 for 2- and 3-bit quantization variants. * Fix docker build I have been sloppy with vector reinterpret casts on ARM_NEON. It seems clang is very forgiving in that regard. * Added forgotten ggml.o dependence on k_quants.h to the Makefile * Had unintentionally committed the Makefile with -Ofast enabled * ggml : rename k_quants -> ggml-quants-k, use lowercase in code --------- Co-authored-by: Iwan Kawrakow <iwan.kawrakow@gmail.com> Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
2023-06-05 19:56:18 +00:00
scale += 4;
if (j == 0) {
qhbits.val[0] = vshrq_n_u8(qhbits.val[0], 4);
qhbits.val[1] = vshrq_n_u8(qhbits.val[1], 4);
}
}
sum += d * isum;
}
*s = sum;
#elif defined __AVX2__
const __m256i m3 = _mm256_set1_epi8(3);
const __m256i mone = _mm256_set1_epi8(1);
const __m128i m32 = _mm_set1_epi8(32);
__m256 acc = _mm256_setzero_ps();
uint32_t aux[3];
for (int i = 0; i < nb; ++i) {
ggml : move FP16 <-> FP32 code to ggml-impl.h (#3861) * ggml : move FP16 <-> FP32 stuff to ggml-impl.h ggml-ci * tests : fix ARM build * ggml : explicitly initialize deprecated type traits * ggml : add math.h to ggml-impl.h * ggml : remove duplicate static assert macros * ggml : prefix lookup tables with ggml_ ggml-ci * ggml-impl : move extern "C" to start of file
2023-10-30 17:19:15 +00:00
const float d = y[i].d * GGML_FP16_TO_FP32(x[i].d);
ggml : add SOTA 2,3,4,5,6 bit k-quantizations (#1684) * Starting to add k-quantization to ggml I think it is better to have quantization separate from ggml. For now just adding the k-quants there, but it would be better to also factor out the existing ggml quantizations. * Adding Q3_K and Q8_K (de)-quantization * Q3_K now working on CUDA and AVX2/scalar CUDA is not ideal - ~50% slower than Q4_0 for single token prediction, about the same in batch mode (perplexity). CPU single token is ~55 ms (on Ryzen 7950X). * Some improvement for Q3_K on CUDA It is now ~22.5 ms/token on my GPU, so ~30% slower than Q4_0. * Some more CUDA optimizations for Q3_K Single token is now 20.5 ms/token (~20% slower than Q4_0). Perplexity is on par with Q4_0. * Adding Q4_K - scalar, AVX2, CUDA Performance is the same or perhaps very slightly better than Q4_0 on the CPU. On the GPU, single token prediction is ~10% better than Q4_0, batch mode (perplexity is about the same). * Adding Q6_K - scalar, AVX2, CUDA Performance is ~40% lower compared to Q4_K on the CPU. This is to be expected, considering that we are memory bound on the CPU and the 6-bit model is ~44% larger than the 4-bit. On the GPU, single token prediction is ~6% lower than Q4_0, batch mode (perplexity) is even closer (but still slower). * Adding Q5_K - scalar, AVX2, CUDA Performance is ~20% lower compared to Q4_K on the CPU. This is to be expected, considering that we are memory bound on the CPU and the 5-bit model is ~22% larger than the 4-bit. On the GPU, single token prediction is about the same as Q4_0 for both, single token and batch prediction. * Per convention, all QX_K quantizations use Q5_K for output.weight * Adding quantization mixes * Quantization mixes: didn't quite get what I wanted in the last commit * Q4_K dot product for ARM_NEON * Q6_K dot product for ARM_NEON * Q5_K dot product for ARM_NEON * Adding Q3_K dot for ARM_NEON It is 22% slower than Q4_K, despite the smaller model size. On x86_64, where we are memory bound, the Q3_K model is quite a bit faster than Q4_K. * A very slightly faster ARM_NEON Q3_K dot * Adding Q2_K - just CUDA for now Token prediction is pretty good - about 15.5 ms on a RTX 4080. Perplexity is about the same as Q4_K. * Adding scalar and AVX2 Q2_K dot * Adding ARM_NEON Q2_K dot About the same performance as Q4_K. * A slightly faster ARM_NEON Q2_K dot Single token prediction is now ~36 ms on M2 Max. The code is much simpler too. * Fixed bug in Q2_K CUDA dot product kernel Stranegly enough, for the few prompts I tried with the 7B model the responses looked perfectly reasonable. Only realized something is not quite right when I tried the larger models and started getting nonse back. In any case, Q2_K single token evaluation time on an RTX 4080 in a Ryzen7950X box iusing CUDA and model fully loaded on the GPU are ~15.5 ms for 7B, ~25.4 ms for 13B, and ~55.8 ms for 30B. The max number of layers that fit in VRAM for The 65B is 32. With that, we get ~330 ms per token, which is not that much faster than just running on the CPU (~470 ms per token). * Don't print zeros/NaNs when no count histogram has been collected * A 10% faster CUDA vector dot kernel for Q3_K Q3_K is now running at ~18.5 ms / token on CUDA, so the gap to Q4_0 is only 10%. It seems memory acccess pattern is more important for performance than the amount of computation the kernel does. * A slightly daster Q4_K AVX2 dot product For perplexity, where we are less memory bound, time per pass drops by ~5%. Barely measurable difference for single token prediction. * A slightly faster ARM_NEON A4_K dot product * Minor * Fix quantization error test We cannot possibly be expecting rmse < 0.002 for 2- and 3-bit quantization variants. * Fix docker build I have been sloppy with vector reinterpret casts on ARM_NEON. It seems clang is very forgiving in that regard. * Added forgotten ggml.o dependence on k_quants.h to the Makefile * Had unintentionally committed the Makefile with -Ofast enabled * ggml : rename k_quants -> ggml-quants-k, use lowercase in code --------- Co-authored-by: Iwan Kawrakow <iwan.kawrakow@gmail.com> Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
2023-06-05 19:56:18 +00:00
const uint8_t * restrict q3 = x[i].qs;
const int8_t * restrict q8 = y[i].qs;
// Set up scales
memcpy(aux, x[i].scales, 12);
__m128i scales128 = _mm_set_epi32(
((aux[1] >> 4) & kmask2) | (((aux[2] >> 6) & kmask1) << 4),
((aux[0] >> 4) & kmask2) | (((aux[2] >> 4) & kmask1) << 4),
(aux[1] & kmask2) | (((aux[2] >> 2) & kmask1) << 4),
(aux[0] & kmask2) | (((aux[2] >> 0) & kmask1) << 4));
scales128 = _mm_sub_epi8(scales128, m32);
const __m256i all_scales = _mm256_cvtepi8_epi16(scales128);
const __m128i l_scales = _mm256_extracti128_si256(all_scales, 0);
const __m128i h_scales = _mm256_extracti128_si256(all_scales, 1);
ggml : workaround for missing _mm256_setr_m128i in GCC < 8 in k_quants.c (#2405)
2023-07-28 18:17:45 +00:00
const __m256i scales[2] = {MM256_SET_M128I(l_scales, l_scales), MM256_SET_M128I(h_scales, h_scales)};
ggml : add SOTA 2,3,4,5,6 bit k-quantizations (#1684) * Starting to add k-quantization to ggml I think it is better to have quantization separate from ggml. For now just adding the k-quants there, but it would be better to also factor out the existing ggml quantizations. * Adding Q3_K and Q8_K (de)-quantization * Q3_K now working on CUDA and AVX2/scalar CUDA is not ideal - ~50% slower than Q4_0 for single token prediction, about the same in batch mode (perplexity). CPU single token is ~55 ms (on Ryzen 7950X). * Some improvement for Q3_K on CUDA It is now ~22.5 ms/token on my GPU, so ~30% slower than Q4_0. * Some more CUDA optimizations for Q3_K Single token is now 20.5 ms/token (~20% slower than Q4_0). Perplexity is on par with Q4_0. * Adding Q4_K - scalar, AVX2, CUDA Performance is the same or perhaps very slightly better than Q4_0 on the CPU. On the GPU, single token prediction is ~10% better than Q4_0, batch mode (perplexity is about the same). * Adding Q6_K - scalar, AVX2, CUDA Performance is ~40% lower compared to Q4_K on the CPU. This is to be expected, considering that we are memory bound on the CPU and the 6-bit model is ~44% larger than the 4-bit. On the GPU, single token prediction is ~6% lower than Q4_0, batch mode (perplexity) is even closer (but still slower). * Adding Q5_K - scalar, AVX2, CUDA Performance is ~20% lower compared to Q4_K on the CPU. This is to be expected, considering that we are memory bound on the CPU and the 5-bit model is ~22% larger than the 4-bit. On the GPU, single token prediction is about the same as Q4_0 for both, single token and batch prediction. * Per convention, all QX_K quantizations use Q5_K for output.weight * Adding quantization mixes * Quantization mixes: didn't quite get what I wanted in the last commit * Q4_K dot product for ARM_NEON * Q6_K dot product for ARM_NEON * Q5_K dot product for ARM_NEON * Adding Q3_K dot for ARM_NEON It is 22% slower than Q4_K, despite the smaller model size. On x86_64, where we are memory bound, the Q3_K model is quite a bit faster than Q4_K. * A very slightly faster ARM_NEON Q3_K dot * Adding Q2_K - just CUDA for now Token prediction is pretty good - about 15.5 ms on a RTX 4080. Perplexity is about the same as Q4_K. * Adding scalar and AVX2 Q2_K dot * Adding ARM_NEON Q2_K dot About the same performance as Q4_K. * A slightly faster ARM_NEON Q2_K dot Single token prediction is now ~36 ms on M2 Max. The code is much simpler too. * Fixed bug in Q2_K CUDA dot product kernel Stranegly enough, for the few prompts I tried with the 7B model the responses looked perfectly reasonable. Only realized something is not quite right when I tried the larger models and started getting nonse back. In any case, Q2_K single token evaluation time on an RTX 4080 in a Ryzen7950X box iusing CUDA and model fully loaded on the GPU are ~15.5 ms for 7B, ~25.4 ms for 13B, and ~55.8 ms for 30B. The max number of layers that fit in VRAM for The 65B is 32. With that, we get ~330 ms per token, which is not that much faster than just running on the CPU (~470 ms per token). * Don't print zeros/NaNs when no count histogram has been collected * A 10% faster CUDA vector dot kernel for Q3_K Q3_K is now running at ~18.5 ms / token on CUDA, so the gap to Q4_0 is only 10%. It seems memory acccess pattern is more important for performance than the amount of computation the kernel does. * A slightly daster Q4_K AVX2 dot product For perplexity, where we are less memory bound, time per pass drops by ~5%. Barely measurable difference for single token prediction. * A slightly faster ARM_NEON A4_K dot product * Minor * Fix quantization error test We cannot possibly be expecting rmse < 0.002 for 2- and 3-bit quantization variants. * Fix docker build I have been sloppy with vector reinterpret casts on ARM_NEON. It seems clang is very forgiving in that regard. * Added forgotten ggml.o dependence on k_quants.h to the Makefile * Had unintentionally committed the Makefile with -Ofast enabled * ggml : rename k_quants -> ggml-quants-k, use lowercase in code --------- Co-authored-by: Iwan Kawrakow <iwan.kawrakow@gmail.com> Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
2023-06-05 19:56:18 +00:00
// high bit
const __m256i hbits = _mm256_loadu_si256((const __m256i*)x[i].hmask);
// integer accumulator
__m256i sumi = _mm256_setzero_si256();
int bit = 0;
int is = 0;
for (int j = 0; j < QK_K/128; ++j) {
// load low 2 bits
const __m256i q3bits = _mm256_loadu_si256((const __m256i*)q3); q3 += 32;
// prepare low and high bits
const __m256i q3l_0 = _mm256_and_si256(q3bits, m3);
const __m256i q3h_0 = _mm256_slli_epi16(_mm256_srli_epi16(_mm256_andnot_si256(hbits, _mm256_slli_epi16(mone, bit)), bit), 2);
++bit;
const __m256i q3l_1 = _mm256_and_si256(_mm256_srli_epi16(q3bits, 2), m3);
const __m256i q3h_1 = _mm256_slli_epi16(_mm256_srli_epi16(_mm256_andnot_si256(hbits, _mm256_slli_epi16(mone, bit)), bit), 2);
++bit;
const __m256i q3l_2 = _mm256_and_si256(_mm256_srli_epi16(q3bits, 4), m3);
const __m256i q3h_2 = _mm256_slli_epi16(_mm256_srli_epi16(_mm256_andnot_si256(hbits, _mm256_slli_epi16(mone, bit)), bit), 2);
++bit;
const __m256i q3l_3 = _mm256_and_si256(_mm256_srli_epi16(q3bits, 6), m3);
const __m256i q3h_3 = _mm256_slli_epi16(_mm256_srli_epi16(_mm256_andnot_si256(hbits, _mm256_slli_epi16(mone, bit)), bit), 2);
++bit;
// load Q8 quants
const __m256i q8_0 = _mm256_loadu_si256((const __m256i*)q8); q8 += 32;
const __m256i q8_1 = _mm256_loadu_si256((const __m256i*)q8); q8 += 32;
const __m256i q8_2 = _mm256_loadu_si256((const __m256i*)q8); q8 += 32;
const __m256i q8_3 = _mm256_loadu_si256((const __m256i*)q8); q8 += 32;
// Dot product: we multiply the 2 low bits and 1 high bit part separately, so we can use _mm256_maddubs_epi16,
// and then subtract. The high bit part has the 2 already subtracted (and so, it is zero if the high bit was not set,
// and 2 if the high bit was set)
__m256i q8s_0 = _mm256_maddubs_epi16(q3h_0, q8_0);
__m256i q8s_1 = _mm256_maddubs_epi16(q3h_1, q8_1);
__m256i q8s_2 = _mm256_maddubs_epi16(q3h_2, q8_2);
__m256i q8s_3 = _mm256_maddubs_epi16(q3h_3, q8_3);
__m256i p16_0 = _mm256_maddubs_epi16(q3l_0, q8_0);
__m256i p16_1 = _mm256_maddubs_epi16(q3l_1, q8_1);
__m256i p16_2 = _mm256_maddubs_epi16(q3l_2, q8_2);
__m256i p16_3 = _mm256_maddubs_epi16(q3l_3, q8_3);
p16_0 = _mm256_sub_epi16(p16_0, q8s_0);
p16_1 = _mm256_sub_epi16(p16_1, q8s_1);
p16_2 = _mm256_sub_epi16(p16_2, q8s_2);
p16_3 = _mm256_sub_epi16(p16_3, q8s_3);
// multiply with scales
p16_0 = _mm256_madd_epi16(_mm256_shuffle_epi8(scales[j], get_scale_shuffle_q3k(is + 0)), p16_0);
p16_1 = _mm256_madd_epi16(_mm256_shuffle_epi8(scales[j], get_scale_shuffle_q3k(is + 1)), p16_1);
p16_2 = _mm256_madd_epi16(_mm256_shuffle_epi8(scales[j], get_scale_shuffle_q3k(is + 2)), p16_2);
p16_3 = _mm256_madd_epi16(_mm256_shuffle_epi8(scales[j], get_scale_shuffle_q3k(is + 3)), p16_3);
// accumulate
p16_0 = _mm256_add_epi32(p16_0, p16_1);
p16_2 = _mm256_add_epi32(p16_2, p16_3);
sumi = _mm256_add_epi32(sumi, _mm256_add_epi32(p16_0, p16_2));
}
// multiply with block scale and accumulate
acc = _mm256_fmadd_ps(_mm256_broadcast_ss(&d), _mm256_cvtepi32_ps(sumi), acc);
}
*s = hsum_float_8(acc);
k-quants : add AVX support to dot functions (#1916) * k_quants : add AVX support * k_quants : apply review comments
2023-06-26 16:46:07 +00:00
#elif defined __AVX__
const __m128i m3 = _mm_set1_epi8(3);
const __m128i mone = _mm_set1_epi8(1);
const __m128i m32 = _mm_set1_epi8(32);
const __m128i m2 = _mm_set1_epi8(2);
__m256 acc = _mm256_setzero_ps();
build : fix most gcc and clang warnings (#2861) * fix most gcc and clang warnings * baby-llama : remove commented opt_params_adam * fix some MinGW warnings * fix more MinGW warnings
2023-09-01 13:34:50 +00:00
const uint32_t *aux;
k-quants : add AVX support to dot functions (#1916) * k_quants : add AVX support * k_quants : apply review comments
2023-06-26 16:46:07 +00:00
for (int i = 0; i < nb; ++i) {
ggml : move FP16 <-> FP32 code to ggml-impl.h (#3861) * ggml : move FP16 <-> FP32 stuff to ggml-impl.h ggml-ci * tests : fix ARM build * ggml : explicitly initialize deprecated type traits * ggml : add math.h to ggml-impl.h * ggml : remove duplicate static assert macros * ggml : prefix lookup tables with ggml_ ggml-ci * ggml-impl : move extern "C" to start of file
2023-10-30 17:19:15 +00:00
const float d = y[i].d * GGML_FP16_TO_FP32(x[i].d);
k-quants : add AVX support to dot functions (#1916) * k_quants : add AVX support * k_quants : apply review comments
2023-06-26 16:46:07 +00:00
const uint8_t * restrict q3 = x[i].qs;
const int8_t * restrict q8 = y[i].qs;
// Set up scales
build : fix most gcc and clang warnings (#2861) * fix most gcc and clang warnings * baby-llama : remove commented opt_params_adam * fix some MinGW warnings * fix more MinGW warnings
2023-09-01 13:34:50 +00:00
aux = (const uint32_t *)x[i].scales;
k-quants : add AVX support to dot functions (#1916) * k_quants : add AVX support * k_quants : apply review comments
2023-06-26 16:46:07 +00:00
__m128i scales128 = _mm_set_epi32(
((aux[1] >> 4) & kmask2) | (((aux[2] >> 6) & kmask1) << 4),
((aux[0] >> 4) & kmask2) | (((aux[2] >> 4) & kmask1) << 4),
(aux[1] & kmask2) | (((aux[2] >> 2) & kmask1) << 4),
(aux[0] & kmask2) | (((aux[2] >> 0) & kmask1) << 4));
scales128 = _mm_sub_epi8(scales128, m32);
const __m128i scales_0 = _mm_cvtepi8_epi16(scales128);
const __m128i scales_1 = _mm_cvtepi8_epi16(_mm_unpackhi_epi64(scales128, scales128));
const __m128i scales[2] = { scales_0, scales_1 };
// high bit *128*2 from block_q3_K.hmask[QK_K/8]
const __m128i hbits_0 = _mm_loadu_si128((const __m128i*)&x[i].hmask[0]);
const __m128i hbits_1 = _mm_loadu_si128((const __m128i*)&x[i].hmask[16]);
// integer accumulator
__m128i sumi_0 = _mm_setzero_si128();
__m128i sumi_1 = _mm_setzero_si128();
for (int j = 0; j < QK_K/128; ++j) {
// load low 2 bits *64*2 from block_q3_K.qs[QK_K/4]
const __m128i q3bits_0 = _mm_loadu_si128((const __m128i*)q3); q3 += 16;
const __m128i q3bits_1 = _mm_loadu_si128((const __m128i*)q3); q3 += 16;
// prepare low and high bits
k-quants : fix indentation
2023-06-26 17:10:52 +00:00
const int bit = j << 2;
k-quants : add AVX support to dot functions (#1916) * k_quants : add AVX support * k_quants : apply review comments
2023-06-26 16:46:07 +00:00
const __m128i q3l_0 = _mm_and_si128(q3bits_0, m3);
const __m128i q3l_1 = _mm_and_si128(q3bits_1, m3);
const __m128i q3h_0 = _mm_slli_epi16(_mm_srli_epi16(_mm_andnot_si128(hbits_0, _mm_slli_epi16(mone, bit)), bit), 2);
const __m128i q3h_1 = _mm_slli_epi16(_mm_srli_epi16(_mm_andnot_si128(hbits_1, _mm_slli_epi16(mone, bit)), bit), 2);
const __m128i q3l_2 = _mm_and_si128(_mm_srli_epi16(q3bits_0, 2), m3);
const __m128i q3l_3 = _mm_and_si128(_mm_srli_epi16(q3bits_1, 2), m3);
const __m128i q3h_2 = _mm_slli_epi16(_mm_srli_epi16(_mm_andnot_si128(hbits_0, _mm_slli_epi16(mone, bit+1)), bit+1), 2);
const __m128i q3h_3 = _mm_slli_epi16(_mm_srli_epi16(_mm_andnot_si128(hbits_1, _mm_slli_epi16(mone, bit+1)), bit+1), 2);
const __m128i q3l_4 = _mm_and_si128(_mm_srli_epi16(q3bits_0, 4), m3);
const __m128i q3l_5 = _mm_and_si128(_mm_srli_epi16(q3bits_1, 4), m3);
const __m128i q3h_4 = _mm_slli_epi16(_mm_srli_epi16(_mm_andnot_si128(hbits_0, _mm_slli_epi16(mone, bit+2)), bit+2), 2);
const __m128i q3h_5 = _mm_slli_epi16(_mm_srli_epi16(_mm_andnot_si128(hbits_1, _mm_slli_epi16(mone, bit+2)), bit+2), 2);
const __m128i q3l_6 = _mm_and_si128(_mm_srli_epi16(q3bits_0, 6), m3);
const __m128i q3l_7 = _mm_and_si128(_mm_srli_epi16(q3bits_1, 6), m3);
const __m128i q3h_6 = _mm_slli_epi16(_mm_srli_epi16(_mm_andnot_si128(hbits_0, _mm_slli_epi16(mone, bit+3)), bit+3), 2);
const __m128i q3h_7 = _mm_slli_epi16(_mm_srli_epi16(_mm_andnot_si128(hbits_1, _mm_slli_epi16(mone, bit+3)), bit+3), 2);
// load Q8 quants from block_q8_K.qs[QK_K]
const __m128i q8_0 = _mm_loadu_si128((const __m128i*)q8); q8 += 16;
const __m128i q8_1 = _mm_loadu_si128((const __m128i*)q8); q8 += 16;
const __m128i q8_2 = _mm_loadu_si128((const __m128i*)q8); q8 += 16;
const __m128i q8_3 = _mm_loadu_si128((const __m128i*)q8); q8 += 16;
const __m128i q8_4 = _mm_loadu_si128((const __m128i*)q8); q8 += 16;
const __m128i q8_5 = _mm_loadu_si128((const __m128i*)q8); q8 += 16;
const __m128i q8_6 = _mm_loadu_si128((const __m128i*)q8); q8 += 16;
const __m128i q8_7 = _mm_loadu_si128((const __m128i*)q8); q8 += 16;
// Dot product: we multiply the 2 low bits and 1 high bit part separately, so we can use _mm256_maddubs_epi16,
// and then subtract. The high bit part has the 2 already subtracted (and so, it is zero if the high bit was not set,
// and 2 if the high bit was set)
__m128i q8s_0 = _mm_maddubs_epi16(q3h_0, q8_0);
__m128i q8s_1 = _mm_maddubs_epi16(q3h_1, q8_1);
__m128i q8s_2 = _mm_maddubs_epi16(q3h_2, q8_2);
__m128i q8s_3 = _mm_maddubs_epi16(q3h_3, q8_3);
__m128i q8s_4 = _mm_maddubs_epi16(q3h_4, q8_4);
__m128i q8s_5 = _mm_maddubs_epi16(q3h_5, q8_5);
__m128i q8s_6 = _mm_maddubs_epi16(q3h_6, q8_6);
__m128i q8s_7 = _mm_maddubs_epi16(q3h_7, q8_7);
__m128i p16_0 = _mm_maddubs_epi16(q3l_0, q8_0);
__m128i p16_1 = _mm_maddubs_epi16(q3l_1, q8_1);
__m128i p16_2 = _mm_maddubs_epi16(q3l_2, q8_2);
__m128i p16_3 = _mm_maddubs_epi16(q3l_3, q8_3);
__m128i p16_4 = _mm_maddubs_epi16(q3l_4, q8_4);
__m128i p16_5 = _mm_maddubs_epi16(q3l_5, q8_5);
__m128i p16_6 = _mm_maddubs_epi16(q3l_6, q8_6);
__m128i p16_7 = _mm_maddubs_epi16(q3l_7, q8_7);
p16_0 = _mm_sub_epi16(p16_0, q8s_0);
p16_1 = _mm_sub_epi16(p16_1, q8s_1);
p16_2 = _mm_sub_epi16(p16_2, q8s_2);
p16_3 = _mm_sub_epi16(p16_3, q8s_3);
p16_4 = _mm_sub_epi16(p16_4, q8s_4);
p16_5 = _mm_sub_epi16(p16_5, q8s_5);
p16_6 = _mm_sub_epi16(p16_6, q8s_6);
p16_7 = _mm_sub_epi16(p16_7, q8s_7);
// multiply with scales
__m128i shuffle = _mm_set1_epi16(0x0100);
p16_0 = _mm_madd_epi16(_mm_shuffle_epi8(scales[j], shuffle), p16_0);
shuffle = _mm_add_epi16(shuffle, m2);
p16_1 = _mm_madd_epi16(_mm_shuffle_epi8(scales[j], shuffle), p16_1);
shuffle = _mm_add_epi16(shuffle, m2);
p16_2 = _mm_madd_epi16(_mm_shuffle_epi8(scales[j], shuffle), p16_2);
shuffle = _mm_add_epi16(shuffle, m2);
p16_3 = _mm_madd_epi16(_mm_shuffle_epi8(scales[j], shuffle), p16_3);
shuffle = _mm_add_epi16(shuffle, m2);
p16_4 = _mm_madd_epi16(_mm_shuffle_epi8(scales[j], shuffle), p16_4);
shuffle = _mm_add_epi16(shuffle, m2);
p16_5 = _mm_madd_epi16(_mm_shuffle_epi8(scales[j], shuffle), p16_5);
shuffle = _mm_add_epi16(shuffle, m2);
p16_6 = _mm_madd_epi16(_mm_shuffle_epi8(scales[j], shuffle), p16_6);
shuffle = _mm_add_epi16(shuffle, m2);
p16_7 = _mm_madd_epi16(_mm_shuffle_epi8(scales[j], shuffle), p16_7);
// accumulate
p16_0 = _mm_add_epi32(p16_0, p16_1);
p16_2 = _mm_add_epi32(p16_2, p16_3);
p16_4 = _mm_add_epi32(p16_4, p16_5);
p16_6 = _mm_add_epi32(p16_6, p16_7);
sumi_0 = _mm_add_epi32(sumi_0, _mm_add_epi32(p16_0, p16_2));
sumi_1 = _mm_add_epi32(sumi_1, _mm_add_epi32(p16_4, p16_6));
}
// multiply with block scale and accumulate
ggml : workaround for missing _mm256_setr_m128i in GCC < 8 in k_quants.c (#2405)
2023-07-28 18:17:45 +00:00
__m256i sumi = MM256_SET_M128I(sumi_1, sumi_0);
k-quants : add AVX support to dot functions (#1916) * k_quants : add AVX support * k_quants : apply review comments
2023-06-26 16:46:07 +00:00
acc = _mm256_add_ps(_mm256_mul_ps(_mm256_broadcast_ss(&d), _mm256_cvtepi32_ps(sumi)), acc);
}
*s = hsum_float_8(acc);
ggml : add RISC-V Vector Support for K-Quants and improved the existing intrinsics (#3453) * Added RVV intrinsics support for Q8 quantize row and also improved the existing dot product function for risc-v. The RVV intrinsics is added for the following quantize row functions quantize_row_q8_0 quantize_row_q8_1 The following dot product functions have also been optimized by using LMUL = 1/2 instead of LMUL = 1 ggml_vec_dot_q4_0_q8_0 ggml_vec_dot_q4_1_q8_1 ggml_vec_dot_q5_0_q8_0 ggml_vec_dot_q5_1_q8_1 And vector initialization in Q5 by temporary array is also replaced by the vid intrinsics Signed-off-by: Ahmad Tameem <ahmad.tameem@10xengineers.ai> * Added RVV intrinsics support for k_quants This adds RISC-V Vector intrinsics support for the following K_quants functions for both QKK = 256 and QKK = 64 ggml_vec_dot_q2_K_q8_K ggml_vec_dot_q3_K_q8_K ggml_vec_dot_q4_K_q8_K ggml_vec_dot_q5_K_q8_K ggml_vec_dot_q6_K_q8_K Signed-off-by: Ahmad Tameem <ahmad.tameem@10xengineers.ai> --------- Signed-off-by: Ahmad Tameem <ahmad.tameem@10xengineers.ai>
2023-10-03 18:38:19 +00:00
#elif defined __riscv_v_intrinsic
uint32_t aux[3];
uint32_t utmp[4];
float sumf = 0;
for (int i = 0; i < nb; ++i) {
const uint8_t * restrict q3 = x[i].qs;
const uint8_t * restrict qh = x[i].hmask;
const int8_t * restrict q8 = y[i].qs;
memcpy(aux, x[i].scales, 12);
utmp[3] = ((aux[1] >> 4) & kmask2) | (((aux[2] >> 6) & kmask1) << 4);
utmp[2] = ((aux[0] >> 4) & kmask2) | (((aux[2] >> 4) & kmask1) << 4);
utmp[1] = (aux[1] & kmask2) | (((aux[2] >> 2) & kmask1) << 4);
utmp[0] = (aux[0] & kmask2) | (((aux[2] >> 0) & kmask1) << 4);
int8_t * scale = (int8_t *)utmp;
for (int j = 0; j < 16; ++j) scale[j] -= 32;
size_t vl = 32;
uint8_t m = 1;
vint32m1_t vzero = __riscv_vmv_v_x_i32m1(0, 1);
vuint8m1_t vqh = __riscv_vle8_v_u8m1(qh, vl);
int sum_t = 0;
for (int j = 0; j < QK_K; j += 128) {
vl = 32;
// load Q3
vuint8m1_t q3_x = __riscv_vle8_v_u8m1(q3, vl);
vint8m1_t q3_0 = __riscv_vreinterpret_v_u8m1_i8m1(__riscv_vand_vx_u8m1(q3_x, 0x03, vl));
vint8m1_t q3_1 = __riscv_vreinterpret_v_u8m1_i8m1(__riscv_vand_vx_u8m1(__riscv_vsrl_vx_u8m1(q3_x, 0x2, vl), 0x03 , vl));
vint8m1_t q3_2 = __riscv_vreinterpret_v_u8m1_i8m1(__riscv_vand_vx_u8m1(__riscv_vsrl_vx_u8m1(q3_x, 0x4, vl), 0x03 , vl));
vint8m1_t q3_3 = __riscv_vreinterpret_v_u8m1_i8m1(__riscv_vand_vx_u8m1(__riscv_vsrl_vx_u8m1(q3_x, 0x6, vl), 0x03 , vl));
// compute mask for subtraction
vuint8m1_t qh_m0 = __riscv_vand_vx_u8m1(vqh, m, vl);
vbool8_t vmask_0 = __riscv_vmseq_vx_u8m1_b8(qh_m0, 0, vl);
vint8m1_t q3_m0 = __riscv_vsub_vx_i8m1_m(vmask_0, q3_0, 0x4, vl);
m <<= 1;
vuint8m1_t qh_m1 = __riscv_vand_vx_u8m1(vqh, m, vl);
vbool8_t vmask_1 = __riscv_vmseq_vx_u8m1_b8(qh_m1, 0, vl);
vint8m1_t q3_m1 = __riscv_vsub_vx_i8m1_m(vmask_1, q3_1, 0x4, vl);
m <<= 1;
vuint8m1_t qh_m2 = __riscv_vand_vx_u8m1(vqh, m, vl);
vbool8_t vmask_2 = __riscv_vmseq_vx_u8m1_b8(qh_m2, 0, vl);
vint8m1_t q3_m2 = __riscv_vsub_vx_i8m1_m(vmask_2, q3_2, 0x4, vl);
m <<= 1;
vuint8m1_t qh_m3 = __riscv_vand_vx_u8m1(vqh, m, vl);
vbool8_t vmask_3 = __riscv_vmseq_vx_u8m1_b8(qh_m3, 0, vl);
vint8m1_t q3_m3 = __riscv_vsub_vx_i8m1_m(vmask_3, q3_3, 0x4, vl);
m <<= 1;
// load Q8 and take product with Q3
vint16m2_t a0 = __riscv_vwmul_vv_i16m2(q3_m0, __riscv_vle8_v_i8m1(q8, vl), vl);
vint16m2_t a1 = __riscv_vwmul_vv_i16m2(q3_m1, __riscv_vle8_v_i8m1(q8+32, vl), vl);
vint16m2_t a2 = __riscv_vwmul_vv_i16m2(q3_m2, __riscv_vle8_v_i8m1(q8+64, vl), vl);
vint16m2_t a3 = __riscv_vwmul_vv_i16m2(q3_m3, __riscv_vle8_v_i8m1(q8+96, vl), vl);
vl = 16;
english : use `typos` to fix comments and logs (#4354)
2023-12-12 09:53:36 +00:00
// retrieve lane to multiply with scale
ggml : add RISC-V Vector Support for K-Quants and improved the existing intrinsics (#3453) * Added RVV intrinsics support for Q8 quantize row and also improved the existing dot product function for risc-v. The RVV intrinsics is added for the following quantize row functions quantize_row_q8_0 quantize_row_q8_1 The following dot product functions have also been optimized by using LMUL = 1/2 instead of LMUL = 1 ggml_vec_dot_q4_0_q8_0 ggml_vec_dot_q4_1_q8_1 ggml_vec_dot_q5_0_q8_0 ggml_vec_dot_q5_1_q8_1 And vector initialization in Q5 by temporary array is also replaced by the vid intrinsics Signed-off-by: Ahmad Tameem <ahmad.tameem@10xengineers.ai> * Added RVV intrinsics support for k_quants This adds RISC-V Vector intrinsics support for the following K_quants functions for both QKK = 256 and QKK = 64 ggml_vec_dot_q2_K_q8_K ggml_vec_dot_q3_K_q8_K ggml_vec_dot_q4_K_q8_K ggml_vec_dot_q5_K_q8_K ggml_vec_dot_q6_K_q8_K Signed-off-by: Ahmad Tameem <ahmad.tameem@10xengineers.ai> --------- Signed-off-by: Ahmad Tameem <ahmad.tameem@10xengineers.ai>
2023-10-03 18:38:19 +00:00
vint32m2_t aux0_0 = __riscv_vwmul_vx_i32m2(__riscv_vget_v_i16m2_i16m1(a0, 0), (scale[0]), vl);
vint32m2_t aux0_1 = __riscv_vwmul_vx_i32m2(__riscv_vget_v_i16m2_i16m1(a0, 1), (scale[1]), vl);
vint32m2_t aux1_0 = __riscv_vwmul_vx_i32m2(__riscv_vget_v_i16m2_i16m1(a1, 0), (scale[2]), vl);
vint32m2_t aux1_1 = __riscv_vwmul_vx_i32m2(__riscv_vget_v_i16m2_i16m1(a1, 1), (scale[3]), vl);
vint32m2_t aux2_0 = __riscv_vwmul_vx_i32m2(__riscv_vget_v_i16m2_i16m1(a2, 0), (scale[4]), vl);
vint32m2_t aux2_1 = __riscv_vwmul_vx_i32m2(__riscv_vget_v_i16m2_i16m1(a2, 1), (scale[5]), vl);
vint32m2_t aux3_0 = __riscv_vwmul_vx_i32m2(__riscv_vget_v_i16m2_i16m1(a3, 0), (scale[6]), vl);
vint32m2_t aux3_1 = __riscv_vwmul_vx_i32m2(__riscv_vget_v_i16m2_i16m1(a3, 1), (scale[7]), vl);
vint32m1_t isum0 = __riscv_vredsum_vs_i32m2_i32m1(__riscv_vadd_vv_i32m2(aux0_0, aux0_1, vl), vzero, vl);
vint32m1_t isum1 = __riscv_vredsum_vs_i32m2_i32m1(__riscv_vadd_vv_i32m2(aux1_0, aux1_1, vl), isum0, vl);
vint32m1_t isum2 = __riscv_vredsum_vs_i32m2_i32m1(__riscv_vadd_vv_i32m2(aux2_0, aux2_1, vl), isum1, vl);
vint32m1_t isum3 = __riscv_vredsum_vs_i32m2_i32m1(__riscv_vadd_vv_i32m2(aux3_0, aux3_1, vl), isum2, vl);
sum_t += __riscv_vmv_x_s_i32m1_i32(isum3);
q3 += 32; q8 += 128; scale += 8;
}
ggml : move FP16 <-> FP32 code to ggml-impl.h (#3861) * ggml : move FP16 <-> FP32 stuff to ggml-impl.h ggml-ci * tests : fix ARM build * ggml : explicitly initialize deprecated type traits * ggml : add math.h to ggml-impl.h * ggml : remove duplicate static assert macros * ggml : prefix lookup tables with ggml_ ggml-ci * ggml-impl : move extern "C" to start of file
2023-10-30 17:19:15 +00:00
const float d = GGML_FP16_TO_FP32(x[i].d) * y[i].d;
ggml : add RISC-V Vector Support for K-Quants and improved the existing intrinsics (#3453) * Added RVV intrinsics support for Q8 quantize row and also improved the existing dot product function for risc-v. The RVV intrinsics is added for the following quantize row functions quantize_row_q8_0 quantize_row_q8_1 The following dot product functions have also been optimized by using LMUL = 1/2 instead of LMUL = 1 ggml_vec_dot_q4_0_q8_0 ggml_vec_dot_q4_1_q8_1 ggml_vec_dot_q5_0_q8_0 ggml_vec_dot_q5_1_q8_1 And vector initialization in Q5 by temporary array is also replaced by the vid intrinsics Signed-off-by: Ahmad Tameem <ahmad.tameem@10xengineers.ai> * Added RVV intrinsics support for k_quants This adds RISC-V Vector intrinsics support for the following K_quants functions for both QKK = 256 and QKK = 64 ggml_vec_dot_q2_K_q8_K ggml_vec_dot_q3_K_q8_K ggml_vec_dot_q4_K_q8_K ggml_vec_dot_q5_K_q8_K ggml_vec_dot_q6_K_q8_K Signed-off-by: Ahmad Tameem <ahmad.tameem@10xengineers.ai> --------- Signed-off-by: Ahmad Tameem <ahmad.tameem@10xengineers.ai>
2023-10-03 18:38:19 +00:00
sumf += d*sum_t;
}
*s = sumf;
ggml : add SOTA 2,3,4,5,6 bit k-quantizations (#1684) * Starting to add k-quantization to ggml I think it is better to have quantization separate from ggml. For now just adding the k-quants there, but it would be better to also factor out the existing ggml quantizations. * Adding Q3_K and Q8_K (de)-quantization * Q3_K now working on CUDA and AVX2/scalar CUDA is not ideal - ~50% slower than Q4_0 for single token prediction, about the same in batch mode (perplexity). CPU single token is ~55 ms (on Ryzen 7950X). * Some improvement for Q3_K on CUDA It is now ~22.5 ms/token on my GPU, so ~30% slower than Q4_0. * Some more CUDA optimizations for Q3_K Single token is now 20.5 ms/token (~20% slower than Q4_0). Perplexity is on par with Q4_0. * Adding Q4_K - scalar, AVX2, CUDA Performance is the same or perhaps very slightly better than Q4_0 on the CPU. On the GPU, single token prediction is ~10% better than Q4_0, batch mode (perplexity is about the same). * Adding Q6_K - scalar, AVX2, CUDA Performance is ~40% lower compared to Q4_K on the CPU. This is to be expected, considering that we are memory bound on the CPU and the 6-bit model is ~44% larger than the 4-bit. On the GPU, single token prediction is ~6% lower than Q4_0, batch mode (perplexity) is even closer (but still slower). * Adding Q5_K - scalar, AVX2, CUDA Performance is ~20% lower compared to Q4_K on the CPU. This is to be expected, considering that we are memory bound on the CPU and the 5-bit model is ~22% larger than the 4-bit. On the GPU, single token prediction is about the same as Q4_0 for both, single token and batch prediction. * Per convention, all QX_K quantizations use Q5_K for output.weight * Adding quantization mixes * Quantization mixes: didn't quite get what I wanted in the last commit * Q4_K dot product for ARM_NEON * Q6_K dot product for ARM_NEON * Q5_K dot product for ARM_NEON * Adding Q3_K dot for ARM_NEON It is 22% slower than Q4_K, despite the smaller model size. On x86_64, where we are memory bound, the Q3_K model is quite a bit faster than Q4_K. * A very slightly faster ARM_NEON Q3_K dot * Adding Q2_K - just CUDA for now Token prediction is pretty good - about 15.5 ms on a RTX 4080. Perplexity is about the same as Q4_K. * Adding scalar and AVX2 Q2_K dot * Adding ARM_NEON Q2_K dot About the same performance as Q4_K. * A slightly faster ARM_NEON Q2_K dot Single token prediction is now ~36 ms on M2 Max. The code is much simpler too. * Fixed bug in Q2_K CUDA dot product kernel Stranegly enough, for the few prompts I tried with the 7B model the responses looked perfectly reasonable. Only realized something is not quite right when I tried the larger models and started getting nonse back. In any case, Q2_K single token evaluation time on an RTX 4080 in a Ryzen7950X box iusing CUDA and model fully loaded on the GPU are ~15.5 ms for 7B, ~25.4 ms for 13B, and ~55.8 ms for 30B. The max number of layers that fit in VRAM for The 65B is 32. With that, we get ~330 ms per token, which is not that much faster than just running on the CPU (~470 ms per token). * Don't print zeros/NaNs when no count histogram has been collected * A 10% faster CUDA vector dot kernel for Q3_K Q3_K is now running at ~18.5 ms / token on CUDA, so the gap to Q4_0 is only 10%. It seems memory acccess pattern is more important for performance than the amount of computation the kernel does. * A slightly daster Q4_K AVX2 dot product For perplexity, where we are less memory bound, time per pass drops by ~5%. Barely measurable difference for single token prediction. * A slightly faster ARM_NEON A4_K dot product * Minor * Fix quantization error test We cannot possibly be expecting rmse < 0.002 for 2- and 3-bit quantization variants. * Fix docker build I have been sloppy with vector reinterpret casts on ARM_NEON. It seems clang is very forgiving in that regard. * Added forgotten ggml.o dependence on k_quants.h to the Makefile * Had unintentionally committed the Makefile with -Ofast enabled * ggml : rename k_quants -> ggml-quants-k, use lowercase in code --------- Co-authored-by: Iwan Kawrakow <iwan.kawrakow@gmail.com> Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
2023-06-05 19:56:18 +00:00
#else
// scalar version
// This function is written like this so the compiler can manage to vectorize most of it
// Using -Ofast, GCC and clang manage to produce code that is within a factor of 2 or so from the
// manually vectorized version above. Every other version I tried would run at least 4 times slower.
// The ideal situation would be if we could just write the code once, and the compiler would
// automatically produce the best possible set of machine instructions, instead of us having to manually
// write vectorized versions for AVX, ARM_NEON, etc.
int8_t aux8[QK_K];
int16_t aux16[8];
float sums [8];
int32_t aux32[8];
memset(sums, 0, 8*sizeof(float));
uint32_t auxs[4];
const int8_t * scales = (const int8_t*)auxs;
float sumf = 0;
for (int i = 0; i < nb; ++i) {
const uint8_t * restrict q3 = x[i].qs;
const uint8_t * restrict hm = x[i].hmask;
const int8_t * restrict q8 = y[i].qs;
memset(aux32, 0, 8*sizeof(int32_t));
int8_t * restrict a = aux8;
uint8_t m = 1;
for (int j = 0; j < QK_K; j += 128) {
for (int l = 0; l < 32; ++l) a[l] = q3[l] & 3;
for (int l = 0; l < 32; ++l) a[l] -= (hm[l] & m ? 0 : 4);
a += 32; m <<= 1;
for (int l = 0; l < 32; ++l) a[l] = (q3[l] >> 2) & 3;
for (int l = 0; l < 32; ++l) a[l] -= (hm[l] & m ? 0 : 4);
a += 32; m <<= 1;
for (int l = 0; l < 32; ++l) a[l] = (q3[l] >> 4) & 3;
for (int l = 0; l < 32; ++l) a[l] -= (hm[l] & m ? 0 : 4);
a += 32; m <<= 1;
for (int l = 0; l < 32; ++l) a[l] = (q3[l] >> 6) & 3;
for (int l = 0; l < 32; ++l) a[l] -= (hm[l] & m ? 0 : 4);
a += 32; m <<= 1;
q3 += 32;
}
a = aux8;
memcpy(auxs, x[i].scales, 12);
uint32_t tmp = auxs[2];
auxs[2] = ((auxs[0] >> 4) & kmask2) | (((tmp >> 4) & kmask1) << 4);
auxs[3] = ((auxs[1] >> 4) & kmask2) | (((tmp >> 6) & kmask1) << 4);
auxs[0] = (auxs[0] & kmask2) | (((tmp >> 0) & kmask1) << 4);
auxs[1] = (auxs[1] & kmask2) | (((tmp >> 2) & kmask1) << 4);
for (int j = 0; j < QK_K/16; ++j) {
for (int l = 0; l < 8; ++l) aux16[l] = q8[l] * a[l];
for (int l = 0; l < 8; ++l) aux32[l] += (scales[j] - 32) * aux16[l];
q8 += 8; a += 8;
for (int l = 0; l < 8; ++l) aux16[l] = q8[l] * a[l];
for (int l = 0; l < 8; ++l) aux32[l] += (scales[j] - 32) * aux16[l];
q8 += 8; a += 8;
}
ggml : move FP16 <-> FP32 code to ggml-impl.h (#3861) * ggml : move FP16 <-> FP32 stuff to ggml-impl.h ggml-ci * tests : fix ARM build * ggml : explicitly initialize deprecated type traits * ggml : add math.h to ggml-impl.h * ggml : remove duplicate static assert macros * ggml : prefix lookup tables with ggml_ ggml-ci * ggml-impl : move extern "C" to start of file
2023-10-30 17:19:15 +00:00
const float d = GGML_FP16_TO_FP32(x[i].d) * y[i].d;
ggml : add SOTA 2,3,4,5,6 bit k-quantizations (#1684) * Starting to add k-quantization to ggml I think it is better to have quantization separate from ggml. For now just adding the k-quants there, but it would be better to also factor out the existing ggml quantizations. * Adding Q3_K and Q8_K (de)-quantization * Q3_K now working on CUDA and AVX2/scalar CUDA is not ideal - ~50% slower than Q4_0 for single token prediction, about the same in batch mode (perplexity). CPU single token is ~55 ms (on Ryzen 7950X). * Some improvement for Q3_K on CUDA It is now ~22.5 ms/token on my GPU, so ~30% slower than Q4_0. * Some more CUDA optimizations for Q3_K Single token is now 20.5 ms/token (~20% slower than Q4_0). Perplexity is on par with Q4_0. * Adding Q4_K - scalar, AVX2, CUDA Performance is the same or perhaps very slightly better than Q4_0 on the CPU. On the GPU, single token prediction is ~10% better than Q4_0, batch mode (perplexity is about the same). * Adding Q6_K - scalar, AVX2, CUDA Performance is ~40% lower compared to Q4_K on the CPU. This is to be expected, considering that we are memory bound on the CPU and the 6-bit model is ~44% larger than the 4-bit. On the GPU, single token prediction is ~6% lower than Q4_0, batch mode (perplexity) is even closer (but still slower). * Adding Q5_K - scalar, AVX2, CUDA Performance is ~20% lower compared to Q4_K on the CPU. This is to be expected, considering that we are memory bound on the CPU and the 5-bit model is ~22% larger than the 4-bit. On the GPU, single token prediction is about the same as Q4_0 for both, single token and batch prediction. * Per convention, all QX_K quantizations use Q5_K for output.weight * Adding quantization mixes * Quantization mixes: didn't quite get what I wanted in the last commit * Q4_K dot product for ARM_NEON * Q6_K dot product for ARM_NEON * Q5_K dot product for ARM_NEON * Adding Q3_K dot for ARM_NEON It is 22% slower than Q4_K, despite the smaller model size. On x86_64, where we are memory bound, the Q3_K model is quite a bit faster than Q4_K. * A very slightly faster ARM_NEON Q3_K dot * Adding Q2_K - just CUDA for now Token prediction is pretty good - about 15.5 ms on a RTX 4080. Perplexity is about the same as Q4_K. * Adding scalar and AVX2 Q2_K dot * Adding ARM_NEON Q2_K dot About the same performance as Q4_K. * A slightly faster ARM_NEON Q2_K dot Single token prediction is now ~36 ms on M2 Max. The code is much simpler too. * Fixed bug in Q2_K CUDA dot product kernel Stranegly enough, for the few prompts I tried with the 7B model the responses looked perfectly reasonable. Only realized something is not quite right when I tried the larger models and started getting nonse back. In any case, Q2_K single token evaluation time on an RTX 4080 in a Ryzen7950X box iusing CUDA and model fully loaded on the GPU are ~15.5 ms for 7B, ~25.4 ms for 13B, and ~55.8 ms for 30B. The max number of layers that fit in VRAM for The 65B is 32. With that, we get ~330 ms per token, which is not that much faster than just running on the CPU (~470 ms per token). * Don't print zeros/NaNs when no count histogram has been collected * A 10% faster CUDA vector dot kernel for Q3_K Q3_K is now running at ~18.5 ms / token on CUDA, so the gap to Q4_0 is only 10%. It seems memory acccess pattern is more important for performance than the amount of computation the kernel does. * A slightly daster Q4_K AVX2 dot product For perplexity, where we are less memory bound, time per pass drops by ~5%. Barely measurable difference for single token prediction. * A slightly faster ARM_NEON A4_K dot product * Minor * Fix quantization error test We cannot possibly be expecting rmse < 0.002 for 2- and 3-bit quantization variants. * Fix docker build I have been sloppy with vector reinterpret casts on ARM_NEON. It seems clang is very forgiving in that regard. * Added forgotten ggml.o dependence on k_quants.h to the Makefile * Had unintentionally committed the Makefile with -Ofast enabled * ggml : rename k_quants -> ggml-quants-k, use lowercase in code --------- Co-authored-by: Iwan Kawrakow <iwan.kawrakow@gmail.com> Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
2023-06-05 19:56:18 +00:00
for (int l = 0; l < 8; ++l) sums[l] += d * aux32[l];
}
for (int l = 0; l < 8; ++l) sumf += sums[l];
*s = sumf;
#endif
}
k-quants : support for super-block size of 64 (#2001) * k_quants: WIP super-blocks with 64 weights * k_quants: WIP super-blocks with 64 weights Q6_K scalar and AVX2 works * k_quants: WIP super-blocks with 64 weights Q4_K scalar and AVX2 works * k_quants: WIP super-blocks with 64 weights Q2_K scalar and AVX2 works. Q2_K is way too slow (it is actually slower than the scalar implementation) * k_quants: WIP super-blocks with 64 weights Q3_K scalar and AVX2 works. * k_quants: WIP super-blocks with 64 weights Q5_K scalar and AVX2 works, and with that all k_quants are done on AVX2 and scalar * k_quants: WIP super-blocks with 64 weights Q6_K working on CUDA. Cannot make it run quite as gast as with super-blocks with 256 weigths: 8% slower on 4080, 20% slower on the 1660 (but there we fit 1 less layer on the GPU because pf the larger model size), so some fraction of these 20% is due to that, * k_quants: WIP super-blocks with 64 weights Q4_K working on CUDA. ~10% slower on GTX-1660, 16% slower on 4080. * k_quants: WIP super-blocks with 64 weights Q2_K working on CUDA. ~3% slower on GTX-1660, 10% slower on 4080. * k_quants: WIP super-blocks with 64 weights Q3_K working on CUDA. * k_quants: WIP super-blocks with 64 weights Q5_K working on CUDA, and with this CUDA is done. * k_quants: WIP super-blocks with 64 weights Q6_K working on ARM_NEON * k_quants: WIP super-blocks with 64 weights Q4_K working on ARM_NEON, but quite a bit slower than 256 weights * k_quants: WIP super-blocks with 64 weights Q2_K working on ARM_NEON, but quite a bit slower than 256 weights * k_quants: WIP super-blocks with 64 weights Q3_K working on ARM_NEON, but quite a bit slower than 256 weights. * k_quants: WIP super-blocks with 64 weights Q5_K working on ARM_NEON, but quite a bit slower than 256 weights. With that, we have full support for ARM_NEON, although performance is not quite there. * k_quants: WIP super-blocks with 64 weights Slightly more efficient Q3_K and Q5_K * k_quants: WIP super-blocks with 64 weights Another small improvement for Q3_K and Q5_K on ARM_NEON * k_quants: WIP super-blocks with 64 weights Yet another speedup for Q5_K on ARM_NEON. We are now within 10% of the QK_K = 256 version. * k_quants: WIP super-blocks with 64 weights * We are able to pass preprocessor macros to the Metal compiler * Q6_K works and is actually slightly more efficient than the QK_K = 256 version (25.2 ms vs 25.8 ms) * k_quants: WIP super-blocks with 64 weights Q4_K works on Metal and is actually slightly faster than QK_K = 256 (21.95 ms vs 24.0 ms). * k_quants: WIP super-blocks with 64 weights Q2_K works on Metal and is very slightly faster than QK_K = 256 (23.8 ms vs 24.2 ms). * k_quants: WIP super-blocks with 64 weights Q3_K works on Metal and is slightly faster than QK_K = 256 (26.6 ms vs 28.3 ms). * k_quants: WIP super-blocks with 64 weights Q5_K works on Metal and is slightly faster than QK_K = 256 (23.7 ms vs 26.3 ms). * k_quants: call them _K, not _k, also on Metal * k_quants: correctly define QK_K in llama.cpp * Fixed bug in q4_K quantization added with the 64-block addition * Simplify via lambda * k_quants: swicth Q3_K to 4-bit scales when QK_K = 64 Otherwise there isn't much benefit from this quantization type. There is some very slight loss in accuracy, but we reduce size by ~7%. E.g., for OpenLLaMA-3B, Q3_K_S perplexity is 8.6131 with 8-bit scales and 8.6352 with 4-bit, while file size decreases from 1.53G to 1.44G. * k_quants: switch Q4_K to 4-bit scales when QK_K = 64 Here the loss in accuracy is greater than for Q3_K, but the Q4_K points still move further to the left on the perplexity vs size curve. * k_quants: forgot to add the Metal changes in last commit * k_quants: change Q5_K to be type 0 when QK_K = 64 Still needs AVX2 implementation * k_quants: AVX2 implementation for new 64-weight Q5_K * k_quants: 10% faster ARM_NEON Q5_K dot product * k_quants: fixed issue caused by merging with master --------- Co-authored-by: Iwan Kawrakow <iwan.kawrakow@gmail.com>
2023-06-26 16:43:07 +00:00
#else
void ggml_vec_dot_q3_K_q8_K(const int n, float * restrict s, const void * restrict vx, const void * restrict vy) {
ggml : add SOTA 2,3,4,5,6 bit k-quantizations (#1684) * Starting to add k-quantization to ggml I think it is better to have quantization separate from ggml. For now just adding the k-quants there, but it would be better to also factor out the existing ggml quantizations. * Adding Q3_K and Q8_K (de)-quantization * Q3_K now working on CUDA and AVX2/scalar CUDA is not ideal - ~50% slower than Q4_0 for single token prediction, about the same in batch mode (perplexity). CPU single token is ~55 ms (on Ryzen 7950X). * Some improvement for Q3_K on CUDA It is now ~22.5 ms/token on my GPU, so ~30% slower than Q4_0. * Some more CUDA optimizations for Q3_K Single token is now 20.5 ms/token (~20% slower than Q4_0). Perplexity is on par with Q4_0. * Adding Q4_K - scalar, AVX2, CUDA Performance is the same or perhaps very slightly better than Q4_0 on the CPU. On the GPU, single token prediction is ~10% better than Q4_0, batch mode (perplexity is about the same). * Adding Q6_K - scalar, AVX2, CUDA Performance is ~40% lower compared to Q4_K on the CPU. This is to be expected, considering that we are memory bound on the CPU and the 6-bit model is ~44% larger than the 4-bit. On the GPU, single token prediction is ~6% lower than Q4_0, batch mode (perplexity) is even closer (but still slower). * Adding Q5_K - scalar, AVX2, CUDA Performance is ~20% lower compared to Q4_K on the CPU. This is to be expected, considering that we are memory bound on the CPU and the 5-bit model is ~22% larger than the 4-bit. On the GPU, single token prediction is about the same as Q4_0 for both, single token and batch prediction. * Per convention, all QX_K quantizations use Q5_K for output.weight * Adding quantization mixes * Quantization mixes: didn't quite get what I wanted in the last commit * Q4_K dot product for ARM_NEON * Q6_K dot product for ARM_NEON * Q5_K dot product for ARM_NEON * Adding Q3_K dot for ARM_NEON It is 22% slower than Q4_K, despite the smaller model size. On x86_64, where we are memory bound, the Q3_K model is quite a bit faster than Q4_K. * A very slightly faster ARM_NEON Q3_K dot * Adding Q2_K - just CUDA for now Token prediction is pretty good - about 15.5 ms on a RTX 4080. Perplexity is about the same as Q4_K. * Adding scalar and AVX2 Q2_K dot * Adding ARM_NEON Q2_K dot About the same performance as Q4_K. * A slightly faster ARM_NEON Q2_K dot Single token prediction is now ~36 ms on M2 Max. The code is much simpler too. * Fixed bug in Q2_K CUDA dot product kernel Stranegly enough, for the few prompts I tried with the 7B model the responses looked perfectly reasonable. Only realized something is not quite right when I tried the larger models and started getting nonse back. In any case, Q2_K single token evaluation time on an RTX 4080 in a Ryzen7950X box iusing CUDA and model fully loaded on the GPU are ~15.5 ms for 7B, ~25.4 ms for 13B, and ~55.8 ms for 30B. The max number of layers that fit in VRAM for The 65B is 32. With that, we get ~330 ms per token, which is not that much faster than just running on the CPU (~470 ms per token). * Don't print zeros/NaNs when no count histogram has been collected * A 10% faster CUDA vector dot kernel for Q3_K Q3_K is now running at ~18.5 ms / token on CUDA, so the gap to Q4_0 is only 10%. It seems memory acccess pattern is more important for performance than the amount of computation the kernel does. * A slightly daster Q4_K AVX2 dot product For perplexity, where we are less memory bound, time per pass drops by ~5%. Barely measurable difference for single token prediction. * A slightly faster ARM_NEON A4_K dot product * Minor * Fix quantization error test We cannot possibly be expecting rmse < 0.002 for 2- and 3-bit quantization variants. * Fix docker build I have been sloppy with vector reinterpret casts on ARM_NEON. It seems clang is very forgiving in that regard. * Added forgotten ggml.o dependence on k_quants.h to the Makefile * Had unintentionally committed the Makefile with -Ofast enabled * ggml : rename k_quants -> ggml-quants-k, use lowercase in code --------- Co-authored-by: Iwan Kawrakow <iwan.kawrakow@gmail.com> Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
2023-06-05 19:56:18 +00:00
assert(n % QK_K == 0);
k-quants : support for super-block size of 64 (#2001) * k_quants: WIP super-blocks with 64 weights * k_quants: WIP super-blocks with 64 weights Q6_K scalar and AVX2 works * k_quants: WIP super-blocks with 64 weights Q4_K scalar and AVX2 works * k_quants: WIP super-blocks with 64 weights Q2_K scalar and AVX2 works. Q2_K is way too slow (it is actually slower than the scalar implementation) * k_quants: WIP super-blocks with 64 weights Q3_K scalar and AVX2 works. * k_quants: WIP super-blocks with 64 weights Q5_K scalar and AVX2 works, and with that all k_quants are done on AVX2 and scalar * k_quants: WIP super-blocks with 64 weights Q6_K working on CUDA. Cannot make it run quite as gast as with super-blocks with 256 weigths: 8% slower on 4080, 20% slower on the 1660 (but there we fit 1 less layer on the GPU because pf the larger model size), so some fraction of these 20% is due to that, * k_quants: WIP super-blocks with 64 weights Q4_K working on CUDA. ~10% slower on GTX-1660, 16% slower on 4080. * k_quants: WIP super-blocks with 64 weights Q2_K working on CUDA. ~3% slower on GTX-1660, 10% slower on 4080. * k_quants: WIP super-blocks with 64 weights Q3_K working on CUDA. * k_quants: WIP super-blocks with 64 weights Q5_K working on CUDA, and with this CUDA is done. * k_quants: WIP super-blocks with 64 weights Q6_K working on ARM_NEON * k_quants: WIP super-blocks with 64 weights Q4_K working on ARM_NEON, but quite a bit slower than 256 weights * k_quants: WIP super-blocks with 64 weights Q2_K working on ARM_NEON, but quite a bit slower than 256 weights * k_quants: WIP super-blocks with 64 weights Q3_K working on ARM_NEON, but quite a bit slower than 256 weights. * k_quants: WIP super-blocks with 64 weights Q5_K working on ARM_NEON, but quite a bit slower than 256 weights. With that, we have full support for ARM_NEON, although performance is not quite there. * k_quants: WIP super-blocks with 64 weights Slightly more efficient Q3_K and Q5_K * k_quants: WIP super-blocks with 64 weights Another small improvement for Q3_K and Q5_K on ARM_NEON * k_quants: WIP super-blocks with 64 weights Yet another speedup for Q5_K on ARM_NEON. We are now within 10% of the QK_K = 256 version. * k_quants: WIP super-blocks with 64 weights * We are able to pass preprocessor macros to the Metal compiler * Q6_K works and is actually slightly more efficient than the QK_K = 256 version (25.2 ms vs 25.8 ms) * k_quants: WIP super-blocks with 64 weights Q4_K works on Metal and is actually slightly faster than QK_K = 256 (21.95 ms vs 24.0 ms). * k_quants: WIP super-blocks with 64 weights Q2_K works on Metal and is very slightly faster than QK_K = 256 (23.8 ms vs 24.2 ms). * k_quants: WIP super-blocks with 64 weights Q3_K works on Metal and is slightly faster than QK_K = 256 (26.6 ms vs 28.3 ms). * k_quants: WIP super-blocks with 64 weights Q5_K works on Metal and is slightly faster than QK_K = 256 (23.7 ms vs 26.3 ms). * k_quants: call them _K, not _k, also on Metal * k_quants: correctly define QK_K in llama.cpp * Fixed bug in q4_K quantization added with the 64-block addition * Simplify via lambda * k_quants: swicth Q3_K to 4-bit scales when QK_K = 64 Otherwise there isn't much benefit from this quantization type. There is some very slight loss in accuracy, but we reduce size by ~7%. E.g., for OpenLLaMA-3B, Q3_K_S perplexity is 8.6131 with 8-bit scales and 8.6352 with 4-bit, while file size decreases from 1.53G to 1.44G. * k_quants: switch Q4_K to 4-bit scales when QK_K = 64 Here the loss in accuracy is greater than for Q3_K, but the Q4_K points still move further to the left on the perplexity vs size curve. * k_quants: forgot to add the Metal changes in last commit * k_quants: change Q5_K to be type 0 when QK_K = 64 Still needs AVX2 implementation * k_quants: AVX2 implementation for new 64-weight Q5_K * k_quants: 10% faster ARM_NEON Q5_K dot product * k_quants: fixed issue caused by merging with master --------- Co-authored-by: Iwan Kawrakow <iwan.kawrakow@gmail.com>
2023-06-26 16:43:07 +00:00
const block_q3_K * restrict x = vx;
k-quants : allow to optionally disable at compile time (#1734) * k-quants : put behind optional compile flag LLAMA_K_QUANTS * build : enable k-quants by default
2023-06-07 07:59:52 +00:00
const block_q8_K * restrict y = vy;
ggml : add SOTA 2,3,4,5,6 bit k-quantizations (#1684) * Starting to add k-quantization to ggml I think it is better to have quantization separate from ggml. For now just adding the k-quants there, but it would be better to also factor out the existing ggml quantizations. * Adding Q3_K and Q8_K (de)-quantization * Q3_K now working on CUDA and AVX2/scalar CUDA is not ideal - ~50% slower than Q4_0 for single token prediction, about the same in batch mode (perplexity). CPU single token is ~55 ms (on Ryzen 7950X). * Some improvement for Q3_K on CUDA It is now ~22.5 ms/token on my GPU, so ~30% slower than Q4_0. * Some more CUDA optimizations for Q3_K Single token is now 20.5 ms/token (~20% slower than Q4_0). Perplexity is on par with Q4_0. * Adding Q4_K - scalar, AVX2, CUDA Performance is the same or perhaps very slightly better than Q4_0 on the CPU. On the GPU, single token prediction is ~10% better than Q4_0, batch mode (perplexity is about the same). * Adding Q6_K - scalar, AVX2, CUDA Performance is ~40% lower compared to Q4_K on the CPU. This is to be expected, considering that we are memory bound on the CPU and the 6-bit model is ~44% larger than the 4-bit. On the GPU, single token prediction is ~6% lower than Q4_0, batch mode (perplexity) is even closer (but still slower). * Adding Q5_K - scalar, AVX2, CUDA Performance is ~20% lower compared to Q4_K on the CPU. This is to be expected, considering that we are memory bound on the CPU and the 5-bit model is ~22% larger than the 4-bit. On the GPU, single token prediction is about the same as Q4_0 for both, single token and batch prediction. * Per convention, all QX_K quantizations use Q5_K for output.weight * Adding quantization mixes * Quantization mixes: didn't quite get what I wanted in the last commit * Q4_K dot product for ARM_NEON * Q6_K dot product for ARM_NEON * Q5_K dot product for ARM_NEON * Adding Q3_K dot for ARM_NEON It is 22% slower than Q4_K, despite the smaller model size. On x86_64, where we are memory bound, the Q3_K model is quite a bit faster than Q4_K. * A very slightly faster ARM_NEON Q3_K dot * Adding Q2_K - just CUDA for now Token prediction is pretty good - about 15.5 ms on a RTX 4080. Perplexity is about the same as Q4_K. * Adding scalar and AVX2 Q2_K dot * Adding ARM_NEON Q2_K dot About the same performance as Q4_K. * A slightly faster ARM_NEON Q2_K dot Single token prediction is now ~36 ms on M2 Max. The code is much simpler too. * Fixed bug in Q2_K CUDA dot product kernel Stranegly enough, for the few prompts I tried with the 7B model the responses looked perfectly reasonable. Only realized something is not quite right when I tried the larger models and started getting nonse back. In any case, Q2_K single token evaluation time on an RTX 4080 in a Ryzen7950X box iusing CUDA and model fully loaded on the GPU are ~15.5 ms for 7B, ~25.4 ms for 13B, and ~55.8 ms for 30B. The max number of layers that fit in VRAM for The 65B is 32. With that, we get ~330 ms per token, which is not that much faster than just running on the CPU (~470 ms per token). * Don't print zeros/NaNs when no count histogram has been collected * A 10% faster CUDA vector dot kernel for Q3_K Q3_K is now running at ~18.5 ms / token on CUDA, so the gap to Q4_0 is only 10%. It seems memory acccess pattern is more important for performance than the amount of computation the kernel does. * A slightly daster Q4_K AVX2 dot product For perplexity, where we are less memory bound, time per pass drops by ~5%. Barely measurable difference for single token prediction. * A slightly faster ARM_NEON A4_K dot product * Minor * Fix quantization error test We cannot possibly be expecting rmse < 0.002 for 2- and 3-bit quantization variants. * Fix docker build I have been sloppy with vector reinterpret casts on ARM_NEON. It seems clang is very forgiving in that regard. * Added forgotten ggml.o dependence on k_quants.h to the Makefile * Had unintentionally committed the Makefile with -Ofast enabled * ggml : rename k_quants -> ggml-quants-k, use lowercase in code --------- Co-authored-by: Iwan Kawrakow <iwan.kawrakow@gmail.com> Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
2023-06-05 19:56:18 +00:00
const int nb = n / QK_K;
#ifdef __ARM_NEON
ggml : fix dot product for ARM (#4630) ggml-ci
2023-12-27 09:02:13 +00:00
const int32x4_t vzero = vdupq_n_s32(0);
ggml : add SOTA 2,3,4,5,6 bit k-quantizations (#1684) * Starting to add k-quantization to ggml I think it is better to have quantization separate from ggml. For now just adding the k-quants there, but it would be better to also factor out the existing ggml quantizations. * Adding Q3_K and Q8_K (de)-quantization * Q3_K now working on CUDA and AVX2/scalar CUDA is not ideal - ~50% slower than Q4_0 for single token prediction, about the same in batch mode (perplexity). CPU single token is ~55 ms (on Ryzen 7950X). * Some improvement for Q3_K on CUDA It is now ~22.5 ms/token on my GPU, so ~30% slower than Q4_0. * Some more CUDA optimizations for Q3_K Single token is now 20.5 ms/token (~20% slower than Q4_0). Perplexity is on par with Q4_0. * Adding Q4_K - scalar, AVX2, CUDA Performance is the same or perhaps very slightly better than Q4_0 on the CPU. On the GPU, single token prediction is ~10% better than Q4_0, batch mode (perplexity is about the same). * Adding Q6_K - scalar, AVX2, CUDA Performance is ~40% lower compared to Q4_K on the CPU. This is to be expected, considering that we are memory bound on the CPU and the 6-bit model is ~44% larger than the 4-bit. On the GPU, single token prediction is ~6% lower than Q4_0, batch mode (perplexity) is even closer (but still slower). * Adding Q5_K - scalar, AVX2, CUDA Performance is ~20% lower compared to Q4_K on the CPU. This is to be expected, considering that we are memory bound on the CPU and the 5-bit model is ~22% larger than the 4-bit. On the GPU, single token prediction is about the same as Q4_0 for both, single token and batch prediction. * Per convention, all QX_K quantizations use Q5_K for output.weight * Adding quantization mixes * Quantization mixes: didn't quite get what I wanted in the last commit * Q4_K dot product for ARM_NEON * Q6_K dot product for ARM_NEON * Q5_K dot product for ARM_NEON * Adding Q3_K dot for ARM_NEON It is 22% slower than Q4_K, despite the smaller model size. On x86_64, where we are memory bound, the Q3_K model is quite a bit faster than Q4_K. * A very slightly faster ARM_NEON Q3_K dot * Adding Q2_K - just CUDA for now Token prediction is pretty good - about 15.5 ms on a RTX 4080. Perplexity is about the same as Q4_K. * Adding scalar and AVX2 Q2_K dot * Adding ARM_NEON Q2_K dot About the same performance as Q4_K. * A slightly faster ARM_NEON Q2_K dot Single token prediction is now ~36 ms on M2 Max. The code is much simpler too. * Fixed bug in Q2_K CUDA dot product kernel Stranegly enough, for the few prompts I tried with the 7B model the responses looked perfectly reasonable. Only realized something is not quite right when I tried the larger models and started getting nonse back. In any case, Q2_K single token evaluation time on an RTX 4080 in a Ryzen7950X box iusing CUDA and model fully loaded on the GPU are ~15.5 ms for 7B, ~25.4 ms for 13B, and ~55.8 ms for 30B. The max number of layers that fit in VRAM for The 65B is 32. With that, we get ~330 ms per token, which is not that much faster than just running on the CPU (~470 ms per token). * Don't print zeros/NaNs when no count histogram has been collected * A 10% faster CUDA vector dot kernel for Q3_K Q3_K is now running at ~18.5 ms / token on CUDA, so the gap to Q4_0 is only 10%. It seems memory acccess pattern is more important for performance than the amount of computation the kernel does. * A slightly daster Q4_K AVX2 dot product For perplexity, where we are less memory bound, time per pass drops by ~5%. Barely measurable difference for single token prediction. * A slightly faster ARM_NEON A4_K dot product * Minor * Fix quantization error test We cannot possibly be expecting rmse < 0.002 for 2- and 3-bit quantization variants. * Fix docker build I have been sloppy with vector reinterpret casts on ARM_NEON. It seems clang is very forgiving in that regard. * Added forgotten ggml.o dependence on k_quants.h to the Makefile * Had unintentionally committed the Makefile with -Ofast enabled * ggml : rename k_quants -> ggml-quants-k, use lowercase in code --------- Co-authored-by: Iwan Kawrakow <iwan.kawrakow@gmail.com> Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
2023-06-05 19:56:18 +00:00
k-quants : support for super-block size of 64 (#2001) * k_quants: WIP super-blocks with 64 weights * k_quants: WIP super-blocks with 64 weights Q6_K scalar and AVX2 works * k_quants: WIP super-blocks with 64 weights Q4_K scalar and AVX2 works * k_quants: WIP super-blocks with 64 weights Q2_K scalar and AVX2 works. Q2_K is way too slow (it is actually slower than the scalar implementation) * k_quants: WIP super-blocks with 64 weights Q3_K scalar and AVX2 works. * k_quants: WIP super-blocks with 64 weights Q5_K scalar and AVX2 works, and with that all k_quants are done on AVX2 and scalar * k_quants: WIP super-blocks with 64 weights Q6_K working on CUDA. Cannot make it run quite as gast as with super-blocks with 256 weigths: 8% slower on 4080, 20% slower on the 1660 (but there we fit 1 less layer on the GPU because pf the larger model size), so some fraction of these 20% is due to that, * k_quants: WIP super-blocks with 64 weights Q4_K working on CUDA. ~10% slower on GTX-1660, 16% slower on 4080. * k_quants: WIP super-blocks with 64 weights Q2_K working on CUDA. ~3% slower on GTX-1660, 10% slower on 4080. * k_quants: WIP super-blocks with 64 weights Q3_K working on CUDA. * k_quants: WIP super-blocks with 64 weights Q5_K working on CUDA, and with this CUDA is done. * k_quants: WIP super-blocks with 64 weights Q6_K working on ARM_NEON * k_quants: WIP super-blocks with 64 weights Q4_K working on ARM_NEON, but quite a bit slower than 256 weights * k_quants: WIP super-blocks with 64 weights Q2_K working on ARM_NEON, but quite a bit slower than 256 weights * k_quants: WIP super-blocks with 64 weights Q3_K working on ARM_NEON, but quite a bit slower than 256 weights. * k_quants: WIP super-blocks with 64 weights Q5_K working on ARM_NEON, but quite a bit slower than 256 weights. With that, we have full support for ARM_NEON, although performance is not quite there. * k_quants: WIP super-blocks with 64 weights Slightly more efficient Q3_K and Q5_K * k_quants: WIP super-blocks with 64 weights Another small improvement for Q3_K and Q5_K on ARM_NEON * k_quants: WIP super-blocks with 64 weights Yet another speedup for Q5_K on ARM_NEON. We are now within 10% of the QK_K = 256 version. * k_quants: WIP super-blocks with 64 weights * We are able to pass preprocessor macros to the Metal compiler * Q6_K works and is actually slightly more efficient than the QK_K = 256 version (25.2 ms vs 25.8 ms) * k_quants: WIP super-blocks with 64 weights Q4_K works on Metal and is actually slightly faster than QK_K = 256 (21.95 ms vs 24.0 ms). * k_quants: WIP super-blocks with 64 weights Q2_K works on Metal and is very slightly faster than QK_K = 256 (23.8 ms vs 24.2 ms). * k_quants: WIP super-blocks with 64 weights Q3_K works on Metal and is slightly faster than QK_K = 256 (26.6 ms vs 28.3 ms). * k_quants: WIP super-blocks with 64 weights Q5_K works on Metal and is slightly faster than QK_K = 256 (23.7 ms vs 26.3 ms). * k_quants: call them _K, not _k, also on Metal * k_quants: correctly define QK_K in llama.cpp * Fixed bug in q4_K quantization added with the 64-block addition * Simplify via lambda * k_quants: swicth Q3_K to 4-bit scales when QK_K = 64 Otherwise there isn't much benefit from this quantization type. There is some very slight loss in accuracy, but we reduce size by ~7%. E.g., for OpenLLaMA-3B, Q3_K_S perplexity is 8.6131 with 8-bit scales and 8.6352 with 4-bit, while file size decreases from 1.53G to 1.44G. * k_quants: switch Q4_K to 4-bit scales when QK_K = 64 Here the loss in accuracy is greater than for Q3_K, but the Q4_K points still move further to the left on the perplexity vs size curve. * k_quants: forgot to add the Metal changes in last commit * k_quants: change Q5_K to be type 0 when QK_K = 64 Still needs AVX2 implementation * k_quants: AVX2 implementation for new 64-weight Q5_K * k_quants: 10% faster ARM_NEON Q5_K dot product * k_quants: fixed issue caused by merging with master --------- Co-authored-by: Iwan Kawrakow <iwan.kawrakow@gmail.com>
2023-06-26 16:43:07 +00:00
const uint8x16_t m3b = vdupq_n_u8(0x3);
const uint8x16_t mh = vdupq_n_u8(4);
ggml : add SOTA 2,3,4,5,6 bit k-quantizations (#1684) * Starting to add k-quantization to ggml I think it is better to have quantization separate from ggml. For now just adding the k-quants there, but it would be better to also factor out the existing ggml quantizations. * Adding Q3_K and Q8_K (de)-quantization * Q3_K now working on CUDA and AVX2/scalar CUDA is not ideal - ~50% slower than Q4_0 for single token prediction, about the same in batch mode (perplexity). CPU single token is ~55 ms (on Ryzen 7950X). * Some improvement for Q3_K on CUDA It is now ~22.5 ms/token on my GPU, so ~30% slower than Q4_0. * Some more CUDA optimizations for Q3_K Single token is now 20.5 ms/token (~20% slower than Q4_0). Perplexity is on par with Q4_0. * Adding Q4_K - scalar, AVX2, CUDA Performance is the same or perhaps very slightly better than Q4_0 on the CPU. On the GPU, single token prediction is ~10% better than Q4_0, batch mode (perplexity is about the same). * Adding Q6_K - scalar, AVX2, CUDA Performance is ~40% lower compared to Q4_K on the CPU. This is to be expected, considering that we are memory bound on the CPU and the 6-bit model is ~44% larger than the 4-bit. On the GPU, single token prediction is ~6% lower than Q4_0, batch mode (perplexity) is even closer (but still slower). * Adding Q5_K - scalar, AVX2, CUDA Performance is ~20% lower compared to Q4_K on the CPU. This is to be expected, considering that we are memory bound on the CPU and the 5-bit model is ~22% larger than the 4-bit. On the GPU, single token prediction is about the same as Q4_0 for both, single token and batch prediction. * Per convention, all QX_K quantizations use Q5_K for output.weight * Adding quantization mixes * Quantization mixes: didn't quite get what I wanted in the last commit * Q4_K dot product for ARM_NEON * Q6_K dot product for ARM_NEON * Q5_K dot product for ARM_NEON * Adding Q3_K dot for ARM_NEON It is 22% slower than Q4_K, despite the smaller model size. On x86_64, where we are memory bound, the Q3_K model is quite a bit faster than Q4_K. * A very slightly faster ARM_NEON Q3_K dot * Adding Q2_K - just CUDA for now Token prediction is pretty good - about 15.5 ms on a RTX 4080. Perplexity is about the same as Q4_K. * Adding scalar and AVX2 Q2_K dot * Adding ARM_NEON Q2_K dot About the same performance as Q4_K. * A slightly faster ARM_NEON Q2_K dot Single token prediction is now ~36 ms on M2 Max. The code is much simpler too. * Fixed bug in Q2_K CUDA dot product kernel Stranegly enough, for the few prompts I tried with the 7B model the responses looked perfectly reasonable. Only realized something is not quite right when I tried the larger models and started getting nonse back. In any case, Q2_K single token evaluation time on an RTX 4080 in a Ryzen7950X box iusing CUDA and model fully loaded on the GPU are ~15.5 ms for 7B, ~25.4 ms for 13B, and ~55.8 ms for 30B. The max number of layers that fit in VRAM for The 65B is 32. With that, we get ~330 ms per token, which is not that much faster than just running on the CPU (~470 ms per token). * Don't print zeros/NaNs when no count histogram has been collected * A 10% faster CUDA vector dot kernel for Q3_K Q3_K is now running at ~18.5 ms / token on CUDA, so the gap to Q4_0 is only 10%. It seems memory acccess pattern is more important for performance than the amount of computation the kernel does. * A slightly daster Q4_K AVX2 dot product For perplexity, where we are less memory bound, time per pass drops by ~5%. Barely measurable difference for single token prediction. * A slightly faster ARM_NEON A4_K dot product * Minor * Fix quantization error test We cannot possibly be expecting rmse < 0.002 for 2- and 3-bit quantization variants. * Fix docker build I have been sloppy with vector reinterpret casts on ARM_NEON. It seems clang is very forgiving in that regard. * Added forgotten ggml.o dependence on k_quants.h to the Makefile * Had unintentionally committed the Makefile with -Ofast enabled * ggml : rename k_quants -> ggml-quants-k, use lowercase in code --------- Co-authored-by: Iwan Kawrakow <iwan.kawrakow@gmail.com> Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
2023-06-05 19:56:18 +00:00
ggml : sync (im2col, GPU conv, 32-bit arm compat) (#4060) ggml-ci
2023-11-13 14:55:52 +00:00
ggml_int8x16x4_t q3bytes;
ggml : add SOTA 2,3,4,5,6 bit k-quantizations (#1684) * Starting to add k-quantization to ggml I think it is better to have quantization separate from ggml. For now just adding the k-quants there, but it would be better to also factor out the existing ggml quantizations. * Adding Q3_K and Q8_K (de)-quantization * Q3_K now working on CUDA and AVX2/scalar CUDA is not ideal - ~50% slower than Q4_0 for single token prediction, about the same in batch mode (perplexity). CPU single token is ~55 ms (on Ryzen 7950X). * Some improvement for Q3_K on CUDA It is now ~22.5 ms/token on my GPU, so ~30% slower than Q4_0. * Some more CUDA optimizations for Q3_K Single token is now 20.5 ms/token (~20% slower than Q4_0). Perplexity is on par with Q4_0. * Adding Q4_K - scalar, AVX2, CUDA Performance is the same or perhaps very slightly better than Q4_0 on the CPU. On the GPU, single token prediction is ~10% better than Q4_0, batch mode (perplexity is about the same). * Adding Q6_K - scalar, AVX2, CUDA Performance is ~40% lower compared to Q4_K on the CPU. This is to be expected, considering that we are memory bound on the CPU and the 6-bit model is ~44% larger than the 4-bit. On the GPU, single token prediction is ~6% lower than Q4_0, batch mode (perplexity) is even closer (but still slower). * Adding Q5_K - scalar, AVX2, CUDA Performance is ~20% lower compared to Q4_K on the CPU. This is to be expected, considering that we are memory bound on the CPU and the 5-bit model is ~22% larger than the 4-bit. On the GPU, single token prediction is about the same as Q4_0 for both, single token and batch prediction. * Per convention, all QX_K quantizations use Q5_K for output.weight * Adding quantization mixes * Quantization mixes: didn't quite get what I wanted in the last commit * Q4_K dot product for ARM_NEON * Q6_K dot product for ARM_NEON * Q5_K dot product for ARM_NEON * Adding Q3_K dot for ARM_NEON It is 22% slower than Q4_K, despite the smaller model size. On x86_64, where we are memory bound, the Q3_K model is quite a bit faster than Q4_K. * A very slightly faster ARM_NEON Q3_K dot * Adding Q2_K - just CUDA for now Token prediction is pretty good - about 15.5 ms on a RTX 4080. Perplexity is about the same as Q4_K. * Adding scalar and AVX2 Q2_K dot * Adding ARM_NEON Q2_K dot About the same performance as Q4_K. * A slightly faster ARM_NEON Q2_K dot Single token prediction is now ~36 ms on M2 Max. The code is much simpler too. * Fixed bug in Q2_K CUDA dot product kernel Stranegly enough, for the few prompts I tried with the 7B model the responses looked perfectly reasonable. Only realized something is not quite right when I tried the larger models and started getting nonse back. In any case, Q2_K single token evaluation time on an RTX 4080 in a Ryzen7950X box iusing CUDA and model fully loaded on the GPU are ~15.5 ms for 7B, ~25.4 ms for 13B, and ~55.8 ms for 30B. The max number of layers that fit in VRAM for The 65B is 32. With that, we get ~330 ms per token, which is not that much faster than just running on the CPU (~470 ms per token). * Don't print zeros/NaNs when no count histogram has been collected * A 10% faster CUDA vector dot kernel for Q3_K Q3_K is now running at ~18.5 ms / token on CUDA, so the gap to Q4_0 is only 10%. It seems memory acccess pattern is more important for performance than the amount of computation the kernel does. * A slightly daster Q4_K AVX2 dot product For perplexity, where we are less memory bound, time per pass drops by ~5%. Barely measurable difference for single token prediction. * A slightly faster ARM_NEON A4_K dot product * Minor * Fix quantization error test We cannot possibly be expecting rmse < 0.002 for 2- and 3-bit quantization variants. * Fix docker build I have been sloppy with vector reinterpret casts on ARM_NEON. It seems clang is very forgiving in that regard. * Added forgotten ggml.o dependence on k_quants.h to the Makefile * Had unintentionally committed the Makefile with -Ofast enabled * ggml : rename k_quants -> ggml-quants-k, use lowercase in code --------- Co-authored-by: Iwan Kawrakow <iwan.kawrakow@gmail.com> Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
2023-06-05 19:56:18 +00:00
k-quants : support for super-block size of 64 (#2001) * k_quants: WIP super-blocks with 64 weights * k_quants: WIP super-blocks with 64 weights Q6_K scalar and AVX2 works * k_quants: WIP super-blocks with 64 weights Q4_K scalar and AVX2 works * k_quants: WIP super-blocks with 64 weights Q2_K scalar and AVX2 works. Q2_K is way too slow (it is actually slower than the scalar implementation) * k_quants: WIP super-blocks with 64 weights Q3_K scalar and AVX2 works. * k_quants: WIP super-blocks with 64 weights Q5_K scalar and AVX2 works, and with that all k_quants are done on AVX2 and scalar * k_quants: WIP super-blocks with 64 weights Q6_K working on CUDA. Cannot make it run quite as gast as with super-blocks with 256 weigths: 8% slower on 4080, 20% slower on the 1660 (but there we fit 1 less layer on the GPU because pf the larger model size), so some fraction of these 20% is due to that, * k_quants: WIP super-blocks with 64 weights Q4_K working on CUDA. ~10% slower on GTX-1660, 16% slower on 4080. * k_quants: WIP super-blocks with 64 weights Q2_K working on CUDA. ~3% slower on GTX-1660, 10% slower on 4080. * k_quants: WIP super-blocks with 64 weights Q3_K working on CUDA. * k_quants: WIP super-blocks with 64 weights Q5_K working on CUDA, and with this CUDA is done. * k_quants: WIP super-blocks with 64 weights Q6_K working on ARM_NEON * k_quants: WIP super-blocks with 64 weights Q4_K working on ARM_NEON, but quite a bit slower than 256 weights * k_quants: WIP super-blocks with 64 weights Q2_K working on ARM_NEON, but quite a bit slower than 256 weights * k_quants: WIP super-blocks with 64 weights Q3_K working on ARM_NEON, but quite a bit slower than 256 weights. * k_quants: WIP super-blocks with 64 weights Q5_K working on ARM_NEON, but quite a bit slower than 256 weights. With that, we have full support for ARM_NEON, although performance is not quite there. * k_quants: WIP super-blocks with 64 weights Slightly more efficient Q3_K and Q5_K * k_quants: WIP super-blocks with 64 weights Another small improvement for Q3_K and Q5_K on ARM_NEON * k_quants: WIP super-blocks with 64 weights Yet another speedup for Q5_K on ARM_NEON. We are now within 10% of the QK_K = 256 version. * k_quants: WIP super-blocks with 64 weights * We are able to pass preprocessor macros to the Metal compiler * Q6_K works and is actually slightly more efficient than the QK_K = 256 version (25.2 ms vs 25.8 ms) * k_quants: WIP super-blocks with 64 weights Q4_K works on Metal and is actually slightly faster than QK_K = 256 (21.95 ms vs 24.0 ms). * k_quants: WIP super-blocks with 64 weights Q2_K works on Metal and is very slightly faster than QK_K = 256 (23.8 ms vs 24.2 ms). * k_quants: WIP super-blocks with 64 weights Q3_K works on Metal and is slightly faster than QK_K = 256 (26.6 ms vs 28.3 ms). * k_quants: WIP super-blocks with 64 weights Q5_K works on Metal and is slightly faster than QK_K = 256 (23.7 ms vs 26.3 ms). * k_quants: call them _K, not _k, also on Metal * k_quants: correctly define QK_K in llama.cpp * Fixed bug in q4_K quantization added with the 64-block addition * Simplify via lambda * k_quants: swicth Q3_K to 4-bit scales when QK_K = 64 Otherwise there isn't much benefit from this quantization type. There is some very slight loss in accuracy, but we reduce size by ~7%. E.g., for OpenLLaMA-3B, Q3_K_S perplexity is 8.6131 with 8-bit scales and 8.6352 with 4-bit, while file size decreases from 1.53G to 1.44G. * k_quants: switch Q4_K to 4-bit scales when QK_K = 64 Here the loss in accuracy is greater than for Q3_K, but the Q4_K points still move further to the left on the perplexity vs size curve. * k_quants: forgot to add the Metal changes in last commit * k_quants: change Q5_K to be type 0 when QK_K = 64 Still needs AVX2 implementation * k_quants: AVX2 implementation for new 64-weight Q5_K * k_quants: 10% faster ARM_NEON Q5_K dot product * k_quants: fixed issue caused by merging with master --------- Co-authored-by: Iwan Kawrakow <iwan.kawrakow@gmail.com>
2023-06-26 16:43:07 +00:00
uint16_t aux16[2];
int8_t * scales = (int8_t *)aux16;
ggml : add SOTA 2,3,4,5,6 bit k-quantizations (#1684) * Starting to add k-quantization to ggml I think it is better to have quantization separate from ggml. For now just adding the k-quants there, but it would be better to also factor out the existing ggml quantizations. * Adding Q3_K and Q8_K (de)-quantization * Q3_K now working on CUDA and AVX2/scalar CUDA is not ideal - ~50% slower than Q4_0 for single token prediction, about the same in batch mode (perplexity). CPU single token is ~55 ms (on Ryzen 7950X). * Some improvement for Q3_K on CUDA It is now ~22.5 ms/token on my GPU, so ~30% slower than Q4_0. * Some more CUDA optimizations for Q3_K Single token is now 20.5 ms/token (~20% slower than Q4_0). Perplexity is on par with Q4_0. * Adding Q4_K - scalar, AVX2, CUDA Performance is the same or perhaps very slightly better than Q4_0 on the CPU. On the GPU, single token prediction is ~10% better than Q4_0, batch mode (perplexity is about the same). * Adding Q6_K - scalar, AVX2, CUDA Performance is ~40% lower compared to Q4_K on the CPU. This is to be expected, considering that we are memory bound on the CPU and the 6-bit model is ~44% larger than the 4-bit. On the GPU, single token prediction is ~6% lower than Q4_0, batch mode (perplexity) is even closer (but still slower). * Adding Q5_K - scalar, AVX2, CUDA Performance is ~20% lower compared to Q4_K on the CPU. This is to be expected, considering that we are memory bound on the CPU and the 5-bit model is ~22% larger than the 4-bit. On the GPU, single token prediction is about the same as Q4_0 for both, single token and batch prediction. * Per convention, all QX_K quantizations use Q5_K for output.weight * Adding quantization mixes * Quantization mixes: didn't quite get what I wanted in the last commit * Q4_K dot product for ARM_NEON * Q6_K dot product for ARM_NEON * Q5_K dot product for ARM_NEON * Adding Q3_K dot for ARM_NEON It is 22% slower than Q4_K, despite the smaller model size. On x86_64, where we are memory bound, the Q3_K model is quite a bit faster than Q4_K. * A very slightly faster ARM_NEON Q3_K dot * Adding Q2_K - just CUDA for now Token prediction is pretty good - about 15.5 ms on a RTX 4080. Perplexity is about the same as Q4_K. * Adding scalar and AVX2 Q2_K dot * Adding ARM_NEON Q2_K dot About the same performance as Q4_K. * A slightly faster ARM_NEON Q2_K dot Single token prediction is now ~36 ms on M2 Max. The code is much simpler too. * Fixed bug in Q2_K CUDA dot product kernel Stranegly enough, for the few prompts I tried with the 7B model the responses looked perfectly reasonable. Only realized something is not quite right when I tried the larger models and started getting nonse back. In any case, Q2_K single token evaluation time on an RTX 4080 in a Ryzen7950X box iusing CUDA and model fully loaded on the GPU are ~15.5 ms for 7B, ~25.4 ms for 13B, and ~55.8 ms for 30B. The max number of layers that fit in VRAM for The 65B is 32. With that, we get ~330 ms per token, which is not that much faster than just running on the CPU (~470 ms per token). * Don't print zeros/NaNs when no count histogram has been collected * A 10% faster CUDA vector dot kernel for Q3_K Q3_K is now running at ~18.5 ms / token on CUDA, so the gap to Q4_0 is only 10%. It seems memory acccess pattern is more important for performance than the amount of computation the kernel does. * A slightly daster Q4_K AVX2 dot product For perplexity, where we are less memory bound, time per pass drops by ~5%. Barely measurable difference for single token prediction. * A slightly faster ARM_NEON A4_K dot product * Minor * Fix quantization error test We cannot possibly be expecting rmse < 0.002 for 2- and 3-bit quantization variants. * Fix docker build I have been sloppy with vector reinterpret casts on ARM_NEON. It seems clang is very forgiving in that regard. * Added forgotten ggml.o dependence on k_quants.h to the Makefile * Had unintentionally committed the Makefile with -Ofast enabled * ggml : rename k_quants -> ggml-quants-k, use lowercase in code --------- Co-authored-by: Iwan Kawrakow <iwan.kawrakow@gmail.com> Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
2023-06-05 19:56:18 +00:00
k-quants : support for super-block size of 64 (#2001) * k_quants: WIP super-blocks with 64 weights * k_quants: WIP super-blocks with 64 weights Q6_K scalar and AVX2 works * k_quants: WIP super-blocks with 64 weights Q4_K scalar and AVX2 works * k_quants: WIP super-blocks with 64 weights Q2_K scalar and AVX2 works. Q2_K is way too slow (it is actually slower than the scalar implementation) * k_quants: WIP super-blocks with 64 weights Q3_K scalar and AVX2 works. * k_quants: WIP super-blocks with 64 weights Q5_K scalar and AVX2 works, and with that all k_quants are done on AVX2 and scalar * k_quants: WIP super-blocks with 64 weights Q6_K working on CUDA. Cannot make it run quite as gast as with super-blocks with 256 weigths: 8% slower on 4080, 20% slower on the 1660 (but there we fit 1 less layer on the GPU because pf the larger model size), so some fraction of these 20% is due to that, * k_quants: WIP super-blocks with 64 weights Q4_K working on CUDA. ~10% slower on GTX-1660, 16% slower on 4080. * k_quants: WIP super-blocks with 64 weights Q2_K working on CUDA. ~3% slower on GTX-1660, 10% slower on 4080. * k_quants: WIP super-blocks with 64 weights Q3_K working on CUDA. * k_quants: WIP super-blocks with 64 weights Q5_K working on CUDA, and with this CUDA is done. * k_quants: WIP super-blocks with 64 weights Q6_K working on ARM_NEON * k_quants: WIP super-blocks with 64 weights Q4_K working on ARM_NEON, but quite a bit slower than 256 weights * k_quants: WIP super-blocks with 64 weights Q2_K working on ARM_NEON, but quite a bit slower than 256 weights * k_quants: WIP super-blocks with 64 weights Q3_K working on ARM_NEON, but quite a bit slower than 256 weights. * k_quants: WIP super-blocks with 64 weights Q5_K working on ARM_NEON, but quite a bit slower than 256 weights. With that, we have full support for ARM_NEON, although performance is not quite there. * k_quants: WIP super-blocks with 64 weights Slightly more efficient Q3_K and Q5_K * k_quants: WIP super-blocks with 64 weights Another small improvement for Q3_K and Q5_K on ARM_NEON * k_quants: WIP super-blocks with 64 weights Yet another speedup for Q5_K on ARM_NEON. We are now within 10% of the QK_K = 256 version. * k_quants: WIP super-blocks with 64 weights * We are able to pass preprocessor macros to the Metal compiler * Q6_K works and is actually slightly more efficient than the QK_K = 256 version (25.2 ms vs 25.8 ms) * k_quants: WIP super-blocks with 64 weights Q4_K works on Metal and is actually slightly faster than QK_K = 256 (21.95 ms vs 24.0 ms). * k_quants: WIP super-blocks with 64 weights Q2_K works on Metal and is very slightly faster than QK_K = 256 (23.8 ms vs 24.2 ms). * k_quants: WIP super-blocks with 64 weights Q3_K works on Metal and is slightly faster than QK_K = 256 (26.6 ms vs 28.3 ms). * k_quants: WIP super-blocks with 64 weights Q5_K works on Metal and is slightly faster than QK_K = 256 (23.7 ms vs 26.3 ms). * k_quants: call them _K, not _k, also on Metal * k_quants: correctly define QK_K in llama.cpp * Fixed bug in q4_K quantization added with the 64-block addition * Simplify via lambda * k_quants: swicth Q3_K to 4-bit scales when QK_K = 64 Otherwise there isn't much benefit from this quantization type. There is some very slight loss in accuracy, but we reduce size by ~7%. E.g., for OpenLLaMA-3B, Q3_K_S perplexity is 8.6131 with 8-bit scales and 8.6352 with 4-bit, while file size decreases from 1.53G to 1.44G. * k_quants: switch Q4_K to 4-bit scales when QK_K = 64 Here the loss in accuracy is greater than for Q3_K, but the Q4_K points still move further to the left on the perplexity vs size curve. * k_quants: forgot to add the Metal changes in last commit * k_quants: change Q5_K to be type 0 when QK_K = 64 Still needs AVX2 implementation * k_quants: AVX2 implementation for new 64-weight Q5_K * k_quants: 10% faster ARM_NEON Q5_K dot product * k_quants: fixed issue caused by merging with master --------- Co-authored-by: Iwan Kawrakow <iwan.kawrakow@gmail.com>
2023-06-26 16:43:07 +00:00
float sum = 0;
ggml : add SOTA 2,3,4,5,6 bit k-quantizations (#1684) * Starting to add k-quantization to ggml I think it is better to have quantization separate from ggml. For now just adding the k-quants there, but it would be better to also factor out the existing ggml quantizations. * Adding Q3_K and Q8_K (de)-quantization * Q3_K now working on CUDA and AVX2/scalar CUDA is not ideal - ~50% slower than Q4_0 for single token prediction, about the same in batch mode (perplexity). CPU single token is ~55 ms (on Ryzen 7950X). * Some improvement for Q3_K on CUDA It is now ~22.5 ms/token on my GPU, so ~30% slower than Q4_0. * Some more CUDA optimizations for Q3_K Single token is now 20.5 ms/token (~20% slower than Q4_0). Perplexity is on par with Q4_0. * Adding Q4_K - scalar, AVX2, CUDA Performance is the same or perhaps very slightly better than Q4_0 on the CPU. On the GPU, single token prediction is ~10% better than Q4_0, batch mode (perplexity is about the same). * Adding Q6_K - scalar, AVX2, CUDA Performance is ~40% lower compared to Q4_K on the CPU. This is to be expected, considering that we are memory bound on the CPU and the 6-bit model is ~44% larger than the 4-bit. On the GPU, single token prediction is ~6% lower than Q4_0, batch mode (perplexity) is even closer (but still slower). * Adding Q5_K - scalar, AVX2, CUDA Performance is ~20% lower compared to Q4_K on the CPU. This is to be expected, considering that we are memory bound on the CPU and the 5-bit model is ~22% larger than the 4-bit. On the GPU, single token prediction is about the same as Q4_0 for both, single token and batch prediction. * Per convention, all QX_K quantizations use Q5_K for output.weight * Adding quantization mixes * Quantization mixes: didn't quite get what I wanted in the last commit * Q4_K dot product for ARM_NEON * Q6_K dot product for ARM_NEON * Q5_K dot product for ARM_NEON * Adding Q3_K dot for ARM_NEON It is 22% slower than Q4_K, despite the smaller model size. On x86_64, where we are memory bound, the Q3_K model is quite a bit faster than Q4_K. * A very slightly faster ARM_NEON Q3_K dot * Adding Q2_K - just CUDA for now Token prediction is pretty good - about 15.5 ms on a RTX 4080. Perplexity is about the same as Q4_K. * Adding scalar and AVX2 Q2_K dot * Adding ARM_NEON Q2_K dot About the same performance as Q4_K. * A slightly faster ARM_NEON Q2_K dot Single token prediction is now ~36 ms on M2 Max. The code is much simpler too. * Fixed bug in Q2_K CUDA dot product kernel Stranegly enough, for the few prompts I tried with the 7B model the responses looked perfectly reasonable. Only realized something is not quite right when I tried the larger models and started getting nonse back. In any case, Q2_K single token evaluation time on an RTX 4080 in a Ryzen7950X box iusing CUDA and model fully loaded on the GPU are ~15.5 ms for 7B, ~25.4 ms for 13B, and ~55.8 ms for 30B. The max number of layers that fit in VRAM for The 65B is 32. With that, we get ~330 ms per token, which is not that much faster than just running on the CPU (~470 ms per token). * Don't print zeros/NaNs when no count histogram has been collected * A 10% faster CUDA vector dot kernel for Q3_K Q3_K is now running at ~18.5 ms / token on CUDA, so the gap to Q4_0 is only 10%. It seems memory acccess pattern is more important for performance than the amount of computation the kernel does. * A slightly daster Q4_K AVX2 dot product For perplexity, where we are less memory bound, time per pass drops by ~5%. Barely measurable difference for single token prediction. * A slightly faster ARM_NEON A4_K dot product * Minor * Fix quantization error test We cannot possibly be expecting rmse < 0.002 for 2- and 3-bit quantization variants. * Fix docker build I have been sloppy with vector reinterpret casts on ARM_NEON. It seems clang is very forgiving in that regard. * Added forgotten ggml.o dependence on k_quants.h to the Makefile * Had unintentionally committed the Makefile with -Ofast enabled * ggml : rename k_quants -> ggml-quants-k, use lowercase in code --------- Co-authored-by: Iwan Kawrakow <iwan.kawrakow@gmail.com> Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
2023-06-05 19:56:18 +00:00
k-quants : support for super-block size of 64 (#2001) * k_quants: WIP super-blocks with 64 weights * k_quants: WIP super-blocks with 64 weights Q6_K scalar and AVX2 works * k_quants: WIP super-blocks with 64 weights Q4_K scalar and AVX2 works * k_quants: WIP super-blocks with 64 weights Q2_K scalar and AVX2 works. Q2_K is way too slow (it is actually slower than the scalar implementation) * k_quants: WIP super-blocks with 64 weights Q3_K scalar and AVX2 works. * k_quants: WIP super-blocks with 64 weights Q5_K scalar and AVX2 works, and with that all k_quants are done on AVX2 and scalar * k_quants: WIP super-blocks with 64 weights Q6_K working on CUDA. Cannot make it run quite as gast as with super-blocks with 256 weigths: 8% slower on 4080, 20% slower on the 1660 (but there we fit 1 less layer on the GPU because pf the larger model size), so some fraction of these 20% is due to that, * k_quants: WIP super-blocks with 64 weights Q4_K working on CUDA. ~10% slower on GTX-1660, 16% slower on 4080. * k_quants: WIP super-blocks with 64 weights Q2_K working on CUDA. ~3% slower on GTX-1660, 10% slower on 4080. * k_quants: WIP super-blocks with 64 weights Q3_K working on CUDA. * k_quants: WIP super-blocks with 64 weights Q5_K working on CUDA, and with this CUDA is done. * k_quants: WIP super-blocks with 64 weights Q6_K working on ARM_NEON * k_quants: WIP super-blocks with 64 weights Q4_K working on ARM_NEON, but quite a bit slower than 256 weights * k_quants: WIP super-blocks with 64 weights Q2_K working on ARM_NEON, but quite a bit slower than 256 weights * k_quants: WIP super-blocks with 64 weights Q3_K working on ARM_NEON, but quite a bit slower than 256 weights. * k_quants: WIP super-blocks with 64 weights Q5_K working on ARM_NEON, but quite a bit slower than 256 weights. With that, we have full support for ARM_NEON, although performance is not quite there. * k_quants: WIP super-blocks with 64 weights Slightly more efficient Q3_K and Q5_K * k_quants: WIP super-blocks with 64 weights Another small improvement for Q3_K and Q5_K on ARM_NEON * k_quants: WIP super-blocks with 64 weights Yet another speedup for Q5_K on ARM_NEON. We are now within 10% of the QK_K = 256 version. * k_quants: WIP super-blocks with 64 weights * We are able to pass preprocessor macros to the Metal compiler * Q6_K works and is actually slightly more efficient than the QK_K = 256 version (25.2 ms vs 25.8 ms) * k_quants: WIP super-blocks with 64 weights Q4_K works on Metal and is actually slightly faster than QK_K = 256 (21.95 ms vs 24.0 ms). * k_quants: WIP super-blocks with 64 weights Q2_K works on Metal and is very slightly faster than QK_K = 256 (23.8 ms vs 24.2 ms). * k_quants: WIP super-blocks with 64 weights Q3_K works on Metal and is slightly faster than QK_K = 256 (26.6 ms vs 28.3 ms). * k_quants: WIP super-blocks with 64 weights Q5_K works on Metal and is slightly faster than QK_K = 256 (23.7 ms vs 26.3 ms). * k_quants: call them _K, not _k, also on Metal * k_quants: correctly define QK_K in llama.cpp * Fixed bug in q4_K quantization added with the 64-block addition * Simplify via lambda * k_quants: swicth Q3_K to 4-bit scales when QK_K = 64 Otherwise there isn't much benefit from this quantization type. There is some very slight loss in accuracy, but we reduce size by ~7%. E.g., for OpenLLaMA-3B, Q3_K_S perplexity is 8.6131 with 8-bit scales and 8.6352 with 4-bit, while file size decreases from 1.53G to 1.44G. * k_quants: switch Q4_K to 4-bit scales when QK_K = 64 Here the loss in accuracy is greater than for Q3_K, but the Q4_K points still move further to the left on the perplexity vs size curve. * k_quants: forgot to add the Metal changes in last commit * k_quants: change Q5_K to be type 0 when QK_K = 64 Still needs AVX2 implementation * k_quants: AVX2 implementation for new 64-weight Q5_K * k_quants: 10% faster ARM_NEON Q5_K dot product * k_quants: fixed issue caused by merging with master --------- Co-authored-by: Iwan Kawrakow <iwan.kawrakow@gmail.com>
2023-06-26 16:43:07 +00:00
for (int i = 0; i < nb; ++i) {
ggml : add SOTA 2,3,4,5,6 bit k-quantizations (#1684) * Starting to add k-quantization to ggml I think it is better to have quantization separate from ggml. For now just adding the k-quants there, but it would be better to also factor out the existing ggml quantizations. * Adding Q3_K and Q8_K (de)-quantization * Q3_K now working on CUDA and AVX2/scalar CUDA is not ideal - ~50% slower than Q4_0 for single token prediction, about the same in batch mode (perplexity). CPU single token is ~55 ms (on Ryzen 7950X). * Some improvement for Q3_K on CUDA It is now ~22.5 ms/token on my GPU, so ~30% slower than Q4_0. * Some more CUDA optimizations for Q3_K Single token is now 20.5 ms/token (~20% slower than Q4_0). Perplexity is on par with Q4_0. * Adding Q4_K - scalar, AVX2, CUDA Performance is the same or perhaps very slightly better than Q4_0 on the CPU. On the GPU, single token prediction is ~10% better than Q4_0, batch mode (perplexity is about the same). * Adding Q6_K - scalar, AVX2, CUDA Performance is ~40% lower compared to Q4_K on the CPU. This is to be expected, considering that we are memory bound on the CPU and the 6-bit model is ~44% larger than the 4-bit. On the GPU, single token prediction is ~6% lower than Q4_0, batch mode (perplexity) is even closer (but still slower). * Adding Q5_K - scalar, AVX2, CUDA Performance is ~20% lower compared to Q4_K on the CPU. This is to be expected, considering that we are memory bound on the CPU and the 5-bit model is ~22% larger than the 4-bit. On the GPU, single token prediction is about the same as Q4_0 for both, single token and batch prediction. * Per convention, all QX_K quantizations use Q5_K for output.weight * Adding quantization mixes * Quantization mixes: didn't quite get what I wanted in the last commit * Q4_K dot product for ARM_NEON * Q6_K dot product for ARM_NEON * Q5_K dot product for ARM_NEON * Adding Q3_K dot for ARM_NEON It is 22% slower than Q4_K, despite the smaller model size. On x86_64, where we are memory bound, the Q3_K model is quite a bit faster than Q4_K. * A very slightly faster ARM_NEON Q3_K dot * Adding Q2_K - just CUDA for now Token prediction is pretty good - about 15.5 ms on a RTX 4080. Perplexity is about the same as Q4_K. * Adding scalar and AVX2 Q2_K dot * Adding ARM_NEON Q2_K dot About the same performance as Q4_K. * A slightly faster ARM_NEON Q2_K dot Single token prediction is now ~36 ms on M2 Max. The code is much simpler too. * Fixed bug in Q2_K CUDA dot product kernel Stranegly enough, for the few prompts I tried with the 7B model the responses looked perfectly reasonable. Only realized something is not quite right when I tried the larger models and started getting nonse back. In any case, Q2_K single token evaluation time on an RTX 4080 in a Ryzen7950X box iusing CUDA and model fully loaded on the GPU are ~15.5 ms for 7B, ~25.4 ms for 13B, and ~55.8 ms for 30B. The max number of layers that fit in VRAM for The 65B is 32. With that, we get ~330 ms per token, which is not that much faster than just running on the CPU (~470 ms per token). * Don't print zeros/NaNs when no count histogram has been collected * A 10% faster CUDA vector dot kernel for Q3_K Q3_K is now running at ~18.5 ms / token on CUDA, so the gap to Q4_0 is only 10%. It seems memory acccess pattern is more important for performance than the amount of computation the kernel does. * A slightly daster Q4_K AVX2 dot product For perplexity, where we are less memory bound, time per pass drops by ~5%. Barely measurable difference for single token prediction. * A slightly faster ARM_NEON A4_K dot product * Minor * Fix quantization error test We cannot possibly be expecting rmse < 0.002 for 2- and 3-bit quantization variants. * Fix docker build I have been sloppy with vector reinterpret casts on ARM_NEON. It seems clang is very forgiving in that regard. * Added forgotten ggml.o dependence on k_quants.h to the Makefile * Had unintentionally committed the Makefile with -Ofast enabled * ggml : rename k_quants -> ggml-quants-k, use lowercase in code --------- Co-authored-by: Iwan Kawrakow <iwan.kawrakow@gmail.com> Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
2023-06-05 19:56:18 +00:00
ggml : sync (im2col, GPU conv, 32-bit arm compat) (#4060) ggml-ci
2023-11-13 14:55:52 +00:00
ggml_uint8x16x4_t q3h;
ggml : add SOTA 2,3,4,5,6 bit k-quantizations (#1684) * Starting to add k-quantization to ggml I think it is better to have quantization separate from ggml. For now just adding the k-quants there, but it would be better to also factor out the existing ggml quantizations. * Adding Q3_K and Q8_K (de)-quantization * Q3_K now working on CUDA and AVX2/scalar CUDA is not ideal - ~50% slower than Q4_0 for single token prediction, about the same in batch mode (perplexity). CPU single token is ~55 ms (on Ryzen 7950X). * Some improvement for Q3_K on CUDA It is now ~22.5 ms/token on my GPU, so ~30% slower than Q4_0. * Some more CUDA optimizations for Q3_K Single token is now 20.5 ms/token (~20% slower than Q4_0). Perplexity is on par with Q4_0. * Adding Q4_K - scalar, AVX2, CUDA Performance is the same or perhaps very slightly better than Q4_0 on the CPU. On the GPU, single token prediction is ~10% better than Q4_0, batch mode (perplexity is about the same). * Adding Q6_K - scalar, AVX2, CUDA Performance is ~40% lower compared to Q4_K on the CPU. This is to be expected, considering that we are memory bound on the CPU and the 6-bit model is ~44% larger than the 4-bit. On the GPU, single token prediction is ~6% lower than Q4_0, batch mode (perplexity) is even closer (but still slower). * Adding Q5_K - scalar, AVX2, CUDA Performance is ~20% lower compared to Q4_K on the CPU. This is to be expected, considering that we are memory bound on the CPU and the 5-bit model is ~22% larger than the 4-bit. On the GPU, single token prediction is about the same as Q4_0 for both, single token and batch prediction. * Per convention, all QX_K quantizations use Q5_K for output.weight * Adding quantization mixes * Quantization mixes: didn't quite get what I wanted in the last commit * Q4_K dot product for ARM_NEON * Q6_K dot product for ARM_NEON * Q5_K dot product for ARM_NEON * Adding Q3_K dot for ARM_NEON It is 22% slower than Q4_K, despite the smaller model size. On x86_64, where we are memory bound, the Q3_K model is quite a bit faster than Q4_K. * A very slightly faster ARM_NEON Q3_K dot * Adding Q2_K - just CUDA for now Token prediction is pretty good - about 15.5 ms on a RTX 4080. Perplexity is about the same as Q4_K. * Adding scalar and AVX2 Q2_K dot * Adding ARM_NEON Q2_K dot About the same performance as Q4_K. * A slightly faster ARM_NEON Q2_K dot Single token prediction is now ~36 ms on M2 Max. The code is much simpler too. * Fixed bug in Q2_K CUDA dot product kernel Stranegly enough, for the few prompts I tried with the 7B model the responses looked perfectly reasonable. Only realized something is not quite right when I tried the larger models and started getting nonse back. In any case, Q2_K single token evaluation time on an RTX 4080 in a Ryzen7950X box iusing CUDA and model fully loaded on the GPU are ~15.5 ms for 7B, ~25.4 ms for 13B, and ~55.8 ms for 30B. The max number of layers that fit in VRAM for The 65B is 32. With that, we get ~330 ms per token, which is not that much faster than just running on the CPU (~470 ms per token). * Don't print zeros/NaNs when no count histogram has been collected * A 10% faster CUDA vector dot kernel for Q3_K Q3_K is now running at ~18.5 ms / token on CUDA, so the gap to Q4_0 is only 10%. It seems memory acccess pattern is more important for performance than the amount of computation the kernel does. * A slightly daster Q4_K AVX2 dot product For perplexity, where we are less memory bound, time per pass drops by ~5%. Barely measurable difference for single token prediction. * A slightly faster ARM_NEON A4_K dot product * Minor * Fix quantization error test We cannot possibly be expecting rmse < 0.002 for 2- and 3-bit quantization variants. * Fix docker build I have been sloppy with vector reinterpret casts on ARM_NEON. It seems clang is very forgiving in that regard. * Added forgotten ggml.o dependence on k_quants.h to the Makefile * Had unintentionally committed the Makefile with -Ofast enabled * ggml : rename k_quants -> ggml-quants-k, use lowercase in code --------- Co-authored-by: Iwan Kawrakow <iwan.kawrakow@gmail.com> Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
2023-06-05 19:56:18 +00:00
k-quants : support for super-block size of 64 (#2001) * k_quants: WIP super-blocks with 64 weights * k_quants: WIP super-blocks with 64 weights Q6_K scalar and AVX2 works * k_quants: WIP super-blocks with 64 weights Q4_K scalar and AVX2 works * k_quants: WIP super-blocks with 64 weights Q2_K scalar and AVX2 works. Q2_K is way too slow (it is actually slower than the scalar implementation) * k_quants: WIP super-blocks with 64 weights Q3_K scalar and AVX2 works. * k_quants: WIP super-blocks with 64 weights Q5_K scalar and AVX2 works, and with that all k_quants are done on AVX2 and scalar * k_quants: WIP super-blocks with 64 weights Q6_K working on CUDA. Cannot make it run quite as gast as with super-blocks with 256 weigths: 8% slower on 4080, 20% slower on the 1660 (but there we fit 1 less layer on the GPU because pf the larger model size), so some fraction of these 20% is due to that, * k_quants: WIP super-blocks with 64 weights Q4_K working on CUDA. ~10% slower on GTX-1660, 16% slower on 4080. * k_quants: WIP super-blocks with 64 weights Q2_K working on CUDA. ~3% slower on GTX-1660, 10% slower on 4080. * k_quants: WIP super-blocks with 64 weights Q3_K working on CUDA. * k_quants: WIP super-blocks with 64 weights Q5_K working on CUDA, and with this CUDA is done. * k_quants: WIP super-blocks with 64 weights Q6_K working on ARM_NEON * k_quants: WIP super-blocks with 64 weights Q4_K working on ARM_NEON, but quite a bit slower than 256 weights * k_quants: WIP super-blocks with 64 weights Q2_K working on ARM_NEON, but quite a bit slower than 256 weights * k_quants: WIP super-blocks with 64 weights Q3_K working on ARM_NEON, but quite a bit slower than 256 weights. * k_quants: WIP super-blocks with 64 weights Q5_K working on ARM_NEON, but quite a bit slower than 256 weights. With that, we have full support for ARM_NEON, although performance is not quite there. * k_quants: WIP super-blocks with 64 weights Slightly more efficient Q3_K and Q5_K * k_quants: WIP super-blocks with 64 weights Another small improvement for Q3_K and Q5_K on ARM_NEON * k_quants: WIP super-blocks with 64 weights Yet another speedup for Q5_K on ARM_NEON. We are now within 10% of the QK_K = 256 version. * k_quants: WIP super-blocks with 64 weights * We are able to pass preprocessor macros to the Metal compiler * Q6_K works and is actually slightly more efficient than the QK_K = 256 version (25.2 ms vs 25.8 ms) * k_quants: WIP super-blocks with 64 weights Q4_K works on Metal and is actually slightly faster than QK_K = 256 (21.95 ms vs 24.0 ms). * k_quants: WIP super-blocks with 64 weights Q2_K works on Metal and is very slightly faster than QK_K = 256 (23.8 ms vs 24.2 ms). * k_quants: WIP super-blocks with 64 weights Q3_K works on Metal and is slightly faster than QK_K = 256 (26.6 ms vs 28.3 ms). * k_quants: WIP super-blocks with 64 weights Q5_K works on Metal and is slightly faster than QK_K = 256 (23.7 ms vs 26.3 ms). * k_quants: call them _K, not _k, also on Metal * k_quants: correctly define QK_K in llama.cpp * Fixed bug in q4_K quantization added with the 64-block addition * Simplify via lambda * k_quants: swicth Q3_K to 4-bit scales when QK_K = 64 Otherwise there isn't much benefit from this quantization type. There is some very slight loss in accuracy, but we reduce size by ~7%. E.g., for OpenLLaMA-3B, Q3_K_S perplexity is 8.6131 with 8-bit scales and 8.6352 with 4-bit, while file size decreases from 1.53G to 1.44G. * k_quants: switch Q4_K to 4-bit scales when QK_K = 64 Here the loss in accuracy is greater than for Q3_K, but the Q4_K points still move further to the left on the perplexity vs size curve. * k_quants: forgot to add the Metal changes in last commit * k_quants: change Q5_K to be type 0 when QK_K = 64 Still needs AVX2 implementation * k_quants: AVX2 implementation for new 64-weight Q5_K * k_quants: 10% faster ARM_NEON Q5_K dot product * k_quants: fixed issue caused by merging with master --------- Co-authored-by: Iwan Kawrakow <iwan.kawrakow@gmail.com>
2023-06-26 16:43:07 +00:00
const uint8x8_t hbits = vld1_u8(x[i].hmask);
const uint8x16_t q3bits = vld1q_u8(x[i].qs);
ggml : sync (im2col, GPU conv, 32-bit arm compat) (#4060) ggml-ci
2023-11-13 14:55:52 +00:00
const ggml_int8x16x4_t q8bytes = ggml_vld1q_s8_x4(y[i].qs);
ggml : add SOTA 2,3,4,5,6 bit k-quantizations (#1684) * Starting to add k-quantization to ggml I think it is better to have quantization separate from ggml. For now just adding the k-quants there, but it would be better to also factor out the existing ggml quantizations. * Adding Q3_K and Q8_K (de)-quantization * Q3_K now working on CUDA and AVX2/scalar CUDA is not ideal - ~50% slower than Q4_0 for single token prediction, about the same in batch mode (perplexity). CPU single token is ~55 ms (on Ryzen 7950X). * Some improvement for Q3_K on CUDA It is now ~22.5 ms/token on my GPU, so ~30% slower than Q4_0. * Some more CUDA optimizations for Q3_K Single token is now 20.5 ms/token (~20% slower than Q4_0). Perplexity is on par with Q4_0. * Adding Q4_K - scalar, AVX2, CUDA Performance is the same or perhaps very slightly better than Q4_0 on the CPU. On the GPU, single token prediction is ~10% better than Q4_0, batch mode (perplexity is about the same). * Adding Q6_K - scalar, AVX2, CUDA Performance is ~40% lower compared to Q4_K on the CPU. This is to be expected, considering that we are memory bound on the CPU and the 6-bit model is ~44% larger than the 4-bit. On the GPU, single token prediction is ~6% lower than Q4_0, batch mode (perplexity) is even closer (but still slower). * Adding Q5_K - scalar, AVX2, CUDA Performance is ~20% lower compared to Q4_K on the CPU. This is to be expected, considering that we are memory bound on the CPU and the 5-bit model is ~22% larger than the 4-bit. On the GPU, single token prediction is about the same as Q4_0 for both, single token and batch prediction. * Per convention, all QX_K quantizations use Q5_K for output.weight * Adding quantization mixes * Quantization mixes: didn't quite get what I wanted in the last commit * Q4_K dot product for ARM_NEON * Q6_K dot product for ARM_NEON * Q5_K dot product for ARM_NEON * Adding Q3_K dot for ARM_NEON It is 22% slower than Q4_K, despite the smaller model size. On x86_64, where we are memory bound, the Q3_K model is quite a bit faster than Q4_K. * A very slightly faster ARM_NEON Q3_K dot * Adding Q2_K - just CUDA for now Token prediction is pretty good - about 15.5 ms on a RTX 4080. Perplexity is about the same as Q4_K. * Adding scalar and AVX2 Q2_K dot * Adding ARM_NEON Q2_K dot About the same performance as Q4_K. * A slightly faster ARM_NEON Q2_K dot Single token prediction is now ~36 ms on M2 Max. The code is much simpler too. * Fixed bug in Q2_K CUDA dot product kernel Stranegly enough, for the few prompts I tried with the 7B model the responses looked perfectly reasonable. Only realized something is not quite right when I tried the larger models and started getting nonse back. In any case, Q2_K single token evaluation time on an RTX 4080 in a Ryzen7950X box iusing CUDA and model fully loaded on the GPU are ~15.5 ms for 7B, ~25.4 ms for 13B, and ~55.8 ms for 30B. The max number of layers that fit in VRAM for The 65B is 32. With that, we get ~330 ms per token, which is not that much faster than just running on the CPU (~470 ms per token). * Don't print zeros/NaNs when no count histogram has been collected * A 10% faster CUDA vector dot kernel for Q3_K Q3_K is now running at ~18.5 ms / token on CUDA, so the gap to Q4_0 is only 10%. It seems memory acccess pattern is more important for performance than the amount of computation the kernel does. * A slightly daster Q4_K AVX2 dot product For perplexity, where we are less memory bound, time per pass drops by ~5%. Barely measurable difference for single token prediction. * A slightly faster ARM_NEON A4_K dot product * Minor * Fix quantization error test We cannot possibly be expecting rmse < 0.002 for 2- and 3-bit quantization variants. * Fix docker build I have been sloppy with vector reinterpret casts on ARM_NEON. It seems clang is very forgiving in that regard. * Added forgotten ggml.o dependence on k_quants.h to the Makefile * Had unintentionally committed the Makefile with -Ofast enabled * ggml : rename k_quants -> ggml-quants-k, use lowercase in code --------- Co-authored-by: Iwan Kawrakow <iwan.kawrakow@gmail.com> Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
2023-06-05 19:56:18 +00:00
k-quants : support for super-block size of 64 (#2001) * k_quants: WIP super-blocks with 64 weights * k_quants: WIP super-blocks with 64 weights Q6_K scalar and AVX2 works * k_quants: WIP super-blocks with 64 weights Q4_K scalar and AVX2 works * k_quants: WIP super-blocks with 64 weights Q2_K scalar and AVX2 works. Q2_K is way too slow (it is actually slower than the scalar implementation) * k_quants: WIP super-blocks with 64 weights Q3_K scalar and AVX2 works. * k_quants: WIP super-blocks with 64 weights Q5_K scalar and AVX2 works, and with that all k_quants are done on AVX2 and scalar * k_quants: WIP super-blocks with 64 weights Q6_K working on CUDA. Cannot make it run quite as gast as with super-blocks with 256 weigths: 8% slower on 4080, 20% slower on the 1660 (but there we fit 1 less layer on the GPU because pf the larger model size), so some fraction of these 20% is due to that, * k_quants: WIP super-blocks with 64 weights Q4_K working on CUDA. ~10% slower on GTX-1660, 16% slower on 4080. * k_quants: WIP super-blocks with 64 weights Q2_K working on CUDA. ~3% slower on GTX-1660, 10% slower on 4080. * k_quants: WIP super-blocks with 64 weights Q3_K working on CUDA. * k_quants: WIP super-blocks with 64 weights Q5_K working on CUDA, and with this CUDA is done. * k_quants: WIP super-blocks with 64 weights Q6_K working on ARM_NEON * k_quants: WIP super-blocks with 64 weights Q4_K working on ARM_NEON, but quite a bit slower than 256 weights * k_quants: WIP super-blocks with 64 weights Q2_K working on ARM_NEON, but quite a bit slower than 256 weights * k_quants: WIP super-blocks with 64 weights Q3_K working on ARM_NEON, but quite a bit slower than 256 weights. * k_quants: WIP super-blocks with 64 weights Q5_K working on ARM_NEON, but quite a bit slower than 256 weights. With that, we have full support for ARM_NEON, although performance is not quite there. * k_quants: WIP super-blocks with 64 weights Slightly more efficient Q3_K and Q5_K * k_quants: WIP super-blocks with 64 weights Another small improvement for Q3_K and Q5_K on ARM_NEON * k_quants: WIP super-blocks with 64 weights Yet another speedup for Q5_K on ARM_NEON. We are now within 10% of the QK_K = 256 version. * k_quants: WIP super-blocks with 64 weights * We are able to pass preprocessor macros to the Metal compiler * Q6_K works and is actually slightly more efficient than the QK_K = 256 version (25.2 ms vs 25.8 ms) * k_quants: WIP super-blocks with 64 weights Q4_K works on Metal and is actually slightly faster than QK_K = 256 (21.95 ms vs 24.0 ms). * k_quants: WIP super-blocks with 64 weights Q2_K works on Metal and is very slightly faster than QK_K = 256 (23.8 ms vs 24.2 ms). * k_quants: WIP super-blocks with 64 weights Q3_K works on Metal and is slightly faster than QK_K = 256 (26.6 ms vs 28.3 ms). * k_quants: WIP super-blocks with 64 weights Q5_K works on Metal and is slightly faster than QK_K = 256 (23.7 ms vs 26.3 ms). * k_quants: call them _K, not _k, also on Metal * k_quants: correctly define QK_K in llama.cpp * Fixed bug in q4_K quantization added with the 64-block addition * Simplify via lambda * k_quants: swicth Q3_K to 4-bit scales when QK_K = 64 Otherwise there isn't much benefit from this quantization type. There is some very slight loss in accuracy, but we reduce size by ~7%. E.g., for OpenLLaMA-3B, Q3_K_S perplexity is 8.6131 with 8-bit scales and 8.6352 with 4-bit, while file size decreases from 1.53G to 1.44G. * k_quants: switch Q4_K to 4-bit scales when QK_K = 64 Here the loss in accuracy is greater than for Q3_K, but the Q4_K points still move further to the left on the perplexity vs size curve. * k_quants: forgot to add the Metal changes in last commit * k_quants: change Q5_K to be type 0 when QK_K = 64 Still needs AVX2 implementation * k_quants: AVX2 implementation for new 64-weight Q5_K * k_quants: 10% faster ARM_NEON Q5_K dot product * k_quants: fixed issue caused by merging with master --------- Co-authored-by: Iwan Kawrakow <iwan.kawrakow@gmail.com>
2023-06-26 16:43:07 +00:00
const uint16_t a = *(const uint16_t *)x[i].scales;
aux16[0] = a & 0x0f0f;
aux16[1] = (a >> 4) & 0x0f0f;
ggml : add SOTA 2,3,4,5,6 bit k-quantizations (#1684) * Starting to add k-quantization to ggml I think it is better to have quantization separate from ggml. For now just adding the k-quants there, but it would be better to also factor out the existing ggml quantizations. * Adding Q3_K and Q8_K (de)-quantization * Q3_K now working on CUDA and AVX2/scalar CUDA is not ideal - ~50% slower than Q4_0 for single token prediction, about the same in batch mode (perplexity). CPU single token is ~55 ms (on Ryzen 7950X). * Some improvement for Q3_K on CUDA It is now ~22.5 ms/token on my GPU, so ~30% slower than Q4_0. * Some more CUDA optimizations for Q3_K Single token is now 20.5 ms/token (~20% slower than Q4_0). Perplexity is on par with Q4_0. * Adding Q4_K - scalar, AVX2, CUDA Performance is the same or perhaps very slightly better than Q4_0 on the CPU. On the GPU, single token prediction is ~10% better than Q4_0, batch mode (perplexity is about the same). * Adding Q6_K - scalar, AVX2, CUDA Performance is ~40% lower compared to Q4_K on the CPU. This is to be expected, considering that we are memory bound on the CPU and the 6-bit model is ~44% larger than the 4-bit. On the GPU, single token prediction is ~6% lower than Q4_0, batch mode (perplexity) is even closer (but still slower). * Adding Q5_K - scalar, AVX2, CUDA Performance is ~20% lower compared to Q4_K on the CPU. This is to be expected, considering that we are memory bound on the CPU and the 5-bit model is ~22% larger than the 4-bit. On the GPU, single token prediction is about the same as Q4_0 for both, single token and batch prediction. * Per convention, all QX_K quantizations use Q5_K for output.weight * Adding quantization mixes * Quantization mixes: didn't quite get what I wanted in the last commit * Q4_K dot product for ARM_NEON * Q6_K dot product for ARM_NEON * Q5_K dot product for ARM_NEON * Adding Q3_K dot for ARM_NEON It is 22% slower than Q4_K, despite the smaller model size. On x86_64, where we are memory bound, the Q3_K model is quite a bit faster than Q4_K. * A very slightly faster ARM_NEON Q3_K dot * Adding Q2_K - just CUDA for now Token prediction is pretty good - about 15.5 ms on a RTX 4080. Perplexity is about the same as Q4_K. * Adding scalar and AVX2 Q2_K dot * Adding ARM_NEON Q2_K dot About the same performance as Q4_K. * A slightly faster ARM_NEON Q2_K dot Single token prediction is now ~36 ms on M2 Max. The code is much simpler too. * Fixed bug in Q2_K CUDA dot product kernel Stranegly enough, for the few prompts I tried with the 7B model the responses looked perfectly reasonable. Only realized something is not quite right when I tried the larger models and started getting nonse back. In any case, Q2_K single token evaluation time on an RTX 4080 in a Ryzen7950X box iusing CUDA and model fully loaded on the GPU are ~15.5 ms for 7B, ~25.4 ms for 13B, and ~55.8 ms for 30B. The max number of layers that fit in VRAM for The 65B is 32. With that, we get ~330 ms per token, which is not that much faster than just running on the CPU (~470 ms per token). * Don't print zeros/NaNs when no count histogram has been collected * A 10% faster CUDA vector dot kernel for Q3_K Q3_K is now running at ~18.5 ms / token on CUDA, so the gap to Q4_0 is only 10%. It seems memory acccess pattern is more important for performance than the amount of computation the kernel does. * A slightly daster Q4_K AVX2 dot product For perplexity, where we are less memory bound, time per pass drops by ~5%. Barely measurable difference for single token prediction. * A slightly faster ARM_NEON A4_K dot product * Minor * Fix quantization error test We cannot possibly be expecting rmse < 0.002 for 2- and 3-bit quantization variants. * Fix docker build I have been sloppy with vector reinterpret casts on ARM_NEON. It seems clang is very forgiving in that regard. * Added forgotten ggml.o dependence on k_quants.h to the Makefile * Had unintentionally committed the Makefile with -Ofast enabled * ggml : rename k_quants -> ggml-quants-k, use lowercase in code --------- Co-authored-by: Iwan Kawrakow <iwan.kawrakow@gmail.com> Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
2023-06-05 19:56:18 +00:00
k-quants : support for super-block size of 64 (#2001) * k_quants: WIP super-blocks with 64 weights * k_quants: WIP super-blocks with 64 weights Q6_K scalar and AVX2 works * k_quants: WIP super-blocks with 64 weights Q4_K scalar and AVX2 works * k_quants: WIP super-blocks with 64 weights Q2_K scalar and AVX2 works. Q2_K is way too slow (it is actually slower than the scalar implementation) * k_quants: WIP super-blocks with 64 weights Q3_K scalar and AVX2 works. * k_quants: WIP super-blocks with 64 weights Q5_K scalar and AVX2 works, and with that all k_quants are done on AVX2 and scalar * k_quants: WIP super-blocks with 64 weights Q6_K working on CUDA. Cannot make it run quite as gast as with super-blocks with 256 weigths: 8% slower on 4080, 20% slower on the 1660 (but there we fit 1 less layer on the GPU because pf the larger model size), so some fraction of these 20% is due to that, * k_quants: WIP super-blocks with 64 weights Q4_K working on CUDA. ~10% slower on GTX-1660, 16% slower on 4080. * k_quants: WIP super-blocks with 64 weights Q2_K working on CUDA. ~3% slower on GTX-1660, 10% slower on 4080. * k_quants: WIP super-blocks with 64 weights Q3_K working on CUDA. * k_quants: WIP super-blocks with 64 weights Q5_K working on CUDA, and with this CUDA is done. * k_quants: WIP super-blocks with 64 weights Q6_K working on ARM_NEON * k_quants: WIP super-blocks with 64 weights Q4_K working on ARM_NEON, but quite a bit slower than 256 weights * k_quants: WIP super-blocks with 64 weights Q2_K working on ARM_NEON, but quite a bit slower than 256 weights * k_quants: WIP super-blocks with 64 weights Q3_K working on ARM_NEON, but quite a bit slower than 256 weights. * k_quants: WIP super-blocks with 64 weights Q5_K working on ARM_NEON, but quite a bit slower than 256 weights. With that, we have full support for ARM_NEON, although performance is not quite there. * k_quants: WIP super-blocks with 64 weights Slightly more efficient Q3_K and Q5_K * k_quants: WIP super-blocks with 64 weights Another small improvement for Q3_K and Q5_K on ARM_NEON * k_quants: WIP super-blocks with 64 weights Yet another speedup for Q5_K on ARM_NEON. We are now within 10% of the QK_K = 256 version. * k_quants: WIP super-blocks with 64 weights * We are able to pass preprocessor macros to the Metal compiler * Q6_K works and is actually slightly more efficient than the QK_K = 256 version (25.2 ms vs 25.8 ms) * k_quants: WIP super-blocks with 64 weights Q4_K works on Metal and is actually slightly faster than QK_K = 256 (21.95 ms vs 24.0 ms). * k_quants: WIP super-blocks with 64 weights Q2_K works on Metal and is very slightly faster than QK_K = 256 (23.8 ms vs 24.2 ms). * k_quants: WIP super-blocks with 64 weights Q3_K works on Metal and is slightly faster than QK_K = 256 (26.6 ms vs 28.3 ms). * k_quants: WIP super-blocks with 64 weights Q5_K works on Metal and is slightly faster than QK_K = 256 (23.7 ms vs 26.3 ms). * k_quants: call them _K, not _k, also on Metal * k_quants: correctly define QK_K in llama.cpp * Fixed bug in q4_K quantization added with the 64-block addition * Simplify via lambda * k_quants: swicth Q3_K to 4-bit scales when QK_K = 64 Otherwise there isn't much benefit from this quantization type. There is some very slight loss in accuracy, but we reduce size by ~7%. E.g., for OpenLLaMA-3B, Q3_K_S perplexity is 8.6131 with 8-bit scales and 8.6352 with 4-bit, while file size decreases from 1.53G to 1.44G. * k_quants: switch Q4_K to 4-bit scales when QK_K = 64 Here the loss in accuracy is greater than for Q3_K, but the Q4_K points still move further to the left on the perplexity vs size curve. * k_quants: forgot to add the Metal changes in last commit * k_quants: change Q5_K to be type 0 when QK_K = 64 Still needs AVX2 implementation * k_quants: AVX2 implementation for new 64-weight Q5_K * k_quants: 10% faster ARM_NEON Q5_K dot product * k_quants: fixed issue caused by merging with master --------- Co-authored-by: Iwan Kawrakow <iwan.kawrakow@gmail.com>
2023-06-26 16:43:07 +00:00
for (int j = 0; j < 4; ++j) scales[j] -= 8;
ggml : add SOTA 2,3,4,5,6 bit k-quantizations (#1684) * Starting to add k-quantization to ggml I think it is better to have quantization separate from ggml. For now just adding the k-quants there, but it would be better to also factor out the existing ggml quantizations. * Adding Q3_K and Q8_K (de)-quantization * Q3_K now working on CUDA and AVX2/scalar CUDA is not ideal - ~50% slower than Q4_0 for single token prediction, about the same in batch mode (perplexity). CPU single token is ~55 ms (on Ryzen 7950X). * Some improvement for Q3_K on CUDA It is now ~22.5 ms/token on my GPU, so ~30% slower than Q4_0. * Some more CUDA optimizations for Q3_K Single token is now 20.5 ms/token (~20% slower than Q4_0). Perplexity is on par with Q4_0. * Adding Q4_K - scalar, AVX2, CUDA Performance is the same or perhaps very slightly better than Q4_0 on the CPU. On the GPU, single token prediction is ~10% better than Q4_0, batch mode (perplexity is about the same). * Adding Q6_K - scalar, AVX2, CUDA Performance is ~40% lower compared to Q4_K on the CPU. This is to be expected, considering that we are memory bound on the CPU and the 6-bit model is ~44% larger than the 4-bit. On the GPU, single token prediction is ~6% lower than Q4_0, batch mode (perplexity) is even closer (but still slower). * Adding Q5_K - scalar, AVX2, CUDA Performance is ~20% lower compared to Q4_K on the CPU. This is to be expected, considering that we are memory bound on the CPU and the 5-bit model is ~22% larger than the 4-bit. On the GPU, single token prediction is about the same as Q4_0 for both, single token and batch prediction. * Per convention, all QX_K quantizations use Q5_K for output.weight * Adding quantization mixes * Quantization mixes: didn't quite get what I wanted in the last commit * Q4_K dot product for ARM_NEON * Q6_K dot product for ARM_NEON * Q5_K dot product for ARM_NEON * Adding Q3_K dot for ARM_NEON It is 22% slower than Q4_K, despite the smaller model size. On x86_64, where we are memory bound, the Q3_K model is quite a bit faster than Q4_K. * A very slightly faster ARM_NEON Q3_K dot * Adding Q2_K - just CUDA for now Token prediction is pretty good - about 15.5 ms on a RTX 4080. Perplexity is about the same as Q4_K. * Adding scalar and AVX2 Q2_K dot * Adding ARM_NEON Q2_K dot About the same performance as Q4_K. * A slightly faster ARM_NEON Q2_K dot Single token prediction is now ~36 ms on M2 Max. The code is much simpler too. * Fixed bug in Q2_K CUDA dot product kernel Stranegly enough, for the few prompts I tried with the 7B model the responses looked perfectly reasonable. Only realized something is not quite right when I tried the larger models and started getting nonse back. In any case, Q2_K single token evaluation time on an RTX 4080 in a Ryzen7950X box iusing CUDA and model fully loaded on the GPU are ~15.5 ms for 7B, ~25.4 ms for 13B, and ~55.8 ms for 30B. The max number of layers that fit in VRAM for The 65B is 32. With that, we get ~330 ms per token, which is not that much faster than just running on the CPU (~470 ms per token). * Don't print zeros/NaNs when no count histogram has been collected * A 10% faster CUDA vector dot kernel for Q3_K Q3_K is now running at ~18.5 ms / token on CUDA, so the gap to Q4_0 is only 10%. It seems memory acccess pattern is more important for performance than the amount of computation the kernel does. * A slightly daster Q4_K AVX2 dot product For perplexity, where we are less memory bound, time per pass drops by ~5%. Barely measurable difference for single token prediction. * A slightly faster ARM_NEON A4_K dot product * Minor * Fix quantization error test We cannot possibly be expecting rmse < 0.002 for 2- and 3-bit quantization variants. * Fix docker build I have been sloppy with vector reinterpret casts on ARM_NEON. It seems clang is very forgiving in that regard. * Added forgotten ggml.o dependence on k_quants.h to the Makefile * Had unintentionally committed the Makefile with -Ofast enabled * ggml : rename k_quants -> ggml-quants-k, use lowercase in code --------- Co-authored-by: Iwan Kawrakow <iwan.kawrakow@gmail.com> Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
2023-06-05 19:56:18 +00:00
k-quants : support for super-block size of 64 (#2001) * k_quants: WIP super-blocks with 64 weights * k_quants: WIP super-blocks with 64 weights Q6_K scalar and AVX2 works * k_quants: WIP super-blocks with 64 weights Q4_K scalar and AVX2 works * k_quants: WIP super-blocks with 64 weights Q2_K scalar and AVX2 works. Q2_K is way too slow (it is actually slower than the scalar implementation) * k_quants: WIP super-blocks with 64 weights Q3_K scalar and AVX2 works. * k_quants: WIP super-blocks with 64 weights Q5_K scalar and AVX2 works, and with that all k_quants are done on AVX2 and scalar * k_quants: WIP super-blocks with 64 weights Q6_K working on CUDA. Cannot make it run quite as gast as with super-blocks with 256 weigths: 8% slower on 4080, 20% slower on the 1660 (but there we fit 1 less layer on the GPU because pf the larger model size), so some fraction of these 20% is due to that, * k_quants: WIP super-blocks with 64 weights Q4_K working on CUDA. ~10% slower on GTX-1660, 16% slower on 4080. * k_quants: WIP super-blocks with 64 weights Q2_K working on CUDA. ~3% slower on GTX-1660, 10% slower on 4080. * k_quants: WIP super-blocks with 64 weights Q3_K working on CUDA. * k_quants: WIP super-blocks with 64 weights Q5_K working on CUDA, and with this CUDA is done. * k_quants: WIP super-blocks with 64 weights Q6_K working on ARM_NEON * k_quants: WIP super-blocks with 64 weights Q4_K working on ARM_NEON, but quite a bit slower than 256 weights * k_quants: WIP super-blocks with 64 weights Q2_K working on ARM_NEON, but quite a bit slower than 256 weights * k_quants: WIP super-blocks with 64 weights Q3_K working on ARM_NEON, but quite a bit slower than 256 weights. * k_quants: WIP super-blocks with 64 weights Q5_K working on ARM_NEON, but quite a bit slower than 256 weights. With that, we have full support for ARM_NEON, although performance is not quite there. * k_quants: WIP super-blocks with 64 weights Slightly more efficient Q3_K and Q5_K * k_quants: WIP super-blocks with 64 weights Another small improvement for Q3_K and Q5_K on ARM_NEON * k_quants: WIP super-blocks with 64 weights Yet another speedup for Q5_K on ARM_NEON. We are now within 10% of the QK_K = 256 version. * k_quants: WIP super-blocks with 64 weights * We are able to pass preprocessor macros to the Metal compiler * Q6_K works and is actually slightly more efficient than the QK_K = 256 version (25.2 ms vs 25.8 ms) * k_quants: WIP super-blocks with 64 weights Q4_K works on Metal and is actually slightly faster than QK_K = 256 (21.95 ms vs 24.0 ms). * k_quants: WIP super-blocks with 64 weights Q2_K works on Metal and is very slightly faster than QK_K = 256 (23.8 ms vs 24.2 ms). * k_quants: WIP super-blocks with 64 weights Q3_K works on Metal and is slightly faster than QK_K = 256 (26.6 ms vs 28.3 ms). * k_quants: WIP super-blocks with 64 weights Q5_K works on Metal and is slightly faster than QK_K = 256 (23.7 ms vs 26.3 ms). * k_quants: call them _K, not _k, also on Metal * k_quants: correctly define QK_K in llama.cpp * Fixed bug in q4_K quantization added with the 64-block addition * Simplify via lambda * k_quants: swicth Q3_K to 4-bit scales when QK_K = 64 Otherwise there isn't much benefit from this quantization type. There is some very slight loss in accuracy, but we reduce size by ~7%. E.g., for OpenLLaMA-3B, Q3_K_S perplexity is 8.6131 with 8-bit scales and 8.6352 with 4-bit, while file size decreases from 1.53G to 1.44G. * k_quants: switch Q4_K to 4-bit scales when QK_K = 64 Here the loss in accuracy is greater than for Q3_K, but the Q4_K points still move further to the left on the perplexity vs size curve. * k_quants: forgot to add the Metal changes in last commit * k_quants: change Q5_K to be type 0 when QK_K = 64 Still needs AVX2 implementation * k_quants: AVX2 implementation for new 64-weight Q5_K * k_quants: 10% faster ARM_NEON Q5_K dot product * k_quants: fixed issue caused by merging with master --------- Co-authored-by: Iwan Kawrakow <iwan.kawrakow@gmail.com>
2023-06-26 16:43:07 +00:00
int32_t isum = -4*(scales[0] * y[i].bsums[0] + scales[2] * y[i].bsums[1] + scales[1] * y[i].bsums[2] + scales[3] * y[i].bsums[3]);
ggml : add SOTA 2,3,4,5,6 bit k-quantizations (#1684) * Starting to add k-quantization to ggml I think it is better to have quantization separate from ggml. For now just adding the k-quants there, but it would be better to also factor out the existing ggml quantizations. * Adding Q3_K and Q8_K (de)-quantization * Q3_K now working on CUDA and AVX2/scalar CUDA is not ideal - ~50% slower than Q4_0 for single token prediction, about the same in batch mode (perplexity). CPU single token is ~55 ms (on Ryzen 7950X). * Some improvement for Q3_K on CUDA It is now ~22.5 ms/token on my GPU, so ~30% slower than Q4_0. * Some more CUDA optimizations for Q3_K Single token is now 20.5 ms/token (~20% slower than Q4_0). Perplexity is on par with Q4_0. * Adding Q4_K - scalar, AVX2, CUDA Performance is the same or perhaps very slightly better than Q4_0 on the CPU. On the GPU, single token prediction is ~10% better than Q4_0, batch mode (perplexity is about the same). * Adding Q6_K - scalar, AVX2, CUDA Performance is ~40% lower compared to Q4_K on the CPU. This is to be expected, considering that we are memory bound on the CPU and the 6-bit model is ~44% larger than the 4-bit. On the GPU, single token prediction is ~6% lower than Q4_0, batch mode (perplexity) is even closer (but still slower). * Adding Q5_K - scalar, AVX2, CUDA Performance is ~20% lower compared to Q4_K on the CPU. This is to be expected, considering that we are memory bound on the CPU and the 5-bit model is ~22% larger than the 4-bit. On the GPU, single token prediction is about the same as Q4_0 for both, single token and batch prediction. * Per convention, all QX_K quantizations use Q5_K for output.weight * Adding quantization mixes * Quantization mixes: didn't quite get what I wanted in the last commit * Q4_K dot product for ARM_NEON * Q6_K dot product for ARM_NEON * Q5_K dot product for ARM_NEON * Adding Q3_K dot for ARM_NEON It is 22% slower than Q4_K, despite the smaller model size. On x86_64, where we are memory bound, the Q3_K model is quite a bit faster than Q4_K. * A very slightly faster ARM_NEON Q3_K dot * Adding Q2_K - just CUDA for now Token prediction is pretty good - about 15.5 ms on a RTX 4080. Perplexity is about the same as Q4_K. * Adding scalar and AVX2 Q2_K dot * Adding ARM_NEON Q2_K dot About the same performance as Q4_K. * A slightly faster ARM_NEON Q2_K dot Single token prediction is now ~36 ms on M2 Max. The code is much simpler too. * Fixed bug in Q2_K CUDA dot product kernel Stranegly enough, for the few prompts I tried with the 7B model the responses looked perfectly reasonable. Only realized something is not quite right when I tried the larger models and started getting nonse back. In any case, Q2_K single token evaluation time on an RTX 4080 in a Ryzen7950X box iusing CUDA and model fully loaded on the GPU are ~15.5 ms for 7B, ~25.4 ms for 13B, and ~55.8 ms for 30B. The max number of layers that fit in VRAM for The 65B is 32. With that, we get ~330 ms per token, which is not that much faster than just running on the CPU (~470 ms per token). * Don't print zeros/NaNs when no count histogram has been collected * A 10% faster CUDA vector dot kernel for Q3_K Q3_K is now running at ~18.5 ms / token on CUDA, so the gap to Q4_0 is only 10%. It seems memory acccess pattern is more important for performance than the amount of computation the kernel does. * A slightly daster Q4_K AVX2 dot product For perplexity, where we are less memory bound, time per pass drops by ~5%. Barely measurable difference for single token prediction. * A slightly faster ARM_NEON A4_K dot product * Minor * Fix quantization error test We cannot possibly be expecting rmse < 0.002 for 2- and 3-bit quantization variants. * Fix docker build I have been sloppy with vector reinterpret casts on ARM_NEON. It seems clang is very forgiving in that regard. * Added forgotten ggml.o dependence on k_quants.h to the Makefile * Had unintentionally committed the Makefile with -Ofast enabled * ggml : rename k_quants -> ggml-quants-k, use lowercase in code --------- Co-authored-by: Iwan Kawrakow <iwan.kawrakow@gmail.com> Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
2023-06-05 19:56:18 +00:00
k-quants : support for super-block size of 64 (#2001) * k_quants: WIP super-blocks with 64 weights * k_quants: WIP super-blocks with 64 weights Q6_K scalar and AVX2 works * k_quants: WIP super-blocks with 64 weights Q4_K scalar and AVX2 works * k_quants: WIP super-blocks with 64 weights Q2_K scalar and AVX2 works. Q2_K is way too slow (it is actually slower than the scalar implementation) * k_quants: WIP super-blocks with 64 weights Q3_K scalar and AVX2 works. * k_quants: WIP super-blocks with 64 weights Q5_K scalar and AVX2 works, and with that all k_quants are done on AVX2 and scalar * k_quants: WIP super-blocks with 64 weights Q6_K working on CUDA. Cannot make it run quite as gast as with super-blocks with 256 weigths: 8% slower on 4080, 20% slower on the 1660 (but there we fit 1 less layer on the GPU because pf the larger model size), so some fraction of these 20% is due to that, * k_quants: WIP super-blocks with 64 weights Q4_K working on CUDA. ~10% slower on GTX-1660, 16% slower on 4080. * k_quants: WIP super-blocks with 64 weights Q2_K working on CUDA. ~3% slower on GTX-1660, 10% slower on 4080. * k_quants: WIP super-blocks with 64 weights Q3_K working on CUDA. * k_quants: WIP super-blocks with 64 weights Q5_K working on CUDA, and with this CUDA is done. * k_quants: WIP super-blocks with 64 weights Q6_K working on ARM_NEON * k_quants: WIP super-blocks with 64 weights Q4_K working on ARM_NEON, but quite a bit slower than 256 weights * k_quants: WIP super-blocks with 64 weights Q2_K working on ARM_NEON, but quite a bit slower than 256 weights * k_quants: WIP super-blocks with 64 weights Q3_K working on ARM_NEON, but quite a bit slower than 256 weights. * k_quants: WIP super-blocks with 64 weights Q5_K working on ARM_NEON, but quite a bit slower than 256 weights. With that, we have full support for ARM_NEON, although performance is not quite there. * k_quants: WIP super-blocks with 64 weights Slightly more efficient Q3_K and Q5_K * k_quants: WIP super-blocks with 64 weights Another small improvement for Q3_K and Q5_K on ARM_NEON * k_quants: WIP super-blocks with 64 weights Yet another speedup for Q5_K on ARM_NEON. We are now within 10% of the QK_K = 256 version. * k_quants: WIP super-blocks with 64 weights * We are able to pass preprocessor macros to the Metal compiler * Q6_K works and is actually slightly more efficient than the QK_K = 256 version (25.2 ms vs 25.8 ms) * k_quants: WIP super-blocks with 64 weights Q4_K works on Metal and is actually slightly faster than QK_K = 256 (21.95 ms vs 24.0 ms). * k_quants: WIP super-blocks with 64 weights Q2_K works on Metal and is very slightly faster than QK_K = 256 (23.8 ms vs 24.2 ms). * k_quants: WIP super-blocks with 64 weights Q3_K works on Metal and is slightly faster than QK_K = 256 (26.6 ms vs 28.3 ms). * k_quants: WIP super-blocks with 64 weights Q5_K works on Metal and is slightly faster than QK_K = 256 (23.7 ms vs 26.3 ms). * k_quants: call them _K, not _k, also on Metal * k_quants: correctly define QK_K in llama.cpp * Fixed bug in q4_K quantization added with the 64-block addition * Simplify via lambda * k_quants: swicth Q3_K to 4-bit scales when QK_K = 64 Otherwise there isn't much benefit from this quantization type. There is some very slight loss in accuracy, but we reduce size by ~7%. E.g., for OpenLLaMA-3B, Q3_K_S perplexity is 8.6131 with 8-bit scales and 8.6352 with 4-bit, while file size decreases from 1.53G to 1.44G. * k_quants: switch Q4_K to 4-bit scales when QK_K = 64 Here the loss in accuracy is greater than for Q3_K, but the Q4_K points still move further to the left on the perplexity vs size curve. * k_quants: forgot to add the Metal changes in last commit * k_quants: change Q5_K to be type 0 when QK_K = 64 Still needs AVX2 implementation * k_quants: AVX2 implementation for new 64-weight Q5_K * k_quants: 10% faster ARM_NEON Q5_K dot product * k_quants: fixed issue caused by merging with master --------- Co-authored-by: Iwan Kawrakow <iwan.kawrakow@gmail.com>
2023-06-26 16:43:07 +00:00
const float d = y[i].d * (float)x[i].d;
ggml : add SOTA 2,3,4,5,6 bit k-quantizations (#1684) * Starting to add k-quantization to ggml I think it is better to have quantization separate from ggml. For now just adding the k-quants there, but it would be better to also factor out the existing ggml quantizations. * Adding Q3_K and Q8_K (de)-quantization * Q3_K now working on CUDA and AVX2/scalar CUDA is not ideal - ~50% slower than Q4_0 for single token prediction, about the same in batch mode (perplexity). CPU single token is ~55 ms (on Ryzen 7950X). * Some improvement for Q3_K on CUDA It is now ~22.5 ms/token on my GPU, so ~30% slower than Q4_0. * Some more CUDA optimizations for Q3_K Single token is now 20.5 ms/token (~20% slower than Q4_0). Perplexity is on par with Q4_0. * Adding Q4_K - scalar, AVX2, CUDA Performance is the same or perhaps very slightly better than Q4_0 on the CPU. On the GPU, single token prediction is ~10% better than Q4_0, batch mode (perplexity is about the same). * Adding Q6_K - scalar, AVX2, CUDA Performance is ~40% lower compared to Q4_K on the CPU. This is to be expected, considering that we are memory bound on the CPU and the 6-bit model is ~44% larger than the 4-bit. On the GPU, single token prediction is ~6% lower than Q4_0, batch mode (perplexity) is even closer (but still slower). * Adding Q5_K - scalar, AVX2, CUDA Performance is ~20% lower compared to Q4_K on the CPU. This is to be expected, considering that we are memory bound on the CPU and the 5-bit model is ~22% larger than the 4-bit. On the GPU, single token prediction is about the same as Q4_0 for both, single token and batch prediction. * Per convention, all QX_K quantizations use Q5_K for output.weight * Adding quantization mixes * Quantization mixes: didn't quite get what I wanted in the last commit * Q4_K dot product for ARM_NEON * Q6_K dot product for ARM_NEON * Q5_K dot product for ARM_NEON * Adding Q3_K dot for ARM_NEON It is 22% slower than Q4_K, despite the smaller model size. On x86_64, where we are memory bound, the Q3_K model is quite a bit faster than Q4_K. * A very slightly faster ARM_NEON Q3_K dot * Adding Q2_K - just CUDA for now Token prediction is pretty good - about 15.5 ms on a RTX 4080. Perplexity is about the same as Q4_K. * Adding scalar and AVX2 Q2_K dot * Adding ARM_NEON Q2_K dot About the same performance as Q4_K. * A slightly faster ARM_NEON Q2_K dot Single token prediction is now ~36 ms on M2 Max. The code is much simpler too. * Fixed bug in Q2_K CUDA dot product kernel Stranegly enough, for the few prompts I tried with the 7B model the responses looked perfectly reasonable. Only realized something is not quite right when I tried the larger models and started getting nonse back. In any case, Q2_K single token evaluation time on an RTX 4080 in a Ryzen7950X box iusing CUDA and model fully loaded on the GPU are ~15.5 ms for 7B, ~25.4 ms for 13B, and ~55.8 ms for 30B. The max number of layers that fit in VRAM for The 65B is 32. With that, we get ~330 ms per token, which is not that much faster than just running on the CPU (~470 ms per token). * Don't print zeros/NaNs when no count histogram has been collected * A 10% faster CUDA vector dot kernel for Q3_K Q3_K is now running at ~18.5 ms / token on CUDA, so the gap to Q4_0 is only 10%. It seems memory acccess pattern is more important for performance than the amount of computation the kernel does. * A slightly daster Q4_K AVX2 dot product For perplexity, where we are less memory bound, time per pass drops by ~5%. Barely measurable difference for single token prediction. * A slightly faster ARM_NEON A4_K dot product * Minor * Fix quantization error test We cannot possibly be expecting rmse < 0.002 for 2- and 3-bit quantization variants. * Fix docker build I have been sloppy with vector reinterpret casts on ARM_NEON. It seems clang is very forgiving in that regard. * Added forgotten ggml.o dependence on k_quants.h to the Makefile * Had unintentionally committed the Makefile with -Ofast enabled * ggml : rename k_quants -> ggml-quants-k, use lowercase in code --------- Co-authored-by: Iwan Kawrakow <iwan.kawrakow@gmail.com> Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
2023-06-05 19:56:18 +00:00
k-quants : support for super-block size of 64 (#2001) * k_quants: WIP super-blocks with 64 weights * k_quants: WIP super-blocks with 64 weights Q6_K scalar and AVX2 works * k_quants: WIP super-blocks with 64 weights Q4_K scalar and AVX2 works * k_quants: WIP super-blocks with 64 weights Q2_K scalar and AVX2 works. Q2_K is way too slow (it is actually slower than the scalar implementation) * k_quants: WIP super-blocks with 64 weights Q3_K scalar and AVX2 works. * k_quants: WIP super-blocks with 64 weights Q5_K scalar and AVX2 works, and with that all k_quants are done on AVX2 and scalar * k_quants: WIP super-blocks with 64 weights Q6_K working on CUDA. Cannot make it run quite as gast as with super-blocks with 256 weigths: 8% slower on 4080, 20% slower on the 1660 (but there we fit 1 less layer on the GPU because pf the larger model size), so some fraction of these 20% is due to that, * k_quants: WIP super-blocks with 64 weights Q4_K working on CUDA. ~10% slower on GTX-1660, 16% slower on 4080. * k_quants: WIP super-blocks with 64 weights Q2_K working on CUDA. ~3% slower on GTX-1660, 10% slower on 4080. * k_quants: WIP super-blocks with 64 weights Q3_K working on CUDA. * k_quants: WIP super-blocks with 64 weights Q5_K working on CUDA, and with this CUDA is done. * k_quants: WIP super-blocks with 64 weights Q6_K working on ARM_NEON * k_quants: WIP super-blocks with 64 weights Q4_K working on ARM_NEON, but quite a bit slower than 256 weights * k_quants: WIP super-blocks with 64 weights Q2_K working on ARM_NEON, but quite a bit slower than 256 weights * k_quants: WIP super-blocks with 64 weights Q3_K working on ARM_NEON, but quite a bit slower than 256 weights. * k_quants: WIP super-blocks with 64 weights Q5_K working on ARM_NEON, but quite a bit slower than 256 weights. With that, we have full support for ARM_NEON, although performance is not quite there. * k_quants: WIP super-blocks with 64 weights Slightly more efficient Q3_K and Q5_K * k_quants: WIP super-blocks with 64 weights Another small improvement for Q3_K and Q5_K on ARM_NEON * k_quants: WIP super-blocks with 64 weights Yet another speedup for Q5_K on ARM_NEON. We are now within 10% of the QK_K = 256 version. * k_quants: WIP super-blocks with 64 weights * We are able to pass preprocessor macros to the Metal compiler * Q6_K works and is actually slightly more efficient than the QK_K = 256 version (25.2 ms vs 25.8 ms) * k_quants: WIP super-blocks with 64 weights Q4_K works on Metal and is actually slightly faster than QK_K = 256 (21.95 ms vs 24.0 ms). * k_quants: WIP super-blocks with 64 weights Q2_K works on Metal and is very slightly faster than QK_K = 256 (23.8 ms vs 24.2 ms). * k_quants: WIP super-blocks with 64 weights Q3_K works on Metal and is slightly faster than QK_K = 256 (26.6 ms vs 28.3 ms). * k_quants: WIP super-blocks with 64 weights Q5_K works on Metal and is slightly faster than QK_K = 256 (23.7 ms vs 26.3 ms). * k_quants: call them _K, not _k, also on Metal * k_quants: correctly define QK_K in llama.cpp * Fixed bug in q4_K quantization added with the 64-block addition * Simplify via lambda * k_quants: swicth Q3_K to 4-bit scales when QK_K = 64 Otherwise there isn't much benefit from this quantization type. There is some very slight loss in accuracy, but we reduce size by ~7%. E.g., for OpenLLaMA-3B, Q3_K_S perplexity is 8.6131 with 8-bit scales and 8.6352 with 4-bit, while file size decreases from 1.53G to 1.44G. * k_quants: switch Q4_K to 4-bit scales when QK_K = 64 Here the loss in accuracy is greater than for Q3_K, but the Q4_K points still move further to the left on the perplexity vs size curve. * k_quants: forgot to add the Metal changes in last commit * k_quants: change Q5_K to be type 0 when QK_K = 64 Still needs AVX2 implementation * k_quants: AVX2 implementation for new 64-weight Q5_K * k_quants: 10% faster ARM_NEON Q5_K dot product * k_quants: fixed issue caused by merging with master --------- Co-authored-by: Iwan Kawrakow <iwan.kawrakow@gmail.com>
2023-06-26 16:43:07 +00:00
const uint8x16_t htmp = vcombine_u8(hbits, vshr_n_u8(hbits, 1));
q3h.val[0] = vandq_u8(mh, vshlq_n_u8(htmp, 2));
q3h.val[1] = vandq_u8(mh, htmp);
q3h.val[2] = vandq_u8(mh, vshrq_n_u8(htmp, 2));
q3h.val[3] = vandq_u8(mh, vshrq_n_u8(htmp, 4));
ggml : add SOTA 2,3,4,5,6 bit k-quantizations (#1684) * Starting to add k-quantization to ggml I think it is better to have quantization separate from ggml. For now just adding the k-quants there, but it would be better to also factor out the existing ggml quantizations. * Adding Q3_K and Q8_K (de)-quantization * Q3_K now working on CUDA and AVX2/scalar CUDA is not ideal - ~50% slower than Q4_0 for single token prediction, about the same in batch mode (perplexity). CPU single token is ~55 ms (on Ryzen 7950X). * Some improvement for Q3_K on CUDA It is now ~22.5 ms/token on my GPU, so ~30% slower than Q4_0. * Some more CUDA optimizations for Q3_K Single token is now 20.5 ms/token (~20% slower than Q4_0). Perplexity is on par with Q4_0. * Adding Q4_K - scalar, AVX2, CUDA Performance is the same or perhaps very slightly better than Q4_0 on the CPU. On the GPU, single token prediction is ~10% better than Q4_0, batch mode (perplexity is about the same). * Adding Q6_K - scalar, AVX2, CUDA Performance is ~40% lower compared to Q4_K on the CPU. This is to be expected, considering that we are memory bound on the CPU and the 6-bit model is ~44% larger than the 4-bit. On the GPU, single token prediction is ~6% lower than Q4_0, batch mode (perplexity) is even closer (but still slower). * Adding Q5_K - scalar, AVX2, CUDA Performance is ~20% lower compared to Q4_K on the CPU. This is to be expected, considering that we are memory bound on the CPU and the 5-bit model is ~22% larger than the 4-bit. On the GPU, single token prediction is about the same as Q4_0 for both, single token and batch prediction. * Per convention, all QX_K quantizations use Q5_K for output.weight * Adding quantization mixes * Quantization mixes: didn't quite get what I wanted in the last commit * Q4_K dot product for ARM_NEON * Q6_K dot product for ARM_NEON * Q5_K dot product for ARM_NEON * Adding Q3_K dot for ARM_NEON It is 22% slower than Q4_K, despite the smaller model size. On x86_64, where we are memory bound, the Q3_K model is quite a bit faster than Q4_K. * A very slightly faster ARM_NEON Q3_K dot * Adding Q2_K - just CUDA for now Token prediction is pretty good - about 15.5 ms on a RTX 4080. Perplexity is about the same as Q4_K. * Adding scalar and AVX2 Q2_K dot * Adding ARM_NEON Q2_K dot About the same performance as Q4_K. * A slightly faster ARM_NEON Q2_K dot Single token prediction is now ~36 ms on M2 Max. The code is much simpler too. * Fixed bug in Q2_K CUDA dot product kernel Stranegly enough, for the few prompts I tried with the 7B model the responses looked perfectly reasonable. Only realized something is not quite right when I tried the larger models and started getting nonse back. In any case, Q2_K single token evaluation time on an RTX 4080 in a Ryzen7950X box iusing CUDA and model fully loaded on the GPU are ~15.5 ms for 7B, ~25.4 ms for 13B, and ~55.8 ms for 30B. The max number of layers that fit in VRAM for The 65B is 32. With that, we get ~330 ms per token, which is not that much faster than just running on the CPU (~470 ms per token). * Don't print zeros/NaNs when no count histogram has been collected * A 10% faster CUDA vector dot kernel for Q3_K Q3_K is now running at ~18.5 ms / token on CUDA, so the gap to Q4_0 is only 10%. It seems memory acccess pattern is more important for performance than the amount of computation the kernel does. * A slightly daster Q4_K AVX2 dot product For perplexity, where we are less memory bound, time per pass drops by ~5%. Barely measurable difference for single token prediction. * A slightly faster ARM_NEON A4_K dot product * Minor * Fix quantization error test We cannot possibly be expecting rmse < 0.002 for 2- and 3-bit quantization variants. * Fix docker build I have been sloppy with vector reinterpret casts on ARM_NEON. It seems clang is very forgiving in that regard. * Added forgotten ggml.o dependence on k_quants.h to the Makefile * Had unintentionally committed the Makefile with -Ofast enabled * ggml : rename k_quants -> ggml-quants-k, use lowercase in code --------- Co-authored-by: Iwan Kawrakow <iwan.kawrakow@gmail.com> Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
2023-06-05 19:56:18 +00:00
k-quants : support for super-block size of 64 (#2001) * k_quants: WIP super-blocks with 64 weights * k_quants: WIP super-blocks with 64 weights Q6_K scalar and AVX2 works * k_quants: WIP super-blocks with 64 weights Q4_K scalar and AVX2 works * k_quants: WIP super-blocks with 64 weights Q2_K scalar and AVX2 works. Q2_K is way too slow (it is actually slower than the scalar implementation) * k_quants: WIP super-blocks with 64 weights Q3_K scalar and AVX2 works. * k_quants: WIP super-blocks with 64 weights Q5_K scalar and AVX2 works, and with that all k_quants are done on AVX2 and scalar * k_quants: WIP super-blocks with 64 weights Q6_K working on CUDA. Cannot make it run quite as gast as with super-blocks with 256 weigths: 8% slower on 4080, 20% slower on the 1660 (but there we fit 1 less layer on the GPU because pf the larger model size), so some fraction of these 20% is due to that, * k_quants: WIP super-blocks with 64 weights Q4_K working on CUDA. ~10% slower on GTX-1660, 16% slower on 4080. * k_quants: WIP super-blocks with 64 weights Q2_K working on CUDA. ~3% slower on GTX-1660, 10% slower on 4080. * k_quants: WIP super-blocks with 64 weights Q3_K working on CUDA. * k_quants: WIP super-blocks with 64 weights Q5_K working on CUDA, and with this CUDA is done. * k_quants: WIP super-blocks with 64 weights Q6_K working on ARM_NEON * k_quants: WIP super-blocks with 64 weights Q4_K working on ARM_NEON, but quite a bit slower than 256 weights * k_quants: WIP super-blocks with 64 weights Q2_K working on ARM_NEON, but quite a bit slower than 256 weights * k_quants: WIP super-blocks with 64 weights Q3_K working on ARM_NEON, but quite a bit slower than 256 weights. * k_quants: WIP super-blocks with 64 weights Q5_K working on ARM_NEON, but quite a bit slower than 256 weights. With that, we have full support for ARM_NEON, although performance is not quite there. * k_quants: WIP super-blocks with 64 weights Slightly more efficient Q3_K and Q5_K * k_quants: WIP super-blocks with 64 weights Another small improvement for Q3_K and Q5_K on ARM_NEON * k_quants: WIP super-blocks with 64 weights Yet another speedup for Q5_K on ARM_NEON. We are now within 10% of the QK_K = 256 version. * k_quants: WIP super-blocks with 64 weights * We are able to pass preprocessor macros to the Metal compiler * Q6_K works and is actually slightly more efficient than the QK_K = 256 version (25.2 ms vs 25.8 ms) * k_quants: WIP super-blocks with 64 weights Q4_K works on Metal and is actually slightly faster than QK_K = 256 (21.95 ms vs 24.0 ms). * k_quants: WIP super-blocks with 64 weights Q2_K works on Metal and is very slightly faster than QK_K = 256 (23.8 ms vs 24.2 ms). * k_quants: WIP super-blocks with 64 weights Q3_K works on Metal and is slightly faster than QK_K = 256 (26.6 ms vs 28.3 ms). * k_quants: WIP super-blocks with 64 weights Q5_K works on Metal and is slightly faster than QK_K = 256 (23.7 ms vs 26.3 ms). * k_quants: call them _K, not _k, also on Metal * k_quants: correctly define QK_K in llama.cpp * Fixed bug in q4_K quantization added with the 64-block addition * Simplify via lambda * k_quants: swicth Q3_K to 4-bit scales when QK_K = 64 Otherwise there isn't much benefit from this quantization type. There is some very slight loss in accuracy, but we reduce size by ~7%. E.g., for OpenLLaMA-3B, Q3_K_S perplexity is 8.6131 with 8-bit scales and 8.6352 with 4-bit, while file size decreases from 1.53G to 1.44G. * k_quants: switch Q4_K to 4-bit scales when QK_K = 64 Here the loss in accuracy is greater than for Q3_K, but the Q4_K points still move further to the left on the perplexity vs size curve. * k_quants: forgot to add the Metal changes in last commit * k_quants: change Q5_K to be type 0 when QK_K = 64 Still needs AVX2 implementation * k_quants: AVX2 implementation for new 64-weight Q5_K * k_quants: 10% faster ARM_NEON Q5_K dot product * k_quants: fixed issue caused by merging with master --------- Co-authored-by: Iwan Kawrakow <iwan.kawrakow@gmail.com>
2023-06-26 16:43:07 +00:00
q3bytes.val[0] = vreinterpretq_s8_u8(vorrq_u8(vandq_u8(q3bits, m3b), q3h.val[0]));
q3bytes.val[1] = vreinterpretq_s8_u8(vorrq_u8(vandq_u8(vshrq_n_u8(q3bits, 2), m3b), q3h.val[1]));
q3bytes.val[2] = vreinterpretq_s8_u8(vorrq_u8(vandq_u8(vshrq_n_u8(q3bits, 4), m3b), q3h.val[2]));
q3bytes.val[3] = vreinterpretq_s8_u8(vorrq_u8(vshrq_n_u8(q3bits, 6), q3h.val[3]));
ggml : add SOTA 2,3,4,5,6 bit k-quantizations (#1684) * Starting to add k-quantization to ggml I think it is better to have quantization separate from ggml. For now just adding the k-quants there, but it would be better to also factor out the existing ggml quantizations. * Adding Q3_K and Q8_K (de)-quantization * Q3_K now working on CUDA and AVX2/scalar CUDA is not ideal - ~50% slower than Q4_0 for single token prediction, about the same in batch mode (perplexity). CPU single token is ~55 ms (on Ryzen 7950X). * Some improvement for Q3_K on CUDA It is now ~22.5 ms/token on my GPU, so ~30% slower than Q4_0. * Some more CUDA optimizations for Q3_K Single token is now 20.5 ms/token (~20% slower than Q4_0). Perplexity is on par with Q4_0. * Adding Q4_K - scalar, AVX2, CUDA Performance is the same or perhaps very slightly better than Q4_0 on the CPU. On the GPU, single token prediction is ~10% better than Q4_0, batch mode (perplexity is about the same). * Adding Q6_K - scalar, AVX2, CUDA Performance is ~40% lower compared to Q4_K on the CPU. This is to be expected, considering that we are memory bound on the CPU and the 6-bit model is ~44% larger than the 4-bit. On the GPU, single token prediction is ~6% lower than Q4_0, batch mode (perplexity) is even closer (but still slower). * Adding Q5_K - scalar, AVX2, CUDA Performance is ~20% lower compared to Q4_K on the CPU. This is to be expected, considering that we are memory bound on the CPU and the 5-bit model is ~22% larger than the 4-bit. On the GPU, single token prediction is about the same as Q4_0 for both, single token and batch prediction. * Per convention, all QX_K quantizations use Q5_K for output.weight * Adding quantization mixes * Quantization mixes: didn't quite get what I wanted in the last commit * Q4_K dot product for ARM_NEON * Q6_K dot product for ARM_NEON * Q5_K dot product for ARM_NEON * Adding Q3_K dot for ARM_NEON It is 22% slower than Q4_K, despite the smaller model size. On x86_64, where we are memory bound, the Q3_K model is quite a bit faster than Q4_K. * A very slightly faster ARM_NEON Q3_K dot * Adding Q2_K - just CUDA for now Token prediction is pretty good - about 15.5 ms on a RTX 4080. Perplexity is about the same as Q4_K. * Adding scalar and AVX2 Q2_K dot * Adding ARM_NEON Q2_K dot About the same performance as Q4_K. * A slightly faster ARM_NEON Q2_K dot Single token prediction is now ~36 ms on M2 Max. The code is much simpler too. * Fixed bug in Q2_K CUDA dot product kernel Stranegly enough, for the few prompts I tried with the 7B model the responses looked perfectly reasonable. Only realized something is not quite right when I tried the larger models and started getting nonse back. In any case, Q2_K single token evaluation time on an RTX 4080 in a Ryzen7950X box iusing CUDA and model fully loaded on the GPU are ~15.5 ms for 7B, ~25.4 ms for 13B, and ~55.8 ms for 30B. The max number of layers that fit in VRAM for The 65B is 32. With that, we get ~330 ms per token, which is not that much faster than just running on the CPU (~470 ms per token). * Don't print zeros/NaNs when no count histogram has been collected * A 10% faster CUDA vector dot kernel for Q3_K Q3_K is now running at ~18.5 ms / token on CUDA, so the gap to Q4_0 is only 10%. It seems memory acccess pattern is more important for performance than the amount of computation the kernel does. * A slightly daster Q4_K AVX2 dot product For perplexity, where we are less memory bound, time per pass drops by ~5%. Barely measurable difference for single token prediction. * A slightly faster ARM_NEON A4_K dot product * Minor * Fix quantization error test We cannot possibly be expecting rmse < 0.002 for 2- and 3-bit quantization variants. * Fix docker build I have been sloppy with vector reinterpret casts on ARM_NEON. It seems clang is very forgiving in that regard. * Added forgotten ggml.o dependence on k_quants.h to the Makefile * Had unintentionally committed the Makefile with -Ofast enabled * ggml : rename k_quants -> ggml-quants-k, use lowercase in code --------- Co-authored-by: Iwan Kawrakow <iwan.kawrakow@gmail.com> Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
2023-06-05 19:56:18 +00:00
ggml : add ggml_vdotq_s32 alias (#4715) ggml-ci
2023-12-31 09:43:31 +00:00
isum += vaddvq_s32(ggml_vdotq_s32(vzero, q3bytes.val[0], q8bytes.val[0])) * scales[0];
isum += vaddvq_s32(ggml_vdotq_s32(vzero, q3bytes.val[1], q8bytes.val[1])) * scales[2];
isum += vaddvq_s32(ggml_vdotq_s32(vzero, q3bytes.val[2], q8bytes.val[2])) * scales[1];
isum += vaddvq_s32(ggml_vdotq_s32(vzero, q3bytes.val[3], q8bytes.val[3])) * scales[3];
ggml : add SOTA 2,3,4,5,6 bit k-quantizations (#1684) * Starting to add k-quantization to ggml I think it is better to have quantization separate from ggml. For now just adding the k-quants there, but it would be better to also factor out the existing ggml quantizations. * Adding Q3_K and Q8_K (de)-quantization * Q3_K now working on CUDA and AVX2/scalar CUDA is not ideal - ~50% slower than Q4_0 for single token prediction, about the same in batch mode (perplexity). CPU single token is ~55 ms (on Ryzen 7950X). * Some improvement for Q3_K on CUDA It is now ~22.5 ms/token on my GPU, so ~30% slower than Q4_0. * Some more CUDA optimizations for Q3_K Single token is now 20.5 ms/token (~20% slower than Q4_0). Perplexity is on par with Q4_0. * Adding Q4_K - scalar, AVX2, CUDA Performance is the same or perhaps very slightly better than Q4_0 on the CPU. On the GPU, single token prediction is ~10% better than Q4_0, batch mode (perplexity is about the same). * Adding Q6_K - scalar, AVX2, CUDA Performance is ~40% lower compared to Q4_K on the CPU. This is to be expected, considering that we are memory bound on the CPU and the 6-bit model is ~44% larger than the 4-bit. On the GPU, single token prediction is ~6% lower than Q4_0, batch mode (perplexity) is even closer (but still slower). * Adding Q5_K - scalar, AVX2, CUDA Performance is ~20% lower compared to Q4_K on the CPU. This is to be expected, considering that we are memory bound on the CPU and the 5-bit model is ~22% larger than the 4-bit. On the GPU, single token prediction is about the same as Q4_0 for both, single token and batch prediction. * Per convention, all QX_K quantizations use Q5_K for output.weight * Adding quantization mixes * Quantization mixes: didn't quite get what I wanted in the last commit * Q4_K dot product for ARM_NEON * Q6_K dot product for ARM_NEON * Q5_K dot product for ARM_NEON * Adding Q3_K dot for ARM_NEON It is 22% slower than Q4_K, despite the smaller model size. On x86_64, where we are memory bound, the Q3_K model is quite a bit faster than Q4_K. * A very slightly faster ARM_NEON Q3_K dot * Adding Q2_K - just CUDA for now Token prediction is pretty good - about 15.5 ms on a RTX 4080. Perplexity is about the same as Q4_K. * Adding scalar and AVX2 Q2_K dot * Adding ARM_NEON Q2_K dot About the same performance as Q4_K. * A slightly faster ARM_NEON Q2_K dot Single token prediction is now ~36 ms on M2 Max. The code is much simpler too. * Fixed bug in Q2_K CUDA dot product kernel Stranegly enough, for the few prompts I tried with the 7B model the responses looked perfectly reasonable. Only realized something is not quite right when I tried the larger models and started getting nonse back. In any case, Q2_K single token evaluation time on an RTX 4080 in a Ryzen7950X box iusing CUDA and model fully loaded on the GPU are ~15.5 ms for 7B, ~25.4 ms for 13B, and ~55.8 ms for 30B. The max number of layers that fit in VRAM for The 65B is 32. With that, we get ~330 ms per token, which is not that much faster than just running on the CPU (~470 ms per token). * Don't print zeros/NaNs when no count histogram has been collected * A 10% faster CUDA vector dot kernel for Q3_K Q3_K is now running at ~18.5 ms / token on CUDA, so the gap to Q4_0 is only 10%. It seems memory acccess pattern is more important for performance than the amount of computation the kernel does. * A slightly daster Q4_K AVX2 dot product For perplexity, where we are less memory bound, time per pass drops by ~5%. Barely measurable difference for single token prediction. * A slightly faster ARM_NEON A4_K dot product * Minor * Fix quantization error test We cannot possibly be expecting rmse < 0.002 for 2- and 3-bit quantization variants. * Fix docker build I have been sloppy with vector reinterpret casts on ARM_NEON. It seems clang is very forgiving in that regard. * Added forgotten ggml.o dependence on k_quants.h to the Makefile * Had unintentionally committed the Makefile with -Ofast enabled * ggml : rename k_quants -> ggml-quants-k, use lowercase in code --------- Co-authored-by: Iwan Kawrakow <iwan.kawrakow@gmail.com> Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
2023-06-05 19:56:18 +00:00
k-quants : support for super-block size of 64 (#2001) * k_quants: WIP super-blocks with 64 weights * k_quants: WIP super-blocks with 64 weights Q6_K scalar and AVX2 works * k_quants: WIP super-blocks with 64 weights Q4_K scalar and AVX2 works * k_quants: WIP super-blocks with 64 weights Q2_K scalar and AVX2 works. Q2_K is way too slow (it is actually slower than the scalar implementation) * k_quants: WIP super-blocks with 64 weights Q3_K scalar and AVX2 works. * k_quants: WIP super-blocks with 64 weights Q5_K scalar and AVX2 works, and with that all k_quants are done on AVX2 and scalar * k_quants: WIP super-blocks with 64 weights Q6_K working on CUDA. Cannot make it run quite as gast as with super-blocks with 256 weigths: 8% slower on 4080, 20% slower on the 1660 (but there we fit 1 less layer on the GPU because pf the larger model size), so some fraction of these 20% is due to that, * k_quants: WIP super-blocks with 64 weights Q4_K working on CUDA. ~10% slower on GTX-1660, 16% slower on 4080. * k_quants: WIP super-blocks with 64 weights Q2_K working on CUDA. ~3% slower on GTX-1660, 10% slower on 4080. * k_quants: WIP super-blocks with 64 weights Q3_K working on CUDA. * k_quants: WIP super-blocks with 64 weights Q5_K working on CUDA, and with this CUDA is done. * k_quants: WIP super-blocks with 64 weights Q6_K working on ARM_NEON * k_quants: WIP super-blocks with 64 weights Q4_K working on ARM_NEON, but quite a bit slower than 256 weights * k_quants: WIP super-blocks with 64 weights Q2_K working on ARM_NEON, but quite a bit slower than 256 weights * k_quants: WIP super-blocks with 64 weights Q3_K working on ARM_NEON, but quite a bit slower than 256 weights. * k_quants: WIP super-blocks with 64 weights Q5_K working on ARM_NEON, but quite a bit slower than 256 weights. With that, we have full support for ARM_NEON, although performance is not quite there. * k_quants: WIP super-blocks with 64 weights Slightly more efficient Q3_K and Q5_K * k_quants: WIP super-blocks with 64 weights Another small improvement for Q3_K and Q5_K on ARM_NEON * k_quants: WIP super-blocks with 64 weights Yet another speedup for Q5_K on ARM_NEON. We are now within 10% of the QK_K = 256 version. * k_quants: WIP super-blocks with 64 weights * We are able to pass preprocessor macros to the Metal compiler * Q6_K works and is actually slightly more efficient than the QK_K = 256 version (25.2 ms vs 25.8 ms) * k_quants: WIP super-blocks with 64 weights Q4_K works on Metal and is actually slightly faster than QK_K = 256 (21.95 ms vs 24.0 ms). * k_quants: WIP super-blocks with 64 weights Q2_K works on Metal and is very slightly faster than QK_K = 256 (23.8 ms vs 24.2 ms). * k_quants: WIP super-blocks with 64 weights Q3_K works on Metal and is slightly faster than QK_K = 256 (26.6 ms vs 28.3 ms). * k_quants: WIP super-blocks with 64 weights Q5_K works on Metal and is slightly faster than QK_K = 256 (23.7 ms vs 26.3 ms). * k_quants: call them _K, not _k, also on Metal * k_quants: correctly define QK_K in llama.cpp * Fixed bug in q4_K quantization added with the 64-block addition * Simplify via lambda * k_quants: swicth Q3_K to 4-bit scales when QK_K = 64 Otherwise there isn't much benefit from this quantization type. There is some very slight loss in accuracy, but we reduce size by ~7%. E.g., for OpenLLaMA-3B, Q3_K_S perplexity is 8.6131 with 8-bit scales and 8.6352 with 4-bit, while file size decreases from 1.53G to 1.44G. * k_quants: switch Q4_K to 4-bit scales when QK_K = 64 Here the loss in accuracy is greater than for Q3_K, but the Q4_K points still move further to the left on the perplexity vs size curve. * k_quants: forgot to add the Metal changes in last commit * k_quants: change Q5_K to be type 0 when QK_K = 64 Still needs AVX2 implementation * k_quants: AVX2 implementation for new 64-weight Q5_K * k_quants: 10% faster ARM_NEON Q5_K dot product * k_quants: fixed issue caused by merging with master --------- Co-authored-by: Iwan Kawrakow <iwan.kawrakow@gmail.com>
2023-06-26 16:43:07 +00:00
sum += d * isum;
ggml : add SOTA 2,3,4,5,6 bit k-quantizations (#1684) * Starting to add k-quantization to ggml I think it is better to have quantization separate from ggml. For now just adding the k-quants there, but it would be better to also factor out the existing ggml quantizations. * Adding Q3_K and Q8_K (de)-quantization * Q3_K now working on CUDA and AVX2/scalar CUDA is not ideal - ~50% slower than Q4_0 for single token prediction, about the same in batch mode (perplexity). CPU single token is ~55 ms (on Ryzen 7950X). * Some improvement for Q3_K on CUDA It is now ~22.5 ms/token on my GPU, so ~30% slower than Q4_0. * Some more CUDA optimizations for Q3_K Single token is now 20.5 ms/token (~20% slower than Q4_0). Perplexity is on par with Q4_0. * Adding Q4_K - scalar, AVX2, CUDA Performance is the same or perhaps very slightly better than Q4_0 on the CPU. On the GPU, single token prediction is ~10% better than Q4_0, batch mode (perplexity is about the same). * Adding Q6_K - scalar, AVX2, CUDA Performance is ~40% lower compared to Q4_K on the CPU. This is to be expected, considering that we are memory bound on the CPU and the 6-bit model is ~44% larger than the 4-bit. On the GPU, single token prediction is ~6% lower than Q4_0, batch mode (perplexity) is even closer (but still slower). * Adding Q5_K - scalar, AVX2, CUDA Performance is ~20% lower compared to Q4_K on the CPU. This is to be expected, considering that we are memory bound on the CPU and the 5-bit model is ~22% larger than the 4-bit. On the GPU, single token prediction is about the same as Q4_0 for both, single token and batch prediction. * Per convention, all QX_K quantizations use Q5_K for output.weight * Adding quantization mixes * Quantization mixes: didn't quite get what I wanted in the last commit * Q4_K dot product for ARM_NEON * Q6_K dot product for ARM_NEON * Q5_K dot product for ARM_NEON * Adding Q3_K dot for ARM_NEON It is 22% slower than Q4_K, despite the smaller model size. On x86_64, where we are memory bound, the Q3_K model is quite a bit faster than Q4_K. * A very slightly faster ARM_NEON Q3_K dot * Adding Q2_K - just CUDA for now Token prediction is pretty good - about 15.5 ms on a RTX 4080. Perplexity is about the same as Q4_K. * Adding scalar and AVX2 Q2_K dot * Adding ARM_NEON Q2_K dot About the same performance as Q4_K. * A slightly faster ARM_NEON Q2_K dot Single token prediction is now ~36 ms on M2 Max. The code is much simpler too. * Fixed bug in Q2_K CUDA dot product kernel Stranegly enough, for the few prompts I tried with the 7B model the responses looked perfectly reasonable. Only realized something is not quite right when I tried the larger models and started getting nonse back. In any case, Q2_K single token evaluation time on an RTX 4080 in a Ryzen7950X box iusing CUDA and model fully loaded on the GPU are ~15.5 ms for 7B, ~25.4 ms for 13B, and ~55.8 ms for 30B. The max number of layers that fit in VRAM for The 65B is 32. With that, we get ~330 ms per token, which is not that much faster than just running on the CPU (~470 ms per token). * Don't print zeros/NaNs when no count histogram has been collected * A 10% faster CUDA vector dot kernel for Q3_K Q3_K is now running at ~18.5 ms / token on CUDA, so the gap to Q4_0 is only 10%. It seems memory acccess pattern is more important for performance than the amount of computation the kernel does. * A slightly daster Q4_K AVX2 dot product For perplexity, where we are less memory bound, time per pass drops by ~5%. Barely measurable difference for single token prediction. * A slightly faster ARM_NEON A4_K dot product * Minor * Fix quantization error test We cannot possibly be expecting rmse < 0.002 for 2- and 3-bit quantization variants. * Fix docker build I have been sloppy with vector reinterpret casts on ARM_NEON. It seems clang is very forgiving in that regard. * Added forgotten ggml.o dependence on k_quants.h to the Makefile * Had unintentionally committed the Makefile with -Ofast enabled * ggml : rename k_quants -> ggml-quants-k, use lowercase in code --------- Co-authored-by: Iwan Kawrakow <iwan.kawrakow@gmail.com> Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
2023-06-05 19:56:18 +00:00
k-quants : support for super-block size of 64 (#2001) * k_quants: WIP super-blocks with 64 weights * k_quants: WIP super-blocks with 64 weights Q6_K scalar and AVX2 works * k_quants: WIP super-blocks with 64 weights Q4_K scalar and AVX2 works * k_quants: WIP super-blocks with 64 weights Q2_K scalar and AVX2 works. Q2_K is way too slow (it is actually slower than the scalar implementation) * k_quants: WIP super-blocks with 64 weights Q3_K scalar and AVX2 works. * k_quants: WIP super-blocks with 64 weights Q5_K scalar and AVX2 works, and with that all k_quants are done on AVX2 and scalar * k_quants: WIP super-blocks with 64 weights Q6_K working on CUDA. Cannot make it run quite as gast as with super-blocks with 256 weigths: 8% slower on 4080, 20% slower on the 1660 (but there we fit 1 less layer on the GPU because pf the larger model size), so some fraction of these 20% is due to that, * k_quants: WIP super-blocks with 64 weights Q4_K working on CUDA. ~10% slower on GTX-1660, 16% slower on 4080. * k_quants: WIP super-blocks with 64 weights Q2_K working on CUDA. ~3% slower on GTX-1660, 10% slower on 4080. * k_quants: WIP super-blocks with 64 weights Q3_K working on CUDA. * k_quants: WIP super-blocks with 64 weights Q5_K working on CUDA, and with this CUDA is done. * k_quants: WIP super-blocks with 64 weights Q6_K working on ARM_NEON * k_quants: WIP super-blocks with 64 weights Q4_K working on ARM_NEON, but quite a bit slower than 256 weights * k_quants: WIP super-blocks with 64 weights Q2_K working on ARM_NEON, but quite a bit slower than 256 weights * k_quants: WIP super-blocks with 64 weights Q3_K working on ARM_NEON, but quite a bit slower than 256 weights. * k_quants: WIP super-blocks with 64 weights Q5_K working on ARM_NEON, but quite a bit slower than 256 weights. With that, we have full support for ARM_NEON, although performance is not quite there. * k_quants: WIP super-blocks with 64 weights Slightly more efficient Q3_K and Q5_K * k_quants: WIP super-blocks with 64 weights Another small improvement for Q3_K and Q5_K on ARM_NEON * k_quants: WIP super-blocks with 64 weights Yet another speedup for Q5_K on ARM_NEON. We are now within 10% of the QK_K = 256 version. * k_quants: WIP super-blocks with 64 weights * We are able to pass preprocessor macros to the Metal compiler * Q6_K works and is actually slightly more efficient than the QK_K = 256 version (25.2 ms vs 25.8 ms) * k_quants: WIP super-blocks with 64 weights Q4_K works on Metal and is actually slightly faster than QK_K = 256 (21.95 ms vs 24.0 ms). * k_quants: WIP super-blocks with 64 weights Q2_K works on Metal and is very slightly faster than QK_K = 256 (23.8 ms vs 24.2 ms). * k_quants: WIP super-blocks with 64 weights Q3_K works on Metal and is slightly faster than QK_K = 256 (26.6 ms vs 28.3 ms). * k_quants: WIP super-blocks with 64 weights Q5_K works on Metal and is slightly faster than QK_K = 256 (23.7 ms vs 26.3 ms). * k_quants: call them _K, not _k, also on Metal * k_quants: correctly define QK_K in llama.cpp * Fixed bug in q4_K quantization added with the 64-block addition * Simplify via lambda * k_quants: swicth Q3_K to 4-bit scales when QK_K = 64 Otherwise there isn't much benefit from this quantization type. There is some very slight loss in accuracy, but we reduce size by ~7%. E.g., for OpenLLaMA-3B, Q3_K_S perplexity is 8.6131 with 8-bit scales and 8.6352 with 4-bit, while file size decreases from 1.53G to 1.44G. * k_quants: switch Q4_K to 4-bit scales when QK_K = 64 Here the loss in accuracy is greater than for Q3_K, but the Q4_K points still move further to the left on the perplexity vs size curve. * k_quants: forgot to add the Metal changes in last commit * k_quants: change Q5_K to be type 0 when QK_K = 64 Still needs AVX2 implementation * k_quants: AVX2 implementation for new 64-weight Q5_K * k_quants: 10% faster ARM_NEON Q5_K dot product * k_quants: fixed issue caused by merging with master --------- Co-authored-by: Iwan Kawrakow <iwan.kawrakow@gmail.com>
2023-06-26 16:43:07 +00:00
}
*s = sum;
#elif defined __AVX2__
const __m256i m3 = _mm256_set1_epi8(3);
const __m256i m1 = _mm256_set1_epi8(1);
__m256 acc = _mm256_setzero_ps();
uint64_t aux64;
uint16_t aux16[2];
const int8_t * aux8 = (const int8_t *)aux16;
for (int i = 0; i < nb; ++i) {
ggml : move FP16 <-> FP32 code to ggml-impl.h (#3861) * ggml : move FP16 <-> FP32 stuff to ggml-impl.h ggml-ci * tests : fix ARM build * ggml : explicitly initialize deprecated type traits * ggml : add math.h to ggml-impl.h * ggml : remove duplicate static assert macros * ggml : prefix lookup tables with ggml_ ggml-ci * ggml-impl : move extern "C" to start of file
2023-10-30 17:19:15 +00:00
const float d = y[i].d * GGML_FP16_TO_FP32(x[i].d);
k-quants : support for super-block size of 64 (#2001) * k_quants: WIP super-blocks with 64 weights * k_quants: WIP super-blocks with 64 weights Q6_K scalar and AVX2 works * k_quants: WIP super-blocks with 64 weights Q4_K scalar and AVX2 works * k_quants: WIP super-blocks with 64 weights Q2_K scalar and AVX2 works. Q2_K is way too slow (it is actually slower than the scalar implementation) * k_quants: WIP super-blocks with 64 weights Q3_K scalar and AVX2 works. * k_quants: WIP super-blocks with 64 weights Q5_K scalar and AVX2 works, and with that all k_quants are done on AVX2 and scalar * k_quants: WIP super-blocks with 64 weights Q6_K working on CUDA. Cannot make it run quite as gast as with super-blocks with 256 weigths: 8% slower on 4080, 20% slower on the 1660 (but there we fit 1 less layer on the GPU because pf the larger model size), so some fraction of these 20% is due to that, * k_quants: WIP super-blocks with 64 weights Q4_K working on CUDA. ~10% slower on GTX-1660, 16% slower on 4080. * k_quants: WIP super-blocks with 64 weights Q2_K working on CUDA. ~3% slower on GTX-1660, 10% slower on 4080. * k_quants: WIP super-blocks with 64 weights Q3_K working on CUDA. * k_quants: WIP super-blocks with 64 weights Q5_K working on CUDA, and with this CUDA is done. * k_quants: WIP super-blocks with 64 weights Q6_K working on ARM_NEON * k_quants: WIP super-blocks with 64 weights Q4_K working on ARM_NEON, but quite a bit slower than 256 weights * k_quants: WIP super-blocks with 64 weights Q2_K working on ARM_NEON, but quite a bit slower than 256 weights * k_quants: WIP super-blocks with 64 weights Q3_K working on ARM_NEON, but quite a bit slower than 256 weights. * k_quants: WIP super-blocks with 64 weights Q5_K working on ARM_NEON, but quite a bit slower than 256 weights. With that, we have full support for ARM_NEON, although performance is not quite there. * k_quants: WIP super-blocks with 64 weights Slightly more efficient Q3_K and Q5_K * k_quants: WIP super-blocks with 64 weights Another small improvement for Q3_K and Q5_K on ARM_NEON * k_quants: WIP super-blocks with 64 weights Yet another speedup for Q5_K on ARM_NEON. We are now within 10% of the QK_K = 256 version. * k_quants: WIP super-blocks with 64 weights * We are able to pass preprocessor macros to the Metal compiler * Q6_K works and is actually slightly more efficient than the QK_K = 256 version (25.2 ms vs 25.8 ms) * k_quants: WIP super-blocks with 64 weights Q4_K works on Metal and is actually slightly faster than QK_K = 256 (21.95 ms vs 24.0 ms). * k_quants: WIP super-blocks with 64 weights Q2_K works on Metal and is very slightly faster than QK_K = 256 (23.8 ms vs 24.2 ms). * k_quants: WIP super-blocks with 64 weights Q3_K works on Metal and is slightly faster than QK_K = 256 (26.6 ms vs 28.3 ms). * k_quants: WIP super-blocks with 64 weights Q5_K works on Metal and is slightly faster than QK_K = 256 (23.7 ms vs 26.3 ms). * k_quants: call them _K, not _k, also on Metal * k_quants: correctly define QK_K in llama.cpp * Fixed bug in q4_K quantization added with the 64-block addition * Simplify via lambda * k_quants: swicth Q3_K to 4-bit scales when QK_K = 64 Otherwise there isn't much benefit from this quantization type. There is some very slight loss in accuracy, but we reduce size by ~7%. E.g., for OpenLLaMA-3B, Q3_K_S perplexity is 8.6131 with 8-bit scales and 8.6352 with 4-bit, while file size decreases from 1.53G to 1.44G. * k_quants: switch Q4_K to 4-bit scales when QK_K = 64 Here the loss in accuracy is greater than for Q3_K, but the Q4_K points still move further to the left on the perplexity vs size curve. * k_quants: forgot to add the Metal changes in last commit * k_quants: change Q5_K to be type 0 when QK_K = 64 Still needs AVX2 implementation * k_quants: AVX2 implementation for new 64-weight Q5_K * k_quants: 10% faster ARM_NEON Q5_K dot product * k_quants: fixed issue caused by merging with master --------- Co-authored-by: Iwan Kawrakow <iwan.kawrakow@gmail.com>
2023-06-26 16:43:07 +00:00
const uint8_t * restrict q3 = x[i].qs;
const int8_t * restrict q8 = y[i].qs;
const uint16_t a = *(const uint16_t *)x[i].scales;
aux16[0] = a & 0x0f0f;
aux16[1] = (a >> 4) & 0x0f0f;
ggml : workaround for missing _mm256_setr_m128i in GCC < 8 in k_quants.c (#2405)
2023-07-28 18:17:45 +00:00
const __m256i scale_0 = MM256_SET_M128I(_mm_set1_epi16(aux8[2] - 8), _mm_set1_epi16(aux8[0] - 8));
const __m256i scale_1 = MM256_SET_M128I(_mm_set1_epi16(aux8[3] - 8), _mm_set1_epi16(aux8[1] - 8));
k-quants : support for super-block size of 64 (#2001) * k_quants: WIP super-blocks with 64 weights * k_quants: WIP super-blocks with 64 weights Q6_K scalar and AVX2 works * k_quants: WIP super-blocks with 64 weights Q4_K scalar and AVX2 works * k_quants: WIP super-blocks with 64 weights Q2_K scalar and AVX2 works. Q2_K is way too slow (it is actually slower than the scalar implementation) * k_quants: WIP super-blocks with 64 weights Q3_K scalar and AVX2 works. * k_quants: WIP super-blocks with 64 weights Q5_K scalar and AVX2 works, and with that all k_quants are done on AVX2 and scalar * k_quants: WIP super-blocks with 64 weights Q6_K working on CUDA. Cannot make it run quite as gast as with super-blocks with 256 weigths: 8% slower on 4080, 20% slower on the 1660 (but there we fit 1 less layer on the GPU because pf the larger model size), so some fraction of these 20% is due to that, * k_quants: WIP super-blocks with 64 weights Q4_K working on CUDA. ~10% slower on GTX-1660, 16% slower on 4080. * k_quants: WIP super-blocks with 64 weights Q2_K working on CUDA. ~3% slower on GTX-1660, 10% slower on 4080. * k_quants: WIP super-blocks with 64 weights Q3_K working on CUDA. * k_quants: WIP super-blocks with 64 weights Q5_K working on CUDA, and with this CUDA is done. * k_quants: WIP super-blocks with 64 weights Q6_K working on ARM_NEON * k_quants: WIP super-blocks with 64 weights Q4_K working on ARM_NEON, but quite a bit slower than 256 weights * k_quants: WIP super-blocks with 64 weights Q2_K working on ARM_NEON, but quite a bit slower than 256 weights * k_quants: WIP super-blocks with 64 weights Q3_K working on ARM_NEON, but quite a bit slower than 256 weights. * k_quants: WIP super-blocks with 64 weights Q5_K working on ARM_NEON, but quite a bit slower than 256 weights. With that, we have full support for ARM_NEON, although performance is not quite there. * k_quants: WIP super-blocks with 64 weights Slightly more efficient Q3_K and Q5_K * k_quants: WIP super-blocks with 64 weights Another small improvement for Q3_K and Q5_K on ARM_NEON * k_quants: WIP super-blocks with 64 weights Yet another speedup for Q5_K on ARM_NEON. We are now within 10% of the QK_K = 256 version. * k_quants: WIP super-blocks with 64 weights * We are able to pass preprocessor macros to the Metal compiler * Q6_K works and is actually slightly more efficient than the QK_K = 256 version (25.2 ms vs 25.8 ms) * k_quants: WIP super-blocks with 64 weights Q4_K works on Metal and is actually slightly faster than QK_K = 256 (21.95 ms vs 24.0 ms). * k_quants: WIP super-blocks with 64 weights Q2_K works on Metal and is very slightly faster than QK_K = 256 (23.8 ms vs 24.2 ms). * k_quants: WIP super-blocks with 64 weights Q3_K works on Metal and is slightly faster than QK_K = 256 (26.6 ms vs 28.3 ms). * k_quants: WIP super-blocks with 64 weights Q5_K works on Metal and is slightly faster than QK_K = 256 (23.7 ms vs 26.3 ms). * k_quants: call them _K, not _k, also on Metal * k_quants: correctly define QK_K in llama.cpp * Fixed bug in q4_K quantization added with the 64-block addition * Simplify via lambda * k_quants: swicth Q3_K to 4-bit scales when QK_K = 64 Otherwise there isn't much benefit from this quantization type. There is some very slight loss in accuracy, but we reduce size by ~7%. E.g., for OpenLLaMA-3B, Q3_K_S perplexity is 8.6131 with 8-bit scales and 8.6352 with 4-bit, while file size decreases from 1.53G to 1.44G. * k_quants: switch Q4_K to 4-bit scales when QK_K = 64 Here the loss in accuracy is greater than for Q3_K, but the Q4_K points still move further to the left on the perplexity vs size curve. * k_quants: forgot to add the Metal changes in last commit * k_quants: change Q5_K to be type 0 when QK_K = 64 Still needs AVX2 implementation * k_quants: AVX2 implementation for new 64-weight Q5_K * k_quants: 10% faster ARM_NEON Q5_K dot product * k_quants: fixed issue caused by merging with master --------- Co-authored-by: Iwan Kawrakow <iwan.kawrakow@gmail.com>
2023-06-26 16:43:07 +00:00
memcpy(&aux64, x[i].hmask, 8);
const __m128i haux = _mm_set_epi64x(aux64 >> 1, aux64 >> 0);
ggml : workaround for missing _mm256_setr_m128i in GCC < 8 in k_quants.c (#2405)
2023-07-28 18:17:45 +00:00
__m256i q3h_0 = MM256_SET_M128I(_mm_srli_epi16(haux, 2), haux);
k-quants : support for super-block size of 64 (#2001) * k_quants: WIP super-blocks with 64 weights * k_quants: WIP super-blocks with 64 weights Q6_K scalar and AVX2 works * k_quants: WIP super-blocks with 64 weights Q4_K scalar and AVX2 works * k_quants: WIP super-blocks with 64 weights Q2_K scalar and AVX2 works. Q2_K is way too slow (it is actually slower than the scalar implementation) * k_quants: WIP super-blocks with 64 weights Q3_K scalar and AVX2 works. * k_quants: WIP super-blocks with 64 weights Q5_K scalar and AVX2 works, and with that all k_quants are done on AVX2 and scalar * k_quants: WIP super-blocks with 64 weights Q6_K working on CUDA. Cannot make it run quite as gast as with super-blocks with 256 weigths: 8% slower on 4080, 20% slower on the 1660 (but there we fit 1 less layer on the GPU because pf the larger model size), so some fraction of these 20% is due to that, * k_quants: WIP super-blocks with 64 weights Q4_K working on CUDA. ~10% slower on GTX-1660, 16% slower on 4080. * k_quants: WIP super-blocks with 64 weights Q2_K working on CUDA. ~3% slower on GTX-1660, 10% slower on 4080. * k_quants: WIP super-blocks with 64 weights Q3_K working on CUDA. * k_quants: WIP super-blocks with 64 weights Q5_K working on CUDA, and with this CUDA is done. * k_quants: WIP super-blocks with 64 weights Q6_K working on ARM_NEON * k_quants: WIP super-blocks with 64 weights Q4_K working on ARM_NEON, but quite a bit slower than 256 weights * k_quants: WIP super-blocks with 64 weights Q2_K working on ARM_NEON, but quite a bit slower than 256 weights * k_quants: WIP super-blocks with 64 weights Q3_K working on ARM_NEON, but quite a bit slower than 256 weights. * k_quants: WIP super-blocks with 64 weights Q5_K working on ARM_NEON, but quite a bit slower than 256 weights. With that, we have full support for ARM_NEON, although performance is not quite there. * k_quants: WIP super-blocks with 64 weights Slightly more efficient Q3_K and Q5_K * k_quants: WIP super-blocks with 64 weights Another small improvement for Q3_K and Q5_K on ARM_NEON * k_quants: WIP super-blocks with 64 weights Yet another speedup for Q5_K on ARM_NEON. We are now within 10% of the QK_K = 256 version. * k_quants: WIP super-blocks with 64 weights * We are able to pass preprocessor macros to the Metal compiler * Q6_K works and is actually slightly more efficient than the QK_K = 256 version (25.2 ms vs 25.8 ms) * k_quants: WIP super-blocks with 64 weights Q4_K works on Metal and is actually slightly faster than QK_K = 256 (21.95 ms vs 24.0 ms). * k_quants: WIP super-blocks with 64 weights Q2_K works on Metal and is very slightly faster than QK_K = 256 (23.8 ms vs 24.2 ms). * k_quants: WIP super-blocks with 64 weights Q3_K works on Metal and is slightly faster than QK_K = 256 (26.6 ms vs 28.3 ms). * k_quants: WIP super-blocks with 64 weights Q5_K works on Metal and is slightly faster than QK_K = 256 (23.7 ms vs 26.3 ms). * k_quants: call them _K, not _k, also on Metal * k_quants: correctly define QK_K in llama.cpp * Fixed bug in q4_K quantization added with the 64-block addition * Simplify via lambda * k_quants: swicth Q3_K to 4-bit scales when QK_K = 64 Otherwise there isn't much benefit from this quantization type. There is some very slight loss in accuracy, but we reduce size by ~7%. E.g., for OpenLLaMA-3B, Q3_K_S perplexity is 8.6131 with 8-bit scales and 8.6352 with 4-bit, while file size decreases from 1.53G to 1.44G. * k_quants: switch Q4_K to 4-bit scales when QK_K = 64 Here the loss in accuracy is greater than for Q3_K, but the Q4_K points still move further to the left on the perplexity vs size curve. * k_quants: forgot to add the Metal changes in last commit * k_quants: change Q5_K to be type 0 when QK_K = 64 Still needs AVX2 implementation * k_quants: AVX2 implementation for new 64-weight Q5_K * k_quants: 10% faster ARM_NEON Q5_K dot product * k_quants: fixed issue caused by merging with master --------- Co-authored-by: Iwan Kawrakow <iwan.kawrakow@gmail.com>
2023-06-26 16:43:07 +00:00
__m256i q3h_1 = _mm256_srli_epi16(q3h_0, 4);
q3h_0 = _mm256_slli_epi16(_mm256_andnot_si256(q3h_0, m1), 2);
q3h_1 = _mm256_slli_epi16(_mm256_andnot_si256(q3h_1, m1), 2);
// load low 2 bits
const __m128i q3bits = _mm_loadu_si128((const __m128i*)q3);
// prepare low and high bits
ggml : workaround for missing _mm256_setr_m128i in GCC < 8 in k_quants.c (#2405)
2023-07-28 18:17:45 +00:00
const __m256i q3aux = MM256_SET_M128I(_mm_srli_epi16(q3bits, 2), q3bits);
k-quants : support for super-block size of 64 (#2001) * k_quants: WIP super-blocks with 64 weights * k_quants: WIP super-blocks with 64 weights Q6_K scalar and AVX2 works * k_quants: WIP super-blocks with 64 weights Q4_K scalar and AVX2 works * k_quants: WIP super-blocks with 64 weights Q2_K scalar and AVX2 works. Q2_K is way too slow (it is actually slower than the scalar implementation) * k_quants: WIP super-blocks with 64 weights Q3_K scalar and AVX2 works. * k_quants: WIP super-blocks with 64 weights Q5_K scalar and AVX2 works, and with that all k_quants are done on AVX2 and scalar * k_quants: WIP super-blocks with 64 weights Q6_K working on CUDA. Cannot make it run quite as gast as with super-blocks with 256 weigths: 8% slower on 4080, 20% slower on the 1660 (but there we fit 1 less layer on the GPU because pf the larger model size), so some fraction of these 20% is due to that, * k_quants: WIP super-blocks with 64 weights Q4_K working on CUDA. ~10% slower on GTX-1660, 16% slower on 4080. * k_quants: WIP super-blocks with 64 weights Q2_K working on CUDA. ~3% slower on GTX-1660, 10% slower on 4080. * k_quants: WIP super-blocks with 64 weights Q3_K working on CUDA. * k_quants: WIP super-blocks with 64 weights Q5_K working on CUDA, and with this CUDA is done. * k_quants: WIP super-blocks with 64 weights Q6_K working on ARM_NEON * k_quants: WIP super-blocks with 64 weights Q4_K working on ARM_NEON, but quite a bit slower than 256 weights * k_quants: WIP super-blocks with 64 weights Q2_K working on ARM_NEON, but quite a bit slower than 256 weights * k_quants: WIP super-blocks with 64 weights Q3_K working on ARM_NEON, but quite a bit slower than 256 weights. * k_quants: WIP super-blocks with 64 weights Q5_K working on ARM_NEON, but quite a bit slower than 256 weights. With that, we have full support for ARM_NEON, although performance is not quite there. * k_quants: WIP super-blocks with 64 weights Slightly more efficient Q3_K and Q5_K * k_quants: WIP super-blocks with 64 weights Another small improvement for Q3_K and Q5_K on ARM_NEON * k_quants: WIP super-blocks with 64 weights Yet another speedup for Q5_K on ARM_NEON. We are now within 10% of the QK_K = 256 version. * k_quants: WIP super-blocks with 64 weights * We are able to pass preprocessor macros to the Metal compiler * Q6_K works and is actually slightly more efficient than the QK_K = 256 version (25.2 ms vs 25.8 ms) * k_quants: WIP super-blocks with 64 weights Q4_K works on Metal and is actually slightly faster than QK_K = 256 (21.95 ms vs 24.0 ms). * k_quants: WIP super-blocks with 64 weights Q2_K works on Metal and is very slightly faster than QK_K = 256 (23.8 ms vs 24.2 ms). * k_quants: WIP super-blocks with 64 weights Q3_K works on Metal and is slightly faster than QK_K = 256 (26.6 ms vs 28.3 ms). * k_quants: WIP super-blocks with 64 weights Q5_K works on Metal and is slightly faster than QK_K = 256 (23.7 ms vs 26.3 ms). * k_quants: call them _K, not _k, also on Metal * k_quants: correctly define QK_K in llama.cpp * Fixed bug in q4_K quantization added with the 64-block addition * Simplify via lambda * k_quants: swicth Q3_K to 4-bit scales when QK_K = 64 Otherwise there isn't much benefit from this quantization type. There is some very slight loss in accuracy, but we reduce size by ~7%. E.g., for OpenLLaMA-3B, Q3_K_S perplexity is 8.6131 with 8-bit scales and 8.6352 with 4-bit, while file size decreases from 1.53G to 1.44G. * k_quants: switch Q4_K to 4-bit scales when QK_K = 64 Here the loss in accuracy is greater than for Q3_K, but the Q4_K points still move further to the left on the perplexity vs size curve. * k_quants: forgot to add the Metal changes in last commit * k_quants: change Q5_K to be type 0 when QK_K = 64 Still needs AVX2 implementation * k_quants: AVX2 implementation for new 64-weight Q5_K * k_quants: 10% faster ARM_NEON Q5_K dot product * k_quants: fixed issue caused by merging with master --------- Co-authored-by: Iwan Kawrakow <iwan.kawrakow@gmail.com>
2023-06-26 16:43:07 +00:00
const __m256i q3l_0 = _mm256_and_si256(q3aux, m3);
const __m256i q3l_1 = _mm256_and_si256(_mm256_srli_epi16(q3aux, 4), m3);
// load Q8 quants
const __m256i q8_0 = _mm256_loadu_si256((const __m256i*)(q8+ 0));
const __m256i q8_1 = _mm256_loadu_si256((const __m256i*)(q8+32));
// Dot product: we multiply the 2 low bits and 1 high bit part separately, so we can use _mm256_maddubs_epi16,
// and then subtract. The high bit part has the 2 already subtracted (and so, it is zero if the high bit was not set,
// and 2 if the high bit was set)
const __m256i q8s_0 = _mm256_maddubs_epi16(q3h_0, q8_0);
const __m256i q8s_1 = _mm256_maddubs_epi16(q3h_1, q8_1);
__m256i p16_0 = _mm256_maddubs_epi16(q3l_0, q8_0);
__m256i p16_1 = _mm256_maddubs_epi16(q3l_1, q8_1);
p16_0 = _mm256_sub_epi16(p16_0, q8s_0);
p16_1 = _mm256_sub_epi16(p16_1, q8s_1);
// multiply with scales
p16_0 = _mm256_madd_epi16(scale_0, p16_0);
p16_1 = _mm256_madd_epi16(scale_1, p16_1);
p16_0 = _mm256_add_epi32(p16_0, p16_1);
// multiply with block scale and accumulate
acc = _mm256_fmadd_ps(_mm256_broadcast_ss(&d), _mm256_cvtepi32_ps(p16_0), acc);
}
*s = hsum_float_8(acc);
k_quants : add AVX support to dot functions with QK_K as 64 (#2339) * add AVX to ggml_vec_dot_q2_K_q8_K() * add AVX to ggml_vec_dot_q3_K_q8_K() * add AVX to ggml_vec_dot_q4_K_q8_K() * add AVX to ggml_vec_dot_q5_K_q8_K() * add AVX to ggml_vec_dot_q6_K_q8_K() * refactor AVX code in ggml_vec_dot_q6_K_q8_K()
2023-07-25 12:13:41 +00:00
#elif defined __AVX__
const __m128i m3 = _mm_set1_epi8(3);
const __m128i m1 = _mm_set1_epi8(1);
__m256 acc = _mm256_setzero_ps();
uint64_t aux64;
uint16_t aux16[2];
const int8_t * aux8 = (const int8_t *)aux16;
for (int i = 0; i < nb; ++i) {
ggml : move FP16 <-> FP32 code to ggml-impl.h (#3861) * ggml : move FP16 <-> FP32 stuff to ggml-impl.h ggml-ci * tests : fix ARM build * ggml : explicitly initialize deprecated type traits * ggml : add math.h to ggml-impl.h * ggml : remove duplicate static assert macros * ggml : prefix lookup tables with ggml_ ggml-ci * ggml-impl : move extern "C" to start of file
2023-10-30 17:19:15 +00:00
const float d = y[i].d * GGML_FP16_TO_FP32(x[i].d);
k_quants : add AVX support to dot functions with QK_K as 64 (#2339) * add AVX to ggml_vec_dot_q2_K_q8_K() * add AVX to ggml_vec_dot_q3_K_q8_K() * add AVX to ggml_vec_dot_q4_K_q8_K() * add AVX to ggml_vec_dot_q5_K_q8_K() * add AVX to ggml_vec_dot_q6_K_q8_K() * refactor AVX code in ggml_vec_dot_q6_K_q8_K()
2023-07-25 12:13:41 +00:00
const uint8_t * restrict q3 = x[i].qs;
const int8_t * restrict q8 = y[i].qs;
const uint16_t a = *(const uint16_t *)x[i].scales;
aux16[0] = a & 0x0f0f;
aux16[1] = (a >> 4) & 0x0f0f;
const __m128i scale_0 = _mm_set1_epi16(aux8[0] - 8);
const __m128i scale_1 = _mm_set1_epi16(aux8[2] - 8);
const __m128i scale_2 = _mm_set1_epi16(aux8[1] - 8);
const __m128i scale_3 = _mm_set1_epi16(aux8[3] - 8);
memcpy(&aux64, x[i].hmask, 8);
__m128i q3h_0 = _mm_set_epi64x(aux64 >> 1, aux64 >> 0);
__m128i q3h_1 = _mm_srli_epi16(q3h_0, 2);
__m128i q3h_2 = _mm_srli_epi16(q3h_0, 4);
__m128i q3h_3 = _mm_srli_epi16(q3h_0, 6);
q3h_0 = _mm_slli_epi16(_mm_andnot_si128(q3h_0, m1), 2);
q3h_1 = _mm_slli_epi16(_mm_andnot_si128(q3h_1, m1), 2);
q3h_2 = _mm_slli_epi16(_mm_andnot_si128(q3h_2, m1), 2);
q3h_3 = _mm_slli_epi16(_mm_andnot_si128(q3h_3, m1), 2);
// load low 2 bits
const __m128i q3bits = _mm_loadu_si128((const __m128i*)q3);
// prepare low and high bits
const __m128i q3l_0 = _mm_and_si128(q3bits, m3);
const __m128i q3l_1 = _mm_and_si128(_mm_srli_epi16(q3bits, 2), m3);
const __m128i q3l_2 = _mm_and_si128(_mm_srli_epi16(q3bits, 4), m3);
const __m128i q3l_3 = _mm_and_si128(_mm_srli_epi16(q3bits, 6), m3);
// load Q8 quants
const __m256i q8_0 = _mm256_loadu_si256((const __m256i*)(q8+ 0));
const __m256i q8_1 = _mm256_loadu_si256((const __m256i*)(q8+32));
// Dot product: we multiply the 2 low bits and 1 high bit part separately, so we can use _mm_maddubs_epi16,
// and then subtract. The high bit part has the 2 already subtracted (and so, it is zero if the high bit was not set,
// and 2 if the high bit was set)
const __m128i q8s_0 = _mm_maddubs_epi16(q3h_0, _mm256_extractf128_si256(q8_0, 0));
const __m128i q8s_1 = _mm_maddubs_epi16(q3h_1, _mm256_extractf128_si256(q8_0, 1));
const __m128i q8s_2 = _mm_maddubs_epi16(q3h_2, _mm256_extractf128_si256(q8_1, 0));
const __m128i q8s_3 = _mm_maddubs_epi16(q3h_3, _mm256_extractf128_si256(q8_1, 1));
__m128i p16_0 = _mm_maddubs_epi16(q3l_0, _mm256_extractf128_si256(q8_0, 0));
__m128i p16_1 = _mm_maddubs_epi16(q3l_1, _mm256_extractf128_si256(q8_0, 1));
__m128i p16_2 = _mm_maddubs_epi16(q3l_2, _mm256_extractf128_si256(q8_1, 0));
__m128i p16_3 = _mm_maddubs_epi16(q3l_3, _mm256_extractf128_si256(q8_1, 1));
p16_0 = _mm_sub_epi16(p16_0, q8s_0);
p16_1 = _mm_sub_epi16(p16_1, q8s_1);
p16_2 = _mm_sub_epi16(p16_2, q8s_2);
p16_3 = _mm_sub_epi16(p16_3, q8s_3);
// multiply with scales
p16_0 = _mm_madd_epi16(scale_0, p16_0);
p16_1 = _mm_madd_epi16(scale_1, p16_1);
p16_2 = _mm_madd_epi16(scale_2, p16_2);
p16_3 = _mm_madd_epi16(scale_3, p16_3);
p16_0 = _mm_add_epi32(p16_0, p16_2);
p16_1 = _mm_add_epi32(p16_1, p16_3);
ggml : workaround for missing _mm256_setr_m128i in GCC < 8 in k_quants.c (#2405)
2023-07-28 18:17:45 +00:00
__m256i p16 = MM256_SET_M128I(p16_1, p16_0);
k_quants : add AVX support to dot functions with QK_K as 64 (#2339) * add AVX to ggml_vec_dot_q2_K_q8_K() * add AVX to ggml_vec_dot_q3_K_q8_K() * add AVX to ggml_vec_dot_q4_K_q8_K() * add AVX to ggml_vec_dot_q5_K_q8_K() * add AVX to ggml_vec_dot_q6_K_q8_K() * refactor AVX code in ggml_vec_dot_q6_K_q8_K()
2023-07-25 12:13:41 +00:00
// multiply with block scale and accumulate
acc = _mm256_add_ps(_mm256_mul_ps(_mm256_broadcast_ss(&d), _mm256_cvtepi32_ps(p16)), acc);
}
*s = hsum_float_8(acc);
ggml : add RISC-V Vector Support for K-Quants and improved the existing intrinsics (#3453) * Added RVV intrinsics support for Q8 quantize row and also improved the existing dot product function for risc-v. The RVV intrinsics is added for the following quantize row functions quantize_row_q8_0 quantize_row_q8_1 The following dot product functions have also been optimized by using LMUL = 1/2 instead of LMUL = 1 ggml_vec_dot_q4_0_q8_0 ggml_vec_dot_q4_1_q8_1 ggml_vec_dot_q5_0_q8_0 ggml_vec_dot_q5_1_q8_1 And vector initialization in Q5 by temporary array is also replaced by the vid intrinsics Signed-off-by: Ahmad Tameem <ahmad.tameem@10xengineers.ai> * Added RVV intrinsics support for k_quants This adds RISC-V Vector intrinsics support for the following K_quants functions for both QKK = 256 and QKK = 64 ggml_vec_dot_q2_K_q8_K ggml_vec_dot_q3_K_q8_K ggml_vec_dot_q4_K_q8_K ggml_vec_dot_q5_K_q8_K ggml_vec_dot_q6_K_q8_K Signed-off-by: Ahmad Tameem <ahmad.tameem@10xengineers.ai> --------- Signed-off-by: Ahmad Tameem <ahmad.tameem@10xengineers.ai>
2023-10-03 18:38:19 +00:00
#elif defined __riscv_v_intrinsic
uint16_t aux16[2];
int8_t * scales = (int8_t *)aux16;
float sumf = 0;
for (int i = 0; i < nb; ++i) {
const uint8_t * restrict q3 = x[i].qs;
const int8_t * restrict q8 = y[i].qs;
const uint16_t a = *(const uint16_t *)x[i].scales;
aux16[0] = a & 0x0f0f;
aux16[1] = (a >> 4) & 0x0f0f;
for (int j = 0; j < 4; ++j) scales[j] -= 8;
int32_t isum = -4*(scales[0] * y[i].bsums[0] + scales[2] * y[i].bsums[1] + scales[1] * y[i].bsums[2] + scales[3] * y[i].bsums[3]);
const float d = y[i].d * (float)x[i].d;
vint32m1_t vzero = __riscv_vmv_v_x_i32m1(0, 1);
// load qh
vuint8mf4_t qh_x1 = __riscv_vle8_v_u8mf4(x[i].hmask, 8);
vuint8mf2_t qh_x2 = __riscv_vlmul_ext_v_u8mf4_u8mf2(__riscv_vsrl_vx_u8mf4(qh_x1, 1, 8));
size_t vl = 16;
// extend and combine both qh_x1 and qh_x2
vuint8mf2_t qh_x = __riscv_vslideup_vx_u8mf2(__riscv_vlmul_ext_v_u8mf4_u8mf2(qh_x1), qh_x2, vl/2, vl);
vuint8mf2_t qh_0 = __riscv_vand_vx_u8mf2(__riscv_vsll_vx_u8mf2(qh_x, 0x2, vl), 0x4, vl);
vuint8mf2_t qh_1 = __riscv_vand_vx_u8mf2(qh_x, 0x4, vl);
vuint8mf2_t qh_2 = __riscv_vand_vx_u8mf2(__riscv_vsrl_vx_u8mf2(qh_x, 0x2, vl), 0x4, vl);
vuint8mf2_t qh_3 = __riscv_vand_vx_u8mf2(__riscv_vsrl_vx_u8mf2(qh_x, 0x4, vl), 0x4, vl);
// load Q3
vuint8mf2_t q3_x = __riscv_vle8_v_u8mf2(q3, vl);
vuint8mf2_t q3h_0 = __riscv_vor_vv_u8mf2(__riscv_vand_vx_u8mf2(q3_x, 0x3, vl), qh_0, vl);
vuint8mf2_t q3h_1 = __riscv_vor_vv_u8mf2(__riscv_vand_vx_u8mf2(__riscv_vsrl_vx_u8mf2(q3_x, 2, vl), 0x3, vl), qh_1, vl);
vuint8mf2_t q3h_2 = __riscv_vor_vv_u8mf2(__riscv_vand_vx_u8mf2(__riscv_vsrl_vx_u8mf2(q3_x, 4, vl), 0x3, vl), qh_2, vl);
vuint8mf2_t q3h_3 = __riscv_vor_vv_u8mf2(__riscv_vsrl_vx_u8mf2(q3_x, 0x6, vl), qh_3, vl);
vint8mf2_t q3_0 = __riscv_vreinterpret_v_u8mf2_i8mf2(q3h_0);
vint8mf2_t q3_1 = __riscv_vreinterpret_v_u8mf2_i8mf2(q3h_1);
vint8mf2_t q3_2 = __riscv_vreinterpret_v_u8mf2_i8mf2(q3h_2);
vint8mf2_t q3_3 = __riscv_vreinterpret_v_u8mf2_i8mf2(q3h_3);
// load Q8 and take product with Q3
vint16m1_t p0 = __riscv_vwmul_vv_i16m1(q3_0, __riscv_vle8_v_i8mf2(q8, vl), vl);
vint16m1_t p1 = __riscv_vwmul_vv_i16m1(q3_1, __riscv_vle8_v_i8mf2(q8+16, vl), vl);
vint16m1_t p2 = __riscv_vwmul_vv_i16m1(q3_2, __riscv_vle8_v_i8mf2(q8+32, vl), vl);
vint16m1_t p3 = __riscv_vwmul_vv_i16m1(q3_3, __riscv_vle8_v_i8mf2(q8+48, vl), vl);
vint32m1_t vs_0 = __riscv_vwredsum_vs_i16m1_i32m1(p0, vzero, vl);
vint32m1_t vs_1 = __riscv_vwredsum_vs_i16m1_i32m1(p1, vzero, vl);
vint32m1_t vs_2 = __riscv_vwredsum_vs_i16m1_i32m1(p2, vzero, vl);
vint32m1_t vs_3 = __riscv_vwredsum_vs_i16m1_i32m1(p3, vzero, vl);
isum += __riscv_vmv_x_s_i32m1_i32(vs_0) * scales[0];
isum += __riscv_vmv_x_s_i32m1_i32(vs_1) * scales[2];
isum += __riscv_vmv_x_s_i32m1_i32(vs_2) * scales[1];
isum += __riscv_vmv_x_s_i32m1_i32(vs_3) * scales[3];
sumf += d * isum;
}
*s = sumf;
k-quants : support for super-block size of 64 (#2001) * k_quants: WIP super-blocks with 64 weights * k_quants: WIP super-blocks with 64 weights Q6_K scalar and AVX2 works * k_quants: WIP super-blocks with 64 weights Q4_K scalar and AVX2 works * k_quants: WIP super-blocks with 64 weights Q2_K scalar and AVX2 works. Q2_K is way too slow (it is actually slower than the scalar implementation) * k_quants: WIP super-blocks with 64 weights Q3_K scalar and AVX2 works. * k_quants: WIP super-blocks with 64 weights Q5_K scalar and AVX2 works, and with that all k_quants are done on AVX2 and scalar * k_quants: WIP super-blocks with 64 weights Q6_K working on CUDA. Cannot make it run quite as gast as with super-blocks with 256 weigths: 8% slower on 4080, 20% slower on the 1660 (but there we fit 1 less layer on the GPU because pf the larger model size), so some fraction of these 20% is due to that, * k_quants: WIP super-blocks with 64 weights Q4_K working on CUDA. ~10% slower on GTX-1660, 16% slower on 4080. * k_quants: WIP super-blocks with 64 weights Q2_K working on CUDA. ~3% slower on GTX-1660, 10% slower on 4080. * k_quants: WIP super-blocks with 64 weights Q3_K working on CUDA. * k_quants: WIP super-blocks with 64 weights Q5_K working on CUDA, and with this CUDA is done. * k_quants: WIP super-blocks with 64 weights Q6_K working on ARM_NEON * k_quants: WIP super-blocks with 64 weights Q4_K working on ARM_NEON, but quite a bit slower than 256 weights * k_quants: WIP super-blocks with 64 weights Q2_K working on ARM_NEON, but quite a bit slower than 256 weights * k_quants: WIP super-blocks with 64 weights Q3_K working on ARM_NEON, but quite a bit slower than 256 weights. * k_quants: WIP super-blocks with 64 weights Q5_K working on ARM_NEON, but quite a bit slower than 256 weights. With that, we have full support for ARM_NEON, although performance is not quite there. * k_quants: WIP super-blocks with 64 weights Slightly more efficient Q3_K and Q5_K * k_quants: WIP super-blocks with 64 weights Another small improvement for Q3_K and Q5_K on ARM_NEON * k_quants: WIP super-blocks with 64 weights Yet another speedup for Q5_K on ARM_NEON. We are now within 10% of the QK_K = 256 version. * k_quants: WIP super-blocks with 64 weights * We are able to pass preprocessor macros to the Metal compiler * Q6_K works and is actually slightly more efficient than the QK_K = 256 version (25.2 ms vs 25.8 ms) * k_quants: WIP super-blocks with 64 weights Q4_K works on Metal and is actually slightly faster than QK_K = 256 (21.95 ms vs 24.0 ms). * k_quants: WIP super-blocks with 64 weights Q2_K works on Metal and is very slightly faster than QK_K = 256 (23.8 ms vs 24.2 ms). * k_quants: WIP super-blocks with 64 weights Q3_K works on Metal and is slightly faster than QK_K = 256 (26.6 ms vs 28.3 ms). * k_quants: WIP super-blocks with 64 weights Q5_K works on Metal and is slightly faster than QK_K = 256 (23.7 ms vs 26.3 ms). * k_quants: call them _K, not _k, also on Metal * k_quants: correctly define QK_K in llama.cpp * Fixed bug in q4_K quantization added with the 64-block addition * Simplify via lambda * k_quants: swicth Q3_K to 4-bit scales when QK_K = 64 Otherwise there isn't much benefit from this quantization type. There is some very slight loss in accuracy, but we reduce size by ~7%. E.g., for OpenLLaMA-3B, Q3_K_S perplexity is 8.6131 with 8-bit scales and 8.6352 with 4-bit, while file size decreases from 1.53G to 1.44G. * k_quants: switch Q4_K to 4-bit scales when QK_K = 64 Here the loss in accuracy is greater than for Q3_K, but the Q4_K points still move further to the left on the perplexity vs size curve. * k_quants: forgot to add the Metal changes in last commit * k_quants: change Q5_K to be type 0 when QK_K = 64 Still needs AVX2 implementation * k_quants: AVX2 implementation for new 64-weight Q5_K * k_quants: 10% faster ARM_NEON Q5_K dot product * k_quants: fixed issue caused by merging with master --------- Co-authored-by: Iwan Kawrakow <iwan.kawrakow@gmail.com>
2023-06-26 16:43:07 +00:00
#else
int8_t aux8[QK_K];
int16_t aux16[8];
float sums [8];
int32_t aux32[8];
int32_t scales[4];
memset(sums, 0, 8*sizeof(float));
float sumf = 0;
for (int i = 0; i < nb; ++i) {
const uint8_t * restrict q3 = x[i].qs;
const uint8_t * restrict hm = x[i].hmask;
const int8_t * restrict q8 = y[i].qs;
int8_t * restrict a = aux8;
for (int l = 0; l < 8; ++l) {
a[l+ 0] = (int8_t)((q3[l+0] >> 0) & 3) - (hm[l] & 0x01 ? 0 : 4);
a[l+ 8] = (int8_t)((q3[l+8] >> 0) & 3) - (hm[l] & 0x02 ? 0 : 4);
a[l+16] = (int8_t)((q3[l+0] >> 2) & 3) - (hm[l] & 0x04 ? 0 : 4);
a[l+24] = (int8_t)((q3[l+8] >> 2) & 3) - (hm[l] & 0x08 ? 0 : 4);
a[l+32] = (int8_t)((q3[l+0] >> 4) & 3) - (hm[l] & 0x10 ? 0 : 4);
a[l+40] = (int8_t)((q3[l+8] >> 4) & 3) - (hm[l] & 0x20 ? 0 : 4);
a[l+48] = (int8_t)((q3[l+0] >> 6) & 3) - (hm[l] & 0x40 ? 0 : 4);
a[l+56] = (int8_t)((q3[l+8] >> 6) & 3) - (hm[l] & 0x80 ? 0 : 4);
}
scales[0] = (x[i].scales[0] & 0xF) - 8;
scales[1] = (x[i].scales[0] >> 4) - 8;
scales[2] = (x[i].scales[1] & 0xF) - 8;
scales[3] = (x[i].scales[1] >> 4) - 8;
memset(aux32, 0, 8*sizeof(int32_t));
for (int j = 0; j < QK_K/16; ++j) {
for (int l = 0; l < 8; ++l) aux16[l] = q8[l] * a[l];
q8 += 8; a += 8;
for (int l = 0; l < 8; ++l) aux16[l] += q8[l] * a[l];
q8 += 8; a += 8;
for (int l = 0; l < 8; ++l) aux32[l] += scales[j] * aux16[l];
}
ggml : move FP16 <-> FP32 code to ggml-impl.h (#3861) * ggml : move FP16 <-> FP32 stuff to ggml-impl.h ggml-ci * tests : fix ARM build * ggml : explicitly initialize deprecated type traits * ggml : add math.h to ggml-impl.h * ggml : remove duplicate static assert macros * ggml : prefix lookup tables with ggml_ ggml-ci * ggml-impl : move extern "C" to start of file
2023-10-30 17:19:15 +00:00
const float d = GGML_FP16_TO_FP32(x[i].d) * y[i].d;
k-quants : support for super-block size of 64 (#2001) * k_quants: WIP super-blocks with 64 weights * k_quants: WIP super-blocks with 64 weights Q6_K scalar and AVX2 works * k_quants: WIP super-blocks with 64 weights Q4_K scalar and AVX2 works * k_quants: WIP super-blocks with 64 weights Q2_K scalar and AVX2 works. Q2_K is way too slow (it is actually slower than the scalar implementation) * k_quants: WIP super-blocks with 64 weights Q3_K scalar and AVX2 works. * k_quants: WIP super-blocks with 64 weights Q5_K scalar and AVX2 works, and with that all k_quants are done on AVX2 and scalar * k_quants: WIP super-blocks with 64 weights Q6_K working on CUDA. Cannot make it run quite as gast as with super-blocks with 256 weigths: 8% slower on 4080, 20% slower on the 1660 (but there we fit 1 less layer on the GPU because pf the larger model size), so some fraction of these 20% is due to that, * k_quants: WIP super-blocks with 64 weights Q4_K working on CUDA. ~10% slower on GTX-1660, 16% slower on 4080. * k_quants: WIP super-blocks with 64 weights Q2_K working on CUDA. ~3% slower on GTX-1660, 10% slower on 4080. * k_quants: WIP super-blocks with 64 weights Q3_K working on CUDA. * k_quants: WIP super-blocks with 64 weights Q5_K working on CUDA, and with this CUDA is done. * k_quants: WIP super-blocks with 64 weights Q6_K working on ARM_NEON * k_quants: WIP super-blocks with 64 weights Q4_K working on ARM_NEON, but quite a bit slower than 256 weights * k_quants: WIP super-blocks with 64 weights Q2_K working on ARM_NEON, but quite a bit slower than 256 weights * k_quants: WIP super-blocks with 64 weights Q3_K working on ARM_NEON, but quite a bit slower than 256 weights. * k_quants: WIP super-blocks with 64 weights Q5_K working on ARM_NEON, but quite a bit slower than 256 weights. With that, we have full support for ARM_NEON, although performance is not quite there. * k_quants: WIP super-blocks with 64 weights Slightly more efficient Q3_K and Q5_K * k_quants: WIP super-blocks with 64 weights Another small improvement for Q3_K and Q5_K on ARM_NEON * k_quants: WIP super-blocks with 64 weights Yet another speedup for Q5_K on ARM_NEON. We are now within 10% of the QK_K = 256 version. * k_quants: WIP super-blocks with 64 weights * We are able to pass preprocessor macros to the Metal compiler * Q6_K works and is actually slightly more efficient than the QK_K = 256 version (25.2 ms vs 25.8 ms) * k_quants: WIP super-blocks with 64 weights Q4_K works on Metal and is actually slightly faster than QK_K = 256 (21.95 ms vs 24.0 ms). * k_quants: WIP super-blocks with 64 weights Q2_K works on Metal and is very slightly faster than QK_K = 256 (23.8 ms vs 24.2 ms). * k_quants: WIP super-blocks with 64 weights Q3_K works on Metal and is slightly faster than QK_K = 256 (26.6 ms vs 28.3 ms). * k_quants: WIP super-blocks with 64 weights Q5_K works on Metal and is slightly faster than QK_K = 256 (23.7 ms vs 26.3 ms). * k_quants: call them _K, not _k, also on Metal * k_quants: correctly define QK_K in llama.cpp * Fixed bug in q4_K quantization added with the 64-block addition * Simplify via lambda * k_quants: swicth Q3_K to 4-bit scales when QK_K = 64 Otherwise there isn't much benefit from this quantization type. There is some very slight loss in accuracy, but we reduce size by ~7%. E.g., for OpenLLaMA-3B, Q3_K_S perplexity is 8.6131 with 8-bit scales and 8.6352 with 4-bit, while file size decreases from 1.53G to 1.44G. * k_quants: switch Q4_K to 4-bit scales when QK_K = 64 Here the loss in accuracy is greater than for Q3_K, but the Q4_K points still move further to the left on the perplexity vs size curve. * k_quants: forgot to add the Metal changes in last commit * k_quants: change Q5_K to be type 0 when QK_K = 64 Still needs AVX2 implementation * k_quants: AVX2 implementation for new 64-weight Q5_K * k_quants: 10% faster ARM_NEON Q5_K dot product * k_quants: fixed issue caused by merging with master --------- Co-authored-by: Iwan Kawrakow <iwan.kawrakow@gmail.com>
2023-06-26 16:43:07 +00:00
for (int l = 0; l < 8; ++l) sums[l] += d * aux32[l];
}
for (int l = 0; l < 8; ++l) sumf += sums[l];
*s = sumf;
#endif
}
#endif
#if QK_K == 256
void ggml_vec_dot_q4_K_q8_K(const int n, float * restrict s, const void * restrict vx, const void * restrict vy) {
assert(n % QK_K == 0);
const block_q4_K * restrict x = vx;
const block_q8_K * restrict y = vy;
const int nb = n / QK_K;
static const uint32_t kmask1 = 0x3f3f3f3f;
static const uint32_t kmask2 = 0x0f0f0f0f;
static const uint32_t kmask3 = 0x03030303;
uint32_t utmp[4];
#ifdef __ARM_NEON
const uint8x16_t m4b = vdupq_n_u8(0xf);
const int32x4_t mzero = vdupq_n_s32(0);
ggml : sync (im2col, GPU conv, 32-bit arm compat) (#4060) ggml-ci
2023-11-13 14:55:52 +00:00
ggml_int8x16x2_t q4bytes;
ggml_int8x16x2_t q8bytes;
k-quants : support for super-block size of 64 (#2001) * k_quants: WIP super-blocks with 64 weights * k_quants: WIP super-blocks with 64 weights Q6_K scalar and AVX2 works * k_quants: WIP super-blocks with 64 weights Q4_K scalar and AVX2 works * k_quants: WIP super-blocks with 64 weights Q2_K scalar and AVX2 works. Q2_K is way too slow (it is actually slower than the scalar implementation) * k_quants: WIP super-blocks with 64 weights Q3_K scalar and AVX2 works. * k_quants: WIP super-blocks with 64 weights Q5_K scalar and AVX2 works, and with that all k_quants are done on AVX2 and scalar * k_quants: WIP super-blocks with 64 weights Q6_K working on CUDA. Cannot make it run quite as gast as with super-blocks with 256 weigths: 8% slower on 4080, 20% slower on the 1660 (but there we fit 1 less layer on the GPU because pf the larger model size), so some fraction of these 20% is due to that, * k_quants: WIP super-blocks with 64 weights Q4_K working on CUDA. ~10% slower on GTX-1660, 16% slower on 4080. * k_quants: WIP super-blocks with 64 weights Q2_K working on CUDA. ~3% slower on GTX-1660, 10% slower on 4080. * k_quants: WIP super-blocks with 64 weights Q3_K working on CUDA. * k_quants: WIP super-blocks with 64 weights Q5_K working on CUDA, and with this CUDA is done. * k_quants: WIP super-blocks with 64 weights Q6_K working on ARM_NEON * k_quants: WIP super-blocks with 64 weights Q4_K working on ARM_NEON, but quite a bit slower than 256 weights * k_quants: WIP super-blocks with 64 weights Q2_K working on ARM_NEON, but quite a bit slower than 256 weights * k_quants: WIP super-blocks with 64 weights Q3_K working on ARM_NEON, but quite a bit slower than 256 weights. * k_quants: WIP super-blocks with 64 weights Q5_K working on ARM_NEON, but quite a bit slower than 256 weights. With that, we have full support for ARM_NEON, although performance is not quite there. * k_quants: WIP super-blocks with 64 weights Slightly more efficient Q3_K and Q5_K * k_quants: WIP super-blocks with 64 weights Another small improvement for Q3_K and Q5_K on ARM_NEON * k_quants: WIP super-blocks with 64 weights Yet another speedup for Q5_K on ARM_NEON. We are now within 10% of the QK_K = 256 version. * k_quants: WIP super-blocks with 64 weights * We are able to pass preprocessor macros to the Metal compiler * Q6_K works and is actually slightly more efficient than the QK_K = 256 version (25.2 ms vs 25.8 ms) * k_quants: WIP super-blocks with 64 weights Q4_K works on Metal and is actually slightly faster than QK_K = 256 (21.95 ms vs 24.0 ms). * k_quants: WIP super-blocks with 64 weights Q2_K works on Metal and is very slightly faster than QK_K = 256 (23.8 ms vs 24.2 ms). * k_quants: WIP super-blocks with 64 weights Q3_K works on Metal and is slightly faster than QK_K = 256 (26.6 ms vs 28.3 ms). * k_quants: WIP super-blocks with 64 weights Q5_K works on Metal and is slightly faster than QK_K = 256 (23.7 ms vs 26.3 ms). * k_quants: call them _K, not _k, also on Metal * k_quants: correctly define QK_K in llama.cpp * Fixed bug in q4_K quantization added with the 64-block addition * Simplify via lambda * k_quants: swicth Q3_K to 4-bit scales when QK_K = 64 Otherwise there isn't much benefit from this quantization type. There is some very slight loss in accuracy, but we reduce size by ~7%. E.g., for OpenLLaMA-3B, Q3_K_S perplexity is 8.6131 with 8-bit scales and 8.6352 with 4-bit, while file size decreases from 1.53G to 1.44G. * k_quants: switch Q4_K to 4-bit scales when QK_K = 64 Here the loss in accuracy is greater than for Q3_K, but the Q4_K points still move further to the left on the perplexity vs size curve. * k_quants: forgot to add the Metal changes in last commit * k_quants: change Q5_K to be type 0 when QK_K = 64 Still needs AVX2 implementation * k_quants: AVX2 implementation for new 64-weight Q5_K * k_quants: 10% faster ARM_NEON Q5_K dot product * k_quants: fixed issue caused by merging with master --------- Co-authored-by: Iwan Kawrakow <iwan.kawrakow@gmail.com>
2023-06-26 16:43:07 +00:00
float sumf = 0;
for (int i = 0; i < nb; ++i) {
ggml : move FP16 <-> FP32 code to ggml-impl.h (#3861) * ggml : move FP16 <-> FP32 stuff to ggml-impl.h ggml-ci * tests : fix ARM build * ggml : explicitly initialize deprecated type traits * ggml : add math.h to ggml-impl.h * ggml : remove duplicate static assert macros * ggml : prefix lookup tables with ggml_ ggml-ci * ggml-impl : move extern "C" to start of file
2023-10-30 17:19:15 +00:00
const float d = y[i].d * GGML_FP16_TO_FP32(x[i].d);
const float dmin = y[i].d * GGML_FP16_TO_FP32(x[i].dmin);
k-quants : support for super-block size of 64 (#2001) * k_quants: WIP super-blocks with 64 weights * k_quants: WIP super-blocks with 64 weights Q6_K scalar and AVX2 works * k_quants: WIP super-blocks with 64 weights Q4_K scalar and AVX2 works * k_quants: WIP super-blocks with 64 weights Q2_K scalar and AVX2 works. Q2_K is way too slow (it is actually slower than the scalar implementation) * k_quants: WIP super-blocks with 64 weights Q3_K scalar and AVX2 works. * k_quants: WIP super-blocks with 64 weights Q5_K scalar and AVX2 works, and with that all k_quants are done on AVX2 and scalar * k_quants: WIP super-blocks with 64 weights Q6_K working on CUDA. Cannot make it run quite as gast as with super-blocks with 256 weigths: 8% slower on 4080, 20% slower on the 1660 (but there we fit 1 less layer on the GPU because pf the larger model size), so some fraction of these 20% is due to that, * k_quants: WIP super-blocks with 64 weights Q4_K working on CUDA. ~10% slower on GTX-1660, 16% slower on 4080. * k_quants: WIP super-blocks with 64 weights Q2_K working on CUDA. ~3% slower on GTX-1660, 10% slower on 4080. * k_quants: WIP super-blocks with 64 weights Q3_K working on CUDA. * k_quants: WIP super-blocks with 64 weights Q5_K working on CUDA, and with this CUDA is done. * k_quants: WIP super-blocks with 64 weights Q6_K working on ARM_NEON * k_quants: WIP super-blocks with 64 weights Q4_K working on ARM_NEON, but quite a bit slower than 256 weights * k_quants: WIP super-blocks with 64 weights Q2_K working on ARM_NEON, but quite a bit slower than 256 weights * k_quants: WIP super-blocks with 64 weights Q3_K working on ARM_NEON, but quite a bit slower than 256 weights. * k_quants: WIP super-blocks with 64 weights Q5_K working on ARM_NEON, but quite a bit slower than 256 weights. With that, we have full support for ARM_NEON, although performance is not quite there. * k_quants: WIP super-blocks with 64 weights Slightly more efficient Q3_K and Q5_K * k_quants: WIP super-blocks with 64 weights Another small improvement for Q3_K and Q5_K on ARM_NEON * k_quants: WIP super-blocks with 64 weights Yet another speedup for Q5_K on ARM_NEON. We are now within 10% of the QK_K = 256 version. * k_quants: WIP super-blocks with 64 weights * We are able to pass preprocessor macros to the Metal compiler * Q6_K works and is actually slightly more efficient than the QK_K = 256 version (25.2 ms vs 25.8 ms) * k_quants: WIP super-blocks with 64 weights Q4_K works on Metal and is actually slightly faster than QK_K = 256 (21.95 ms vs 24.0 ms). * k_quants: WIP super-blocks with 64 weights Q2_K works on Metal and is very slightly faster than QK_K = 256 (23.8 ms vs 24.2 ms). * k_quants: WIP super-blocks with 64 weights Q3_K works on Metal and is slightly faster than QK_K = 256 (26.6 ms vs 28.3 ms). * k_quants: WIP super-blocks with 64 weights Q5_K works on Metal and is slightly faster than QK_K = 256 (23.7 ms vs 26.3 ms). * k_quants: call them _K, not _k, also on Metal * k_quants: correctly define QK_K in llama.cpp * Fixed bug in q4_K quantization added with the 64-block addition * Simplify via lambda * k_quants: swicth Q3_K to 4-bit scales when QK_K = 64 Otherwise there isn't much benefit from this quantization type. There is some very slight loss in accuracy, but we reduce size by ~7%. E.g., for OpenLLaMA-3B, Q3_K_S perplexity is 8.6131 with 8-bit scales and 8.6352 with 4-bit, while file size decreases from 1.53G to 1.44G. * k_quants: switch Q4_K to 4-bit scales when QK_K = 64 Here the loss in accuracy is greater than for Q3_K, but the Q4_K points still move further to the left on the perplexity vs size curve. * k_quants: forgot to add the Metal changes in last commit * k_quants: change Q5_K to be type 0 when QK_K = 64 Still needs AVX2 implementation * k_quants: AVX2 implementation for new 64-weight Q5_K * k_quants: 10% faster ARM_NEON Q5_K dot product * k_quants: fixed issue caused by merging with master --------- Co-authored-by: Iwan Kawrakow <iwan.kawrakow@gmail.com>
2023-06-26 16:43:07 +00:00
const int16x8_t q8sums = vpaddq_s16(vld1q_s16(y[i].bsums), vld1q_s16(y[i].bsums + 8));
memcpy(utmp, x[i].scales, 12);
arm64 support for windows (#3007) Co-authored-by: Cebtenzzre <cebtenzzre@gmail.com>
2023-09-13 01:54:20 +00:00
uint32x2_t mins8 = { 0 };
mins8 = vset_lane_u32(utmp[1] & kmask1, mins8, 0);
mins8 = vset_lane_u32(((utmp[2] >> 4) & kmask2) | (((utmp[1] >> 6) & kmask3) << 4), mins8, 1);
k-quants : support for super-block size of 64 (#2001) * k_quants: WIP super-blocks with 64 weights * k_quants: WIP super-blocks with 64 weights Q6_K scalar and AVX2 works * k_quants: WIP super-blocks with 64 weights Q4_K scalar and AVX2 works * k_quants: WIP super-blocks with 64 weights Q2_K scalar and AVX2 works. Q2_K is way too slow (it is actually slower than the scalar implementation) * k_quants: WIP super-blocks with 64 weights Q3_K scalar and AVX2 works. * k_quants: WIP super-blocks with 64 weights Q5_K scalar and AVX2 works, and with that all k_quants are done on AVX2 and scalar * k_quants: WIP super-blocks with 64 weights Q6_K working on CUDA. Cannot make it run quite as gast as with super-blocks with 256 weigths: 8% slower on 4080, 20% slower on the 1660 (but there we fit 1 less layer on the GPU because pf the larger model size), so some fraction of these 20% is due to that, * k_quants: WIP super-blocks with 64 weights Q4_K working on CUDA. ~10% slower on GTX-1660, 16% slower on 4080. * k_quants: WIP super-blocks with 64 weights Q2_K working on CUDA. ~3% slower on GTX-1660, 10% slower on 4080. * k_quants: WIP super-blocks with 64 weights Q3_K working on CUDA. * k_quants: WIP super-blocks with 64 weights Q5_K working on CUDA, and with this CUDA is done. * k_quants: WIP super-blocks with 64 weights Q6_K working on ARM_NEON * k_quants: WIP super-blocks with 64 weights Q4_K working on ARM_NEON, but quite a bit slower than 256 weights * k_quants: WIP super-blocks with 64 weights Q2_K working on ARM_NEON, but quite a bit slower than 256 weights * k_quants: WIP super-blocks with 64 weights Q3_K working on ARM_NEON, but quite a bit slower than 256 weights. * k_quants: WIP super-blocks with 64 weights Q5_K working on ARM_NEON, but quite a bit slower than 256 weights. With that, we have full support for ARM_NEON, although performance is not quite there. * k_quants: WIP super-blocks with 64 weights Slightly more efficient Q3_K and Q5_K * k_quants: WIP super-blocks with 64 weights Another small improvement for Q3_K and Q5_K on ARM_NEON * k_quants: WIP super-blocks with 64 weights Yet another speedup for Q5_K on ARM_NEON. We are now within 10% of the QK_K = 256 version. * k_quants: WIP super-blocks with 64 weights * We are able to pass preprocessor macros to the Metal compiler * Q6_K works and is actually slightly more efficient than the QK_K = 256 version (25.2 ms vs 25.8 ms) * k_quants: WIP super-blocks with 64 weights Q4_K works on Metal and is actually slightly faster than QK_K = 256 (21.95 ms vs 24.0 ms). * k_quants: WIP super-blocks with 64 weights Q2_K works on Metal and is very slightly faster than QK_K = 256 (23.8 ms vs 24.2 ms). * k_quants: WIP super-blocks with 64 weights Q3_K works on Metal and is slightly faster than QK_K = 256 (26.6 ms vs 28.3 ms). * k_quants: WIP super-blocks with 64 weights Q5_K works on Metal and is slightly faster than QK_K = 256 (23.7 ms vs 26.3 ms). * k_quants: call them _K, not _k, also on Metal * k_quants: correctly define QK_K in llama.cpp * Fixed bug in q4_K quantization added with the 64-block addition * Simplify via lambda * k_quants: swicth Q3_K to 4-bit scales when QK_K = 64 Otherwise there isn't much benefit from this quantization type. There is some very slight loss in accuracy, but we reduce size by ~7%. E.g., for OpenLLaMA-3B, Q3_K_S perplexity is 8.6131 with 8-bit scales and 8.6352 with 4-bit, while file size decreases from 1.53G to 1.44G. * k_quants: switch Q4_K to 4-bit scales when QK_K = 64 Here the loss in accuracy is greater than for Q3_K, but the Q4_K points still move further to the left on the perplexity vs size curve. * k_quants: forgot to add the Metal changes in last commit * k_quants: change Q5_K to be type 0 when QK_K = 64 Still needs AVX2 implementation * k_quants: AVX2 implementation for new 64-weight Q5_K * k_quants: 10% faster ARM_NEON Q5_K dot product * k_quants: fixed issue caused by merging with master --------- Co-authored-by: Iwan Kawrakow <iwan.kawrakow@gmail.com>
2023-06-26 16:43:07 +00:00
utmp[1] = (utmp[2] & kmask2) | (((utmp[0] >> 6) & kmask3) << 4);
utmp[0] &= kmask1;
const int16x8_t mins = vreinterpretq_s16_u16(vmovl_u8(vreinterpret_u8_u32(mins8)));
const int32x4_t prod = vaddq_s32(vmull_s16(vget_low_s16 (q8sums), vget_low_s16 (mins)),
vmull_s16(vget_high_s16(q8sums), vget_high_s16(mins)));
sumf -= dmin * vaddvq_s32(prod);
const uint8_t * scales = (const uint8_t *)utmp;
const uint8_t * restrict q4 = x[i].qs;
const int8_t * restrict q8 = y[i].qs;
int32_t sumi1 = 0;
int32_t sumi2 = 0;
for (int j = 0; j < QK_K/64; ++j) {
ggml : sync (im2col, GPU conv, 32-bit arm compat) (#4060) ggml-ci
2023-11-13 14:55:52 +00:00
const ggml_uint8x16x2_t q4bits = ggml_vld1q_u8_x2(q4); q4 += 32;
k-quants : support for super-block size of 64 (#2001) * k_quants: WIP super-blocks with 64 weights * k_quants: WIP super-blocks with 64 weights Q6_K scalar and AVX2 works * k_quants: WIP super-blocks with 64 weights Q4_K scalar and AVX2 works * k_quants: WIP super-blocks with 64 weights Q2_K scalar and AVX2 works. Q2_K is way too slow (it is actually slower than the scalar implementation) * k_quants: WIP super-blocks with 64 weights Q3_K scalar and AVX2 works. * k_quants: WIP super-blocks with 64 weights Q5_K scalar and AVX2 works, and with that all k_quants are done on AVX2 and scalar * k_quants: WIP super-blocks with 64 weights Q6_K working on CUDA. Cannot make it run quite as gast as with super-blocks with 256 weigths: 8% slower on 4080, 20% slower on the 1660 (but there we fit 1 less layer on the GPU because pf the larger model size), so some fraction of these 20% is due to that, * k_quants: WIP super-blocks with 64 weights Q4_K working on CUDA. ~10% slower on GTX-1660, 16% slower on 4080. * k_quants: WIP super-blocks with 64 weights Q2_K working on CUDA. ~3% slower on GTX-1660, 10% slower on 4080. * k_quants: WIP super-blocks with 64 weights Q3_K working on CUDA. * k_quants: WIP super-blocks with 64 weights Q5_K working on CUDA, and with this CUDA is done. * k_quants: WIP super-blocks with 64 weights Q6_K working on ARM_NEON * k_quants: WIP super-blocks with 64 weights Q4_K working on ARM_NEON, but quite a bit slower than 256 weights * k_quants: WIP super-blocks with 64 weights Q2_K working on ARM_NEON, but quite a bit slower than 256 weights * k_quants: WIP super-blocks with 64 weights Q3_K working on ARM_NEON, but quite a bit slower than 256 weights. * k_quants: WIP super-blocks with 64 weights Q5_K working on ARM_NEON, but quite a bit slower than 256 weights. With that, we have full support for ARM_NEON, although performance is not quite there. * k_quants: WIP super-blocks with 64 weights Slightly more efficient Q3_K and Q5_K * k_quants: WIP super-blocks with 64 weights Another small improvement for Q3_K and Q5_K on ARM_NEON * k_quants: WIP super-blocks with 64 weights Yet another speedup for Q5_K on ARM_NEON. We are now within 10% of the QK_K = 256 version. * k_quants: WIP super-blocks with 64 weights * We are able to pass preprocessor macros to the Metal compiler * Q6_K works and is actually slightly more efficient than the QK_K = 256 version (25.2 ms vs 25.8 ms) * k_quants: WIP super-blocks with 64 weights Q4_K works on Metal and is actually slightly faster than QK_K = 256 (21.95 ms vs 24.0 ms). * k_quants: WIP super-blocks with 64 weights Q2_K works on Metal and is very slightly faster than QK_K = 256 (23.8 ms vs 24.2 ms). * k_quants: WIP super-blocks with 64 weights Q3_K works on Metal and is slightly faster than QK_K = 256 (26.6 ms vs 28.3 ms). * k_quants: WIP super-blocks with 64 weights Q5_K works on Metal and is slightly faster than QK_K = 256 (23.7 ms vs 26.3 ms). * k_quants: call them _K, not _k, also on Metal * k_quants: correctly define QK_K in llama.cpp * Fixed bug in q4_K quantization added with the 64-block addition * Simplify via lambda * k_quants: swicth Q3_K to 4-bit scales when QK_K = 64 Otherwise there isn't much benefit from this quantization type. There is some very slight loss in accuracy, but we reduce size by ~7%. E.g., for OpenLLaMA-3B, Q3_K_S perplexity is 8.6131 with 8-bit scales and 8.6352 with 4-bit, while file size decreases from 1.53G to 1.44G. * k_quants: switch Q4_K to 4-bit scales when QK_K = 64 Here the loss in accuracy is greater than for Q3_K, but the Q4_K points still move further to the left on the perplexity vs size curve. * k_quants: forgot to add the Metal changes in last commit * k_quants: change Q5_K to be type 0 when QK_K = 64 Still needs AVX2 implementation * k_quants: AVX2 implementation for new 64-weight Q5_K * k_quants: 10% faster ARM_NEON Q5_K dot product * k_quants: fixed issue caused by merging with master --------- Co-authored-by: Iwan Kawrakow <iwan.kawrakow@gmail.com>
2023-06-26 16:43:07 +00:00
ggml : sync (im2col, GPU conv, 32-bit arm compat) (#4060) ggml-ci
2023-11-13 14:55:52 +00:00
q8bytes = ggml_vld1q_s8_x2(q8); q8 += 32;
k-quants : support for super-block size of 64 (#2001) * k_quants: WIP super-blocks with 64 weights * k_quants: WIP super-blocks with 64 weights Q6_K scalar and AVX2 works * k_quants: WIP super-blocks with 64 weights Q4_K scalar and AVX2 works * k_quants: WIP super-blocks with 64 weights Q2_K scalar and AVX2 works. Q2_K is way too slow (it is actually slower than the scalar implementation) * k_quants: WIP super-blocks with 64 weights Q3_K scalar and AVX2 works. * k_quants: WIP super-blocks with 64 weights Q5_K scalar and AVX2 works, and with that all k_quants are done on AVX2 and scalar * k_quants: WIP super-blocks with 64 weights Q6_K working on CUDA. Cannot make it run quite as gast as with super-blocks with 256 weigths: 8% slower on 4080, 20% slower on the 1660 (but there we fit 1 less layer on the GPU because pf the larger model size), so some fraction of these 20% is due to that, * k_quants: WIP super-blocks with 64 weights Q4_K working on CUDA. ~10% slower on GTX-1660, 16% slower on 4080. * k_quants: WIP super-blocks with 64 weights Q2_K working on CUDA. ~3% slower on GTX-1660, 10% slower on 4080. * k_quants: WIP super-blocks with 64 weights Q3_K working on CUDA. * k_quants: WIP super-blocks with 64 weights Q5_K working on CUDA, and with this CUDA is done. * k_quants: WIP super-blocks with 64 weights Q6_K working on ARM_NEON * k_quants: WIP super-blocks with 64 weights Q4_K working on ARM_NEON, but quite a bit slower than 256 weights * k_quants: WIP super-blocks with 64 weights Q2_K working on ARM_NEON, but quite a bit slower than 256 weights * k_quants: WIP super-blocks with 64 weights Q3_K working on ARM_NEON, but quite a bit slower than 256 weights. * k_quants: WIP super-blocks with 64 weights Q5_K working on ARM_NEON, but quite a bit slower than 256 weights. With that, we have full support for ARM_NEON, although performance is not quite there. * k_quants: WIP super-blocks with 64 weights Slightly more efficient Q3_K and Q5_K * k_quants: WIP super-blocks with 64 weights Another small improvement for Q3_K and Q5_K on ARM_NEON * k_quants: WIP super-blocks with 64 weights Yet another speedup for Q5_K on ARM_NEON. We are now within 10% of the QK_K = 256 version. * k_quants: WIP super-blocks with 64 weights * We are able to pass preprocessor macros to the Metal compiler * Q6_K works and is actually slightly more efficient than the QK_K = 256 version (25.2 ms vs 25.8 ms) * k_quants: WIP super-blocks with 64 weights Q4_K works on Metal and is actually slightly faster than QK_K = 256 (21.95 ms vs 24.0 ms). * k_quants: WIP super-blocks with 64 weights Q2_K works on Metal and is very slightly faster than QK_K = 256 (23.8 ms vs 24.2 ms). * k_quants: WIP super-blocks with 64 weights Q3_K works on Metal and is slightly faster than QK_K = 256 (26.6 ms vs 28.3 ms). * k_quants: WIP super-blocks with 64 weights Q5_K works on Metal and is slightly faster than QK_K = 256 (23.7 ms vs 26.3 ms). * k_quants: call them _K, not _k, also on Metal * k_quants: correctly define QK_K in llama.cpp * Fixed bug in q4_K quantization added with the 64-block addition * Simplify via lambda * k_quants: swicth Q3_K to 4-bit scales when QK_K = 64 Otherwise there isn't much benefit from this quantization type. There is some very slight loss in accuracy, but we reduce size by ~7%. E.g., for OpenLLaMA-3B, Q3_K_S perplexity is 8.6131 with 8-bit scales and 8.6352 with 4-bit, while file size decreases from 1.53G to 1.44G. * k_quants: switch Q4_K to 4-bit scales when QK_K = 64 Here the loss in accuracy is greater than for Q3_K, but the Q4_K points still move further to the left on the perplexity vs size curve. * k_quants: forgot to add the Metal changes in last commit * k_quants: change Q5_K to be type 0 when QK_K = 64 Still needs AVX2 implementation * k_quants: AVX2 implementation for new 64-weight Q5_K * k_quants: 10% faster ARM_NEON Q5_K dot product * k_quants: fixed issue caused by merging with master --------- Co-authored-by: Iwan Kawrakow <iwan.kawrakow@gmail.com>
2023-06-26 16:43:07 +00:00
q4bytes.val[0] = vreinterpretq_s8_u8(vandq_u8 (q4bits.val[0], m4b));
q4bytes.val[1] = vreinterpretq_s8_u8(vandq_u8 (q4bits.val[1], m4b));
ggml : add ggml_vdotq_s32 alias (#4715) ggml-ci
2023-12-31 09:43:31 +00:00
const int32x4_t p1 = ggml_vdotq_s32(ggml_vdotq_s32(mzero, q4bytes.val[0], q8bytes.val[0]), q4bytes.val[1], q8bytes.val[1]);
k-quants : support for super-block size of 64 (#2001) * k_quants: WIP super-blocks with 64 weights * k_quants: WIP super-blocks with 64 weights Q6_K scalar and AVX2 works * k_quants: WIP super-blocks with 64 weights Q4_K scalar and AVX2 works * k_quants: WIP super-blocks with 64 weights Q2_K scalar and AVX2 works. Q2_K is way too slow (it is actually slower than the scalar implementation) * k_quants: WIP super-blocks with 64 weights Q3_K scalar and AVX2 works. * k_quants: WIP super-blocks with 64 weights Q5_K scalar and AVX2 works, and with that all k_quants are done on AVX2 and scalar * k_quants: WIP super-blocks with 64 weights Q6_K working on CUDA. Cannot make it run quite as gast as with super-blocks with 256 weigths: 8% slower on 4080, 20% slower on the 1660 (but there we fit 1 less layer on the GPU because pf the larger model size), so some fraction of these 20% is due to that, * k_quants: WIP super-blocks with 64 weights Q4_K working on CUDA. ~10% slower on GTX-1660, 16% slower on 4080. * k_quants: WIP super-blocks with 64 weights Q2_K working on CUDA. ~3% slower on GTX-1660, 10% slower on 4080. * k_quants: WIP super-blocks with 64 weights Q3_K working on CUDA. * k_quants: WIP super-blocks with 64 weights Q5_K working on CUDA, and with this CUDA is done. * k_quants: WIP super-blocks with 64 weights Q6_K working on ARM_NEON * k_quants: WIP super-blocks with 64 weights Q4_K working on ARM_NEON, but quite a bit slower than 256 weights * k_quants: WIP super-blocks with 64 weights Q2_K working on ARM_NEON, but quite a bit slower than 256 weights * k_quants: WIP super-blocks with 64 weights Q3_K working on ARM_NEON, but quite a bit slower than 256 weights. * k_quants: WIP super-blocks with 64 weights Q5_K working on ARM_NEON, but quite a bit slower than 256 weights. With that, we have full support for ARM_NEON, although performance is not quite there. * k_quants: WIP super-blocks with 64 weights Slightly more efficient Q3_K and Q5_K * k_quants: WIP super-blocks with 64 weights Another small improvement for Q3_K and Q5_K on ARM_NEON * k_quants: WIP super-blocks with 64 weights Yet another speedup for Q5_K on ARM_NEON. We are now within 10% of the QK_K = 256 version. * k_quants: WIP super-blocks with 64 weights * We are able to pass preprocessor macros to the Metal compiler * Q6_K works and is actually slightly more efficient than the QK_K = 256 version (25.2 ms vs 25.8 ms) * k_quants: WIP super-blocks with 64 weights Q4_K works on Metal and is actually slightly faster than QK_K = 256 (21.95 ms vs 24.0 ms). * k_quants: WIP super-blocks with 64 weights Q2_K works on Metal and is very slightly faster than QK_K = 256 (23.8 ms vs 24.2 ms). * k_quants: WIP super-blocks with 64 weights Q3_K works on Metal and is slightly faster than QK_K = 256 (26.6 ms vs 28.3 ms). * k_quants: WIP super-blocks with 64 weights Q5_K works on Metal and is slightly faster than QK_K = 256 (23.7 ms vs 26.3 ms). * k_quants: call them _K, not _k, also on Metal * k_quants: correctly define QK_K in llama.cpp * Fixed bug in q4_K quantization added with the 64-block addition * Simplify via lambda * k_quants: swicth Q3_K to 4-bit scales when QK_K = 64 Otherwise there isn't much benefit from this quantization type. There is some very slight loss in accuracy, but we reduce size by ~7%. E.g., for OpenLLaMA-3B, Q3_K_S perplexity is 8.6131 with 8-bit scales and 8.6352 with 4-bit, while file size decreases from 1.53G to 1.44G. * k_quants: switch Q4_K to 4-bit scales when QK_K = 64 Here the loss in accuracy is greater than for Q3_K, but the Q4_K points still move further to the left on the perplexity vs size curve. * k_quants: forgot to add the Metal changes in last commit * k_quants: change Q5_K to be type 0 when QK_K = 64 Still needs AVX2 implementation * k_quants: AVX2 implementation for new 64-weight Q5_K * k_quants: 10% faster ARM_NEON Q5_K dot product * k_quants: fixed issue caused by merging with master --------- Co-authored-by: Iwan Kawrakow <iwan.kawrakow@gmail.com>
2023-06-26 16:43:07 +00:00
sumi1 += vaddvq_s32(p1) * scales[2*j+0];
ggml : add SOTA 2,3,4,5,6 bit k-quantizations (#1684) * Starting to add k-quantization to ggml I think it is better to have quantization separate from ggml. For now just adding the k-quants there, but it would be better to also factor out the existing ggml quantizations. * Adding Q3_K and Q8_K (de)-quantization * Q3_K now working on CUDA and AVX2/scalar CUDA is not ideal - ~50% slower than Q4_0 for single token prediction, about the same in batch mode (perplexity). CPU single token is ~55 ms (on Ryzen 7950X). * Some improvement for Q3_K on CUDA It is now ~22.5 ms/token on my GPU, so ~30% slower than Q4_0. * Some more CUDA optimizations for Q3_K Single token is now 20.5 ms/token (~20% slower than Q4_0). Perplexity is on par with Q4_0. * Adding Q4_K - scalar, AVX2, CUDA Performance is the same or perhaps very slightly better than Q4_0 on the CPU. On the GPU, single token prediction is ~10% better than Q4_0, batch mode (perplexity is about the same). * Adding Q6_K - scalar, AVX2, CUDA Performance is ~40% lower compared to Q4_K on the CPU. This is to be expected, considering that we are memory bound on the CPU and the 6-bit model is ~44% larger than the 4-bit. On the GPU, single token prediction is ~6% lower than Q4_0, batch mode (perplexity) is even closer (but still slower). * Adding Q5_K - scalar, AVX2, CUDA Performance is ~20% lower compared to Q4_K on the CPU. This is to be expected, considering that we are memory bound on the CPU and the 5-bit model is ~22% larger than the 4-bit. On the GPU, single token prediction is about the same as Q4_0 for both, single token and batch prediction. * Per convention, all QX_K quantizations use Q5_K for output.weight * Adding quantization mixes * Quantization mixes: didn't quite get what I wanted in the last commit * Q4_K dot product for ARM_NEON * Q6_K dot product for ARM_NEON * Q5_K dot product for ARM_NEON * Adding Q3_K dot for ARM_NEON It is 22% slower than Q4_K, despite the smaller model size. On x86_64, where we are memory bound, the Q3_K model is quite a bit faster than Q4_K. * A very slightly faster ARM_NEON Q3_K dot * Adding Q2_K - just CUDA for now Token prediction is pretty good - about 15.5 ms on a RTX 4080. Perplexity is about the same as Q4_K. * Adding scalar and AVX2 Q2_K dot * Adding ARM_NEON Q2_K dot About the same performance as Q4_K. * A slightly faster ARM_NEON Q2_K dot Single token prediction is now ~36 ms on M2 Max. The code is much simpler too. * Fixed bug in Q2_K CUDA dot product kernel Stranegly enough, for the few prompts I tried with the 7B model the responses looked perfectly reasonable. Only realized something is not quite right when I tried the larger models and started getting nonse back. In any case, Q2_K single token evaluation time on an RTX 4080 in a Ryzen7950X box iusing CUDA and model fully loaded on the GPU are ~15.5 ms for 7B, ~25.4 ms for 13B, and ~55.8 ms for 30B. The max number of layers that fit in VRAM for The 65B is 32. With that, we get ~330 ms per token, which is not that much faster than just running on the CPU (~470 ms per token). * Don't print zeros/NaNs when no count histogram has been collected * A 10% faster CUDA vector dot kernel for Q3_K Q3_K is now running at ~18.5 ms / token on CUDA, so the gap to Q4_0 is only 10%. It seems memory acccess pattern is more important for performance than the amount of computation the kernel does. * A slightly daster Q4_K AVX2 dot product For perplexity, where we are less memory bound, time per pass drops by ~5%. Barely measurable difference for single token prediction. * A slightly faster ARM_NEON A4_K dot product * Minor * Fix quantization error test We cannot possibly be expecting rmse < 0.002 for 2- and 3-bit quantization variants. * Fix docker build I have been sloppy with vector reinterpret casts on ARM_NEON. It seems clang is very forgiving in that regard. * Added forgotten ggml.o dependence on k_quants.h to the Makefile * Had unintentionally committed the Makefile with -Ofast enabled * ggml : rename k_quants -> ggml-quants-k, use lowercase in code --------- Co-authored-by: Iwan Kawrakow <iwan.kawrakow@gmail.com> Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
2023-06-05 19:56:18 +00:00
ggml : sync (im2col, GPU conv, 32-bit arm compat) (#4060) ggml-ci
2023-11-13 14:55:52 +00:00
q8bytes = ggml_vld1q_s8_x2(q8); q8 += 32;
ggml : add SOTA 2,3,4,5,6 bit k-quantizations (#1684) * Starting to add k-quantization to ggml I think it is better to have quantization separate from ggml. For now just adding the k-quants there, but it would be better to also factor out the existing ggml quantizations. * Adding Q3_K and Q8_K (de)-quantization * Q3_K now working on CUDA and AVX2/scalar CUDA is not ideal - ~50% slower than Q4_0 for single token prediction, about the same in batch mode (perplexity). CPU single token is ~55 ms (on Ryzen 7950X). * Some improvement for Q3_K on CUDA It is now ~22.5 ms/token on my GPU, so ~30% slower than Q4_0. * Some more CUDA optimizations for Q3_K Single token is now 20.5 ms/token (~20% slower than Q4_0). Perplexity is on par with Q4_0. * Adding Q4_K - scalar, AVX2, CUDA Performance is the same or perhaps very slightly better than Q4_0 on the CPU. On the GPU, single token prediction is ~10% better than Q4_0, batch mode (perplexity is about the same). * Adding Q6_K - scalar, AVX2, CUDA Performance is ~40% lower compared to Q4_K on the CPU. This is to be expected, considering that we are memory bound on the CPU and the 6-bit model is ~44% larger than the 4-bit. On the GPU, single token prediction is ~6% lower than Q4_0, batch mode (perplexity) is even closer (but still slower). * Adding Q5_K - scalar, AVX2, CUDA Performance is ~20% lower compared to Q4_K on the CPU. This is to be expected, considering that we are memory bound on the CPU and the 5-bit model is ~22% larger than the 4-bit. On the GPU, single token prediction is about the same as Q4_0 for both, single token and batch prediction. * Per convention, all QX_K quantizations use Q5_K for output.weight * Adding quantization mixes * Quantization mixes: didn't quite get what I wanted in the last commit * Q4_K dot product for ARM_NEON * Q6_K dot product for ARM_NEON * Q5_K dot product for ARM_NEON * Adding Q3_K dot for ARM_NEON It is 22% slower than Q4_K, despite the smaller model size. On x86_64, where we are memory bound, the Q3_K model is quite a bit faster than Q4_K. * A very slightly faster ARM_NEON Q3_K dot * Adding Q2_K - just CUDA for now Token prediction is pretty good - about 15.5 ms on a RTX 4080. Perplexity is about the same as Q4_K. * Adding scalar and AVX2 Q2_K dot * Adding ARM_NEON Q2_K dot About the same performance as Q4_K. * A slightly faster ARM_NEON Q2_K dot Single token prediction is now ~36 ms on M2 Max. The code is much simpler too. * Fixed bug in Q2_K CUDA dot product kernel Stranegly enough, for the few prompts I tried with the 7B model the responses looked perfectly reasonable. Only realized something is not quite right when I tried the larger models and started getting nonse back. In any case, Q2_K single token evaluation time on an RTX 4080 in a Ryzen7950X box iusing CUDA and model fully loaded on the GPU are ~15.5 ms for 7B, ~25.4 ms for 13B, and ~55.8 ms for 30B. The max number of layers that fit in VRAM for The 65B is 32. With that, we get ~330 ms per token, which is not that much faster than just running on the CPU (~470 ms per token). * Don't print zeros/NaNs when no count histogram has been collected * A 10% faster CUDA vector dot kernel for Q3_K Q3_K is now running at ~18.5 ms / token on CUDA, so the gap to Q4_0 is only 10%. It seems memory acccess pattern is more important for performance than the amount of computation the kernel does. * A slightly daster Q4_K AVX2 dot product For perplexity, where we are less memory bound, time per pass drops by ~5%. Barely measurable difference for single token prediction. * A slightly faster ARM_NEON A4_K dot product * Minor * Fix quantization error test We cannot possibly be expecting rmse < 0.002 for 2- and 3-bit quantization variants. * Fix docker build I have been sloppy with vector reinterpret casts on ARM_NEON. It seems clang is very forgiving in that regard. * Added forgotten ggml.o dependence on k_quants.h to the Makefile * Had unintentionally committed the Makefile with -Ofast enabled * ggml : rename k_quants -> ggml-quants-k, use lowercase in code --------- Co-authored-by: Iwan Kawrakow <iwan.kawrakow@gmail.com> Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
2023-06-05 19:56:18 +00:00
q4bytes.val[0] = vreinterpretq_s8_u8(vshrq_n_u8(q4bits.val[0], 4));
q4bytes.val[1] = vreinterpretq_s8_u8(vshrq_n_u8(q4bits.val[1], 4));
ggml : add ggml_vdotq_s32 alias (#4715) ggml-ci
2023-12-31 09:43:31 +00:00
const int32x4_t p2 = ggml_vdotq_s32(ggml_vdotq_s32(mzero, q4bytes.val[0], q8bytes.val[0]), q4bytes.val[1], q8bytes.val[1]);
ggml : add SOTA 2,3,4,5,6 bit k-quantizations (#1684) * Starting to add k-quantization to ggml I think it is better to have quantization separate from ggml. For now just adding the k-quants there, but it would be better to also factor out the existing ggml quantizations. * Adding Q3_K and Q8_K (de)-quantization * Q3_K now working on CUDA and AVX2/scalar CUDA is not ideal - ~50% slower than Q4_0 for single token prediction, about the same in batch mode (perplexity). CPU single token is ~55 ms (on Ryzen 7950X). * Some improvement for Q3_K on CUDA It is now ~22.5 ms/token on my GPU, so ~30% slower than Q4_0. * Some more CUDA optimizations for Q3_K Single token is now 20.5 ms/token (~20% slower than Q4_0). Perplexity is on par with Q4_0. * Adding Q4_K - scalar, AVX2, CUDA Performance is the same or perhaps very slightly better than Q4_0 on the CPU. On the GPU, single token prediction is ~10% better than Q4_0, batch mode (perplexity is about the same). * Adding Q6_K - scalar, AVX2, CUDA Performance is ~40% lower compared to Q4_K on the CPU. This is to be expected, considering that we are memory bound on the CPU and the 6-bit model is ~44% larger than the 4-bit. On the GPU, single token prediction is ~6% lower than Q4_0, batch mode (perplexity) is even closer (but still slower). * Adding Q5_K - scalar, AVX2, CUDA Performance is ~20% lower compared to Q4_K on the CPU. This is to be expected, considering that we are memory bound on the CPU and the 5-bit model is ~22% larger than the 4-bit. On the GPU, single token prediction is about the same as Q4_0 for both, single token and batch prediction. * Per convention, all QX_K quantizations use Q5_K for output.weight * Adding quantization mixes * Quantization mixes: didn't quite get what I wanted in the last commit * Q4_K dot product for ARM_NEON * Q6_K dot product for ARM_NEON * Q5_K dot product for ARM_NEON * Adding Q3_K dot for ARM_NEON It is 22% slower than Q4_K, despite the smaller model size. On x86_64, where we are memory bound, the Q3_K model is quite a bit faster than Q4_K. * A very slightly faster ARM_NEON Q3_K dot * Adding Q2_K - just CUDA for now Token prediction is pretty good - about 15.5 ms on a RTX 4080. Perplexity is about the same as Q4_K. * Adding scalar and AVX2 Q2_K dot * Adding ARM_NEON Q2_K dot About the same performance as Q4_K. * A slightly faster ARM_NEON Q2_K dot Single token prediction is now ~36 ms on M2 Max. The code is much simpler too. * Fixed bug in Q2_K CUDA dot product kernel Stranegly enough, for the few prompts I tried with the 7B model the responses looked perfectly reasonable. Only realized something is not quite right when I tried the larger models and started getting nonse back. In any case, Q2_K single token evaluation time on an RTX 4080 in a Ryzen7950X box iusing CUDA and model fully loaded on the GPU are ~15.5 ms for 7B, ~25.4 ms for 13B, and ~55.8 ms for 30B. The max number of layers that fit in VRAM for The 65B is 32. With that, we get ~330 ms per token, which is not that much faster than just running on the CPU (~470 ms per token). * Don't print zeros/NaNs when no count histogram has been collected * A 10% faster CUDA vector dot kernel for Q3_K Q3_K is now running at ~18.5 ms / token on CUDA, so the gap to Q4_0 is only 10%. It seems memory acccess pattern is more important for performance than the amount of computation the kernel does. * A slightly daster Q4_K AVX2 dot product For perplexity, where we are less memory bound, time per pass drops by ~5%. Barely measurable difference for single token prediction. * A slightly faster ARM_NEON A4_K dot product * Minor * Fix quantization error test We cannot possibly be expecting rmse < 0.002 for 2- and 3-bit quantization variants. * Fix docker build I have been sloppy with vector reinterpret casts on ARM_NEON. It seems clang is very forgiving in that regard. * Added forgotten ggml.o dependence on k_quants.h to the Makefile * Had unintentionally committed the Makefile with -Ofast enabled * ggml : rename k_quants -> ggml-quants-k, use lowercase in code --------- Co-authored-by: Iwan Kawrakow <iwan.kawrakow@gmail.com> Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
2023-06-05 19:56:18 +00:00
sumi2 += vaddvq_s32(p2) * scales[2*j+1];
}
sumf += d * (sumi1 + sumi2);
}
*s = sumf;
#elif defined __AVX2__
const __m256i m4 = _mm256_set1_epi8(0xF);
__m256 acc = _mm256_setzero_ps();
__m128 acc_m = _mm_setzero_ps();
for (int i = 0; i < nb; ++i) {
ggml : move FP16 <-> FP32 code to ggml-impl.h (#3861) * ggml : move FP16 <-> FP32 stuff to ggml-impl.h ggml-ci * tests : fix ARM build * ggml : explicitly initialize deprecated type traits * ggml : add math.h to ggml-impl.h * ggml : remove duplicate static assert macros * ggml : prefix lookup tables with ggml_ ggml-ci * ggml-impl : move extern "C" to start of file
2023-10-30 17:19:15 +00:00
const float d = y[i].d * GGML_FP16_TO_FP32(x[i].d);
const float dmin = -y[i].d * GGML_FP16_TO_FP32(x[i].dmin);
ggml : add SOTA 2,3,4,5,6 bit k-quantizations (#1684) * Starting to add k-quantization to ggml I think it is better to have quantization separate from ggml. For now just adding the k-quants there, but it would be better to also factor out the existing ggml quantizations. * Adding Q3_K and Q8_K (de)-quantization * Q3_K now working on CUDA and AVX2/scalar CUDA is not ideal - ~50% slower than Q4_0 for single token prediction, about the same in batch mode (perplexity). CPU single token is ~55 ms (on Ryzen 7950X). * Some improvement for Q3_K on CUDA It is now ~22.5 ms/token on my GPU, so ~30% slower than Q4_0. * Some more CUDA optimizations for Q3_K Single token is now 20.5 ms/token (~20% slower than Q4_0). Perplexity is on par with Q4_0. * Adding Q4_K - scalar, AVX2, CUDA Performance is the same or perhaps very slightly better than Q4_0 on the CPU. On the GPU, single token prediction is ~10% better than Q4_0, batch mode (perplexity is about the same). * Adding Q6_K - scalar, AVX2, CUDA Performance is ~40% lower compared to Q4_K on the CPU. This is to be expected, considering that we are memory bound on the CPU and the 6-bit model is ~44% larger than the 4-bit. On the GPU, single token prediction is ~6% lower than Q4_0, batch mode (perplexity) is even closer (but still slower). * Adding Q5_K - scalar, AVX2, CUDA Performance is ~20% lower compared to Q4_K on the CPU. This is to be expected, considering that we are memory bound on the CPU and the 5-bit model is ~22% larger than the 4-bit. On the GPU, single token prediction is about the same as Q4_0 for both, single token and batch prediction. * Per convention, all QX_K quantizations use Q5_K for output.weight * Adding quantization mixes * Quantization mixes: didn't quite get what I wanted in the last commit * Q4_K dot product for ARM_NEON * Q6_K dot product for ARM_NEON * Q5_K dot product for ARM_NEON * Adding Q3_K dot for ARM_NEON It is 22% slower than Q4_K, despite the smaller model size. On x86_64, where we are memory bound, the Q3_K model is quite a bit faster than Q4_K. * A very slightly faster ARM_NEON Q3_K dot * Adding Q2_K - just CUDA for now Token prediction is pretty good - about 15.5 ms on a RTX 4080. Perplexity is about the same as Q4_K. * Adding scalar and AVX2 Q2_K dot * Adding ARM_NEON Q2_K dot About the same performance as Q4_K. * A slightly faster ARM_NEON Q2_K dot Single token prediction is now ~36 ms on M2 Max. The code is much simpler too. * Fixed bug in Q2_K CUDA dot product kernel Stranegly enough, for the few prompts I tried with the 7B model the responses looked perfectly reasonable. Only realized something is not quite right when I tried the larger models and started getting nonse back. In any case, Q2_K single token evaluation time on an RTX 4080 in a Ryzen7950X box iusing CUDA and model fully loaded on the GPU are ~15.5 ms for 7B, ~25.4 ms for 13B, and ~55.8 ms for 30B. The max number of layers that fit in VRAM for The 65B is 32. With that, we get ~330 ms per token, which is not that much faster than just running on the CPU (~470 ms per token). * Don't print zeros/NaNs when no count histogram has been collected * A 10% faster CUDA vector dot kernel for Q3_K Q3_K is now running at ~18.5 ms / token on CUDA, so the gap to Q4_0 is only 10%. It seems memory acccess pattern is more important for performance than the amount of computation the kernel does. * A slightly daster Q4_K AVX2 dot product For perplexity, where we are less memory bound, time per pass drops by ~5%. Barely measurable difference for single token prediction. * A slightly faster ARM_NEON A4_K dot product * Minor * Fix quantization error test We cannot possibly be expecting rmse < 0.002 for 2- and 3-bit quantization variants. * Fix docker build I have been sloppy with vector reinterpret casts on ARM_NEON. It seems clang is very forgiving in that regard. * Added forgotten ggml.o dependence on k_quants.h to the Makefile * Had unintentionally committed the Makefile with -Ofast enabled * ggml : rename k_quants -> ggml-quants-k, use lowercase in code --------- Co-authored-by: Iwan Kawrakow <iwan.kawrakow@gmail.com> Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
2023-06-05 19:56:18 +00:00
memcpy(utmp, x[i].scales, 12);
utmp[3] = ((utmp[2] >> 4) & kmask2) | (((utmp[1] >> 6) & kmask3) << 4);
const uint32_t uaux = utmp[1] & kmask1;
utmp[1] = (utmp[2] & kmask2) | (((utmp[0] >> 6) & kmask3) << 4);
utmp[2] = uaux;
utmp[0] &= kmask1;
k-quants : support for super-block size of 64 (#2001) * k_quants: WIP super-blocks with 64 weights * k_quants: WIP super-blocks with 64 weights Q6_K scalar and AVX2 works * k_quants: WIP super-blocks with 64 weights Q4_K scalar and AVX2 works * k_quants: WIP super-blocks with 64 weights Q2_K scalar and AVX2 works. Q2_K is way too slow (it is actually slower than the scalar implementation) * k_quants: WIP super-blocks with 64 weights Q3_K scalar and AVX2 works. * k_quants: WIP super-blocks with 64 weights Q5_K scalar and AVX2 works, and with that all k_quants are done on AVX2 and scalar * k_quants: WIP super-blocks with 64 weights Q6_K working on CUDA. Cannot make it run quite as gast as with super-blocks with 256 weigths: 8% slower on 4080, 20% slower on the 1660 (but there we fit 1 less layer on the GPU because pf the larger model size), so some fraction of these 20% is due to that, * k_quants: WIP super-blocks with 64 weights Q4_K working on CUDA. ~10% slower on GTX-1660, 16% slower on 4080. * k_quants: WIP super-blocks with 64 weights Q2_K working on CUDA. ~3% slower on GTX-1660, 10% slower on 4080. * k_quants: WIP super-blocks with 64 weights Q3_K working on CUDA. * k_quants: WIP super-blocks with 64 weights Q5_K working on CUDA, and with this CUDA is done. * k_quants: WIP super-blocks with 64 weights Q6_K working on ARM_NEON * k_quants: WIP super-blocks with 64 weights Q4_K working on ARM_NEON, but quite a bit slower than 256 weights * k_quants: WIP super-blocks with 64 weights Q2_K working on ARM_NEON, but quite a bit slower than 256 weights * k_quants: WIP super-blocks with 64 weights Q3_K working on ARM_NEON, but quite a bit slower than 256 weights. * k_quants: WIP super-blocks with 64 weights Q5_K working on ARM_NEON, but quite a bit slower than 256 weights. With that, we have full support for ARM_NEON, although performance is not quite there. * k_quants: WIP super-blocks with 64 weights Slightly more efficient Q3_K and Q5_K * k_quants: WIP super-blocks with 64 weights Another small improvement for Q3_K and Q5_K on ARM_NEON * k_quants: WIP super-blocks with 64 weights Yet another speedup for Q5_K on ARM_NEON. We are now within 10% of the QK_K = 256 version. * k_quants: WIP super-blocks with 64 weights * We are able to pass preprocessor macros to the Metal compiler * Q6_K works and is actually slightly more efficient than the QK_K = 256 version (25.2 ms vs 25.8 ms) * k_quants: WIP super-blocks with 64 weights Q4_K works on Metal and is actually slightly faster than QK_K = 256 (21.95 ms vs 24.0 ms). * k_quants: WIP super-blocks with 64 weights Q2_K works on Metal and is very slightly faster than QK_K = 256 (23.8 ms vs 24.2 ms). * k_quants: WIP super-blocks with 64 weights Q3_K works on Metal and is slightly faster than QK_K = 256 (26.6 ms vs 28.3 ms). * k_quants: WIP super-blocks with 64 weights Q5_K works on Metal and is slightly faster than QK_K = 256 (23.7 ms vs 26.3 ms). * k_quants: call them _K, not _k, also on Metal * k_quants: correctly define QK_K in llama.cpp * Fixed bug in q4_K quantization added with the 64-block addition * Simplify via lambda * k_quants: swicth Q3_K to 4-bit scales when QK_K = 64 Otherwise there isn't much benefit from this quantization type. There is some very slight loss in accuracy, but we reduce size by ~7%. E.g., for OpenLLaMA-3B, Q3_K_S perplexity is 8.6131 with 8-bit scales and 8.6352 with 4-bit, while file size decreases from 1.53G to 1.44G. * k_quants: switch Q4_K to 4-bit scales when QK_K = 64 Here the loss in accuracy is greater than for Q3_K, but the Q4_K points still move further to the left on the perplexity vs size curve. * k_quants: forgot to add the Metal changes in last commit * k_quants: change Q5_K to be type 0 when QK_K = 64 Still needs AVX2 implementation * k_quants: AVX2 implementation for new 64-weight Q5_K * k_quants: 10% faster ARM_NEON Q5_K dot product * k_quants: fixed issue caused by merging with master --------- Co-authored-by: Iwan Kawrakow <iwan.kawrakow@gmail.com>
2023-06-26 16:43:07 +00:00
const uint8_t * restrict q4 = x[i].qs;
const int8_t * restrict q8 = y[i].qs;
ggml : add SOTA 2,3,4,5,6 bit k-quantizations (#1684) * Starting to add k-quantization to ggml I think it is better to have quantization separate from ggml. For now just adding the k-quants there, but it would be better to also factor out the existing ggml quantizations. * Adding Q3_K and Q8_K (de)-quantization * Q3_K now working on CUDA and AVX2/scalar CUDA is not ideal - ~50% slower than Q4_0 for single token prediction, about the same in batch mode (perplexity). CPU single token is ~55 ms (on Ryzen 7950X). * Some improvement for Q3_K on CUDA It is now ~22.5 ms/token on my GPU, so ~30% slower than Q4_0. * Some more CUDA optimizations for Q3_K Single token is now 20.5 ms/token (~20% slower than Q4_0). Perplexity is on par with Q4_0. * Adding Q4_K - scalar, AVX2, CUDA Performance is the same or perhaps very slightly better than Q4_0 on the CPU. On the GPU, single token prediction is ~10% better than Q4_0, batch mode (perplexity is about the same). * Adding Q6_K - scalar, AVX2, CUDA Performance is ~40% lower compared to Q4_K on the CPU. This is to be expected, considering that we are memory bound on the CPU and the 6-bit model is ~44% larger than the 4-bit. On the GPU, single token prediction is ~6% lower than Q4_0, batch mode (perplexity) is even closer (but still slower). * Adding Q5_K - scalar, AVX2, CUDA Performance is ~20% lower compared to Q4_K on the CPU. This is to be expected, considering that we are memory bound on the CPU and the 5-bit model is ~22% larger than the 4-bit. On the GPU, single token prediction is about the same as Q4_0 for both, single token and batch prediction. * Per convention, all QX_K quantizations use Q5_K for output.weight * Adding quantization mixes * Quantization mixes: didn't quite get what I wanted in the last commit * Q4_K dot product for ARM_NEON * Q6_K dot product for ARM_NEON * Q5_K dot product for ARM_NEON * Adding Q3_K dot for ARM_NEON It is 22% slower than Q4_K, despite the smaller model size. On x86_64, where we are memory bound, the Q3_K model is quite a bit faster than Q4_K. * A very slightly faster ARM_NEON Q3_K dot * Adding Q2_K - just CUDA for now Token prediction is pretty good - about 15.5 ms on a RTX 4080. Perplexity is about the same as Q4_K. * Adding scalar and AVX2 Q2_K dot * Adding ARM_NEON Q2_K dot About the same performance as Q4_K. * A slightly faster ARM_NEON Q2_K dot Single token prediction is now ~36 ms on M2 Max. The code is much simpler too. * Fixed bug in Q2_K CUDA dot product kernel Stranegly enough, for the few prompts I tried with the 7B model the responses looked perfectly reasonable. Only realized something is not quite right when I tried the larger models and started getting nonse back. In any case, Q2_K single token evaluation time on an RTX 4080 in a Ryzen7950X box iusing CUDA and model fully loaded on the GPU are ~15.5 ms for 7B, ~25.4 ms for 13B, and ~55.8 ms for 30B. The max number of layers that fit in VRAM for The 65B is 32. With that, we get ~330 ms per token, which is not that much faster than just running on the CPU (~470 ms per token). * Don't print zeros/NaNs when no count histogram has been collected * A 10% faster CUDA vector dot kernel for Q3_K Q3_K is now running at ~18.5 ms / token on CUDA, so the gap to Q4_0 is only 10%. It seems memory acccess pattern is more important for performance than the amount of computation the kernel does. * A slightly daster Q4_K AVX2 dot product For perplexity, where we are less memory bound, time per pass drops by ~5%. Barely measurable difference for single token prediction. * A slightly faster ARM_NEON A4_K dot product * Minor * Fix quantization error test We cannot possibly be expecting rmse < 0.002 for 2- and 3-bit quantization variants. * Fix docker build I have been sloppy with vector reinterpret casts on ARM_NEON. It seems clang is very forgiving in that regard. * Added forgotten ggml.o dependence on k_quants.h to the Makefile * Had unintentionally committed the Makefile with -Ofast enabled * ggml : rename k_quants -> ggml-quants-k, use lowercase in code --------- Co-authored-by: Iwan Kawrakow <iwan.kawrakow@gmail.com> Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
2023-06-05 19:56:18 +00:00
const __m256i mins_and_scales = _mm256_cvtepu8_epi16(_mm_set_epi32(utmp[3], utmp[2], utmp[1], utmp[0]));
const __m256i q8sums = _mm256_loadu_si256((const __m256i*)y[i].bsums);
const __m128i q8s = _mm_hadd_epi16(_mm256_extracti128_si256(q8sums, 0), _mm256_extracti128_si256(q8sums, 1));
const __m128i prod = _mm_madd_epi16(_mm256_extracti128_si256(mins_and_scales, 1), q8s);
acc_m = _mm_fmadd_ps(_mm_set1_ps(dmin), _mm_cvtepi32_ps(prod), acc_m);
const __m128i sc128 = _mm256_extracti128_si256(mins_and_scales, 0);
ggml : workaround for missing _mm256_setr_m128i in GCC < 8 in k_quants.c (#2405)
2023-07-28 18:17:45 +00:00
const __m256i scales = MM256_SET_M128I(sc128, sc128);
ggml : add SOTA 2,3,4,5,6 bit k-quantizations (#1684) * Starting to add k-quantization to ggml I think it is better to have quantization separate from ggml. For now just adding the k-quants there, but it would be better to also factor out the existing ggml quantizations. * Adding Q3_K and Q8_K (de)-quantization * Q3_K now working on CUDA and AVX2/scalar CUDA is not ideal - ~50% slower than Q4_0 for single token prediction, about the same in batch mode (perplexity). CPU single token is ~55 ms (on Ryzen 7950X). * Some improvement for Q3_K on CUDA It is now ~22.5 ms/token on my GPU, so ~30% slower than Q4_0. * Some more CUDA optimizations for Q3_K Single token is now 20.5 ms/token (~20% slower than Q4_0). Perplexity is on par with Q4_0. * Adding Q4_K - scalar, AVX2, CUDA Performance is the same or perhaps very slightly better than Q4_0 on the CPU. On the GPU, single token prediction is ~10% better than Q4_0, batch mode (perplexity is about the same). * Adding Q6_K - scalar, AVX2, CUDA Performance is ~40% lower compared to Q4_K on the CPU. This is to be expected, considering that we are memory bound on the CPU and the 6-bit model is ~44% larger than the 4-bit. On the GPU, single token prediction is ~6% lower than Q4_0, batch mode (perplexity) is even closer (but still slower). * Adding Q5_K - scalar, AVX2, CUDA Performance is ~20% lower compared to Q4_K on the CPU. This is to be expected, considering that we are memory bound on the CPU and the 5-bit model is ~22% larger than the 4-bit. On the GPU, single token prediction is about the same as Q4_0 for both, single token and batch prediction. * Per convention, all QX_K quantizations use Q5_K for output.weight * Adding quantization mixes * Quantization mixes: didn't quite get what I wanted in the last commit * Q4_K dot product for ARM_NEON * Q6_K dot product for ARM_NEON * Q5_K dot product for ARM_NEON * Adding Q3_K dot for ARM_NEON It is 22% slower than Q4_K, despite the smaller model size. On x86_64, where we are memory bound, the Q3_K model is quite a bit faster than Q4_K. * A very slightly faster ARM_NEON Q3_K dot * Adding Q2_K - just CUDA for now Token prediction is pretty good - about 15.5 ms on a RTX 4080. Perplexity is about the same as Q4_K. * Adding scalar and AVX2 Q2_K dot * Adding ARM_NEON Q2_K dot About the same performance as Q4_K. * A slightly faster ARM_NEON Q2_K dot Single token prediction is now ~36 ms on M2 Max. The code is much simpler too. * Fixed bug in Q2_K CUDA dot product kernel Stranegly enough, for the few prompts I tried with the 7B model the responses looked perfectly reasonable. Only realized something is not quite right when I tried the larger models and started getting nonse back. In any case, Q2_K single token evaluation time on an RTX 4080 in a Ryzen7950X box iusing CUDA and model fully loaded on the GPU are ~15.5 ms for 7B, ~25.4 ms for 13B, and ~55.8 ms for 30B. The max number of layers that fit in VRAM for The 65B is 32. With that, we get ~330 ms per token, which is not that much faster than just running on the CPU (~470 ms per token). * Don't print zeros/NaNs when no count histogram has been collected * A 10% faster CUDA vector dot kernel for Q3_K Q3_K is now running at ~18.5 ms / token on CUDA, so the gap to Q4_0 is only 10%. It seems memory acccess pattern is more important for performance than the amount of computation the kernel does. * A slightly daster Q4_K AVX2 dot product For perplexity, where we are less memory bound, time per pass drops by ~5%. Barely measurable difference for single token prediction. * A slightly faster ARM_NEON A4_K dot product * Minor * Fix quantization error test We cannot possibly be expecting rmse < 0.002 for 2- and 3-bit quantization variants. * Fix docker build I have been sloppy with vector reinterpret casts on ARM_NEON. It seems clang is very forgiving in that regard. * Added forgotten ggml.o dependence on k_quants.h to the Makefile * Had unintentionally committed the Makefile with -Ofast enabled * ggml : rename k_quants -> ggml-quants-k, use lowercase in code --------- Co-authored-by: Iwan Kawrakow <iwan.kawrakow@gmail.com> Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
2023-06-05 19:56:18 +00:00
__m256i sumi = _mm256_setzero_si256();
for (int j = 0; j < QK_K/64; ++j) {
const __m256i scale_l = _mm256_shuffle_epi8(scales, get_scale_shuffle_k4(2*j+0));
const __m256i scale_h = _mm256_shuffle_epi8(scales, get_scale_shuffle_k4(2*j+1));
const __m256i q4bits = _mm256_loadu_si256((const __m256i*)q4); q4 += 32;
const __m256i q4l = _mm256_and_si256(q4bits, m4);
const __m256i q4h = _mm256_and_si256(_mm256_srli_epi16(q4bits, 4), m4);
const __m256i q8l = _mm256_loadu_si256((const __m256i*)q8); q8 += 32;
__m256i p16l = _mm256_maddubs_epi16(q4l, q8l);
p16l = _mm256_madd_epi16(scale_l, p16l);
const __m256i q8h = _mm256_loadu_si256((const __m256i*)q8); q8 += 32;
__m256i p16h = _mm256_maddubs_epi16(q4h, q8h);
p16h = _mm256_madd_epi16(scale_h, p16h);
ggml : tiny ggml_vec_dot_q4_K_q8_K AVX2 improvement (#2819)
2023-08-28 12:51:08 +00:00
const __m256i sumj = _mm256_add_epi32(p16l, p16h);
ggml : add SOTA 2,3,4,5,6 bit k-quantizations (#1684) * Starting to add k-quantization to ggml I think it is better to have quantization separate from ggml. For now just adding the k-quants there, but it would be better to also factor out the existing ggml quantizations. * Adding Q3_K and Q8_K (de)-quantization * Q3_K now working on CUDA and AVX2/scalar CUDA is not ideal - ~50% slower than Q4_0 for single token prediction, about the same in batch mode (perplexity). CPU single token is ~55 ms (on Ryzen 7950X). * Some improvement for Q3_K on CUDA It is now ~22.5 ms/token on my GPU, so ~30% slower than Q4_0. * Some more CUDA optimizations for Q3_K Single token is now 20.5 ms/token (~20% slower than Q4_0). Perplexity is on par with Q4_0. * Adding Q4_K - scalar, AVX2, CUDA Performance is the same or perhaps very slightly better than Q4_0 on the CPU. On the GPU, single token prediction is ~10% better than Q4_0, batch mode (perplexity is about the same). * Adding Q6_K - scalar, AVX2, CUDA Performance is ~40% lower compared to Q4_K on the CPU. This is to be expected, considering that we are memory bound on the CPU and the 6-bit model is ~44% larger than the 4-bit. On the GPU, single token prediction is ~6% lower than Q4_0, batch mode (perplexity) is even closer (but still slower). * Adding Q5_K - scalar, AVX2, CUDA Performance is ~20% lower compared to Q4_K on the CPU. This is to be expected, considering that we are memory bound on the CPU and the 5-bit model is ~22% larger than the 4-bit. On the GPU, single token prediction is about the same as Q4_0 for both, single token and batch prediction. * Per convention, all QX_K quantizations use Q5_K for output.weight * Adding quantization mixes * Quantization mixes: didn't quite get what I wanted in the last commit * Q4_K dot product for ARM_NEON * Q6_K dot product for ARM_NEON * Q5_K dot product for ARM_NEON * Adding Q3_K dot for ARM_NEON It is 22% slower than Q4_K, despite the smaller model size. On x86_64, where we are memory bound, the Q3_K model is quite a bit faster than Q4_K. * A very slightly faster ARM_NEON Q3_K dot * Adding Q2_K - just CUDA for now Token prediction is pretty good - about 15.5 ms on a RTX 4080. Perplexity is about the same as Q4_K. * Adding scalar and AVX2 Q2_K dot * Adding ARM_NEON Q2_K dot About the same performance as Q4_K. * A slightly faster ARM_NEON Q2_K dot Single token prediction is now ~36 ms on M2 Max. The code is much simpler too. * Fixed bug in Q2_K CUDA dot product kernel Stranegly enough, for the few prompts I tried with the 7B model the responses looked perfectly reasonable. Only realized something is not quite right when I tried the larger models and started getting nonse back. In any case, Q2_K single token evaluation time on an RTX 4080 in a Ryzen7950X box iusing CUDA and model fully loaded on the GPU are ~15.5 ms for 7B, ~25.4 ms for 13B, and ~55.8 ms for 30B. The max number of layers that fit in VRAM for The 65B is 32. With that, we get ~330 ms per token, which is not that much faster than just running on the CPU (~470 ms per token). * Don't print zeros/NaNs when no count histogram has been collected * A 10% faster CUDA vector dot kernel for Q3_K Q3_K is now running at ~18.5 ms / token on CUDA, so the gap to Q4_0 is only 10%. It seems memory acccess pattern is more important for performance than the amount of computation the kernel does. * A slightly daster Q4_K AVX2 dot product For perplexity, where we are less memory bound, time per pass drops by ~5%. Barely measurable difference for single token prediction. * A slightly faster ARM_NEON A4_K dot product * Minor * Fix quantization error test We cannot possibly be expecting rmse < 0.002 for 2- and 3-bit quantization variants. * Fix docker build I have been sloppy with vector reinterpret casts on ARM_NEON. It seems clang is very forgiving in that regard. * Added forgotten ggml.o dependence on k_quants.h to the Makefile * Had unintentionally committed the Makefile with -Ofast enabled * ggml : rename k_quants -> ggml-quants-k, use lowercase in code --------- Co-authored-by: Iwan Kawrakow <iwan.kawrakow@gmail.com> Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
2023-06-05 19:56:18 +00:00
ggml : tiny ggml_vec_dot_q4_K_q8_K AVX2 improvement (#2819)
2023-08-28 12:51:08 +00:00
sumi = _mm256_add_epi32(sumi, sumj);
ggml : add SOTA 2,3,4,5,6 bit k-quantizations (#1684) * Starting to add k-quantization to ggml I think it is better to have quantization separate from ggml. For now just adding the k-quants there, but it would be better to also factor out the existing ggml quantizations. * Adding Q3_K and Q8_K (de)-quantization * Q3_K now working on CUDA and AVX2/scalar CUDA is not ideal - ~50% slower than Q4_0 for single token prediction, about the same in batch mode (perplexity). CPU single token is ~55 ms (on Ryzen 7950X). * Some improvement for Q3_K on CUDA It is now ~22.5 ms/token on my GPU, so ~30% slower than Q4_0. * Some more CUDA optimizations for Q3_K Single token is now 20.5 ms/token (~20% slower than Q4_0). Perplexity is on par with Q4_0. * Adding Q4_K - scalar, AVX2, CUDA Performance is the same or perhaps very slightly better than Q4_0 on the CPU. On the GPU, single token prediction is ~10% better than Q4_0, batch mode (perplexity is about the same). * Adding Q6_K - scalar, AVX2, CUDA Performance is ~40% lower compared to Q4_K on the CPU. This is to be expected, considering that we are memory bound on the CPU and the 6-bit model is ~44% larger than the 4-bit. On the GPU, single token prediction is ~6% lower than Q4_0, batch mode (perplexity) is even closer (but still slower). * Adding Q5_K - scalar, AVX2, CUDA Performance is ~20% lower compared to Q4_K on the CPU. This is to be expected, considering that we are memory bound on the CPU and the 5-bit model is ~22% larger than the 4-bit. On the GPU, single token prediction is about the same as Q4_0 for both, single token and batch prediction. * Per convention, all QX_K quantizations use Q5_K for output.weight * Adding quantization mixes * Quantization mixes: didn't quite get what I wanted in the last commit * Q4_K dot product for ARM_NEON * Q6_K dot product for ARM_NEON * Q5_K dot product for ARM_NEON * Adding Q3_K dot for ARM_NEON It is 22% slower than Q4_K, despite the smaller model size. On x86_64, where we are memory bound, the Q3_K model is quite a bit faster than Q4_K. * A very slightly faster ARM_NEON Q3_K dot * Adding Q2_K - just CUDA for now Token prediction is pretty good - about 15.5 ms on a RTX 4080. Perplexity is about the same as Q4_K. * Adding scalar and AVX2 Q2_K dot * Adding ARM_NEON Q2_K dot About the same performance as Q4_K. * A slightly faster ARM_NEON Q2_K dot Single token prediction is now ~36 ms on M2 Max. The code is much simpler too. * Fixed bug in Q2_K CUDA dot product kernel Stranegly enough, for the few prompts I tried with the 7B model the responses looked perfectly reasonable. Only realized something is not quite right when I tried the larger models and started getting nonse back. In any case, Q2_K single token evaluation time on an RTX 4080 in a Ryzen7950X box iusing CUDA and model fully loaded on the GPU are ~15.5 ms for 7B, ~25.4 ms for 13B, and ~55.8 ms for 30B. The max number of layers that fit in VRAM for The 65B is 32. With that, we get ~330 ms per token, which is not that much faster than just running on the CPU (~470 ms per token). * Don't print zeros/NaNs when no count histogram has been collected * A 10% faster CUDA vector dot kernel for Q3_K Q3_K is now running at ~18.5 ms / token on CUDA, so the gap to Q4_0 is only 10%. It seems memory acccess pattern is more important for performance than the amount of computation the kernel does. * A slightly daster Q4_K AVX2 dot product For perplexity, where we are less memory bound, time per pass drops by ~5%. Barely measurable difference for single token prediction. * A slightly faster ARM_NEON A4_K dot product * Minor * Fix quantization error test We cannot possibly be expecting rmse < 0.002 for 2- and 3-bit quantization variants. * Fix docker build I have been sloppy with vector reinterpret casts on ARM_NEON. It seems clang is very forgiving in that regard. * Added forgotten ggml.o dependence on k_quants.h to the Makefile * Had unintentionally committed the Makefile with -Ofast enabled * ggml : rename k_quants -> ggml-quants-k, use lowercase in code --------- Co-authored-by: Iwan Kawrakow <iwan.kawrakow@gmail.com> Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
2023-06-05 19:56:18 +00:00
}
__m256 vd = _mm256_set1_ps(d);
acc = _mm256_fmadd_ps(vd, _mm256_cvtepi32_ps(sumi), acc);
}
acc_m = _mm_add_ps(acc_m, _mm_movehl_ps(acc_m, acc_m));
acc_m = _mm_add_ss(acc_m, _mm_movehdup_ps(acc_m));
*s = hsum_float_8(acc) + _mm_cvtss_f32(acc_m);
k-quants : add AVX support to dot functions (#1916) * k_quants : add AVX support * k_quants : apply review comments
2023-06-26 16:46:07 +00:00
#elif defined __AVX__
const __m128i m4 = _mm_set1_epi8(0xF);
const __m128i m2 = _mm_set1_epi8(0x2);
__m256 acc = _mm256_setzero_ps();
__m128 acc_m = _mm_setzero_ps();
for (int i = 0; i < nb; ++i) {
ggml : move FP16 <-> FP32 code to ggml-impl.h (#3861) * ggml : move FP16 <-> FP32 stuff to ggml-impl.h ggml-ci * tests : fix ARM build * ggml : explicitly initialize deprecated type traits * ggml : add math.h to ggml-impl.h * ggml : remove duplicate static assert macros * ggml : prefix lookup tables with ggml_ ggml-ci * ggml-impl : move extern "C" to start of file
2023-10-30 17:19:15 +00:00
const float d = y[i].d * GGML_FP16_TO_FP32(x[i].d);
const float dmin = -y[i].d * GGML_FP16_TO_FP32(x[i].dmin);
k-quants : add AVX support to dot functions (#1916) * k_quants : add AVX support * k_quants : apply review comments
2023-06-26 16:46:07 +00:00
const uint8_t * restrict q4 = x[i].qs;
const int8_t * restrict q8 = y[i].qs;
memcpy(utmp, x[i].scales, 12);
utmp[3] = ((utmp[2] >> 4) & kmask2) | (((utmp[1] >> 6) & kmask3) << 4);
const uint32_t uaux = utmp[1] & kmask1;
utmp[1] = (utmp[2] & kmask2) | (((utmp[0] >> 6) & kmask3) << 4);
utmp[2] = uaux;
utmp[0] &= kmask1;
const __m128i utmps = _mm_set_epi32(utmp[3], utmp[2], utmp[1], utmp[0]);
const __m128i scales = _mm_cvtepu8_epi16(utmps);
const __m128i mins = _mm_cvtepu8_epi16(_mm_unpackhi_epi64(utmps, utmps));
const __m128i q8sums_0 = _mm_loadu_si128((const __m128i*)&y[i].bsums[0]);
const __m128i q8sums_1 = _mm_loadu_si128((const __m128i*)&y[i].bsums[8]);
const __m128i q8s = _mm_hadd_epi16(q8sums_0, q8sums_1);
const __m128i prod = _mm_madd_epi16(mins, q8s);
acc_m = _mm_add_ps(_mm_mul_ps(_mm_set1_ps(dmin), _mm_cvtepi32_ps(prod)), acc_m);
__m128i sumi_0 = _mm_setzero_si128();
__m128i sumi_1 = _mm_setzero_si128();
__m128i shuffle = _mm_set1_epi16(0x0100);
for (int j = 0; j < QK_K/64; ++j) {
const __m128i scale_l = _mm_shuffle_epi8(scales, shuffle);
shuffle = _mm_add_epi16(shuffle, m2);
const __m128i scale_h = _mm_shuffle_epi8(scales, shuffle);
shuffle = _mm_add_epi16(shuffle, m2);
__m128i q4bits = _mm_loadu_si128((const __m128i*)q4); q4 += 16;
const __m128i q4l_0 = _mm_and_si128(q4bits, m4);
const __m128i q4h_0 = _mm_and_si128(_mm_srli_epi16(q4bits, 4), m4);
q4bits = _mm_loadu_si128((const __m128i*)q4); q4 += 16;
const __m128i q4l_1 = _mm_and_si128(q4bits, m4);
const __m128i q4h_1 = _mm_and_si128(_mm_srli_epi16(q4bits, 4), m4);
const __m128i q8l_0 = _mm_loadu_si128((const __m128i*)q8); q8 += 16;
__m128i p16l = _mm_maddubs_epi16(q4l_0, q8l_0);
p16l = _mm_madd_epi16(scale_l, p16l);
sumi_0 = _mm_add_epi32(sumi_0, p16l);
const __m128i q8l_1 = _mm_loadu_si128((const __m128i*)q8); q8 += 16;
p16l = _mm_maddubs_epi16(q4l_1, q8l_1);
p16l = _mm_madd_epi16(scale_l, p16l);
sumi_1 = _mm_add_epi32(sumi_1, p16l);
const __m128i q8h_0 = _mm_loadu_si128((const __m128i*)q8); q8 += 16;
__m128i p16h = _mm_maddubs_epi16(q4h_0, q8h_0);
p16h = _mm_madd_epi16(scale_h, p16h);
sumi_0 = _mm_add_epi32(sumi_0, p16h);
const __m128i q8h_1 = _mm_loadu_si128((const __m128i*)q8); q8 += 16;
p16h = _mm_maddubs_epi16(q4h_1, q8h_1);
p16h = _mm_madd_epi16(scale_h, p16h);
sumi_1 = _mm_add_epi32(sumi_1, p16h);
}
__m256 vd = _mm256_set1_ps(d);
ggml : workaround for missing _mm256_setr_m128i in GCC < 8 in k_quants.c (#2405)
2023-07-28 18:17:45 +00:00
__m256i sumi = MM256_SET_M128I(sumi_1, sumi_0);
k-quants : add AVX support to dot functions (#1916) * k_quants : add AVX support * k_quants : apply review comments
2023-06-26 16:46:07 +00:00
acc = _mm256_add_ps(_mm256_mul_ps(vd, _mm256_cvtepi32_ps(sumi)), acc);
}
acc_m = _mm_add_ps(acc_m, _mm_movehl_ps(acc_m, acc_m));
acc_m = _mm_add_ss(acc_m, _mm_movehdup_ps(acc_m));
*s = hsum_float_8(acc) + _mm_cvtss_f32(acc_m);
ggml : add RISC-V Vector Support for K-Quants and improved the existing intrinsics (#3453) * Added RVV intrinsics support for Q8 quantize row and also improved the existing dot product function for risc-v. The RVV intrinsics is added for the following quantize row functions quantize_row_q8_0 quantize_row_q8_1 The following dot product functions have also been optimized by using LMUL = 1/2 instead of LMUL = 1 ggml_vec_dot_q4_0_q8_0 ggml_vec_dot_q4_1_q8_1 ggml_vec_dot_q5_0_q8_0 ggml_vec_dot_q5_1_q8_1 And vector initialization in Q5 by temporary array is also replaced by the vid intrinsics Signed-off-by: Ahmad Tameem <ahmad.tameem@10xengineers.ai> * Added RVV intrinsics support for k_quants This adds RISC-V Vector intrinsics support for the following K_quants functions for both QKK = 256 and QKK = 64 ggml_vec_dot_q2_K_q8_K ggml_vec_dot_q3_K_q8_K ggml_vec_dot_q4_K_q8_K ggml_vec_dot_q5_K_q8_K ggml_vec_dot_q6_K_q8_K Signed-off-by: Ahmad Tameem <ahmad.tameem@10xengineers.ai> --------- Signed-off-by: Ahmad Tameem <ahmad.tameem@10xengineers.ai>
2023-10-03 18:38:19 +00:00
#elif defined __riscv_v_intrinsic
const uint8_t * scales = (const uint8_t*)&utmp[0];
const uint8_t * mins = (const uint8_t*)&utmp[2];
float sumf = 0;
for (int i = 0; i < nb; ++i) {
size_t vl = 8;
ggml : move FP16 <-> FP32 code to ggml-impl.h (#3861) * ggml : move FP16 <-> FP32 stuff to ggml-impl.h ggml-ci * tests : fix ARM build * ggml : explicitly initialize deprecated type traits * ggml : add math.h to ggml-impl.h * ggml : remove duplicate static assert macros * ggml : prefix lookup tables with ggml_ ggml-ci * ggml-impl : move extern "C" to start of file
2023-10-30 17:19:15 +00:00
const float d = y[i].d * GGML_FP16_TO_FP32(x[i].d);
const float dmin = y[i].d * GGML_FP16_TO_FP32(x[i].dmin);
ggml : add RISC-V Vector Support for K-Quants and improved the existing intrinsics (#3453) * Added RVV intrinsics support for Q8 quantize row and also improved the existing dot product function for risc-v. The RVV intrinsics is added for the following quantize row functions quantize_row_q8_0 quantize_row_q8_1 The following dot product functions have also been optimized by using LMUL = 1/2 instead of LMUL = 1 ggml_vec_dot_q4_0_q8_0 ggml_vec_dot_q4_1_q8_1 ggml_vec_dot_q5_0_q8_0 ggml_vec_dot_q5_1_q8_1 And vector initialization in Q5 by temporary array is also replaced by the vid intrinsics Signed-off-by: Ahmad Tameem <ahmad.tameem@10xengineers.ai> * Added RVV intrinsics support for k_quants This adds RISC-V Vector intrinsics support for the following K_quants functions for both QKK = 256 and QKK = 64 ggml_vec_dot_q2_K_q8_K ggml_vec_dot_q3_K_q8_K ggml_vec_dot_q4_K_q8_K ggml_vec_dot_q5_K_q8_K ggml_vec_dot_q6_K_q8_K Signed-off-by: Ahmad Tameem <ahmad.tameem@10xengineers.ai> --------- Signed-off-by: Ahmad Tameem <ahmad.tameem@10xengineers.ai>
2023-10-03 18:38:19 +00:00
vint16mf2_t q8sums_0 = __riscv_vlse16_v_i16mf2(y[i].bsums, 4, vl);
vint16mf2_t q8sums_1 = __riscv_vlse16_v_i16mf2(y[i].bsums+1, 4, vl);
vint16mf2_t q8sums = __riscv_vadd_vv_i16mf2(q8sums_0, q8sums_1, vl);
memcpy(utmp, x[i].scales, 12);
utmp[3] = ((utmp[2] >> 4) & kmask2) | (((utmp[1] >> 6) & kmask3) << 4);
const uint32_t uaux = utmp[1] & kmask1;
utmp[1] = (utmp[2] & kmask2) | (((utmp[0] >> 6) & kmask3) << 4);
utmp[2] = uaux;
utmp[0] &= kmask1;
vuint8mf4_t mins8 = __riscv_vle8_v_u8mf4(mins, vl);
vint16mf2_t v_mins = __riscv_vreinterpret_v_u16mf2_i16mf2(__riscv_vzext_vf2_u16mf2(mins8, vl));
vint32m1_t prod = __riscv_vwmul_vv_i32m1(q8sums, v_mins, vl);
vint32m1_t sumi = __riscv_vredsum_vs_i32m1_i32m1(prod, __riscv_vmv_v_x_i32m1(0, 1), vl);
sumf -= dmin * __riscv_vmv_x_s_i32m1_i32(sumi);
const uint8_t * restrict q4 = x[i].qs;
const int8_t * restrict q8 = y[i].qs;
vl = 32;
int32_t sum_1 = 0;
int32_t sum_2 = 0;
vint16m1_t vzero = __riscv_vmv_v_x_i16m1(0, 1);
for (int j = 0; j < QK_K/64; ++j) {
// load Q4
vuint8m1_t q4_x = __riscv_vle8_v_u8m1(q4, vl);
// load Q8 and multiply it with lower Q4 nibble
vint8m1_t q8_0 = __riscv_vle8_v_i8m1(q8, vl);
vint8m1_t q4_0 = __riscv_vreinterpret_v_u8m1_i8m1(__riscv_vand_vx_u8m1(q4_x, 0x0F, vl));
vint16m2_t qv_0 = __riscv_vwmul_vv_i16m2(q4_0, q8_0, vl);
vint16m1_t vs_0 = __riscv_vredsum_vs_i16m2_i16m1(qv_0, vzero, vl);
sum_1 += __riscv_vmv_x_s_i16m1_i16(vs_0) * scales[2*j+0];
// load Q8 and multiply it with upper Q4 nibble
vint8m1_t q8_1 = __riscv_vle8_v_i8m1(q8+32, vl);
vint8m1_t q4_1 = __riscv_vreinterpret_v_u8m1_i8m1(__riscv_vsrl_vx_u8m1(q4_x, 0x04, vl));
vint16m2_t qv_1 = __riscv_vwmul_vv_i16m2(q4_1, q8_1, vl);
vint16m1_t vs_1 = __riscv_vredsum_vs_i16m2_i16m1(qv_1, vzero, vl);
sum_2 += __riscv_vmv_x_s_i16m1_i16(vs_1) * scales[2*j+1];
q4 += 32; q8 += 64;
}
sumf += d*(sum_1 + sum_2);
}
*s = sumf;
ggml : add SOTA 2,3,4,5,6 bit k-quantizations (#1684) * Starting to add k-quantization to ggml I think it is better to have quantization separate from ggml. For now just adding the k-quants there, but it would be better to also factor out the existing ggml quantizations. * Adding Q3_K and Q8_K (de)-quantization * Q3_K now working on CUDA and AVX2/scalar CUDA is not ideal - ~50% slower than Q4_0 for single token prediction, about the same in batch mode (perplexity). CPU single token is ~55 ms (on Ryzen 7950X). * Some improvement for Q3_K on CUDA It is now ~22.5 ms/token on my GPU, so ~30% slower than Q4_0. * Some more CUDA optimizations for Q3_K Single token is now 20.5 ms/token (~20% slower than Q4_0). Perplexity is on par with Q4_0. * Adding Q4_K - scalar, AVX2, CUDA Performance is the same or perhaps very slightly better than Q4_0 on the CPU. On the GPU, single token prediction is ~10% better than Q4_0, batch mode (perplexity is about the same). * Adding Q6_K - scalar, AVX2, CUDA Performance is ~40% lower compared to Q4_K on the CPU. This is to be expected, considering that we are memory bound on the CPU and the 6-bit model is ~44% larger than the 4-bit. On the GPU, single token prediction is ~6% lower than Q4_0, batch mode (perplexity) is even closer (but still slower). * Adding Q5_K - scalar, AVX2, CUDA Performance is ~20% lower compared to Q4_K on the CPU. This is to be expected, considering that we are memory bound on the CPU and the 5-bit model is ~22% larger than the 4-bit. On the GPU, single token prediction is about the same as Q4_0 for both, single token and batch prediction. * Per convention, all QX_K quantizations use Q5_K for output.weight * Adding quantization mixes * Quantization mixes: didn't quite get what I wanted in the last commit * Q4_K dot product for ARM_NEON * Q6_K dot product for ARM_NEON * Q5_K dot product for ARM_NEON * Adding Q3_K dot for ARM_NEON It is 22% slower than Q4_K, despite the smaller model size. On x86_64, where we are memory bound, the Q3_K model is quite a bit faster than Q4_K. * A very slightly faster ARM_NEON Q3_K dot * Adding Q2_K - just CUDA for now Token prediction is pretty good - about 15.5 ms on a RTX 4080. Perplexity is about the same as Q4_K. * Adding scalar and AVX2 Q2_K dot * Adding ARM_NEON Q2_K dot About the same performance as Q4_K. * A slightly faster ARM_NEON Q2_K dot Single token prediction is now ~36 ms on M2 Max. The code is much simpler too. * Fixed bug in Q2_K CUDA dot product kernel Stranegly enough, for the few prompts I tried with the 7B model the responses looked perfectly reasonable. Only realized something is not quite right when I tried the larger models and started getting nonse back. In any case, Q2_K single token evaluation time on an RTX 4080 in a Ryzen7950X box iusing CUDA and model fully loaded on the GPU are ~15.5 ms for 7B, ~25.4 ms for 13B, and ~55.8 ms for 30B. The max number of layers that fit in VRAM for The 65B is 32. With that, we get ~330 ms per token, which is not that much faster than just running on the CPU (~470 ms per token). * Don't print zeros/NaNs when no count histogram has been collected * A 10% faster CUDA vector dot kernel for Q3_K Q3_K is now running at ~18.5 ms / token on CUDA, so the gap to Q4_0 is only 10%. It seems memory acccess pattern is more important for performance than the amount of computation the kernel does. * A slightly daster Q4_K AVX2 dot product For perplexity, where we are less memory bound, time per pass drops by ~5%. Barely measurable difference for single token prediction. * A slightly faster ARM_NEON A4_K dot product * Minor * Fix quantization error test We cannot possibly be expecting rmse < 0.002 for 2- and 3-bit quantization variants. * Fix docker build I have been sloppy with vector reinterpret casts on ARM_NEON. It seems clang is very forgiving in that regard. * Added forgotten ggml.o dependence on k_quants.h to the Makefile * Had unintentionally committed the Makefile with -Ofast enabled * ggml : rename k_quants -> ggml-quants-k, use lowercase in code --------- Co-authored-by: Iwan Kawrakow <iwan.kawrakow@gmail.com> Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
2023-06-05 19:56:18 +00:00
#else
const uint8_t * scales = (const uint8_t*)&utmp[0];
const uint8_t * mins = (const uint8_t*)&utmp[2];
int8_t aux8[QK_K];
int16_t aux16[8];
float sums [8];
int32_t aux32[8];
memset(sums, 0, 8*sizeof(float));
float sumf = 0;
for (int i = 0; i < nb; ++i) {
const uint8_t * restrict q4 = x[i].qs;
const int8_t * restrict q8 = y[i].qs;
memset(aux32, 0, 8*sizeof(int32_t));
int8_t * restrict a = aux8;
for (int j = 0; j < QK_K/64; ++j) {
for (int l = 0; l < 32; ++l) a[l] = (int8_t)(q4[l] & 0xF);
a += 32;
for (int l = 0; l < 32; ++l) a[l] = (int8_t)(q4[l] >> 4);
a += 32; q4 += 32;
}
memcpy(utmp, x[i].scales, 12);
utmp[3] = ((utmp[2] >> 4) & kmask2) | (((utmp[1] >> 6) & kmask3) << 4);
const uint32_t uaux = utmp[1] & kmask1;
utmp[1] = (utmp[2] & kmask2) | (((utmp[0] >> 6) & kmask3) << 4);
utmp[2] = uaux;
utmp[0] &= kmask1;
int sumi = 0;
for (int j = 0; j < QK_K/16; ++j) sumi += y[i].bsums[j] * mins[j/2];
a = aux8;
int is = 0;
for (int j = 0; j < QK_K/32; ++j) {
int32_t scale = scales[is++];
for (int l = 0; l < 8; ++l) aux16[l] = q8[l] * a[l];
for (int l = 0; l < 8; ++l) aux32[l] += scale * aux16[l];
q8 += 8; a += 8;
for (int l = 0; l < 8; ++l) aux16[l] = q8[l] * a[l];
for (int l = 0; l < 8; ++l) aux32[l] += scale * aux16[l];
q8 += 8; a += 8;
for (int l = 0; l < 8; ++l) aux16[l] = q8[l] * a[l];
for (int l = 0; l < 8; ++l) aux32[l] += scale * aux16[l];
q8 += 8; a += 8;
for (int l = 0; l < 8; ++l) aux16[l] = q8[l] * a[l];
for (int l = 0; l < 8; ++l) aux32[l] += scale * aux16[l];
q8 += 8; a += 8;
}
ggml : move FP16 <-> FP32 code to ggml-impl.h (#3861) * ggml : move FP16 <-> FP32 stuff to ggml-impl.h ggml-ci * tests : fix ARM build * ggml : explicitly initialize deprecated type traits * ggml : add math.h to ggml-impl.h * ggml : remove duplicate static assert macros * ggml : prefix lookup tables with ggml_ ggml-ci * ggml-impl : move extern "C" to start of file
2023-10-30 17:19:15 +00:00
const float d = GGML_FP16_TO_FP32(x[i].d) * y[i].d;
ggml : add SOTA 2,3,4,5,6 bit k-quantizations (#1684) * Starting to add k-quantization to ggml I think it is better to have quantization separate from ggml. For now just adding the k-quants there, but it would be better to also factor out the existing ggml quantizations. * Adding Q3_K and Q8_K (de)-quantization * Q3_K now working on CUDA and AVX2/scalar CUDA is not ideal - ~50% slower than Q4_0 for single token prediction, about the same in batch mode (perplexity). CPU single token is ~55 ms (on Ryzen 7950X). * Some improvement for Q3_K on CUDA It is now ~22.5 ms/token on my GPU, so ~30% slower than Q4_0. * Some more CUDA optimizations for Q3_K Single token is now 20.5 ms/token (~20% slower than Q4_0). Perplexity is on par with Q4_0. * Adding Q4_K - scalar, AVX2, CUDA Performance is the same or perhaps very slightly better than Q4_0 on the CPU. On the GPU, single token prediction is ~10% better than Q4_0, batch mode (perplexity is about the same). * Adding Q6_K - scalar, AVX2, CUDA Performance is ~40% lower compared to Q4_K on the CPU. This is to be expected, considering that we are memory bound on the CPU and the 6-bit model is ~44% larger than the 4-bit. On the GPU, single token prediction is ~6% lower than Q4_0, batch mode (perplexity) is even closer (but still slower). * Adding Q5_K - scalar, AVX2, CUDA Performance is ~20% lower compared to Q4_K on the CPU. This is to be expected, considering that we are memory bound on the CPU and the 5-bit model is ~22% larger than the 4-bit. On the GPU, single token prediction is about the same as Q4_0 for both, single token and batch prediction. * Per convention, all QX_K quantizations use Q5_K for output.weight * Adding quantization mixes * Quantization mixes: didn't quite get what I wanted in the last commit * Q4_K dot product for ARM_NEON * Q6_K dot product for ARM_NEON * Q5_K dot product for ARM_NEON * Adding Q3_K dot for ARM_NEON It is 22% slower than Q4_K, despite the smaller model size. On x86_64, where we are memory bound, the Q3_K model is quite a bit faster than Q4_K. * A very slightly faster ARM_NEON Q3_K dot * Adding Q2_K - just CUDA for now Token prediction is pretty good - about 15.5 ms on a RTX 4080. Perplexity is about the same as Q4_K. * Adding scalar and AVX2 Q2_K dot * Adding ARM_NEON Q2_K dot About the same performance as Q4_K. * A slightly faster ARM_NEON Q2_K dot Single token prediction is now ~36 ms on M2 Max. The code is much simpler too. * Fixed bug in Q2_K CUDA dot product kernel Stranegly enough, for the few prompts I tried with the 7B model the responses looked perfectly reasonable. Only realized something is not quite right when I tried the larger models and started getting nonse back. In any case, Q2_K single token evaluation time on an RTX 4080 in a Ryzen7950X box iusing CUDA and model fully loaded on the GPU are ~15.5 ms for 7B, ~25.4 ms for 13B, and ~55.8 ms for 30B. The max number of layers that fit in VRAM for The 65B is 32. With that, we get ~330 ms per token, which is not that much faster than just running on the CPU (~470 ms per token). * Don't print zeros/NaNs when no count histogram has been collected * A 10% faster CUDA vector dot kernel for Q3_K Q3_K is now running at ~18.5 ms / token on CUDA, so the gap to Q4_0 is only 10%. It seems memory acccess pattern is more important for performance than the amount of computation the kernel does. * A slightly daster Q4_K AVX2 dot product For perplexity, where we are less memory bound, time per pass drops by ~5%. Barely measurable difference for single token prediction. * A slightly faster ARM_NEON A4_K dot product * Minor * Fix quantization error test We cannot possibly be expecting rmse < 0.002 for 2- and 3-bit quantization variants. * Fix docker build I have been sloppy with vector reinterpret casts on ARM_NEON. It seems clang is very forgiving in that regard. * Added forgotten ggml.o dependence on k_quants.h to the Makefile * Had unintentionally committed the Makefile with -Ofast enabled * ggml : rename k_quants -> ggml-quants-k, use lowercase in code --------- Co-authored-by: Iwan Kawrakow <iwan.kawrakow@gmail.com> Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
2023-06-05 19:56:18 +00:00
for (int l = 0; l < 8; ++l) sums[l] += d * aux32[l];
ggml : move FP16 <-> FP32 code to ggml-impl.h (#3861) * ggml : move FP16 <-> FP32 stuff to ggml-impl.h ggml-ci * tests : fix ARM build * ggml : explicitly initialize deprecated type traits * ggml : add math.h to ggml-impl.h * ggml : remove duplicate static assert macros * ggml : prefix lookup tables with ggml_ ggml-ci * ggml-impl : move extern "C" to start of file
2023-10-30 17:19:15 +00:00
const float dmin = GGML_FP16_TO_FP32(x[i].dmin) * y[i].d;
ggml : add SOTA 2,3,4,5,6 bit k-quantizations (#1684) * Starting to add k-quantization to ggml I think it is better to have quantization separate from ggml. For now just adding the k-quants there, but it would be better to also factor out the existing ggml quantizations. * Adding Q3_K and Q8_K (de)-quantization * Q3_K now working on CUDA and AVX2/scalar CUDA is not ideal - ~50% slower than Q4_0 for single token prediction, about the same in batch mode (perplexity). CPU single token is ~55 ms (on Ryzen 7950X). * Some improvement for Q3_K on CUDA It is now ~22.5 ms/token on my GPU, so ~30% slower than Q4_0. * Some more CUDA optimizations for Q3_K Single token is now 20.5 ms/token (~20% slower than Q4_0). Perplexity is on par with Q4_0. * Adding Q4_K - scalar, AVX2, CUDA Performance is the same or perhaps very slightly better than Q4_0 on the CPU. On the GPU, single token prediction is ~10% better than Q4_0, batch mode (perplexity is about the same). * Adding Q6_K - scalar, AVX2, CUDA Performance is ~40% lower compared to Q4_K on the CPU. This is to be expected, considering that we are memory bound on the CPU and the 6-bit model is ~44% larger than the 4-bit. On the GPU, single token prediction is ~6% lower than Q4_0, batch mode (perplexity) is even closer (but still slower). * Adding Q5_K - scalar, AVX2, CUDA Performance is ~20% lower compared to Q4_K on the CPU. This is to be expected, considering that we are memory bound on the CPU and the 5-bit model is ~22% larger than the 4-bit. On the GPU, single token prediction is about the same as Q4_0 for both, single token and batch prediction. * Per convention, all QX_K quantizations use Q5_K for output.weight * Adding quantization mixes * Quantization mixes: didn't quite get what I wanted in the last commit * Q4_K dot product for ARM_NEON * Q6_K dot product for ARM_NEON * Q5_K dot product for ARM_NEON * Adding Q3_K dot for ARM_NEON It is 22% slower than Q4_K, despite the smaller model size. On x86_64, where we are memory bound, the Q3_K model is quite a bit faster than Q4_K. * A very slightly faster ARM_NEON Q3_K dot * Adding Q2_K - just CUDA for now Token prediction is pretty good - about 15.5 ms on a RTX 4080. Perplexity is about the same as Q4_K. * Adding scalar and AVX2 Q2_K dot * Adding ARM_NEON Q2_K dot About the same performance as Q4_K. * A slightly faster ARM_NEON Q2_K dot Single token prediction is now ~36 ms on M2 Max. The code is much simpler too. * Fixed bug in Q2_K CUDA dot product kernel Stranegly enough, for the few prompts I tried with the 7B model the responses looked perfectly reasonable. Only realized something is not quite right when I tried the larger models and started getting nonse back. In any case, Q2_K single token evaluation time on an RTX 4080 in a Ryzen7950X box iusing CUDA and model fully loaded on the GPU are ~15.5 ms for 7B, ~25.4 ms for 13B, and ~55.8 ms for 30B. The max number of layers that fit in VRAM for The 65B is 32. With that, we get ~330 ms per token, which is not that much faster than just running on the CPU (~470 ms per token). * Don't print zeros/NaNs when no count histogram has been collected * A 10% faster CUDA vector dot kernel for Q3_K Q3_K is now running at ~18.5 ms / token on CUDA, so the gap to Q4_0 is only 10%. It seems memory acccess pattern is more important for performance than the amount of computation the kernel does. * A slightly daster Q4_K AVX2 dot product For perplexity, where we are less memory bound, time per pass drops by ~5%. Barely measurable difference for single token prediction. * A slightly faster ARM_NEON A4_K dot product * Minor * Fix quantization error test We cannot possibly be expecting rmse < 0.002 for 2- and 3-bit quantization variants. * Fix docker build I have been sloppy with vector reinterpret casts on ARM_NEON. It seems clang is very forgiving in that regard. * Added forgotten ggml.o dependence on k_quants.h to the Makefile * Had unintentionally committed the Makefile with -Ofast enabled * ggml : rename k_quants -> ggml-quants-k, use lowercase in code --------- Co-authored-by: Iwan Kawrakow <iwan.kawrakow@gmail.com> Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
2023-06-05 19:56:18 +00:00
sumf -= dmin * sumi;
}
for (int l = 0; l < 8; ++l) sumf += sums[l];
*s = sumf;
#endif
}
k-quants : support for super-block size of 64 (#2001) * k_quants: WIP super-blocks with 64 weights * k_quants: WIP super-blocks with 64 weights Q6_K scalar and AVX2 works * k_quants: WIP super-blocks with 64 weights Q4_K scalar and AVX2 works * k_quants: WIP super-blocks with 64 weights Q2_K scalar and AVX2 works. Q2_K is way too slow (it is actually slower than the scalar implementation) * k_quants: WIP super-blocks with 64 weights Q3_K scalar and AVX2 works. * k_quants: WIP super-blocks with 64 weights Q5_K scalar and AVX2 works, and with that all k_quants are done on AVX2 and scalar * k_quants: WIP super-blocks with 64 weights Q6_K working on CUDA. Cannot make it run quite as gast as with super-blocks with 256 weigths: 8% slower on 4080, 20% slower on the 1660 (but there we fit 1 less layer on the GPU because pf the larger model size), so some fraction of these 20% is due to that, * k_quants: WIP super-blocks with 64 weights Q4_K working on CUDA. ~10% slower on GTX-1660, 16% slower on 4080. * k_quants: WIP super-blocks with 64 weights Q2_K working on CUDA. ~3% slower on GTX-1660, 10% slower on 4080. * k_quants: WIP super-blocks with 64 weights Q3_K working on CUDA. * k_quants: WIP super-blocks with 64 weights Q5_K working on CUDA, and with this CUDA is done. * k_quants: WIP super-blocks with 64 weights Q6_K working on ARM_NEON * k_quants: WIP super-blocks with 64 weights Q4_K working on ARM_NEON, but quite a bit slower than 256 weights * k_quants: WIP super-blocks with 64 weights Q2_K working on ARM_NEON, but quite a bit slower than 256 weights * k_quants: WIP super-blocks with 64 weights Q3_K working on ARM_NEON, but quite a bit slower than 256 weights. * k_quants: WIP super-blocks with 64 weights Q5_K working on ARM_NEON, but quite a bit slower than 256 weights. With that, we have full support for ARM_NEON, although performance is not quite there. * k_quants: WIP super-blocks with 64 weights Slightly more efficient Q3_K and Q5_K * k_quants: WIP super-blocks with 64 weights Another small improvement for Q3_K and Q5_K on ARM_NEON * k_quants: WIP super-blocks with 64 weights Yet another speedup for Q5_K on ARM_NEON. We are now within 10% of the QK_K = 256 version. * k_quants: WIP super-blocks with 64 weights * We are able to pass preprocessor macros to the Metal compiler * Q6_K works and is actually slightly more efficient than the QK_K = 256 version (25.2 ms vs 25.8 ms) * k_quants: WIP super-blocks with 64 weights Q4_K works on Metal and is actually slightly faster than QK_K = 256 (21.95 ms vs 24.0 ms). * k_quants: WIP super-blocks with 64 weights Q2_K works on Metal and is very slightly faster than QK_K = 256 (23.8 ms vs 24.2 ms). * k_quants: WIP super-blocks with 64 weights Q3_K works on Metal and is slightly faster than QK_K = 256 (26.6 ms vs 28.3 ms). * k_quants: WIP super-blocks with 64 weights Q5_K works on Metal and is slightly faster than QK_K = 256 (23.7 ms vs 26.3 ms). * k_quants: call them _K, not _k, also on Metal * k_quants: correctly define QK_K in llama.cpp * Fixed bug in q4_K quantization added with the 64-block addition * Simplify via lambda * k_quants: swicth Q3_K to 4-bit scales when QK_K = 64 Otherwise there isn't much benefit from this quantization type. There is some very slight loss in accuracy, but we reduce size by ~7%. E.g., for OpenLLaMA-3B, Q3_K_S perplexity is 8.6131 with 8-bit scales and 8.6352 with 4-bit, while file size decreases from 1.53G to 1.44G. * k_quants: switch Q4_K to 4-bit scales when QK_K = 64 Here the loss in accuracy is greater than for Q3_K, but the Q4_K points still move further to the left on the perplexity vs size curve. * k_quants: forgot to add the Metal changes in last commit * k_quants: change Q5_K to be type 0 when QK_K = 64 Still needs AVX2 implementation * k_quants: AVX2 implementation for new 64-weight Q5_K * k_quants: 10% faster ARM_NEON Q5_K dot product * k_quants: fixed issue caused by merging with master --------- Co-authored-by: Iwan Kawrakow <iwan.kawrakow@gmail.com>
2023-06-26 16:43:07 +00:00
#else
void ggml_vec_dot_q4_K_q8_K(const int n, float * restrict s, const void * restrict vx, const void * restrict vy) {
assert(n % QK_K == 0);
const block_q4_K * restrict x = vx;
const block_q8_K * restrict y = vy;
const int nb = n / QK_K;
#ifdef __ARM_NEON
const uint8x16_t m4b = vdupq_n_u8(0xf);
const int32x4_t mzero = vdupq_n_s32(0);
float sumf = 0;
ggml : sync (im2col, GPU conv, 32-bit arm compat) (#4060) ggml-ci
2023-11-13 14:55:52 +00:00
ggml_int8x16x2_t q4bytes;
ggml_int8x16x4_t q8bytes;
k-quants : support for super-block size of 64 (#2001) * k_quants: WIP super-blocks with 64 weights * k_quants: WIP super-blocks with 64 weights Q6_K scalar and AVX2 works * k_quants: WIP super-blocks with 64 weights Q4_K scalar and AVX2 works * k_quants: WIP super-blocks with 64 weights Q2_K scalar and AVX2 works. Q2_K is way too slow (it is actually slower than the scalar implementation) * k_quants: WIP super-blocks with 64 weights Q3_K scalar and AVX2 works. * k_quants: WIP super-blocks with 64 weights Q5_K scalar and AVX2 works, and with that all k_quants are done on AVX2 and scalar * k_quants: WIP super-blocks with 64 weights Q6_K working on CUDA. Cannot make it run quite as gast as with super-blocks with 256 weigths: 8% slower on 4080, 20% slower on the 1660 (but there we fit 1 less layer on the GPU because pf the larger model size), so some fraction of these 20% is due to that, * k_quants: WIP super-blocks with 64 weights Q4_K working on CUDA. ~10% slower on GTX-1660, 16% slower on 4080. * k_quants: WIP super-blocks with 64 weights Q2_K working on CUDA. ~3% slower on GTX-1660, 10% slower on 4080. * k_quants: WIP super-blocks with 64 weights Q3_K working on CUDA. * k_quants: WIP super-blocks with 64 weights Q5_K working on CUDA, and with this CUDA is done. * k_quants: WIP super-blocks with 64 weights Q6_K working on ARM_NEON * k_quants: WIP super-blocks with 64 weights Q4_K working on ARM_NEON, but quite a bit slower than 256 weights * k_quants: WIP super-blocks with 64 weights Q2_K working on ARM_NEON, but quite a bit slower than 256 weights * k_quants: WIP super-blocks with 64 weights Q3_K working on ARM_NEON, but quite a bit slower than 256 weights. * k_quants: WIP super-blocks with 64 weights Q5_K working on ARM_NEON, but quite a bit slower than 256 weights. With that, we have full support for ARM_NEON, although performance is not quite there. * k_quants: WIP super-blocks with 64 weights Slightly more efficient Q3_K and Q5_K * k_quants: WIP super-blocks with 64 weights Another small improvement for Q3_K and Q5_K on ARM_NEON * k_quants: WIP super-blocks with 64 weights Yet another speedup for Q5_K on ARM_NEON. We are now within 10% of the QK_K = 256 version. * k_quants: WIP super-blocks with 64 weights * We are able to pass preprocessor macros to the Metal compiler * Q6_K works and is actually slightly more efficient than the QK_K = 256 version (25.2 ms vs 25.8 ms) * k_quants: WIP super-blocks with 64 weights Q4_K works on Metal and is actually slightly faster than QK_K = 256 (21.95 ms vs 24.0 ms). * k_quants: WIP super-blocks with 64 weights Q2_K works on Metal and is very slightly faster than QK_K = 256 (23.8 ms vs 24.2 ms). * k_quants: WIP super-blocks with 64 weights Q3_K works on Metal and is slightly faster than QK_K = 256 (26.6 ms vs 28.3 ms). * k_quants: WIP super-blocks with 64 weights Q5_K works on Metal and is slightly faster than QK_K = 256 (23.7 ms vs 26.3 ms). * k_quants: call them _K, not _k, also on Metal * k_quants: correctly define QK_K in llama.cpp * Fixed bug in q4_K quantization added with the 64-block addition * Simplify via lambda * k_quants: swicth Q3_K to 4-bit scales when QK_K = 64 Otherwise there isn't much benefit from this quantization type. There is some very slight loss in accuracy, but we reduce size by ~7%. E.g., for OpenLLaMA-3B, Q3_K_S perplexity is 8.6131 with 8-bit scales and 8.6352 with 4-bit, while file size decreases from 1.53G to 1.44G. * k_quants: switch Q4_K to 4-bit scales when QK_K = 64 Here the loss in accuracy is greater than for Q3_K, but the Q4_K points still move further to the left on the perplexity vs size curve. * k_quants: forgot to add the Metal changes in last commit * k_quants: change Q5_K to be type 0 when QK_K = 64 Still needs AVX2 implementation * k_quants: AVX2 implementation for new 64-weight Q5_K * k_quants: 10% faster ARM_NEON Q5_K dot product * k_quants: fixed issue caused by merging with master --------- Co-authored-by: Iwan Kawrakow <iwan.kawrakow@gmail.com>
2023-06-26 16:43:07 +00:00
float sum_mins = 0.f;
uint16_t aux16[2];
const uint8_t * restrict scales = (const uint8_t *)aux16;
for (int i = 0; i < nb; ++i) {
const uint8_t * restrict q4 = x[i].qs;
const int8_t * restrict q8 = y[i].qs;
const uint16_t * restrict a = (const uint16_t *)x[i].scales;
aux16[0] = a[0] & 0x0f0f;
aux16[1] = (a[0] >> 4) & 0x0f0f;
const int32_t summi = scales[2] * (y[i].bsums[0] + y[i].bsums[1]) + scales[3] * (y[i].bsums[2] + y[i].bsums[3]);
sum_mins += y[i].d * (float)x[i].d[1] * summi;
const float d = y[i].d * (float)x[i].d[0];
ggml : add SOTA 2,3,4,5,6 bit k-quantizations (#1684) * Starting to add k-quantization to ggml I think it is better to have quantization separate from ggml. For now just adding the k-quants there, but it would be better to also factor out the existing ggml quantizations. * Adding Q3_K and Q8_K (de)-quantization * Q3_K now working on CUDA and AVX2/scalar CUDA is not ideal - ~50% slower than Q4_0 for single token prediction, about the same in batch mode (perplexity). CPU single token is ~55 ms (on Ryzen 7950X). * Some improvement for Q3_K on CUDA It is now ~22.5 ms/token on my GPU, so ~30% slower than Q4_0. * Some more CUDA optimizations for Q3_K Single token is now 20.5 ms/token (~20% slower than Q4_0). Perplexity is on par with Q4_0. * Adding Q4_K - scalar, AVX2, CUDA Performance is the same or perhaps very slightly better than Q4_0 on the CPU. On the GPU, single token prediction is ~10% better than Q4_0, batch mode (perplexity is about the same). * Adding Q6_K - scalar, AVX2, CUDA Performance is ~40% lower compared to Q4_K on the CPU. This is to be expected, considering that we are memory bound on the CPU and the 6-bit model is ~44% larger than the 4-bit. On the GPU, single token prediction is ~6% lower than Q4_0, batch mode (perplexity) is even closer (but still slower). * Adding Q5_K - scalar, AVX2, CUDA Performance is ~20% lower compared to Q4_K on the CPU. This is to be expected, considering that we are memory bound on the CPU and the 5-bit model is ~22% larger than the 4-bit. On the GPU, single token prediction is about the same as Q4_0 for both, single token and batch prediction. * Per convention, all QX_K quantizations use Q5_K for output.weight * Adding quantization mixes * Quantization mixes: didn't quite get what I wanted in the last commit * Q4_K dot product for ARM_NEON * Q6_K dot product for ARM_NEON * Q5_K dot product for ARM_NEON * Adding Q3_K dot for ARM_NEON It is 22% slower than Q4_K, despite the smaller model size. On x86_64, where we are memory bound, the Q3_K model is quite a bit faster than Q4_K. * A very slightly faster ARM_NEON Q3_K dot * Adding Q2_K - just CUDA for now Token prediction is pretty good - about 15.5 ms on a RTX 4080. Perplexity is about the same as Q4_K. * Adding scalar and AVX2 Q2_K dot * Adding ARM_NEON Q2_K dot About the same performance as Q4_K. * A slightly faster ARM_NEON Q2_K dot Single token prediction is now ~36 ms on M2 Max. The code is much simpler too. * Fixed bug in Q2_K CUDA dot product kernel Stranegly enough, for the few prompts I tried with the 7B model the responses looked perfectly reasonable. Only realized something is not quite right when I tried the larger models and started getting nonse back. In any case, Q2_K single token evaluation time on an RTX 4080 in a Ryzen7950X box iusing CUDA and model fully loaded on the GPU are ~15.5 ms for 7B, ~25.4 ms for 13B, and ~55.8 ms for 30B. The max number of layers that fit in VRAM for The 65B is 32. With that, we get ~330 ms per token, which is not that much faster than just running on the CPU (~470 ms per token). * Don't print zeros/NaNs when no count histogram has been collected * A 10% faster CUDA vector dot kernel for Q3_K Q3_K is now running at ~18.5 ms / token on CUDA, so the gap to Q4_0 is only 10%. It seems memory acccess pattern is more important for performance than the amount of computation the kernel does. * A slightly daster Q4_K AVX2 dot product For perplexity, where we are less memory bound, time per pass drops by ~5%. Barely measurable difference for single token prediction. * A slightly faster ARM_NEON A4_K dot product * Minor * Fix quantization error test We cannot possibly be expecting rmse < 0.002 for 2- and 3-bit quantization variants. * Fix docker build I have been sloppy with vector reinterpret casts on ARM_NEON. It seems clang is very forgiving in that regard. * Added forgotten ggml.o dependence on k_quants.h to the Makefile * Had unintentionally committed the Makefile with -Ofast enabled * ggml : rename k_quants -> ggml-quants-k, use lowercase in code --------- Co-authored-by: Iwan Kawrakow <iwan.kawrakow@gmail.com> Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
2023-06-05 19:56:18 +00:00
ggml : sync (im2col, GPU conv, 32-bit arm compat) (#4060) ggml-ci
2023-11-13 14:55:52 +00:00
const ggml_uint8x16x2_t q4bits = ggml_vld1q_u8_x2(q4);
k-quants : support for super-block size of 64 (#2001) * k_quants: WIP super-blocks with 64 weights * k_quants: WIP super-blocks with 64 weights Q6_K scalar and AVX2 works * k_quants: WIP super-blocks with 64 weights Q4_K scalar and AVX2 works * k_quants: WIP super-blocks with 64 weights Q2_K scalar and AVX2 works. Q2_K is way too slow (it is actually slower than the scalar implementation) * k_quants: WIP super-blocks with 64 weights Q3_K scalar and AVX2 works. * k_quants: WIP super-blocks with 64 weights Q5_K scalar and AVX2 works, and with that all k_quants are done on AVX2 and scalar * k_quants: WIP super-blocks with 64 weights Q6_K working on CUDA. Cannot make it run quite as gast as with super-blocks with 256 weigths: 8% slower on 4080, 20% slower on the 1660 (but there we fit 1 less layer on the GPU because pf the larger model size), so some fraction of these 20% is due to that, * k_quants: WIP super-blocks with 64 weights Q4_K working on CUDA. ~10% slower on GTX-1660, 16% slower on 4080. * k_quants: WIP super-blocks with 64 weights Q2_K working on CUDA. ~3% slower on GTX-1660, 10% slower on 4080. * k_quants: WIP super-blocks with 64 weights Q3_K working on CUDA. * k_quants: WIP super-blocks with 64 weights Q5_K working on CUDA, and with this CUDA is done. * k_quants: WIP super-blocks with 64 weights Q6_K working on ARM_NEON * k_quants: WIP super-blocks with 64 weights Q4_K working on ARM_NEON, but quite a bit slower than 256 weights * k_quants: WIP super-blocks with 64 weights Q2_K working on ARM_NEON, but quite a bit slower than 256 weights * k_quants: WIP super-blocks with 64 weights Q3_K working on ARM_NEON, but quite a bit slower than 256 weights. * k_quants: WIP super-blocks with 64 weights Q5_K working on ARM_NEON, but quite a bit slower than 256 weights. With that, we have full support for ARM_NEON, although performance is not quite there. * k_quants: WIP super-blocks with 64 weights Slightly more efficient Q3_K and Q5_K * k_quants: WIP super-blocks with 64 weights Another small improvement for Q3_K and Q5_K on ARM_NEON * k_quants: WIP super-blocks with 64 weights Yet another speedup for Q5_K on ARM_NEON. We are now within 10% of the QK_K = 256 version. * k_quants: WIP super-blocks with 64 weights * We are able to pass preprocessor macros to the Metal compiler * Q6_K works and is actually slightly more efficient than the QK_K = 256 version (25.2 ms vs 25.8 ms) * k_quants: WIP super-blocks with 64 weights Q4_K works on Metal and is actually slightly faster than QK_K = 256 (21.95 ms vs 24.0 ms). * k_quants: WIP super-blocks with 64 weights Q2_K works on Metal and is very slightly faster than QK_K = 256 (23.8 ms vs 24.2 ms). * k_quants: WIP super-blocks with 64 weights Q3_K works on Metal and is slightly faster than QK_K = 256 (26.6 ms vs 28.3 ms). * k_quants: WIP super-blocks with 64 weights Q5_K works on Metal and is slightly faster than QK_K = 256 (23.7 ms vs 26.3 ms). * k_quants: call them _K, not _k, also on Metal * k_quants: correctly define QK_K in llama.cpp * Fixed bug in q4_K quantization added with the 64-block addition * Simplify via lambda * k_quants: swicth Q3_K to 4-bit scales when QK_K = 64 Otherwise there isn't much benefit from this quantization type. There is some very slight loss in accuracy, but we reduce size by ~7%. E.g., for OpenLLaMA-3B, Q3_K_S perplexity is 8.6131 with 8-bit scales and 8.6352 with 4-bit, while file size decreases from 1.53G to 1.44G. * k_quants: switch Q4_K to 4-bit scales when QK_K = 64 Here the loss in accuracy is greater than for Q3_K, but the Q4_K points still move further to the left on the perplexity vs size curve. * k_quants: forgot to add the Metal changes in last commit * k_quants: change Q5_K to be type 0 when QK_K = 64 Still needs AVX2 implementation * k_quants: AVX2 implementation for new 64-weight Q5_K * k_quants: 10% faster ARM_NEON Q5_K dot product * k_quants: fixed issue caused by merging with master --------- Co-authored-by: Iwan Kawrakow <iwan.kawrakow@gmail.com>
2023-06-26 16:43:07 +00:00
ggml : sync (im2col, GPU conv, 32-bit arm compat) (#4060) ggml-ci
2023-11-13 14:55:52 +00:00
q8bytes = ggml_vld1q_s8_x4(q8);
k-quants : support for super-block size of 64 (#2001) * k_quants: WIP super-blocks with 64 weights * k_quants: WIP super-blocks with 64 weights Q6_K scalar and AVX2 works * k_quants: WIP super-blocks with 64 weights Q4_K scalar and AVX2 works * k_quants: WIP super-blocks with 64 weights Q2_K scalar and AVX2 works. Q2_K is way too slow (it is actually slower than the scalar implementation) * k_quants: WIP super-blocks with 64 weights Q3_K scalar and AVX2 works. * k_quants: WIP super-blocks with 64 weights Q5_K scalar and AVX2 works, and with that all k_quants are done on AVX2 and scalar * k_quants: WIP super-blocks with 64 weights Q6_K working on CUDA. Cannot make it run quite as gast as with super-blocks with 256 weigths: 8% slower on 4080, 20% slower on the 1660 (but there we fit 1 less layer on the GPU because pf the larger model size), so some fraction of these 20% is due to that, * k_quants: WIP super-blocks with 64 weights Q4_K working on CUDA. ~10% slower on GTX-1660, 16% slower on 4080. * k_quants: WIP super-blocks with 64 weights Q2_K working on CUDA. ~3% slower on GTX-1660, 10% slower on 4080. * k_quants: WIP super-blocks with 64 weights Q3_K working on CUDA. * k_quants: WIP super-blocks with 64 weights Q5_K working on CUDA, and with this CUDA is done. * k_quants: WIP super-blocks with 64 weights Q6_K working on ARM_NEON * k_quants: WIP super-blocks with 64 weights Q4_K working on ARM_NEON, but quite a bit slower than 256 weights * k_quants: WIP super-blocks with 64 weights Q2_K working on ARM_NEON, but quite a bit slower than 256 weights * k_quants: WIP super-blocks with 64 weights Q3_K working on ARM_NEON, but quite a bit slower than 256 weights. * k_quants: WIP super-blocks with 64 weights Q5_K working on ARM_NEON, but quite a bit slower than 256 weights. With that, we have full support for ARM_NEON, although performance is not quite there. * k_quants: WIP super-blocks with 64 weights Slightly more efficient Q3_K and Q5_K * k_quants: WIP super-blocks with 64 weights Another small improvement for Q3_K and Q5_K on ARM_NEON * k_quants: WIP super-blocks with 64 weights Yet another speedup for Q5_K on ARM_NEON. We are now within 10% of the QK_K = 256 version. * k_quants: WIP super-blocks with 64 weights * We are able to pass preprocessor macros to the Metal compiler * Q6_K works and is actually slightly more efficient than the QK_K = 256 version (25.2 ms vs 25.8 ms) * k_quants: WIP super-blocks with 64 weights Q4_K works on Metal and is actually slightly faster than QK_K = 256 (21.95 ms vs 24.0 ms). * k_quants: WIP super-blocks with 64 weights Q2_K works on Metal and is very slightly faster than QK_K = 256 (23.8 ms vs 24.2 ms). * k_quants: WIP super-blocks with 64 weights Q3_K works on Metal and is slightly faster than QK_K = 256 (26.6 ms vs 28.3 ms). * k_quants: WIP super-blocks with 64 weights Q5_K works on Metal and is slightly faster than QK_K = 256 (23.7 ms vs 26.3 ms). * k_quants: call them _K, not _k, also on Metal * k_quants: correctly define QK_K in llama.cpp * Fixed bug in q4_K quantization added with the 64-block addition * Simplify via lambda * k_quants: swicth Q3_K to 4-bit scales when QK_K = 64 Otherwise there isn't much benefit from this quantization type. There is some very slight loss in accuracy, but we reduce size by ~7%. E.g., for OpenLLaMA-3B, Q3_K_S perplexity is 8.6131 with 8-bit scales and 8.6352 with 4-bit, while file size decreases from 1.53G to 1.44G. * k_quants: switch Q4_K to 4-bit scales when QK_K = 64 Here the loss in accuracy is greater than for Q3_K, but the Q4_K points still move further to the left on the perplexity vs size curve. * k_quants: forgot to add the Metal changes in last commit * k_quants: change Q5_K to be type 0 when QK_K = 64 Still needs AVX2 implementation * k_quants: AVX2 implementation for new 64-weight Q5_K * k_quants: 10% faster ARM_NEON Q5_K dot product * k_quants: fixed issue caused by merging with master --------- Co-authored-by: Iwan Kawrakow <iwan.kawrakow@gmail.com>
2023-06-26 16:43:07 +00:00
q4bytes.val[0] = vreinterpretq_s8_u8(vandq_u8 (q4bits.val[0], m4b));
q4bytes.val[1] = vreinterpretq_s8_u8(vandq_u8 (q4bits.val[1], m4b));
ggml : add ggml_vdotq_s32 alias (#4715) ggml-ci
2023-12-31 09:43:31 +00:00
const int32x4_t p1 = ggml_vdotq_s32(ggml_vdotq_s32(mzero, q4bytes.val[0], q8bytes.val[0]), q4bytes.val[1], q8bytes.val[1]);
k-quants : support for super-block size of 64 (#2001) * k_quants: WIP super-blocks with 64 weights * k_quants: WIP super-blocks with 64 weights Q6_K scalar and AVX2 works * k_quants: WIP super-blocks with 64 weights Q4_K scalar and AVX2 works * k_quants: WIP super-blocks with 64 weights Q2_K scalar and AVX2 works. Q2_K is way too slow (it is actually slower than the scalar implementation) * k_quants: WIP super-blocks with 64 weights Q3_K scalar and AVX2 works. * k_quants: WIP super-blocks with 64 weights Q5_K scalar and AVX2 works, and with that all k_quants are done on AVX2 and scalar * k_quants: WIP super-blocks with 64 weights Q6_K working on CUDA. Cannot make it run quite as gast as with super-blocks with 256 weigths: 8% slower on 4080, 20% slower on the 1660 (but there we fit 1 less layer on the GPU because pf the larger model size), so some fraction of these 20% is due to that, * k_quants: WIP super-blocks with 64 weights Q4_K working on CUDA. ~10% slower on GTX-1660, 16% slower on 4080. * k_quants: WIP super-blocks with 64 weights Q2_K working on CUDA. ~3% slower on GTX-1660, 10% slower on 4080. * k_quants: WIP super-blocks with 64 weights Q3_K working on CUDA. * k_quants: WIP super-blocks with 64 weights Q5_K working on CUDA, and with this CUDA is done. * k_quants: WIP super-blocks with 64 weights Q6_K working on ARM_NEON * k_quants: WIP super-blocks with 64 weights Q4_K working on ARM_NEON, but quite a bit slower than 256 weights * k_quants: WIP super-blocks with 64 weights Q2_K working on ARM_NEON, but quite a bit slower than 256 weights * k_quants: WIP super-blocks with 64 weights Q3_K working on ARM_NEON, but quite a bit slower than 256 weights. * k_quants: WIP super-blocks with 64 weights Q5_K working on ARM_NEON, but quite a bit slower than 256 weights. With that, we have full support for ARM_NEON, although performance is not quite there. * k_quants: WIP super-blocks with 64 weights Slightly more efficient Q3_K and Q5_K * k_quants: WIP super-blocks with 64 weights Another small improvement for Q3_K and Q5_K on ARM_NEON * k_quants: WIP super-blocks with 64 weights Yet another speedup for Q5_K on ARM_NEON. We are now within 10% of the QK_K = 256 version. * k_quants: WIP super-blocks with 64 weights * We are able to pass preprocessor macros to the Metal compiler * Q6_K works and is actually slightly more efficient than the QK_K = 256 version (25.2 ms vs 25.8 ms) * k_quants: WIP super-blocks with 64 weights Q4_K works on Metal and is actually slightly faster than QK_K = 256 (21.95 ms vs 24.0 ms). * k_quants: WIP super-blocks with 64 weights Q2_K works on Metal and is very slightly faster than QK_K = 256 (23.8 ms vs 24.2 ms). * k_quants: WIP super-blocks with 64 weights Q3_K works on Metal and is slightly faster than QK_K = 256 (26.6 ms vs 28.3 ms). * k_quants: WIP super-blocks with 64 weights Q5_K works on Metal and is slightly faster than QK_K = 256 (23.7 ms vs 26.3 ms). * k_quants: call them _K, not _k, also on Metal * k_quants: correctly define QK_K in llama.cpp * Fixed bug in q4_K quantization added with the 64-block addition * Simplify via lambda * k_quants: swicth Q3_K to 4-bit scales when QK_K = 64 Otherwise there isn't much benefit from this quantization type. There is some very slight loss in accuracy, but we reduce size by ~7%. E.g., for OpenLLaMA-3B, Q3_K_S perplexity is 8.6131 with 8-bit scales and 8.6352 with 4-bit, while file size decreases from 1.53G to 1.44G. * k_quants: switch Q4_K to 4-bit scales when QK_K = 64 Here the loss in accuracy is greater than for Q3_K, but the Q4_K points still move further to the left on the perplexity vs size curve. * k_quants: forgot to add the Metal changes in last commit * k_quants: change Q5_K to be type 0 when QK_K = 64 Still needs AVX2 implementation * k_quants: AVX2 implementation for new 64-weight Q5_K * k_quants: 10% faster ARM_NEON Q5_K dot product * k_quants: fixed issue caused by merging with master --------- Co-authored-by: Iwan Kawrakow <iwan.kawrakow@gmail.com>
2023-06-26 16:43:07 +00:00
const int32_t sumi1 = vaddvq_s32(p1) * scales[0];
q4bytes.val[0] = vreinterpretq_s8_u8(vshrq_n_u8(q4bits.val[0], 4));
q4bytes.val[1] = vreinterpretq_s8_u8(vshrq_n_u8(q4bits.val[1], 4));
ggml : add ggml_vdotq_s32 alias (#4715) ggml-ci
2023-12-31 09:43:31 +00:00
const int32x4_t p2 = ggml_vdotq_s32(ggml_vdotq_s32(mzero, q4bytes.val[0], q8bytes.val[2]), q4bytes.val[1], q8bytes.val[3]);
k-quants : support for super-block size of 64 (#2001) * k_quants: WIP super-blocks with 64 weights * k_quants: WIP super-blocks with 64 weights Q6_K scalar and AVX2 works * k_quants: WIP super-blocks with 64 weights Q4_K scalar and AVX2 works * k_quants: WIP super-blocks with 64 weights Q2_K scalar and AVX2 works. Q2_K is way too slow (it is actually slower than the scalar implementation) * k_quants: WIP super-blocks with 64 weights Q3_K scalar and AVX2 works. * k_quants: WIP super-blocks with 64 weights Q5_K scalar and AVX2 works, and with that all k_quants are done on AVX2 and scalar * k_quants: WIP super-blocks with 64 weights Q6_K working on CUDA. Cannot make it run quite as gast as with super-blocks with 256 weigths: 8% slower on 4080, 20% slower on the 1660 (but there we fit 1 less layer on the GPU because pf the larger model size), so some fraction of these 20% is due to that, * k_quants: WIP super-blocks with 64 weights Q4_K working on CUDA. ~10% slower on GTX-1660, 16% slower on 4080. * k_quants: WIP super-blocks with 64 weights Q2_K working on CUDA. ~3% slower on GTX-1660, 10% slower on 4080. * k_quants: WIP super-blocks with 64 weights Q3_K working on CUDA. * k_quants: WIP super-blocks with 64 weights Q5_K working on CUDA, and with this CUDA is done. * k_quants: WIP super-blocks with 64 weights Q6_K working on ARM_NEON * k_quants: WIP super-blocks with 64 weights Q4_K working on ARM_NEON, but quite a bit slower than 256 weights * k_quants: WIP super-blocks with 64 weights Q2_K working on ARM_NEON, but quite a bit slower than 256 weights * k_quants: WIP super-blocks with 64 weights Q3_K working on ARM_NEON, but quite a bit slower than 256 weights. * k_quants: WIP super-blocks with 64 weights Q5_K working on ARM_NEON, but quite a bit slower than 256 weights. With that, we have full support for ARM_NEON, although performance is not quite there. * k_quants: WIP super-blocks with 64 weights Slightly more efficient Q3_K and Q5_K * k_quants: WIP super-blocks with 64 weights Another small improvement for Q3_K and Q5_K on ARM_NEON * k_quants: WIP super-blocks with 64 weights Yet another speedup for Q5_K on ARM_NEON. We are now within 10% of the QK_K = 256 version. * k_quants: WIP super-blocks with 64 weights * We are able to pass preprocessor macros to the Metal compiler * Q6_K works and is actually slightly more efficient than the QK_K = 256 version (25.2 ms vs 25.8 ms) * k_quants: WIP super-blocks with 64 weights Q4_K works on Metal and is actually slightly faster than QK_K = 256 (21.95 ms vs 24.0 ms). * k_quants: WIP super-blocks with 64 weights Q2_K works on Metal and is very slightly faster than QK_K = 256 (23.8 ms vs 24.2 ms). * k_quants: WIP super-blocks with 64 weights Q3_K works on Metal and is slightly faster than QK_K = 256 (26.6 ms vs 28.3 ms). * k_quants: WIP super-blocks with 64 weights Q5_K works on Metal and is slightly faster than QK_K = 256 (23.7 ms vs 26.3 ms). * k_quants: call them _K, not _k, also on Metal * k_quants: correctly define QK_K in llama.cpp * Fixed bug in q4_K quantization added with the 64-block addition * Simplify via lambda * k_quants: swicth Q3_K to 4-bit scales when QK_K = 64 Otherwise there isn't much benefit from this quantization type. There is some very slight loss in accuracy, but we reduce size by ~7%. E.g., for OpenLLaMA-3B, Q3_K_S perplexity is 8.6131 with 8-bit scales and 8.6352 with 4-bit, while file size decreases from 1.53G to 1.44G. * k_quants: switch Q4_K to 4-bit scales when QK_K = 64 Here the loss in accuracy is greater than for Q3_K, but the Q4_K points still move further to the left on the perplexity vs size curve. * k_quants: forgot to add the Metal changes in last commit * k_quants: change Q5_K to be type 0 when QK_K = 64 Still needs AVX2 implementation * k_quants: AVX2 implementation for new 64-weight Q5_K * k_quants: 10% faster ARM_NEON Q5_K dot product * k_quants: fixed issue caused by merging with master --------- Co-authored-by: Iwan Kawrakow <iwan.kawrakow@gmail.com>
2023-06-26 16:43:07 +00:00
const int32_t sumi2 = vaddvq_s32(p2) * scales[1];
sumf += d * (sumi1 + sumi2);
}
*s = sumf - sum_mins;
#elif defined __AVX2__
const __m256i m4 = _mm256_set1_epi8(0xF);
__m256 acc = _mm256_setzero_ps();
float summs = 0;
uint16_t aux16[2];
const uint8_t * scales = (const uint8_t *)aux16;
for (int i = 0; i < nb; ++i) {
ggml : move FP16 <-> FP32 code to ggml-impl.h (#3861) * ggml : move FP16 <-> FP32 stuff to ggml-impl.h ggml-ci * tests : fix ARM build * ggml : explicitly initialize deprecated type traits * ggml : add math.h to ggml-impl.h * ggml : remove duplicate static assert macros * ggml : prefix lookup tables with ggml_ ggml-ci * ggml-impl : move extern "C" to start of file
2023-10-30 17:19:15 +00:00
const float d = GGML_FP16_TO_FP32(x[i].d[0]) * y[i].d;
const float m = GGML_FP16_TO_FP32(x[i].d[1]) * y[i].d;
k-quants : support for super-block size of 64 (#2001) * k_quants: WIP super-blocks with 64 weights * k_quants: WIP super-blocks with 64 weights Q6_K scalar and AVX2 works * k_quants: WIP super-blocks with 64 weights Q4_K scalar and AVX2 works * k_quants: WIP super-blocks with 64 weights Q2_K scalar and AVX2 works. Q2_K is way too slow (it is actually slower than the scalar implementation) * k_quants: WIP super-blocks with 64 weights Q3_K scalar and AVX2 works. * k_quants: WIP super-blocks with 64 weights Q5_K scalar and AVX2 works, and with that all k_quants are done on AVX2 and scalar * k_quants: WIP super-blocks with 64 weights Q6_K working on CUDA. Cannot make it run quite as gast as with super-blocks with 256 weigths: 8% slower on 4080, 20% slower on the 1660 (but there we fit 1 less layer on the GPU because pf the larger model size), so some fraction of these 20% is due to that, * k_quants: WIP super-blocks with 64 weights Q4_K working on CUDA. ~10% slower on GTX-1660, 16% slower on 4080. * k_quants: WIP super-blocks with 64 weights Q2_K working on CUDA. ~3% slower on GTX-1660, 10% slower on 4080. * k_quants: WIP super-blocks with 64 weights Q3_K working on CUDA. * k_quants: WIP super-blocks with 64 weights Q5_K working on CUDA, and with this CUDA is done. * k_quants: WIP super-blocks with 64 weights Q6_K working on ARM_NEON * k_quants: WIP super-blocks with 64 weights Q4_K working on ARM_NEON, but quite a bit slower than 256 weights * k_quants: WIP super-blocks with 64 weights Q2_K working on ARM_NEON, but quite a bit slower than 256 weights * k_quants: WIP super-blocks with 64 weights Q3_K working on ARM_NEON, but quite a bit slower than 256 weights. * k_quants: WIP super-blocks with 64 weights Q5_K working on ARM_NEON, but quite a bit slower than 256 weights. With that, we have full support for ARM_NEON, although performance is not quite there. * k_quants: WIP super-blocks with 64 weights Slightly more efficient Q3_K and Q5_K * k_quants: WIP super-blocks with 64 weights Another small improvement for Q3_K and Q5_K on ARM_NEON * k_quants: WIP super-blocks with 64 weights Yet another speedup for Q5_K on ARM_NEON. We are now within 10% of the QK_K = 256 version. * k_quants: WIP super-blocks with 64 weights * We are able to pass preprocessor macros to the Metal compiler * Q6_K works and is actually slightly more efficient than the QK_K = 256 version (25.2 ms vs 25.8 ms) * k_quants: WIP super-blocks with 64 weights Q4_K works on Metal and is actually slightly faster than QK_K = 256 (21.95 ms vs 24.0 ms). * k_quants: WIP super-blocks with 64 weights Q2_K works on Metal and is very slightly faster than QK_K = 256 (23.8 ms vs 24.2 ms). * k_quants: WIP super-blocks with 64 weights Q3_K works on Metal and is slightly faster than QK_K = 256 (26.6 ms vs 28.3 ms). * k_quants: WIP super-blocks with 64 weights Q5_K works on Metal and is slightly faster than QK_K = 256 (23.7 ms vs 26.3 ms). * k_quants: call them _K, not _k, also on Metal * k_quants: correctly define QK_K in llama.cpp * Fixed bug in q4_K quantization added with the 64-block addition * Simplify via lambda * k_quants: swicth Q3_K to 4-bit scales when QK_K = 64 Otherwise there isn't much benefit from this quantization type. There is some very slight loss in accuracy, but we reduce size by ~7%. E.g., for OpenLLaMA-3B, Q3_K_S perplexity is 8.6131 with 8-bit scales and 8.6352 with 4-bit, while file size decreases from 1.53G to 1.44G. * k_quants: switch Q4_K to 4-bit scales when QK_K = 64 Here the loss in accuracy is greater than for Q3_K, but the Q4_K points still move further to the left on the perplexity vs size curve. * k_quants: forgot to add the Metal changes in last commit * k_quants: change Q5_K to be type 0 when QK_K = 64 Still needs AVX2 implementation * k_quants: AVX2 implementation for new 64-weight Q5_K * k_quants: 10% faster ARM_NEON Q5_K dot product * k_quants: fixed issue caused by merging with master --------- Co-authored-by: Iwan Kawrakow <iwan.kawrakow@gmail.com>
2023-06-26 16:43:07 +00:00
const __m256 vd = _mm256_set1_ps(d);
const uint16_t * a = (const uint16_t *)x[i].scales;
aux16[0] = a[0] & 0x0f0f;
aux16[1] = (a[0] >> 4) & 0x0f0f;
summs += m * (scales[2] * (y[i].bsums[0] + y[i].bsums[1]) + scales[3] * (y[i].bsums[2] + y[i].bsums[3]));
const uint8_t * restrict q4 = x[i].qs;
const int8_t * restrict q8 = y[i].qs;
const __m256i q4bits = _mm256_loadu_si256((const __m256i*)q4);
const __m256i q4l = _mm256_and_si256(q4bits, m4);
const __m256i q4h = _mm256_and_si256(_mm256_srli_epi16(q4bits, 4), m4);
const __m256i q8l = _mm256_loadu_si256((const __m256i*)(q8+ 0));
const __m256i q8h = _mm256_loadu_si256((const __m256i*)(q8+32));
const __m256i p16l = _mm256_maddubs_epi16(q4l, q8l);
const __m256i p16h = _mm256_maddubs_epi16(q4h, q8h);
const __m256i p32l = _mm256_madd_epi16(_mm256_set1_epi16(scales[0]), p16l);
acc = _mm256_fmadd_ps(vd, _mm256_cvtepi32_ps(p32l), acc);
const __m256i p32h = _mm256_madd_epi16(_mm256_set1_epi16(scales[1]), p16h);
acc = _mm256_fmadd_ps(vd, _mm256_cvtepi32_ps(p32h), acc);
}
*s = hsum_float_8(acc) - summs;
k_quants : add AVX support to dot functions with QK_K as 64 (#2339) * add AVX to ggml_vec_dot_q2_K_q8_K() * add AVX to ggml_vec_dot_q3_K_q8_K() * add AVX to ggml_vec_dot_q4_K_q8_K() * add AVX to ggml_vec_dot_q5_K_q8_K() * add AVX to ggml_vec_dot_q6_K_q8_K() * refactor AVX code in ggml_vec_dot_q6_K_q8_K()
2023-07-25 12:13:41 +00:00
#elif defined __AVX__
const __m128i m4 = _mm_set1_epi8(0xF);
__m256 acc = _mm256_setzero_ps();
float summs = 0;
uint16_t aux16[2];
const uint8_t * scales = (const uint8_t *)aux16;
for (int i = 0; i < nb; ++i) {
ggml : move FP16 <-> FP32 code to ggml-impl.h (#3861) * ggml : move FP16 <-> FP32 stuff to ggml-impl.h ggml-ci * tests : fix ARM build * ggml : explicitly initialize deprecated type traits * ggml : add math.h to ggml-impl.h * ggml : remove duplicate static assert macros * ggml : prefix lookup tables with ggml_ ggml-ci * ggml-impl : move extern "C" to start of file
2023-10-30 17:19:15 +00:00
const float d = GGML_FP16_TO_FP32(x[i].d[0]) * y[i].d;
const float m = GGML_FP16_TO_FP32(x[i].d[1]) * y[i].d;
k_quants : add AVX support to dot functions with QK_K as 64 (#2339) * add AVX to ggml_vec_dot_q2_K_q8_K() * add AVX to ggml_vec_dot_q3_K_q8_K() * add AVX to ggml_vec_dot_q4_K_q8_K() * add AVX to ggml_vec_dot_q5_K_q8_K() * add AVX to ggml_vec_dot_q6_K_q8_K() * refactor AVX code in ggml_vec_dot_q6_K_q8_K()
2023-07-25 12:13:41 +00:00
const __m256 vd = _mm256_set1_ps(d);
const uint16_t * a = (const uint16_t *)x[i].scales;
aux16[0] = a[0] & 0x0f0f;
aux16[1] = (a[0] >> 4) & 0x0f0f;
summs += m * (scales[2] * (y[i].bsums[0] + y[i].bsums[1]) + scales[3] * (y[i].bsums[2] + y[i].bsums[3]));
const uint8_t * restrict q4 = x[i].qs;
const int8_t * restrict q8 = y[i].qs;
const __m256i q4bits = _mm256_loadu_si256((const __m256i*)q4);
const __m128i q4bits_0 = _mm256_extractf128_si256(q4bits, 0);
const __m128i q4bits_1 = _mm256_extractf128_si256(q4bits, 1);
const __m128i q4_0 = _mm_and_si128(q4bits_0, m4);
const __m128i q4_1 = _mm_and_si128(q4bits_1, m4);
const __m128i q4_2 = _mm_and_si128(_mm_srli_epi16(q4bits_0, 4), m4);
const __m128i q4_3 = _mm_and_si128(_mm_srli_epi16(q4bits_1, 4), m4);
const __m256i q8_0 = _mm256_loadu_si256((const __m256i*)(q8+ 0));
const __m256i q8_1 = _mm256_loadu_si256((const __m256i*)(q8+32));
const __m128i p16_0 = _mm_maddubs_epi16(q4_0, _mm256_extractf128_si256(q8_0, 0));
const __m128i p16_1 = _mm_maddubs_epi16(q4_1, _mm256_extractf128_si256(q8_0, 1));
const __m128i p16_2 = _mm_maddubs_epi16(q4_2, _mm256_extractf128_si256(q8_1, 0));
const __m128i p16_3 = _mm_maddubs_epi16(q4_3, _mm256_extractf128_si256(q8_1, 1));
const __m128i p32_0 = _mm_madd_epi16(_mm_set1_epi16(scales[0]), p16_0);
const __m128i p32_1 = _mm_madd_epi16(_mm_set1_epi16(scales[0]), p16_1);
ggml : workaround for missing _mm256_setr_m128i in GCC < 8 in k_quants.c (#2405)
2023-07-28 18:17:45 +00:00
acc = _mm256_add_ps(_mm256_mul_ps(vd, _mm256_cvtepi32_ps(MM256_SET_M128I(p32_1, p32_0))), acc);
k_quants : add AVX support to dot functions with QK_K as 64 (#2339) * add AVX to ggml_vec_dot_q2_K_q8_K() * add AVX to ggml_vec_dot_q3_K_q8_K() * add AVX to ggml_vec_dot_q4_K_q8_K() * add AVX to ggml_vec_dot_q5_K_q8_K() * add AVX to ggml_vec_dot_q6_K_q8_K() * refactor AVX code in ggml_vec_dot_q6_K_q8_K()
2023-07-25 12:13:41 +00:00
const __m128i p32_2 = _mm_madd_epi16(_mm_set1_epi16(scales[1]), p16_2);
const __m128i p32_3 = _mm_madd_epi16(_mm_set1_epi16(scales[1]), p16_3);
ggml : workaround for missing _mm256_setr_m128i in GCC < 8 in k_quants.c (#2405)
2023-07-28 18:17:45 +00:00
acc = _mm256_add_ps(_mm256_mul_ps(vd, _mm256_cvtepi32_ps(MM256_SET_M128I(p32_3, p32_2))), acc);
k_quants : add AVX support to dot functions with QK_K as 64 (#2339) * add AVX to ggml_vec_dot_q2_K_q8_K() * add AVX to ggml_vec_dot_q3_K_q8_K() * add AVX to ggml_vec_dot_q4_K_q8_K() * add AVX to ggml_vec_dot_q5_K_q8_K() * add AVX to ggml_vec_dot_q6_K_q8_K() * refactor AVX code in ggml_vec_dot_q6_K_q8_K()
2023-07-25 12:13:41 +00:00
}
*s = hsum_float_8(acc) - summs;
ggml : add RISC-V Vector Support for K-Quants and improved the existing intrinsics (#3453) * Added RVV intrinsics support for Q8 quantize row and also improved the existing dot product function for risc-v. The RVV intrinsics is added for the following quantize row functions quantize_row_q8_0 quantize_row_q8_1 The following dot product functions have also been optimized by using LMUL = 1/2 instead of LMUL = 1 ggml_vec_dot_q4_0_q8_0 ggml_vec_dot_q4_1_q8_1 ggml_vec_dot_q5_0_q8_0 ggml_vec_dot_q5_1_q8_1 And vector initialization in Q5 by temporary array is also replaced by the vid intrinsics Signed-off-by: Ahmad Tameem <ahmad.tameem@10xengineers.ai> * Added RVV intrinsics support for k_quants This adds RISC-V Vector intrinsics support for the following K_quants functions for both QKK = 256 and QKK = 64 ggml_vec_dot_q2_K_q8_K ggml_vec_dot_q3_K_q8_K ggml_vec_dot_q4_K_q8_K ggml_vec_dot_q5_K_q8_K ggml_vec_dot_q6_K_q8_K Signed-off-by: Ahmad Tameem <ahmad.tameem@10xengineers.ai> --------- Signed-off-by: Ahmad Tameem <ahmad.tameem@10xengineers.ai>
2023-10-03 18:38:19 +00:00
#elif defined __riscv_v_intrinsic
uint16_t s16[2];
const uint8_t * restrict scales = (const uint8_t *)s16;
float sumf = 0;
for (int i = 0; i < nb; ++i) {
const uint8_t * restrict q4 = x[i].qs;
const int8_t * restrict q8 = y[i].qs;
const uint16_t * restrict b = (const uint16_t *)x[i].scales;
s16[0] = b[0] & 0x0f0f;
s16[1] = (b[0] >> 4) & 0x0f0f;
ggml : move FP16 <-> FP32 code to ggml-impl.h (#3861) * ggml : move FP16 <-> FP32 stuff to ggml-impl.h ggml-ci * tests : fix ARM build * ggml : explicitly initialize deprecated type traits * ggml : add math.h to ggml-impl.h * ggml : remove duplicate static assert macros * ggml : prefix lookup tables with ggml_ ggml-ci * ggml-impl : move extern "C" to start of file
2023-10-30 17:19:15 +00:00
sumf -= y[i].d * GGML_FP16_TO_FP32(x[i].d[1]) * (scales[2] * (y[i].bsums[0] + y[i].bsums[1]) + scales[3] * (y[i].bsums[2] + y[i].bsums[3]));
const float d = y[i].d * GGML_FP16_TO_FP32(x[i].d[0]);
ggml : add RISC-V Vector Support for K-Quants and improved the existing intrinsics (#3453) * Added RVV intrinsics support for Q8 quantize row and also improved the existing dot product function for risc-v. The RVV intrinsics is added for the following quantize row functions quantize_row_q8_0 quantize_row_q8_1 The following dot product functions have also been optimized by using LMUL = 1/2 instead of LMUL = 1 ggml_vec_dot_q4_0_q8_0 ggml_vec_dot_q4_1_q8_1 ggml_vec_dot_q5_0_q8_0 ggml_vec_dot_q5_1_q8_1 And vector initialization in Q5 by temporary array is also replaced by the vid intrinsics Signed-off-by: Ahmad Tameem <ahmad.tameem@10xengineers.ai> * Added RVV intrinsics support for k_quants This adds RISC-V Vector intrinsics support for the following K_quants functions for both QKK = 256 and QKK = 64 ggml_vec_dot_q2_K_q8_K ggml_vec_dot_q3_K_q8_K ggml_vec_dot_q4_K_q8_K ggml_vec_dot_q5_K_q8_K ggml_vec_dot_q6_K_q8_K Signed-off-by: Ahmad Tameem <ahmad.tameem@10xengineers.ai> --------- Signed-off-by: Ahmad Tameem <ahmad.tameem@10xengineers.ai>
2023-10-03 18:38:19 +00:00
size_t vl = 32;
vint16m1_t vzero = __riscv_vmv_v_x_i16m1(0, 1);
// load Q4
vuint8m1_t q4_x = __riscv_vle8_v_u8m1(q4, vl);
// load Q8 and multiply it with lower Q4 nibble
vint8m1_t q4_a = __riscv_vreinterpret_v_u8m1_i8m1(__riscv_vand_vx_u8m1(q4_x, 0x0F, vl));
vint16m2_t va_0 = __riscv_vwmul_vv_i16m2(q4_a, __riscv_vle8_v_i8m1(q8, vl), vl);
vint16m1_t aux1 = __riscv_vredsum_vs_i16m2_i16m1(va_0, vzero, vl);
sumf += d*scales[0]*__riscv_vmv_x_s_i16m1_i16(aux1);
// load Q8 and multiply it with upper Q4 nibble
vint8m1_t q4_s = __riscv_vreinterpret_v_u8m1_i8m1(__riscv_vsrl_vx_u8m1(q4_x, 0x04, vl));
vint16m2_t va_1 = __riscv_vwmul_vv_i16m2(q4_s, __riscv_vle8_v_i8m1(q8+32, vl), vl);
vint16m1_t aux2 = __riscv_vredsum_vs_i16m2_i16m1(va_1, vzero, vl);
sumf += d*scales[1]*__riscv_vmv_x_s_i16m1_i16(aux2);
}
*s = sumf;
k-quants : support for super-block size of 64 (#2001) * k_quants: WIP super-blocks with 64 weights * k_quants: WIP super-blocks with 64 weights Q6_K scalar and AVX2 works * k_quants: WIP super-blocks with 64 weights Q4_K scalar and AVX2 works * k_quants: WIP super-blocks with 64 weights Q2_K scalar and AVX2 works. Q2_K is way too slow (it is actually slower than the scalar implementation) * k_quants: WIP super-blocks with 64 weights Q3_K scalar and AVX2 works. * k_quants: WIP super-blocks with 64 weights Q5_K scalar and AVX2 works, and with that all k_quants are done on AVX2 and scalar * k_quants: WIP super-blocks with 64 weights Q6_K working on CUDA. Cannot make it run quite as gast as with super-blocks with 256 weigths: 8% slower on 4080, 20% slower on the 1660 (but there we fit 1 less layer on the GPU because pf the larger model size), so some fraction of these 20% is due to that, * k_quants: WIP super-blocks with 64 weights Q4_K working on CUDA. ~10% slower on GTX-1660, 16% slower on 4080. * k_quants: WIP super-blocks with 64 weights Q2_K working on CUDA. ~3% slower on GTX-1660, 10% slower on 4080. * k_quants: WIP super-blocks with 64 weights Q3_K working on CUDA. * k_quants: WIP super-blocks with 64 weights Q5_K working on CUDA, and with this CUDA is done. * k_quants: WIP super-blocks with 64 weights Q6_K working on ARM_NEON * k_quants: WIP super-blocks with 64 weights Q4_K working on ARM_NEON, but quite a bit slower than 256 weights * k_quants: WIP super-blocks with 64 weights Q2_K working on ARM_NEON, but quite a bit slower than 256 weights * k_quants: WIP super-blocks with 64 weights Q3_K working on ARM_NEON, but quite a bit slower than 256 weights. * k_quants: WIP super-blocks with 64 weights Q5_K working on ARM_NEON, but quite a bit slower than 256 weights. With that, we have full support for ARM_NEON, although performance is not quite there. * k_quants: WIP super-blocks with 64 weights Slightly more efficient Q3_K and Q5_K * k_quants: WIP super-blocks with 64 weights Another small improvement for Q3_K and Q5_K on ARM_NEON * k_quants: WIP super-blocks with 64 weights Yet another speedup for Q5_K on ARM_NEON. We are now within 10% of the QK_K = 256 version. * k_quants: WIP super-blocks with 64 weights * We are able to pass preprocessor macros to the Metal compiler * Q6_K works and is actually slightly more efficient than the QK_K = 256 version (25.2 ms vs 25.8 ms) * k_quants: WIP super-blocks with 64 weights Q4_K works on Metal and is actually slightly faster than QK_K = 256 (21.95 ms vs 24.0 ms). * k_quants: WIP super-blocks with 64 weights Q2_K works on Metal and is very slightly faster than QK_K = 256 (23.8 ms vs 24.2 ms). * k_quants: WIP super-blocks with 64 weights Q3_K works on Metal and is slightly faster than QK_K = 256 (26.6 ms vs 28.3 ms). * k_quants: WIP super-blocks with 64 weights Q5_K works on Metal and is slightly faster than QK_K = 256 (23.7 ms vs 26.3 ms). * k_quants: call them _K, not _k, also on Metal * k_quants: correctly define QK_K in llama.cpp * Fixed bug in q4_K quantization added with the 64-block addition * Simplify via lambda * k_quants: swicth Q3_K to 4-bit scales when QK_K = 64 Otherwise there isn't much benefit from this quantization type. There is some very slight loss in accuracy, but we reduce size by ~7%. E.g., for OpenLLaMA-3B, Q3_K_S perplexity is 8.6131 with 8-bit scales and 8.6352 with 4-bit, while file size decreases from 1.53G to 1.44G. * k_quants: switch Q4_K to 4-bit scales when QK_K = 64 Here the loss in accuracy is greater than for Q3_K, but the Q4_K points still move further to the left on the perplexity vs size curve. * k_quants: forgot to add the Metal changes in last commit * k_quants: change Q5_K to be type 0 when QK_K = 64 Still needs AVX2 implementation * k_quants: AVX2 implementation for new 64-weight Q5_K * k_quants: 10% faster ARM_NEON Q5_K dot product * k_quants: fixed issue caused by merging with master --------- Co-authored-by: Iwan Kawrakow <iwan.kawrakow@gmail.com>
2023-06-26 16:43:07 +00:00
#else
uint8_t aux8[QK_K];
int16_t aux16[16];
float sums [8];
memset(sums, 0, 8*sizeof(float));
uint16_t s16[2];
const uint8_t * restrict scales = (const uint8_t *)s16;
float sumf = 0;
for (int i = 0; i < nb; ++i) {
const uint8_t * restrict q4 = x[i].qs;
const int8_t * restrict q8 = y[i].qs;
uint8_t * restrict a = aux8;
for (int l = 0; l < 32; ++l) a[l+ 0] = q4[l] & 0xF;
for (int l = 0; l < 32; ++l) a[l+32] = q4[l] >> 4;
const uint16_t * restrict b = (const uint16_t *)x[i].scales;
s16[0] = b[0] & 0x0f0f;
s16[1] = (b[0] >> 4) & 0x0f0f;
ggml : move FP16 <-> FP32 code to ggml-impl.h (#3861) * ggml : move FP16 <-> FP32 stuff to ggml-impl.h ggml-ci * tests : fix ARM build * ggml : explicitly initialize deprecated type traits * ggml : add math.h to ggml-impl.h * ggml : remove duplicate static assert macros * ggml : prefix lookup tables with ggml_ ggml-ci * ggml-impl : move extern "C" to start of file
2023-10-30 17:19:15 +00:00
sumf -= y[i].d * GGML_FP16_TO_FP32(x[i].d[1]) * (scales[2] * (y[i].bsums[0] + y[i].bsums[1]) + scales[3] * (y[i].bsums[2] + y[i].bsums[3]));
k-quants : support for super-block size of 64 (#2001) * k_quants: WIP super-blocks with 64 weights * k_quants: WIP super-blocks with 64 weights Q6_K scalar and AVX2 works * k_quants: WIP super-blocks with 64 weights Q4_K scalar and AVX2 works * k_quants: WIP super-blocks with 64 weights Q2_K scalar and AVX2 works. Q2_K is way too slow (it is actually slower than the scalar implementation) * k_quants: WIP super-blocks with 64 weights Q3_K scalar and AVX2 works. * k_quants: WIP super-blocks with 64 weights Q5_K scalar and AVX2 works, and with that all k_quants are done on AVX2 and scalar * k_quants: WIP super-blocks with 64 weights Q6_K working on CUDA. Cannot make it run quite as gast as with super-blocks with 256 weigths: 8% slower on 4080, 20% slower on the 1660 (but there we fit 1 less layer on the GPU because pf the larger model size), so some fraction of these 20% is due to that, * k_quants: WIP super-blocks with 64 weights Q4_K working on CUDA. ~10% slower on GTX-1660, 16% slower on 4080. * k_quants: WIP super-blocks with 64 weights Q2_K working on CUDA. ~3% slower on GTX-1660, 10% slower on 4080. * k_quants: WIP super-blocks with 64 weights Q3_K working on CUDA. * k_quants: WIP super-blocks with 64 weights Q5_K working on CUDA, and with this CUDA is done. * k_quants: WIP super-blocks with 64 weights Q6_K working on ARM_NEON * k_quants: WIP super-blocks with 64 weights Q4_K working on ARM_NEON, but quite a bit slower than 256 weights * k_quants: WIP super-blocks with 64 weights Q2_K working on ARM_NEON, but quite a bit slower than 256 weights * k_quants: WIP super-blocks with 64 weights Q3_K working on ARM_NEON, but quite a bit slower than 256 weights. * k_quants: WIP super-blocks with 64 weights Q5_K working on ARM_NEON, but quite a bit slower than 256 weights. With that, we have full support for ARM_NEON, although performance is not quite there. * k_quants: WIP super-blocks with 64 weights Slightly more efficient Q3_K and Q5_K * k_quants: WIP super-blocks with 64 weights Another small improvement for Q3_K and Q5_K on ARM_NEON * k_quants: WIP super-blocks with 64 weights Yet another speedup for Q5_K on ARM_NEON. We are now within 10% of the QK_K = 256 version. * k_quants: WIP super-blocks with 64 weights * We are able to pass preprocessor macros to the Metal compiler * Q6_K works and is actually slightly more efficient than the QK_K = 256 version (25.2 ms vs 25.8 ms) * k_quants: WIP super-blocks with 64 weights Q4_K works on Metal and is actually slightly faster than QK_K = 256 (21.95 ms vs 24.0 ms). * k_quants: WIP super-blocks with 64 weights Q2_K works on Metal and is very slightly faster than QK_K = 256 (23.8 ms vs 24.2 ms). * k_quants: WIP super-blocks with 64 weights Q3_K works on Metal and is slightly faster than QK_K = 256 (26.6 ms vs 28.3 ms). * k_quants: WIP super-blocks with 64 weights Q5_K works on Metal and is slightly faster than QK_K = 256 (23.7 ms vs 26.3 ms). * k_quants: call them _K, not _k, also on Metal * k_quants: correctly define QK_K in llama.cpp * Fixed bug in q4_K quantization added with the 64-block addition * Simplify via lambda * k_quants: swicth Q3_K to 4-bit scales when QK_K = 64 Otherwise there isn't much benefit from this quantization type. There is some very slight loss in accuracy, but we reduce size by ~7%. E.g., for OpenLLaMA-3B, Q3_K_S perplexity is 8.6131 with 8-bit scales and 8.6352 with 4-bit, while file size decreases from 1.53G to 1.44G. * k_quants: switch Q4_K to 4-bit scales when QK_K = 64 Here the loss in accuracy is greater than for Q3_K, but the Q4_K points still move further to the left on the perplexity vs size curve. * k_quants: forgot to add the Metal changes in last commit * k_quants: change Q5_K to be type 0 when QK_K = 64 Still needs AVX2 implementation * k_quants: AVX2 implementation for new 64-weight Q5_K * k_quants: 10% faster ARM_NEON Q5_K dot product * k_quants: fixed issue caused by merging with master --------- Co-authored-by: Iwan Kawrakow <iwan.kawrakow@gmail.com>
2023-06-26 16:43:07 +00:00
ggml : move FP16 <-> FP32 code to ggml-impl.h (#3861) * ggml : move FP16 <-> FP32 stuff to ggml-impl.h ggml-ci * tests : fix ARM build * ggml : explicitly initialize deprecated type traits * ggml : add math.h to ggml-impl.h * ggml : remove duplicate static assert macros * ggml : prefix lookup tables with ggml_ ggml-ci * ggml-impl : move extern "C" to start of file
2023-10-30 17:19:15 +00:00
const float d = y[i].d * GGML_FP16_TO_FP32(x[i].d[0]);
k-quants : support for super-block size of 64 (#2001) * k_quants: WIP super-blocks with 64 weights * k_quants: WIP super-blocks with 64 weights Q6_K scalar and AVX2 works * k_quants: WIP super-blocks with 64 weights Q4_K scalar and AVX2 works * k_quants: WIP super-blocks with 64 weights Q2_K scalar and AVX2 works. Q2_K is way too slow (it is actually slower than the scalar implementation) * k_quants: WIP super-blocks with 64 weights Q3_K scalar and AVX2 works. * k_quants: WIP super-blocks with 64 weights Q5_K scalar and AVX2 works, and with that all k_quants are done on AVX2 and scalar * k_quants: WIP super-blocks with 64 weights Q6_K working on CUDA. Cannot make it run quite as gast as with super-blocks with 256 weigths: 8% slower on 4080, 20% slower on the 1660 (but there we fit 1 less layer on the GPU because pf the larger model size), so some fraction of these 20% is due to that, * k_quants: WIP super-blocks with 64 weights Q4_K working on CUDA. ~10% slower on GTX-1660, 16% slower on 4080. * k_quants: WIP super-blocks with 64 weights Q2_K working on CUDA. ~3% slower on GTX-1660, 10% slower on 4080. * k_quants: WIP super-blocks with 64 weights Q3_K working on CUDA. * k_quants: WIP super-blocks with 64 weights Q5_K working on CUDA, and with this CUDA is done. * k_quants: WIP super-blocks with 64 weights Q6_K working on ARM_NEON * k_quants: WIP super-blocks with 64 weights Q4_K working on ARM_NEON, but quite a bit slower than 256 weights * k_quants: WIP super-blocks with 64 weights Q2_K working on ARM_NEON, but quite a bit slower than 256 weights * k_quants: WIP super-blocks with 64 weights Q3_K working on ARM_NEON, but quite a bit slower than 256 weights. * k_quants: WIP super-blocks with 64 weights Q5_K working on ARM_NEON, but quite a bit slower than 256 weights. With that, we have full support for ARM_NEON, although performance is not quite there. * k_quants: WIP super-blocks with 64 weights Slightly more efficient Q3_K and Q5_K * k_quants: WIP super-blocks with 64 weights Another small improvement for Q3_K and Q5_K on ARM_NEON * k_quants: WIP super-blocks with 64 weights Yet another speedup for Q5_K on ARM_NEON. We are now within 10% of the QK_K = 256 version. * k_quants: WIP super-blocks with 64 weights * We are able to pass preprocessor macros to the Metal compiler * Q6_K works and is actually slightly more efficient than the QK_K = 256 version (25.2 ms vs 25.8 ms) * k_quants: WIP super-blocks with 64 weights Q4_K works on Metal and is actually slightly faster than QK_K = 256 (21.95 ms vs 24.0 ms). * k_quants: WIP super-blocks with 64 weights Q2_K works on Metal and is very slightly faster than QK_K = 256 (23.8 ms vs 24.2 ms). * k_quants: WIP super-blocks with 64 weights Q3_K works on Metal and is slightly faster than QK_K = 256 (26.6 ms vs 28.3 ms). * k_quants: WIP super-blocks with 64 weights Q5_K works on Metal and is slightly faster than QK_K = 256 (23.7 ms vs 26.3 ms). * k_quants: call them _K, not _k, also on Metal * k_quants: correctly define QK_K in llama.cpp * Fixed bug in q4_K quantization added with the 64-block addition * Simplify via lambda * k_quants: swicth Q3_K to 4-bit scales when QK_K = 64 Otherwise there isn't much benefit from this quantization type. There is some very slight loss in accuracy, but we reduce size by ~7%. E.g., for OpenLLaMA-3B, Q3_K_S perplexity is 8.6131 with 8-bit scales and 8.6352 with 4-bit, while file size decreases from 1.53G to 1.44G. * k_quants: switch Q4_K to 4-bit scales when QK_K = 64 Here the loss in accuracy is greater than for Q3_K, but the Q4_K points still move further to the left on the perplexity vs size curve. * k_quants: forgot to add the Metal changes in last commit * k_quants: change Q5_K to be type 0 when QK_K = 64 Still needs AVX2 implementation * k_quants: AVX2 implementation for new 64-weight Q5_K * k_quants: 10% faster ARM_NEON Q5_K dot product * k_quants: fixed issue caused by merging with master --------- Co-authored-by: Iwan Kawrakow <iwan.kawrakow@gmail.com>
2023-06-26 16:43:07 +00:00
for (int j = 0; j < QK_K/32; ++j) {
for (int l = 0; l < 16; ++l) aux16[l] = q8[l] * a[l];
q8 += 16; a += 16;
for (int l = 0; l < 16; ++l) aux16[l] += q8[l] * a[l];
q8 += 16; a += 16;
const float dl = d * scales[j];
for (int l = 0; l < 8; ++l) sums[l] += dl * (aux16[l] + aux16[l+8]);
}
}
for (int l = 0; l < 8; ++l) sumf += sums[l];
*s = sumf;
#endif
}
#endif
#if QK_K == 256
k-quants : allow to optionally disable at compile time (#1734) * k-quants : put behind optional compile flag LLAMA_K_QUANTS * build : enable k-quants by default
2023-06-07 07:59:52 +00:00
void ggml_vec_dot_q5_K_q8_K(const int n, float * restrict s, const void * restrict vx, const void * restrict vy) {
ggml : add SOTA 2,3,4,5,6 bit k-quantizations (#1684) * Starting to add k-quantization to ggml I think it is better to have quantization separate from ggml. For now just adding the k-quants there, but it would be better to also factor out the existing ggml quantizations. * Adding Q3_K and Q8_K (de)-quantization * Q3_K now working on CUDA and AVX2/scalar CUDA is not ideal - ~50% slower than Q4_0 for single token prediction, about the same in batch mode (perplexity). CPU single token is ~55 ms (on Ryzen 7950X). * Some improvement for Q3_K on CUDA It is now ~22.5 ms/token on my GPU, so ~30% slower than Q4_0. * Some more CUDA optimizations for Q3_K Single token is now 20.5 ms/token (~20% slower than Q4_0). Perplexity is on par with Q4_0. * Adding Q4_K - scalar, AVX2, CUDA Performance is the same or perhaps very slightly better than Q4_0 on the CPU. On the GPU, single token prediction is ~10% better than Q4_0, batch mode (perplexity is about the same). * Adding Q6_K - scalar, AVX2, CUDA Performance is ~40% lower compared to Q4_K on the CPU. This is to be expected, considering that we are memory bound on the CPU and the 6-bit model is ~44% larger than the 4-bit. On the GPU, single token prediction is ~6% lower than Q4_0, batch mode (perplexity) is even closer (but still slower). * Adding Q5_K - scalar, AVX2, CUDA Performance is ~20% lower compared to Q4_K on the CPU. This is to be expected, considering that we are memory bound on the CPU and the 5-bit model is ~22% larger than the 4-bit. On the GPU, single token prediction is about the same as Q4_0 for both, single token and batch prediction. * Per convention, all QX_K quantizations use Q5_K for output.weight * Adding quantization mixes * Quantization mixes: didn't quite get what I wanted in the last commit * Q4_K dot product for ARM_NEON * Q6_K dot product for ARM_NEON * Q5_K dot product for ARM_NEON * Adding Q3_K dot for ARM_NEON It is 22% slower than Q4_K, despite the smaller model size. On x86_64, where we are memory bound, the Q3_K model is quite a bit faster than Q4_K. * A very slightly faster ARM_NEON Q3_K dot * Adding Q2_K - just CUDA for now Token prediction is pretty good - about 15.5 ms on a RTX 4080. Perplexity is about the same as Q4_K. * Adding scalar and AVX2 Q2_K dot * Adding ARM_NEON Q2_K dot About the same performance as Q4_K. * A slightly faster ARM_NEON Q2_K dot Single token prediction is now ~36 ms on M2 Max. The code is much simpler too. * Fixed bug in Q2_K CUDA dot product kernel Stranegly enough, for the few prompts I tried with the 7B model the responses looked perfectly reasonable. Only realized something is not quite right when I tried the larger models and started getting nonse back. In any case, Q2_K single token evaluation time on an RTX 4080 in a Ryzen7950X box iusing CUDA and model fully loaded on the GPU are ~15.5 ms for 7B, ~25.4 ms for 13B, and ~55.8 ms for 30B. The max number of layers that fit in VRAM for The 65B is 32. With that, we get ~330 ms per token, which is not that much faster than just running on the CPU (~470 ms per token). * Don't print zeros/NaNs when no count histogram has been collected * A 10% faster CUDA vector dot kernel for Q3_K Q3_K is now running at ~18.5 ms / token on CUDA, so the gap to Q4_0 is only 10%. It seems memory acccess pattern is more important for performance than the amount of computation the kernel does. * A slightly daster Q4_K AVX2 dot product For perplexity, where we are less memory bound, time per pass drops by ~5%. Barely measurable difference for single token prediction. * A slightly faster ARM_NEON A4_K dot product * Minor * Fix quantization error test We cannot possibly be expecting rmse < 0.002 for 2- and 3-bit quantization variants. * Fix docker build I have been sloppy with vector reinterpret casts on ARM_NEON. It seems clang is very forgiving in that regard. * Added forgotten ggml.o dependence on k_quants.h to the Makefile * Had unintentionally committed the Makefile with -Ofast enabled * ggml : rename k_quants -> ggml-quants-k, use lowercase in code --------- Co-authored-by: Iwan Kawrakow <iwan.kawrakow@gmail.com> Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
2023-06-05 19:56:18 +00:00
assert(n % QK_K == 0);
k-quants : allow to optionally disable at compile time (#1734) * k-quants : put behind optional compile flag LLAMA_K_QUANTS * build : enable k-quants by default
2023-06-07 07:59:52 +00:00
const block_q5_K * restrict x = vx;
const block_q8_K * restrict y = vy;
ggml : add SOTA 2,3,4,5,6 bit k-quantizations (#1684) * Starting to add k-quantization to ggml I think it is better to have quantization separate from ggml. For now just adding the k-quants there, but it would be better to also factor out the existing ggml quantizations. * Adding Q3_K and Q8_K (de)-quantization * Q3_K now working on CUDA and AVX2/scalar CUDA is not ideal - ~50% slower than Q4_0 for single token prediction, about the same in batch mode (perplexity). CPU single token is ~55 ms (on Ryzen 7950X). * Some improvement for Q3_K on CUDA It is now ~22.5 ms/token on my GPU, so ~30% slower than Q4_0. * Some more CUDA optimizations for Q3_K Single token is now 20.5 ms/token (~20% slower than Q4_0). Perplexity is on par with Q4_0. * Adding Q4_K - scalar, AVX2, CUDA Performance is the same or perhaps very slightly better than Q4_0 on the CPU. On the GPU, single token prediction is ~10% better than Q4_0, batch mode (perplexity is about the same). * Adding Q6_K - scalar, AVX2, CUDA Performance is ~40% lower compared to Q4_K on the CPU. This is to be expected, considering that we are memory bound on the CPU and the 6-bit model is ~44% larger than the 4-bit. On the GPU, single token prediction is ~6% lower than Q4_0, batch mode (perplexity) is even closer (but still slower). * Adding Q5_K - scalar, AVX2, CUDA Performance is ~20% lower compared to Q4_K on the CPU. This is to be expected, considering that we are memory bound on the CPU and the 5-bit model is ~22% larger than the 4-bit. On the GPU, single token prediction is about the same as Q4_0 for both, single token and batch prediction. * Per convention, all QX_K quantizations use Q5_K for output.weight * Adding quantization mixes * Quantization mixes: didn't quite get what I wanted in the last commit * Q4_K dot product for ARM_NEON * Q6_K dot product for ARM_NEON * Q5_K dot product for ARM_NEON * Adding Q3_K dot for ARM_NEON It is 22% slower than Q4_K, despite the smaller model size. On x86_64, where we are memory bound, the Q3_K model is quite a bit faster than Q4_K. * A very slightly faster ARM_NEON Q3_K dot * Adding Q2_K - just CUDA for now Token prediction is pretty good - about 15.5 ms on a RTX 4080. Perplexity is about the same as Q4_K. * Adding scalar and AVX2 Q2_K dot * Adding ARM_NEON Q2_K dot About the same performance as Q4_K. * A slightly faster ARM_NEON Q2_K dot Single token prediction is now ~36 ms on M2 Max. The code is much simpler too. * Fixed bug in Q2_K CUDA dot product kernel Stranegly enough, for the few prompts I tried with the 7B model the responses looked perfectly reasonable. Only realized something is not quite right when I tried the larger models and started getting nonse back. In any case, Q2_K single token evaluation time on an RTX 4080 in a Ryzen7950X box iusing CUDA and model fully loaded on the GPU are ~15.5 ms for 7B, ~25.4 ms for 13B, and ~55.8 ms for 30B. The max number of layers that fit in VRAM for The 65B is 32. With that, we get ~330 ms per token, which is not that much faster than just running on the CPU (~470 ms per token). * Don't print zeros/NaNs when no count histogram has been collected * A 10% faster CUDA vector dot kernel for Q3_K Q3_K is now running at ~18.5 ms / token on CUDA, so the gap to Q4_0 is only 10%. It seems memory acccess pattern is more important for performance than the amount of computation the kernel does. * A slightly daster Q4_K AVX2 dot product For perplexity, where we are less memory bound, time per pass drops by ~5%. Barely measurable difference for single token prediction. * A slightly faster ARM_NEON A4_K dot product * Minor * Fix quantization error test We cannot possibly be expecting rmse < 0.002 for 2- and 3-bit quantization variants. * Fix docker build I have been sloppy with vector reinterpret casts on ARM_NEON. It seems clang is very forgiving in that regard. * Added forgotten ggml.o dependence on k_quants.h to the Makefile * Had unintentionally committed the Makefile with -Ofast enabled * ggml : rename k_quants -> ggml-quants-k, use lowercase in code --------- Co-authored-by: Iwan Kawrakow <iwan.kawrakow@gmail.com> Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
2023-06-05 19:56:18 +00:00
const int nb = n / QK_K;
static const uint32_t kmask1 = 0x3f3f3f3f;
static const uint32_t kmask2 = 0x0f0f0f0f;
static const uint32_t kmask3 = 0x03030303;
uint32_t utmp[4];
#ifdef __ARM_NEON
const uint8x16_t m4b = vdupq_n_u8(0xf);
const uint8x16_t mone = vdupq_n_u8(1);
const uint8x16_t mtwo = vdupq_n_u8(2);
k-quants : fix build on armv7 (android only) (#2920) * k-quants : fix build on armv7 * ggml : cleanup unused arm32 specific impl * k-quants : avoid some unused vzero / mzero define * ggml-alloc : use 4g for MEASURE_MAX_SIZE in 32-bit arm
2023-09-02 12:23:45 +00:00
const int32x4_t mzero = vdupq_n_s32(0);
ggml : add SOTA 2,3,4,5,6 bit k-quantizations (#1684) * Starting to add k-quantization to ggml I think it is better to have quantization separate from ggml. For now just adding the k-quants there, but it would be better to also factor out the existing ggml quantizations. * Adding Q3_K and Q8_K (de)-quantization * Q3_K now working on CUDA and AVX2/scalar CUDA is not ideal - ~50% slower than Q4_0 for single token prediction, about the same in batch mode (perplexity). CPU single token is ~55 ms (on Ryzen 7950X). * Some improvement for Q3_K on CUDA It is now ~22.5 ms/token on my GPU, so ~30% slower than Q4_0. * Some more CUDA optimizations for Q3_K Single token is now 20.5 ms/token (~20% slower than Q4_0). Perplexity is on par with Q4_0. * Adding Q4_K - scalar, AVX2, CUDA Performance is the same or perhaps very slightly better than Q4_0 on the CPU. On the GPU, single token prediction is ~10% better than Q4_0, batch mode (perplexity is about the same). * Adding Q6_K - scalar, AVX2, CUDA Performance is ~40% lower compared to Q4_K on the CPU. This is to be expected, considering that we are memory bound on the CPU and the 6-bit model is ~44% larger than the 4-bit. On the GPU, single token prediction is ~6% lower than Q4_0, batch mode (perplexity) is even closer (but still slower). * Adding Q5_K - scalar, AVX2, CUDA Performance is ~20% lower compared to Q4_K on the CPU. This is to be expected, considering that we are memory bound on the CPU and the 5-bit model is ~22% larger than the 4-bit. On the GPU, single token prediction is about the same as Q4_0 for both, single token and batch prediction. * Per convention, all QX_K quantizations use Q5_K for output.weight * Adding quantization mixes * Quantization mixes: didn't quite get what I wanted in the last commit * Q4_K dot product for ARM_NEON * Q6_K dot product for ARM_NEON * Q5_K dot product for ARM_NEON * Adding Q3_K dot for ARM_NEON It is 22% slower than Q4_K, despite the smaller model size. On x86_64, where we are memory bound, the Q3_K model is quite a bit faster than Q4_K. * A very slightly faster ARM_NEON Q3_K dot * Adding Q2_K - just CUDA for now Token prediction is pretty good - about 15.5 ms on a RTX 4080. Perplexity is about the same as Q4_K. * Adding scalar and AVX2 Q2_K dot * Adding ARM_NEON Q2_K dot About the same performance as Q4_K. * A slightly faster ARM_NEON Q2_K dot Single token prediction is now ~36 ms on M2 Max. The code is much simpler too. * Fixed bug in Q2_K CUDA dot product kernel Stranegly enough, for the few prompts I tried with the 7B model the responses looked perfectly reasonable. Only realized something is not quite right when I tried the larger models and started getting nonse back. In any case, Q2_K single token evaluation time on an RTX 4080 in a Ryzen7950X box iusing CUDA and model fully loaded on the GPU are ~15.5 ms for 7B, ~25.4 ms for 13B, and ~55.8 ms for 30B. The max number of layers that fit in VRAM for The 65B is 32. With that, we get ~330 ms per token, which is not that much faster than just running on the CPU (~470 ms per token). * Don't print zeros/NaNs when no count histogram has been collected * A 10% faster CUDA vector dot kernel for Q3_K Q3_K is now running at ~18.5 ms / token on CUDA, so the gap to Q4_0 is only 10%. It seems memory acccess pattern is more important for performance than the amount of computation the kernel does. * A slightly daster Q4_K AVX2 dot product For perplexity, where we are less memory bound, time per pass drops by ~5%. Barely measurable difference for single token prediction. * A slightly faster ARM_NEON A4_K dot product * Minor * Fix quantization error test We cannot possibly be expecting rmse < 0.002 for 2- and 3-bit quantization variants. * Fix docker build I have been sloppy with vector reinterpret casts on ARM_NEON. It seems clang is very forgiving in that regard. * Added forgotten ggml.o dependence on k_quants.h to the Makefile * Had unintentionally committed the Makefile with -Ofast enabled * ggml : rename k_quants -> ggml-quants-k, use lowercase in code --------- Co-authored-by: Iwan Kawrakow <iwan.kawrakow@gmail.com> Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
2023-06-05 19:56:18 +00:00
ggml : sync (im2col, GPU conv, 32-bit arm compat) (#4060) ggml-ci
2023-11-13 14:55:52 +00:00
ggml_int8x16x4_t q5bytes;
ggml : add SOTA 2,3,4,5,6 bit k-quantizations (#1684) * Starting to add k-quantization to ggml I think it is better to have quantization separate from ggml. For now just adding the k-quants there, but it would be better to also factor out the existing ggml quantizations. * Adding Q3_K and Q8_K (de)-quantization * Q3_K now working on CUDA and AVX2/scalar CUDA is not ideal - ~50% slower than Q4_0 for single token prediction, about the same in batch mode (perplexity). CPU single token is ~55 ms (on Ryzen 7950X). * Some improvement for Q3_K on CUDA It is now ~22.5 ms/token on my GPU, so ~30% slower than Q4_0. * Some more CUDA optimizations for Q3_K Single token is now 20.5 ms/token (~20% slower than Q4_0). Perplexity is on par with Q4_0. * Adding Q4_K - scalar, AVX2, CUDA Performance is the same or perhaps very slightly better than Q4_0 on the CPU. On the GPU, single token prediction is ~10% better than Q4_0, batch mode (perplexity is about the same). * Adding Q6_K - scalar, AVX2, CUDA Performance is ~40% lower compared to Q4_K on the CPU. This is to be expected, considering that we are memory bound on the CPU and the 6-bit model is ~44% larger than the 4-bit. On the GPU, single token prediction is ~6% lower than Q4_0, batch mode (perplexity) is even closer (but still slower). * Adding Q5_K - scalar, AVX2, CUDA Performance is ~20% lower compared to Q4_K on the CPU. This is to be expected, considering that we are memory bound on the CPU and the 5-bit model is ~22% larger than the 4-bit. On the GPU, single token prediction is about the same as Q4_0 for both, single token and batch prediction. * Per convention, all QX_K quantizations use Q5_K for output.weight * Adding quantization mixes * Quantization mixes: didn't quite get what I wanted in the last commit * Q4_K dot product for ARM_NEON * Q6_K dot product for ARM_NEON * Q5_K dot product for ARM_NEON * Adding Q3_K dot for ARM_NEON It is 22% slower than Q4_K, despite the smaller model size. On x86_64, where we are memory bound, the Q3_K model is quite a bit faster than Q4_K. * A very slightly faster ARM_NEON Q3_K dot * Adding Q2_K - just CUDA for now Token prediction is pretty good - about 15.5 ms on a RTX 4080. Perplexity is about the same as Q4_K. * Adding scalar and AVX2 Q2_K dot * Adding ARM_NEON Q2_K dot About the same performance as Q4_K. * A slightly faster ARM_NEON Q2_K dot Single token prediction is now ~36 ms on M2 Max. The code is much simpler too. * Fixed bug in Q2_K CUDA dot product kernel Stranegly enough, for the few prompts I tried with the 7B model the responses looked perfectly reasonable. Only realized something is not quite right when I tried the larger models and started getting nonse back. In any case, Q2_K single token evaluation time on an RTX 4080 in a Ryzen7950X box iusing CUDA and model fully loaded on the GPU are ~15.5 ms for 7B, ~25.4 ms for 13B, and ~55.8 ms for 30B. The max number of layers that fit in VRAM for The 65B is 32. With that, we get ~330 ms per token, which is not that much faster than just running on the CPU (~470 ms per token). * Don't print zeros/NaNs when no count histogram has been collected * A 10% faster CUDA vector dot kernel for Q3_K Q3_K is now running at ~18.5 ms / token on CUDA, so the gap to Q4_0 is only 10%. It seems memory acccess pattern is more important for performance than the amount of computation the kernel does. * A slightly daster Q4_K AVX2 dot product For perplexity, where we are less memory bound, time per pass drops by ~5%. Barely measurable difference for single token prediction. * A slightly faster ARM_NEON A4_K dot product * Minor * Fix quantization error test We cannot possibly be expecting rmse < 0.002 for 2- and 3-bit quantization variants. * Fix docker build I have been sloppy with vector reinterpret casts on ARM_NEON. It seems clang is very forgiving in that regard. * Added forgotten ggml.o dependence on k_quants.h to the Makefile * Had unintentionally committed the Makefile with -Ofast enabled * ggml : rename k_quants -> ggml-quants-k, use lowercase in code --------- Co-authored-by: Iwan Kawrakow <iwan.kawrakow@gmail.com> Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
2023-06-05 19:56:18 +00:00
float sumf = 0;
for (int i = 0; i < nb; ++i) {
ggml : move FP16 <-> FP32 code to ggml-impl.h (#3861) * ggml : move FP16 <-> FP32 stuff to ggml-impl.h ggml-ci * tests : fix ARM build * ggml : explicitly initialize deprecated type traits * ggml : add math.h to ggml-impl.h * ggml : remove duplicate static assert macros * ggml : prefix lookup tables with ggml_ ggml-ci * ggml-impl : move extern "C" to start of file
2023-10-30 17:19:15 +00:00
const float d = y[i].d * GGML_FP16_TO_FP32(x[i].d);
const float dmin = y[i].d * GGML_FP16_TO_FP32(x[i].dmin);
ggml : add SOTA 2,3,4,5,6 bit k-quantizations (#1684) * Starting to add k-quantization to ggml I think it is better to have quantization separate from ggml. For now just adding the k-quants there, but it would be better to also factor out the existing ggml quantizations. * Adding Q3_K and Q8_K (de)-quantization * Q3_K now working on CUDA and AVX2/scalar CUDA is not ideal - ~50% slower than Q4_0 for single token prediction, about the same in batch mode (perplexity). CPU single token is ~55 ms (on Ryzen 7950X). * Some improvement for Q3_K on CUDA It is now ~22.5 ms/token on my GPU, so ~30% slower than Q4_0. * Some more CUDA optimizations for Q3_K Single token is now 20.5 ms/token (~20% slower than Q4_0). Perplexity is on par with Q4_0. * Adding Q4_K - scalar, AVX2, CUDA Performance is the same or perhaps very slightly better than Q4_0 on the CPU. On the GPU, single token prediction is ~10% better than Q4_0, batch mode (perplexity is about the same). * Adding Q6_K - scalar, AVX2, CUDA Performance is ~40% lower compared to Q4_K on the CPU. This is to be expected, considering that we are memory bound on the CPU and the 6-bit model is ~44% larger than the 4-bit. On the GPU, single token prediction is ~6% lower than Q4_0, batch mode (perplexity) is even closer (but still slower). * Adding Q5_K - scalar, AVX2, CUDA Performance is ~20% lower compared to Q4_K on the CPU. This is to be expected, considering that we are memory bound on the CPU and the 5-bit model is ~22% larger than the 4-bit. On the GPU, single token prediction is about the same as Q4_0 for both, single token and batch prediction. * Per convention, all QX_K quantizations use Q5_K for output.weight * Adding quantization mixes * Quantization mixes: didn't quite get what I wanted in the last commit * Q4_K dot product for ARM_NEON * Q6_K dot product for ARM_NEON * Q5_K dot product for ARM_NEON * Adding Q3_K dot for ARM_NEON It is 22% slower than Q4_K, despite the smaller model size. On x86_64, where we are memory bound, the Q3_K model is quite a bit faster than Q4_K. * A very slightly faster ARM_NEON Q3_K dot * Adding Q2_K - just CUDA for now Token prediction is pretty good - about 15.5 ms on a RTX 4080. Perplexity is about the same as Q4_K. * Adding scalar and AVX2 Q2_K dot * Adding ARM_NEON Q2_K dot About the same performance as Q4_K. * A slightly faster ARM_NEON Q2_K dot Single token prediction is now ~36 ms on M2 Max. The code is much simpler too. * Fixed bug in Q2_K CUDA dot product kernel Stranegly enough, for the few prompts I tried with the 7B model the responses looked perfectly reasonable. Only realized something is not quite right when I tried the larger models and started getting nonse back. In any case, Q2_K single token evaluation time on an RTX 4080 in a Ryzen7950X box iusing CUDA and model fully loaded on the GPU are ~15.5 ms for 7B, ~25.4 ms for 13B, and ~55.8 ms for 30B. The max number of layers that fit in VRAM for The 65B is 32. With that, we get ~330 ms per token, which is not that much faster than just running on the CPU (~470 ms per token). * Don't print zeros/NaNs when no count histogram has been collected * A 10% faster CUDA vector dot kernel for Q3_K Q3_K is now running at ~18.5 ms / token on CUDA, so the gap to Q4_0 is only 10%. It seems memory acccess pattern is more important for performance than the amount of computation the kernel does. * A slightly daster Q4_K AVX2 dot product For perplexity, where we are less memory bound, time per pass drops by ~5%. Barely measurable difference for single token prediction. * A slightly faster ARM_NEON A4_K dot product * Minor * Fix quantization error test We cannot possibly be expecting rmse < 0.002 for 2- and 3-bit quantization variants. * Fix docker build I have been sloppy with vector reinterpret casts on ARM_NEON. It seems clang is very forgiving in that regard. * Added forgotten ggml.o dependence on k_quants.h to the Makefile * Had unintentionally committed the Makefile with -Ofast enabled * ggml : rename k_quants -> ggml-quants-k, use lowercase in code --------- Co-authored-by: Iwan Kawrakow <iwan.kawrakow@gmail.com> Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
2023-06-05 19:56:18 +00:00
const int16x8_t q8sums = vpaddq_s16(vld1q_s16(y[i].bsums), vld1q_s16(y[i].bsums + 8));
memcpy(utmp, x[i].scales, 12);
utmp[3] = ((utmp[2] >> 4) & kmask2) | (((utmp[1] >> 6) & kmask3) << 4);
const uint32_t uaux = utmp[1] & kmask1;
utmp[1] = (utmp[2] & kmask2) | (((utmp[0] >> 6) & kmask3) << 4);
utmp[2] = uaux;
utmp[0] &= kmask1;
const uint8x8_t mins8 = vld1_u8((const uint8_t*)utmp + 8);
const int16x8_t mins = vreinterpretq_s16_u16(vmovl_u8(mins8));
const int32x4_t prod = vaddq_s32(vmull_s16(vget_low_s16 (q8sums), vget_low_s16 (mins)),
vmull_s16(vget_high_s16(q8sums), vget_high_s16(mins)));
int32_t sumi_mins = vaddvq_s32(prod);
const uint8_t * scales = (const uint8_t *)utmp;
const uint8_t * restrict q5 = x[i].qs;
const uint8_t * restrict qh = x[i].qh;
const int8_t * restrict q8 = y[i].qs;
ggml : sync (im2col, GPU conv, 32-bit arm compat) (#4060) ggml-ci
2023-11-13 14:55:52 +00:00
ggml_uint8x16x2_t qhbits = ggml_vld1q_u8_x2(qh);
ggml : add SOTA 2,3,4,5,6 bit k-quantizations (#1684) * Starting to add k-quantization to ggml I think it is better to have quantization separate from ggml. For now just adding the k-quants there, but it would be better to also factor out the existing ggml quantizations. * Adding Q3_K and Q8_K (de)-quantization * Q3_K now working on CUDA and AVX2/scalar CUDA is not ideal - ~50% slower than Q4_0 for single token prediction, about the same in batch mode (perplexity). CPU single token is ~55 ms (on Ryzen 7950X). * Some improvement for Q3_K on CUDA It is now ~22.5 ms/token on my GPU, so ~30% slower than Q4_0. * Some more CUDA optimizations for Q3_K Single token is now 20.5 ms/token (~20% slower than Q4_0). Perplexity is on par with Q4_0. * Adding Q4_K - scalar, AVX2, CUDA Performance is the same or perhaps very slightly better than Q4_0 on the CPU. On the GPU, single token prediction is ~10% better than Q4_0, batch mode (perplexity is about the same). * Adding Q6_K - scalar, AVX2, CUDA Performance is ~40% lower compared to Q4_K on the CPU. This is to be expected, considering that we are memory bound on the CPU and the 6-bit model is ~44% larger than the 4-bit. On the GPU, single token prediction is ~6% lower than Q4_0, batch mode (perplexity) is even closer (but still slower). * Adding Q5_K - scalar, AVX2, CUDA Performance is ~20% lower compared to Q4_K on the CPU. This is to be expected, considering that we are memory bound on the CPU and the 5-bit model is ~22% larger than the 4-bit. On the GPU, single token prediction is about the same as Q4_0 for both, single token and batch prediction. * Per convention, all QX_K quantizations use Q5_K for output.weight * Adding quantization mixes * Quantization mixes: didn't quite get what I wanted in the last commit * Q4_K dot product for ARM_NEON * Q6_K dot product for ARM_NEON * Q5_K dot product for ARM_NEON * Adding Q3_K dot for ARM_NEON It is 22% slower than Q4_K, despite the smaller model size. On x86_64, where we are memory bound, the Q3_K model is quite a bit faster than Q4_K. * A very slightly faster ARM_NEON Q3_K dot * Adding Q2_K - just CUDA for now Token prediction is pretty good - about 15.5 ms on a RTX 4080. Perplexity is about the same as Q4_K. * Adding scalar and AVX2 Q2_K dot * Adding ARM_NEON Q2_K dot About the same performance as Q4_K. * A slightly faster ARM_NEON Q2_K dot Single token prediction is now ~36 ms on M2 Max. The code is much simpler too. * Fixed bug in Q2_K CUDA dot product kernel Stranegly enough, for the few prompts I tried with the 7B model the responses looked perfectly reasonable. Only realized something is not quite right when I tried the larger models and started getting nonse back. In any case, Q2_K single token evaluation time on an RTX 4080 in a Ryzen7950X box iusing CUDA and model fully loaded on the GPU are ~15.5 ms for 7B, ~25.4 ms for 13B, and ~55.8 ms for 30B. The max number of layers that fit in VRAM for The 65B is 32. With that, we get ~330 ms per token, which is not that much faster than just running on the CPU (~470 ms per token). * Don't print zeros/NaNs when no count histogram has been collected * A 10% faster CUDA vector dot kernel for Q3_K Q3_K is now running at ~18.5 ms / token on CUDA, so the gap to Q4_0 is only 10%. It seems memory acccess pattern is more important for performance than the amount of computation the kernel does. * A slightly daster Q4_K AVX2 dot product For perplexity, where we are less memory bound, time per pass drops by ~5%. Barely measurable difference for single token prediction. * A slightly faster ARM_NEON A4_K dot product * Minor * Fix quantization error test We cannot possibly be expecting rmse < 0.002 for 2- and 3-bit quantization variants. * Fix docker build I have been sloppy with vector reinterpret casts on ARM_NEON. It seems clang is very forgiving in that regard. * Added forgotten ggml.o dependence on k_quants.h to the Makefile * Had unintentionally committed the Makefile with -Ofast enabled * ggml : rename k_quants -> ggml-quants-k, use lowercase in code --------- Co-authored-by: Iwan Kawrakow <iwan.kawrakow@gmail.com> Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
2023-06-05 19:56:18 +00:00
ggml : sync (im2col, GPU conv, 32-bit arm compat) (#4060) ggml-ci
2023-11-13 14:55:52 +00:00
ggml_uint8x16x4_t q5h;
ggml : add SOTA 2,3,4,5,6 bit k-quantizations (#1684) * Starting to add k-quantization to ggml I think it is better to have quantization separate from ggml. For now just adding the k-quants there, but it would be better to also factor out the existing ggml quantizations. * Adding Q3_K and Q8_K (de)-quantization * Q3_K now working on CUDA and AVX2/scalar CUDA is not ideal - ~50% slower than Q4_0 for single token prediction, about the same in batch mode (perplexity). CPU single token is ~55 ms (on Ryzen 7950X). * Some improvement for Q3_K on CUDA It is now ~22.5 ms/token on my GPU, so ~30% slower than Q4_0. * Some more CUDA optimizations for Q3_K Single token is now 20.5 ms/token (~20% slower than Q4_0). Perplexity is on par with Q4_0. * Adding Q4_K - scalar, AVX2, CUDA Performance is the same or perhaps very slightly better than Q4_0 on the CPU. On the GPU, single token prediction is ~10% better than Q4_0, batch mode (perplexity is about the same). * Adding Q6_K - scalar, AVX2, CUDA Performance is ~40% lower compared to Q4_K on the CPU. This is to be expected, considering that we are memory bound on the CPU and the 6-bit model is ~44% larger than the 4-bit. On the GPU, single token prediction is ~6% lower than Q4_0, batch mode (perplexity) is even closer (but still slower). * Adding Q5_K - scalar, AVX2, CUDA Performance is ~20% lower compared to Q4_K on the CPU. This is to be expected, considering that we are memory bound on the CPU and the 5-bit model is ~22% larger than the 4-bit. On the GPU, single token prediction is about the same as Q4_0 for both, single token and batch prediction. * Per convention, all QX_K quantizations use Q5_K for output.weight * Adding quantization mixes * Quantization mixes: didn't quite get what I wanted in the last commit * Q4_K dot product for ARM_NEON * Q6_K dot product for ARM_NEON * Q5_K dot product for ARM_NEON * Adding Q3_K dot for ARM_NEON It is 22% slower than Q4_K, despite the smaller model size. On x86_64, where we are memory bound, the Q3_K model is quite a bit faster than Q4_K. * A very slightly faster ARM_NEON Q3_K dot * Adding Q2_K - just CUDA for now Token prediction is pretty good - about 15.5 ms on a RTX 4080. Perplexity is about the same as Q4_K. * Adding scalar and AVX2 Q2_K dot * Adding ARM_NEON Q2_K dot About the same performance as Q4_K. * A slightly faster ARM_NEON Q2_K dot Single token prediction is now ~36 ms on M2 Max. The code is much simpler too. * Fixed bug in Q2_K CUDA dot product kernel Stranegly enough, for the few prompts I tried with the 7B model the responses looked perfectly reasonable. Only realized something is not quite right when I tried the larger models and started getting nonse back. In any case, Q2_K single token evaluation time on an RTX 4080 in a Ryzen7950X box iusing CUDA and model fully loaded on the GPU are ~15.5 ms for 7B, ~25.4 ms for 13B, and ~55.8 ms for 30B. The max number of layers that fit in VRAM for The 65B is 32. With that, we get ~330 ms per token, which is not that much faster than just running on the CPU (~470 ms per token). * Don't print zeros/NaNs when no count histogram has been collected * A 10% faster CUDA vector dot kernel for Q3_K Q3_K is now running at ~18.5 ms / token on CUDA, so the gap to Q4_0 is only 10%. It seems memory acccess pattern is more important for performance than the amount of computation the kernel does. * A slightly daster Q4_K AVX2 dot product For perplexity, where we are less memory bound, time per pass drops by ~5%. Barely measurable difference for single token prediction. * A slightly faster ARM_NEON A4_K dot product * Minor * Fix quantization error test We cannot possibly be expecting rmse < 0.002 for 2- and 3-bit quantization variants. * Fix docker build I have been sloppy with vector reinterpret casts on ARM_NEON. It seems clang is very forgiving in that regard. * Added forgotten ggml.o dependence on k_quants.h to the Makefile * Had unintentionally committed the Makefile with -Ofast enabled * ggml : rename k_quants -> ggml-quants-k, use lowercase in code --------- Co-authored-by: Iwan Kawrakow <iwan.kawrakow@gmail.com> Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
2023-06-05 19:56:18 +00:00
int32_t sumi = 0;
for (int j = 0; j < QK_K/64; ++j) {
ggml : sync (im2col, GPU conv, 32-bit arm compat) (#4060) ggml-ci
2023-11-13 14:55:52 +00:00
const ggml_uint8x16x2_t q5bits = ggml_vld1q_u8_x2(q5); q5 += 32;
const ggml_int8x16x4_t q8bytes = ggml_vld1q_s8_x4(q8); q8 += 64;
ggml : add SOTA 2,3,4,5,6 bit k-quantizations (#1684) * Starting to add k-quantization to ggml I think it is better to have quantization separate from ggml. For now just adding the k-quants there, but it would be better to also factor out the existing ggml quantizations. * Adding Q3_K and Q8_K (de)-quantization * Q3_K now working on CUDA and AVX2/scalar CUDA is not ideal - ~50% slower than Q4_0 for single token prediction, about the same in batch mode (perplexity). CPU single token is ~55 ms (on Ryzen 7950X). * Some improvement for Q3_K on CUDA It is now ~22.5 ms/token on my GPU, so ~30% slower than Q4_0. * Some more CUDA optimizations for Q3_K Single token is now 20.5 ms/token (~20% slower than Q4_0). Perplexity is on par with Q4_0. * Adding Q4_K - scalar, AVX2, CUDA Performance is the same or perhaps very slightly better than Q4_0 on the CPU. On the GPU, single token prediction is ~10% better than Q4_0, batch mode (perplexity is about the same). * Adding Q6_K - scalar, AVX2, CUDA Performance is ~40% lower compared to Q4_K on the CPU. This is to be expected, considering that we are memory bound on the CPU and the 6-bit model is ~44% larger than the 4-bit. On the GPU, single token prediction is ~6% lower than Q4_0, batch mode (perplexity) is even closer (but still slower). * Adding Q5_K - scalar, AVX2, CUDA Performance is ~20% lower compared to Q4_K on the CPU. This is to be expected, considering that we are memory bound on the CPU and the 5-bit model is ~22% larger than the 4-bit. On the GPU, single token prediction is about the same as Q4_0 for both, single token and batch prediction. * Per convention, all QX_K quantizations use Q5_K for output.weight * Adding quantization mixes * Quantization mixes: didn't quite get what I wanted in the last commit * Q4_K dot product for ARM_NEON * Q6_K dot product for ARM_NEON * Q5_K dot product for ARM_NEON * Adding Q3_K dot for ARM_NEON It is 22% slower than Q4_K, despite the smaller model size. On x86_64, where we are memory bound, the Q3_K model is quite a bit faster than Q4_K. * A very slightly faster ARM_NEON Q3_K dot * Adding Q2_K - just CUDA for now Token prediction is pretty good - about 15.5 ms on a RTX 4080. Perplexity is about the same as Q4_K. * Adding scalar and AVX2 Q2_K dot * Adding ARM_NEON Q2_K dot About the same performance as Q4_K. * A slightly faster ARM_NEON Q2_K dot Single token prediction is now ~36 ms on M2 Max. The code is much simpler too. * Fixed bug in Q2_K CUDA dot product kernel Stranegly enough, for the few prompts I tried with the 7B model the responses looked perfectly reasonable. Only realized something is not quite right when I tried the larger models and started getting nonse back. In any case, Q2_K single token evaluation time on an RTX 4080 in a Ryzen7950X box iusing CUDA and model fully loaded on the GPU are ~15.5 ms for 7B, ~25.4 ms for 13B, and ~55.8 ms for 30B. The max number of layers that fit in VRAM for The 65B is 32. With that, we get ~330 ms per token, which is not that much faster than just running on the CPU (~470 ms per token). * Don't print zeros/NaNs when no count histogram has been collected * A 10% faster CUDA vector dot kernel for Q3_K Q3_K is now running at ~18.5 ms / token on CUDA, so the gap to Q4_0 is only 10%. It seems memory acccess pattern is more important for performance than the amount of computation the kernel does. * A slightly daster Q4_K AVX2 dot product For perplexity, where we are less memory bound, time per pass drops by ~5%. Barely measurable difference for single token prediction. * A slightly faster ARM_NEON A4_K dot product * Minor * Fix quantization error test We cannot possibly be expecting rmse < 0.002 for 2- and 3-bit quantization variants. * Fix docker build I have been sloppy with vector reinterpret casts on ARM_NEON. It seems clang is very forgiving in that regard. * Added forgotten ggml.o dependence on k_quants.h to the Makefile * Had unintentionally committed the Makefile with -Ofast enabled * ggml : rename k_quants -> ggml-quants-k, use lowercase in code --------- Co-authored-by: Iwan Kawrakow <iwan.kawrakow@gmail.com> Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
2023-06-05 19:56:18 +00:00
q5h.val[0] = vshlq_n_u8(vandq_u8(mone, qhbits.val[0]), 4);
q5h.val[1] = vshlq_n_u8(vandq_u8(mone, qhbits.val[1]), 4);
q5h.val[2] = vshlq_n_u8(vandq_u8(mtwo, qhbits.val[0]), 3);
q5h.val[3] = vshlq_n_u8(vandq_u8(mtwo, qhbits.val[1]), 3);
qhbits.val[0] = vshrq_n_u8(qhbits.val[0], 2);
qhbits.val[1] = vshrq_n_u8(qhbits.val[1], 2);
q5bytes.val[0] = vreinterpretq_s8_u8(vorrq_u8(vandq_u8(q5bits.val[0], m4b), q5h.val[0]));
q5bytes.val[1] = vreinterpretq_s8_u8(vorrq_u8(vandq_u8(q5bits.val[1], m4b), q5h.val[1]));
q5bytes.val[2] = vreinterpretq_s8_u8(vorrq_u8(vshrq_n_u8(q5bits.val[0], 4), q5h.val[2]));
q5bytes.val[3] = vreinterpretq_s8_u8(vorrq_u8(vshrq_n_u8(q5bits.val[1], 4), q5h.val[3]));
ggml : add ggml_vdotq_s32 alias (#4715) ggml-ci
2023-12-31 09:43:31 +00:00
sumi += vaddvq_s32(ggml_vdotq_s32(ggml_vdotq_s32(mzero, q5bytes.val[0], q8bytes.val[0]), q5bytes.val[1], q8bytes.val[1])) * *scales++;
sumi += vaddvq_s32(ggml_vdotq_s32(ggml_vdotq_s32(mzero, q5bytes.val[2], q8bytes.val[2]), q5bytes.val[3], q8bytes.val[3])) * *scales++;
ggml : add SOTA 2,3,4,5,6 bit k-quantizations (#1684) * Starting to add k-quantization to ggml I think it is better to have quantization separate from ggml. For now just adding the k-quants there, but it would be better to also factor out the existing ggml quantizations. * Adding Q3_K and Q8_K (de)-quantization * Q3_K now working on CUDA and AVX2/scalar CUDA is not ideal - ~50% slower than Q4_0 for single token prediction, about the same in batch mode (perplexity). CPU single token is ~55 ms (on Ryzen 7950X). * Some improvement for Q3_K on CUDA It is now ~22.5 ms/token on my GPU, so ~30% slower than Q4_0. * Some more CUDA optimizations for Q3_K Single token is now 20.5 ms/token (~20% slower than Q4_0). Perplexity is on par with Q4_0. * Adding Q4_K - scalar, AVX2, CUDA Performance is the same or perhaps very slightly better than Q4_0 on the CPU. On the GPU, single token prediction is ~10% better than Q4_0, batch mode (perplexity is about the same). * Adding Q6_K - scalar, AVX2, CUDA Performance is ~40% lower compared to Q4_K on the CPU. This is to be expected, considering that we are memory bound on the CPU and the 6-bit model is ~44% larger than the 4-bit. On the GPU, single token prediction is ~6% lower than Q4_0, batch mode (perplexity) is even closer (but still slower). * Adding Q5_K - scalar, AVX2, CUDA Performance is ~20% lower compared to Q4_K on the CPU. This is to be expected, considering that we are memory bound on the CPU and the 5-bit model is ~22% larger than the 4-bit. On the GPU, single token prediction is about the same as Q4_0 for both, single token and batch prediction. * Per convention, all QX_K quantizations use Q5_K for output.weight * Adding quantization mixes * Quantization mixes: didn't quite get what I wanted in the last commit * Q4_K dot product for ARM_NEON * Q6_K dot product for ARM_NEON * Q5_K dot product for ARM_NEON * Adding Q3_K dot for ARM_NEON It is 22% slower than Q4_K, despite the smaller model size. On x86_64, where we are memory bound, the Q3_K model is quite a bit faster than Q4_K. * A very slightly faster ARM_NEON Q3_K dot * Adding Q2_K - just CUDA for now Token prediction is pretty good - about 15.5 ms on a RTX 4080. Perplexity is about the same as Q4_K. * Adding scalar and AVX2 Q2_K dot * Adding ARM_NEON Q2_K dot About the same performance as Q4_K. * A slightly faster ARM_NEON Q2_K dot Single token prediction is now ~36 ms on M2 Max. The code is much simpler too. * Fixed bug in Q2_K CUDA dot product kernel Stranegly enough, for the few prompts I tried with the 7B model the responses looked perfectly reasonable. Only realized something is not quite right when I tried the larger models and started getting nonse back. In any case, Q2_K single token evaluation time on an RTX 4080 in a Ryzen7950X box iusing CUDA and model fully loaded on the GPU are ~15.5 ms for 7B, ~25.4 ms for 13B, and ~55.8 ms for 30B. The max number of layers that fit in VRAM for The 65B is 32. With that, we get ~330 ms per token, which is not that much faster than just running on the CPU (~470 ms per token). * Don't print zeros/NaNs when no count histogram has been collected * A 10% faster CUDA vector dot kernel for Q3_K Q3_K is now running at ~18.5 ms / token on CUDA, so the gap to Q4_0 is only 10%. It seems memory acccess pattern is more important for performance than the amount of computation the kernel does. * A slightly daster Q4_K AVX2 dot product For perplexity, where we are less memory bound, time per pass drops by ~5%. Barely measurable difference for single token prediction. * A slightly faster ARM_NEON A4_K dot product * Minor * Fix quantization error test We cannot possibly be expecting rmse < 0.002 for 2- and 3-bit quantization variants. * Fix docker build I have been sloppy with vector reinterpret casts on ARM_NEON. It seems clang is very forgiving in that regard. * Added forgotten ggml.o dependence on k_quants.h to the Makefile * Had unintentionally committed the Makefile with -Ofast enabled * ggml : rename k_quants -> ggml-quants-k, use lowercase in code --------- Co-authored-by: Iwan Kawrakow <iwan.kawrakow@gmail.com> Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
2023-06-05 19:56:18 +00:00
}
sumf += d * sumi - dmin * sumi_mins;
}
*s = sumf;
#elif defined __AVX2__
const __m256i m4 = _mm256_set1_epi8(0xF);
const __m128i mzero = _mm_setzero_si128();
const __m256i mone = _mm256_set1_epi8(1);
__m256 acc = _mm256_setzero_ps();
float summs = 0.f;
for (int i = 0; i < nb; ++i) {
const uint8_t * restrict q5 = x[i].qs;
const int8_t * restrict q8 = y[i].qs;
k-quants : support for super-block size of 64 (#2001) * k_quants: WIP super-blocks with 64 weights * k_quants: WIP super-blocks with 64 weights Q6_K scalar and AVX2 works * k_quants: WIP super-blocks with 64 weights Q4_K scalar and AVX2 works * k_quants: WIP super-blocks with 64 weights Q2_K scalar and AVX2 works. Q2_K is way too slow (it is actually slower than the scalar implementation) * k_quants: WIP super-blocks with 64 weights Q3_K scalar and AVX2 works. * k_quants: WIP super-blocks with 64 weights Q5_K scalar and AVX2 works, and with that all k_quants are done on AVX2 and scalar * k_quants: WIP super-blocks with 64 weights Q6_K working on CUDA. Cannot make it run quite as gast as with super-blocks with 256 weigths: 8% slower on 4080, 20% slower on the 1660 (but there we fit 1 less layer on the GPU because pf the larger model size), so some fraction of these 20% is due to that, * k_quants: WIP super-blocks with 64 weights Q4_K working on CUDA. ~10% slower on GTX-1660, 16% slower on 4080. * k_quants: WIP super-blocks with 64 weights Q2_K working on CUDA. ~3% slower on GTX-1660, 10% slower on 4080. * k_quants: WIP super-blocks with 64 weights Q3_K working on CUDA. * k_quants: WIP super-blocks with 64 weights Q5_K working on CUDA, and with this CUDA is done. * k_quants: WIP super-blocks with 64 weights Q6_K working on ARM_NEON * k_quants: WIP super-blocks with 64 weights Q4_K working on ARM_NEON, but quite a bit slower than 256 weights * k_quants: WIP super-blocks with 64 weights Q2_K working on ARM_NEON, but quite a bit slower than 256 weights * k_quants: WIP super-blocks with 64 weights Q3_K working on ARM_NEON, but quite a bit slower than 256 weights. * k_quants: WIP super-blocks with 64 weights Q5_K working on ARM_NEON, but quite a bit slower than 256 weights. With that, we have full support for ARM_NEON, although performance is not quite there. * k_quants: WIP super-blocks with 64 weights Slightly more efficient Q3_K and Q5_K * k_quants: WIP super-blocks with 64 weights Another small improvement for Q3_K and Q5_K on ARM_NEON * k_quants: WIP super-blocks with 64 weights Yet another speedup for Q5_K on ARM_NEON. We are now within 10% of the QK_K = 256 version. * k_quants: WIP super-blocks with 64 weights * We are able to pass preprocessor macros to the Metal compiler * Q6_K works and is actually slightly more efficient than the QK_K = 256 version (25.2 ms vs 25.8 ms) * k_quants: WIP super-blocks with 64 weights Q4_K works on Metal and is actually slightly faster than QK_K = 256 (21.95 ms vs 24.0 ms). * k_quants: WIP super-blocks with 64 weights Q2_K works on Metal and is very slightly faster than QK_K = 256 (23.8 ms vs 24.2 ms). * k_quants: WIP super-blocks with 64 weights Q3_K works on Metal and is slightly faster than QK_K = 256 (26.6 ms vs 28.3 ms). * k_quants: WIP super-blocks with 64 weights Q5_K works on Metal and is slightly faster than QK_K = 256 (23.7 ms vs 26.3 ms). * k_quants: call them _K, not _k, also on Metal * k_quants: correctly define QK_K in llama.cpp * Fixed bug in q4_K quantization added with the 64-block addition * Simplify via lambda * k_quants: swicth Q3_K to 4-bit scales when QK_K = 64 Otherwise there isn't much benefit from this quantization type. There is some very slight loss in accuracy, but we reduce size by ~7%. E.g., for OpenLLaMA-3B, Q3_K_S perplexity is 8.6131 with 8-bit scales and 8.6352 with 4-bit, while file size decreases from 1.53G to 1.44G. * k_quants: switch Q4_K to 4-bit scales when QK_K = 64 Here the loss in accuracy is greater than for Q3_K, but the Q4_K points still move further to the left on the perplexity vs size curve. * k_quants: forgot to add the Metal changes in last commit * k_quants: change Q5_K to be type 0 when QK_K = 64 Still needs AVX2 implementation * k_quants: AVX2 implementation for new 64-weight Q5_K * k_quants: 10% faster ARM_NEON Q5_K dot product * k_quants: fixed issue caused by merging with master --------- Co-authored-by: Iwan Kawrakow <iwan.kawrakow@gmail.com>
2023-06-26 16:43:07 +00:00
#if QK_K == 256
ggml : move FP16 <-> FP32 code to ggml-impl.h (#3861) * ggml : move FP16 <-> FP32 stuff to ggml-impl.h ggml-ci * tests : fix ARM build * ggml : explicitly initialize deprecated type traits * ggml : add math.h to ggml-impl.h * ggml : remove duplicate static assert macros * ggml : prefix lookup tables with ggml_ ggml-ci * ggml-impl : move extern "C" to start of file
2023-10-30 17:19:15 +00:00
const float d = y[i].d * GGML_FP16_TO_FP32(x[i].d);
const float dmin = -y[i].d * GGML_FP16_TO_FP32(x[i].dmin);
k-quants : support for super-block size of 64 (#2001) * k_quants: WIP super-blocks with 64 weights * k_quants: WIP super-blocks with 64 weights Q6_K scalar and AVX2 works * k_quants: WIP super-blocks with 64 weights Q4_K scalar and AVX2 works * k_quants: WIP super-blocks with 64 weights Q2_K scalar and AVX2 works. Q2_K is way too slow (it is actually slower than the scalar implementation) * k_quants: WIP super-blocks with 64 weights Q3_K scalar and AVX2 works. * k_quants: WIP super-blocks with 64 weights Q5_K scalar and AVX2 works, and with that all k_quants are done on AVX2 and scalar * k_quants: WIP super-blocks with 64 weights Q6_K working on CUDA. Cannot make it run quite as gast as with super-blocks with 256 weigths: 8% slower on 4080, 20% slower on the 1660 (but there we fit 1 less layer on the GPU because pf the larger model size), so some fraction of these 20% is due to that, * k_quants: WIP super-blocks with 64 weights Q4_K working on CUDA. ~10% slower on GTX-1660, 16% slower on 4080. * k_quants: WIP super-blocks with 64 weights Q2_K working on CUDA. ~3% slower on GTX-1660, 10% slower on 4080. * k_quants: WIP super-blocks with 64 weights Q3_K working on CUDA. * k_quants: WIP super-blocks with 64 weights Q5_K working on CUDA, and with this CUDA is done. * k_quants: WIP super-blocks with 64 weights Q6_K working on ARM_NEON * k_quants: WIP super-blocks with 64 weights Q4_K working on ARM_NEON, but quite a bit slower than 256 weights * k_quants: WIP super-blocks with 64 weights Q2_K working on ARM_NEON, but quite a bit slower than 256 weights * k_quants: WIP super-blocks with 64 weights Q3_K working on ARM_NEON, but quite a bit slower than 256 weights. * k_quants: WIP super-blocks with 64 weights Q5_K working on ARM_NEON, but quite a bit slower than 256 weights. With that, we have full support for ARM_NEON, although performance is not quite there. * k_quants: WIP super-blocks with 64 weights Slightly more efficient Q3_K and Q5_K * k_quants: WIP super-blocks with 64 weights Another small improvement for Q3_K and Q5_K on ARM_NEON * k_quants: WIP super-blocks with 64 weights Yet another speedup for Q5_K on ARM_NEON. We are now within 10% of the QK_K = 256 version. * k_quants: WIP super-blocks with 64 weights * We are able to pass preprocessor macros to the Metal compiler * Q6_K works and is actually slightly more efficient than the QK_K = 256 version (25.2 ms vs 25.8 ms) * k_quants: WIP super-blocks with 64 weights Q4_K works on Metal and is actually slightly faster than QK_K = 256 (21.95 ms vs 24.0 ms). * k_quants: WIP super-blocks with 64 weights Q2_K works on Metal and is very slightly faster than QK_K = 256 (23.8 ms vs 24.2 ms). * k_quants: WIP super-blocks with 64 weights Q3_K works on Metal and is slightly faster than QK_K = 256 (26.6 ms vs 28.3 ms). * k_quants: WIP super-blocks with 64 weights Q5_K works on Metal and is slightly faster than QK_K = 256 (23.7 ms vs 26.3 ms). * k_quants: call them _K, not _k, also on Metal * k_quants: correctly define QK_K in llama.cpp * Fixed bug in q4_K quantization added with the 64-block addition * Simplify via lambda * k_quants: swicth Q3_K to 4-bit scales when QK_K = 64 Otherwise there isn't much benefit from this quantization type. There is some very slight loss in accuracy, but we reduce size by ~7%. E.g., for OpenLLaMA-3B, Q3_K_S perplexity is 8.6131 with 8-bit scales and 8.6352 with 4-bit, while file size decreases from 1.53G to 1.44G. * k_quants: switch Q4_K to 4-bit scales when QK_K = 64 Here the loss in accuracy is greater than for Q3_K, but the Q4_K points still move further to the left on the perplexity vs size curve. * k_quants: forgot to add the Metal changes in last commit * k_quants: change Q5_K to be type 0 when QK_K = 64 Still needs AVX2 implementation * k_quants: AVX2 implementation for new 64-weight Q5_K * k_quants: 10% faster ARM_NEON Q5_K dot product * k_quants: fixed issue caused by merging with master --------- Co-authored-by: Iwan Kawrakow <iwan.kawrakow@gmail.com>
2023-06-26 16:43:07 +00:00
ggml : add SOTA 2,3,4,5,6 bit k-quantizations (#1684) * Starting to add k-quantization to ggml I think it is better to have quantization separate from ggml. For now just adding the k-quants there, but it would be better to also factor out the existing ggml quantizations. * Adding Q3_K and Q8_K (de)-quantization * Q3_K now working on CUDA and AVX2/scalar CUDA is not ideal - ~50% slower than Q4_0 for single token prediction, about the same in batch mode (perplexity). CPU single token is ~55 ms (on Ryzen 7950X). * Some improvement for Q3_K on CUDA It is now ~22.5 ms/token on my GPU, so ~30% slower than Q4_0. * Some more CUDA optimizations for Q3_K Single token is now 20.5 ms/token (~20% slower than Q4_0). Perplexity is on par with Q4_0. * Adding Q4_K - scalar, AVX2, CUDA Performance is the same or perhaps very slightly better than Q4_0 on the CPU. On the GPU, single token prediction is ~10% better than Q4_0, batch mode (perplexity is about the same). * Adding Q6_K - scalar, AVX2, CUDA Performance is ~40% lower compared to Q4_K on the CPU. This is to be expected, considering that we are memory bound on the CPU and the 6-bit model is ~44% larger than the 4-bit. On the GPU, single token prediction is ~6% lower than Q4_0, batch mode (perplexity) is even closer (but still slower). * Adding Q5_K - scalar, AVX2, CUDA Performance is ~20% lower compared to Q4_K on the CPU. This is to be expected, considering that we are memory bound on the CPU and the 5-bit model is ~22% larger than the 4-bit. On the GPU, single token prediction is about the same as Q4_0 for both, single token and batch prediction. * Per convention, all QX_K quantizations use Q5_K for output.weight * Adding quantization mixes * Quantization mixes: didn't quite get what I wanted in the last commit * Q4_K dot product for ARM_NEON * Q6_K dot product for ARM_NEON * Q5_K dot product for ARM_NEON * Adding Q3_K dot for ARM_NEON It is 22% slower than Q4_K, despite the smaller model size. On x86_64, where we are memory bound, the Q3_K model is quite a bit faster than Q4_K. * A very slightly faster ARM_NEON Q3_K dot * Adding Q2_K - just CUDA for now Token prediction is pretty good - about 15.5 ms on a RTX 4080. Perplexity is about the same as Q4_K. * Adding scalar and AVX2 Q2_K dot * Adding ARM_NEON Q2_K dot About the same performance as Q4_K. * A slightly faster ARM_NEON Q2_K dot Single token prediction is now ~36 ms on M2 Max. The code is much simpler too. * Fixed bug in Q2_K CUDA dot product kernel Stranegly enough, for the few prompts I tried with the 7B model the responses looked perfectly reasonable. Only realized something is not quite right when I tried the larger models and started getting nonse back. In any case, Q2_K single token evaluation time on an RTX 4080 in a Ryzen7950X box iusing CUDA and model fully loaded on the GPU are ~15.5 ms for 7B, ~25.4 ms for 13B, and ~55.8 ms for 30B. The max number of layers that fit in VRAM for The 65B is 32. With that, we get ~330 ms per token, which is not that much faster than just running on the CPU (~470 ms per token). * Don't print zeros/NaNs when no count histogram has been collected * A 10% faster CUDA vector dot kernel for Q3_K Q3_K is now running at ~18.5 ms / token on CUDA, so the gap to Q4_0 is only 10%. It seems memory acccess pattern is more important for performance than the amount of computation the kernel does. * A slightly daster Q4_K AVX2 dot product For perplexity, where we are less memory bound, time per pass drops by ~5%. Barely measurable difference for single token prediction. * A slightly faster ARM_NEON A4_K dot product * Minor * Fix quantization error test We cannot possibly be expecting rmse < 0.002 for 2- and 3-bit quantization variants. * Fix docker build I have been sloppy with vector reinterpret casts on ARM_NEON. It seems clang is very forgiving in that regard. * Added forgotten ggml.o dependence on k_quants.h to the Makefile * Had unintentionally committed the Makefile with -Ofast enabled * ggml : rename k_quants -> ggml-quants-k, use lowercase in code --------- Co-authored-by: Iwan Kawrakow <iwan.kawrakow@gmail.com> Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
2023-06-05 19:56:18 +00:00
memcpy(utmp, x[i].scales, 12);
utmp[3] = ((utmp[2] >> 4) & kmask2) | (((utmp[1] >> 6) & kmask3) << 4);
const uint32_t uaux = utmp[1] & kmask1;
utmp[1] = (utmp[2] & kmask2) | (((utmp[0] >> 6) & kmask3) << 4);
utmp[2] = uaux;
utmp[0] &= kmask1;
k-quants : support for super-block size of 64 (#2001) * k_quants: WIP super-blocks with 64 weights * k_quants: WIP super-blocks with 64 weights Q6_K scalar and AVX2 works * k_quants: WIP super-blocks with 64 weights Q4_K scalar and AVX2 works * k_quants: WIP super-blocks with 64 weights Q2_K scalar and AVX2 works. Q2_K is way too slow (it is actually slower than the scalar implementation) * k_quants: WIP super-blocks with 64 weights Q3_K scalar and AVX2 works. * k_quants: WIP super-blocks with 64 weights Q5_K scalar and AVX2 works, and with that all k_quants are done on AVX2 and scalar * k_quants: WIP super-blocks with 64 weights Q6_K working on CUDA. Cannot make it run quite as gast as with super-blocks with 256 weigths: 8% slower on 4080, 20% slower on the 1660 (but there we fit 1 less layer on the GPU because pf the larger model size), so some fraction of these 20% is due to that, * k_quants: WIP super-blocks with 64 weights Q4_K working on CUDA. ~10% slower on GTX-1660, 16% slower on 4080. * k_quants: WIP super-blocks with 64 weights Q2_K working on CUDA. ~3% slower on GTX-1660, 10% slower on 4080. * k_quants: WIP super-blocks with 64 weights Q3_K working on CUDA. * k_quants: WIP super-blocks with 64 weights Q5_K working on CUDA, and with this CUDA is done. * k_quants: WIP super-blocks with 64 weights Q6_K working on ARM_NEON * k_quants: WIP super-blocks with 64 weights Q4_K working on ARM_NEON, but quite a bit slower than 256 weights * k_quants: WIP super-blocks with 64 weights Q2_K working on ARM_NEON, but quite a bit slower than 256 weights * k_quants: WIP super-blocks with 64 weights Q3_K working on ARM_NEON, but quite a bit slower than 256 weights. * k_quants: WIP super-blocks with 64 weights Q5_K working on ARM_NEON, but quite a bit slower than 256 weights. With that, we have full support for ARM_NEON, although performance is not quite there. * k_quants: WIP super-blocks with 64 weights Slightly more efficient Q3_K and Q5_K * k_quants: WIP super-blocks with 64 weights Another small improvement for Q3_K and Q5_K on ARM_NEON * k_quants: WIP super-blocks with 64 weights Yet another speedup for Q5_K on ARM_NEON. We are now within 10% of the QK_K = 256 version. * k_quants: WIP super-blocks with 64 weights * We are able to pass preprocessor macros to the Metal compiler * Q6_K works and is actually slightly more efficient than the QK_K = 256 version (25.2 ms vs 25.8 ms) * k_quants: WIP super-blocks with 64 weights Q4_K works on Metal and is actually slightly faster than QK_K = 256 (21.95 ms vs 24.0 ms). * k_quants: WIP super-blocks with 64 weights Q2_K works on Metal and is very slightly faster than QK_K = 256 (23.8 ms vs 24.2 ms). * k_quants: WIP super-blocks with 64 weights Q3_K works on Metal and is slightly faster than QK_K = 256 (26.6 ms vs 28.3 ms). * k_quants: WIP super-blocks with 64 weights Q5_K works on Metal and is slightly faster than QK_K = 256 (23.7 ms vs 26.3 ms). * k_quants: call them _K, not _k, also on Metal * k_quants: correctly define QK_K in llama.cpp * Fixed bug in q4_K quantization added with the 64-block addition * Simplify via lambda * k_quants: swicth Q3_K to 4-bit scales when QK_K = 64 Otherwise there isn't much benefit from this quantization type. There is some very slight loss in accuracy, but we reduce size by ~7%. E.g., for OpenLLaMA-3B, Q3_K_S perplexity is 8.6131 with 8-bit scales and 8.6352 with 4-bit, while file size decreases from 1.53G to 1.44G. * k_quants: switch Q4_K to 4-bit scales when QK_K = 64 Here the loss in accuracy is greater than for Q3_K, but the Q4_K points still move further to the left on the perplexity vs size curve. * k_quants: forgot to add the Metal changes in last commit * k_quants: change Q5_K to be type 0 when QK_K = 64 Still needs AVX2 implementation * k_quants: AVX2 implementation for new 64-weight Q5_K * k_quants: 10% faster ARM_NEON Q5_K dot product * k_quants: fixed issue caused by merging with master --------- Co-authored-by: Iwan Kawrakow <iwan.kawrakow@gmail.com>
2023-06-26 16:43:07 +00:00
#else
// TODO
const float d = 0, dmin = 0;
#endif
ggml : add SOTA 2,3,4,5,6 bit k-quantizations (#1684) * Starting to add k-quantization to ggml I think it is better to have quantization separate from ggml. For now just adding the k-quants there, but it would be better to also factor out the existing ggml quantizations. * Adding Q3_K and Q8_K (de)-quantization * Q3_K now working on CUDA and AVX2/scalar CUDA is not ideal - ~50% slower than Q4_0 for single token prediction, about the same in batch mode (perplexity). CPU single token is ~55 ms (on Ryzen 7950X). * Some improvement for Q3_K on CUDA It is now ~22.5 ms/token on my GPU, so ~30% slower than Q4_0. * Some more CUDA optimizations for Q3_K Single token is now 20.5 ms/token (~20% slower than Q4_0). Perplexity is on par with Q4_0. * Adding Q4_K - scalar, AVX2, CUDA Performance is the same or perhaps very slightly better than Q4_0 on the CPU. On the GPU, single token prediction is ~10% better than Q4_0, batch mode (perplexity is about the same). * Adding Q6_K - scalar, AVX2, CUDA Performance is ~40% lower compared to Q4_K on the CPU. This is to be expected, considering that we are memory bound on the CPU and the 6-bit model is ~44% larger than the 4-bit. On the GPU, single token prediction is ~6% lower than Q4_0, batch mode (perplexity) is even closer (but still slower). * Adding Q5_K - scalar, AVX2, CUDA Performance is ~20% lower compared to Q4_K on the CPU. This is to be expected, considering that we are memory bound on the CPU and the 5-bit model is ~22% larger than the 4-bit. On the GPU, single token prediction is about the same as Q4_0 for both, single token and batch prediction. * Per convention, all QX_K quantizations use Q5_K for output.weight * Adding quantization mixes * Quantization mixes: didn't quite get what I wanted in the last commit * Q4_K dot product for ARM_NEON * Q6_K dot product for ARM_NEON * Q5_K dot product for ARM_NEON * Adding Q3_K dot for ARM_NEON It is 22% slower than Q4_K, despite the smaller model size. On x86_64, where we are memory bound, the Q3_K model is quite a bit faster than Q4_K. * A very slightly faster ARM_NEON Q3_K dot * Adding Q2_K - just CUDA for now Token prediction is pretty good - about 15.5 ms on a RTX 4080. Perplexity is about the same as Q4_K. * Adding scalar and AVX2 Q2_K dot * Adding ARM_NEON Q2_K dot About the same performance as Q4_K. * A slightly faster ARM_NEON Q2_K dot Single token prediction is now ~36 ms on M2 Max. The code is much simpler too. * Fixed bug in Q2_K CUDA dot product kernel Stranegly enough, for the few prompts I tried with the 7B model the responses looked perfectly reasonable. Only realized something is not quite right when I tried the larger models and started getting nonse back. In any case, Q2_K single token evaluation time on an RTX 4080 in a Ryzen7950X box iusing CUDA and model fully loaded on the GPU are ~15.5 ms for 7B, ~25.4 ms for 13B, and ~55.8 ms for 30B. The max number of layers that fit in VRAM for The 65B is 32. With that, we get ~330 ms per token, which is not that much faster than just running on the CPU (~470 ms per token). * Don't print zeros/NaNs when no count histogram has been collected * A 10% faster CUDA vector dot kernel for Q3_K Q3_K is now running at ~18.5 ms / token on CUDA, so the gap to Q4_0 is only 10%. It seems memory acccess pattern is more important for performance than the amount of computation the kernel does. * A slightly daster Q4_K AVX2 dot product For perplexity, where we are less memory bound, time per pass drops by ~5%. Barely measurable difference for single token prediction. * A slightly faster ARM_NEON A4_K dot product * Minor * Fix quantization error test We cannot possibly be expecting rmse < 0.002 for 2- and 3-bit quantization variants. * Fix docker build I have been sloppy with vector reinterpret casts on ARM_NEON. It seems clang is very forgiving in that regard. * Added forgotten ggml.o dependence on k_quants.h to the Makefile * Had unintentionally committed the Makefile with -Ofast enabled * ggml : rename k_quants -> ggml-quants-k, use lowercase in code --------- Co-authored-by: Iwan Kawrakow <iwan.kawrakow@gmail.com> Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
2023-06-05 19:56:18 +00:00
const __m256i mins_and_scales = _mm256_cvtepu8_epi16(_mm_set_epi32(utmp[3], utmp[2], utmp[1], utmp[0]));
const __m256i q8sums = _mm256_loadu_si256((const __m256i*)y[i].bsums);
const __m128i q8s = _mm_hadd_epi16(_mm256_extracti128_si256(q8sums, 0), _mm256_extracti128_si256(q8sums, 1));
const __m128i prod = _mm_madd_epi16(_mm256_extracti128_si256(mins_and_scales, 1), q8s);
const __m128i hsum = _mm_hadd_epi32(_mm_hadd_epi32(prod, mzero), mzero);
summs += dmin * _mm_extract_epi32(hsum, 0);
const __m128i sc128 = _mm256_extracti128_si256(mins_and_scales, 0);
ggml : workaround for missing _mm256_setr_m128i in GCC < 8 in k_quants.c (#2405)
2023-07-28 18:17:45 +00:00
const __m256i scales = MM256_SET_M128I(sc128, sc128);
ggml : add SOTA 2,3,4,5,6 bit k-quantizations (#1684) * Starting to add k-quantization to ggml I think it is better to have quantization separate from ggml. For now just adding the k-quants there, but it would be better to also factor out the existing ggml quantizations. * Adding Q3_K and Q8_K (de)-quantization * Q3_K now working on CUDA and AVX2/scalar CUDA is not ideal - ~50% slower than Q4_0 for single token prediction, about the same in batch mode (perplexity). CPU single token is ~55 ms (on Ryzen 7950X). * Some improvement for Q3_K on CUDA It is now ~22.5 ms/token on my GPU, so ~30% slower than Q4_0. * Some more CUDA optimizations for Q3_K Single token is now 20.5 ms/token (~20% slower than Q4_0). Perplexity is on par with Q4_0. * Adding Q4_K - scalar, AVX2, CUDA Performance is the same or perhaps very slightly better than Q4_0 on the CPU. On the GPU, single token prediction is ~10% better than Q4_0, batch mode (perplexity is about the same). * Adding Q6_K - scalar, AVX2, CUDA Performance is ~40% lower compared to Q4_K on the CPU. This is to be expected, considering that we are memory bound on the CPU and the 6-bit model is ~44% larger than the 4-bit. On the GPU, single token prediction is ~6% lower than Q4_0, batch mode (perplexity) is even closer (but still slower). * Adding Q5_K - scalar, AVX2, CUDA Performance is ~20% lower compared to Q4_K on the CPU. This is to be expected, considering that we are memory bound on the CPU and the 5-bit model is ~22% larger than the 4-bit. On the GPU, single token prediction is about the same as Q4_0 for both, single token and batch prediction. * Per convention, all QX_K quantizations use Q5_K for output.weight * Adding quantization mixes * Quantization mixes: didn't quite get what I wanted in the last commit * Q4_K dot product for ARM_NEON * Q6_K dot product for ARM_NEON * Q5_K dot product for ARM_NEON * Adding Q3_K dot for ARM_NEON It is 22% slower than Q4_K, despite the smaller model size. On x86_64, where we are memory bound, the Q3_K model is quite a bit faster than Q4_K. * A very slightly faster ARM_NEON Q3_K dot * Adding Q2_K - just CUDA for now Token prediction is pretty good - about 15.5 ms on a RTX 4080. Perplexity is about the same as Q4_K. * Adding scalar and AVX2 Q2_K dot * Adding ARM_NEON Q2_K dot About the same performance as Q4_K. * A slightly faster ARM_NEON Q2_K dot Single token prediction is now ~36 ms on M2 Max. The code is much simpler too. * Fixed bug in Q2_K CUDA dot product kernel Stranegly enough, for the few prompts I tried with the 7B model the responses looked perfectly reasonable. Only realized something is not quite right when I tried the larger models and started getting nonse back. In any case, Q2_K single token evaluation time on an RTX 4080 in a Ryzen7950X box iusing CUDA and model fully loaded on the GPU are ~15.5 ms for 7B, ~25.4 ms for 13B, and ~55.8 ms for 30B. The max number of layers that fit in VRAM for The 65B is 32. With that, we get ~330 ms per token, which is not that much faster than just running on the CPU (~470 ms per token). * Don't print zeros/NaNs when no count histogram has been collected * A 10% faster CUDA vector dot kernel for Q3_K Q3_K is now running at ~18.5 ms / token on CUDA, so the gap to Q4_0 is only 10%. It seems memory acccess pattern is more important for performance than the amount of computation the kernel does. * A slightly daster Q4_K AVX2 dot product For perplexity, where we are less memory bound, time per pass drops by ~5%. Barely measurable difference for single token prediction. * A slightly faster ARM_NEON A4_K dot product * Minor * Fix quantization error test We cannot possibly be expecting rmse < 0.002 for 2- and 3-bit quantization variants. * Fix docker build I have been sloppy with vector reinterpret casts on ARM_NEON. It seems clang is very forgiving in that regard. * Added forgotten ggml.o dependence on k_quants.h to the Makefile * Had unintentionally committed the Makefile with -Ofast enabled * ggml : rename k_quants -> ggml-quants-k, use lowercase in code --------- Co-authored-by: Iwan Kawrakow <iwan.kawrakow@gmail.com> Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
2023-06-05 19:56:18 +00:00
const __m256i hbits = _mm256_loadu_si256((const __m256i*)x[i].qh);
__m256i hmask = mone;
__m256i sumi = _mm256_setzero_si256();
int bit = 0;
for (int j = 0; j < QK_K/64; ++j) {
const __m256i scale_0 = _mm256_shuffle_epi8(scales, get_scale_shuffle_k4(2*j+0));
const __m256i scale_1 = _mm256_shuffle_epi8(scales, get_scale_shuffle_k4(2*j+1));
const __m256i q5bits = _mm256_loadu_si256((const __m256i*)q5); q5 += 32;
const __m256i q5l_0 = _mm256_and_si256(q5bits, m4);
const __m256i q5h_0 = _mm256_slli_epi16(_mm256_srli_epi16(_mm256_and_si256(hbits, hmask), bit++), 4);
const __m256i q5_0 = _mm256_add_epi8(q5l_0, q5h_0);
hmask = _mm256_slli_epi16(hmask, 1);
const __m256i q5l_1 = _mm256_and_si256(_mm256_srli_epi16(q5bits, 4), m4);
const __m256i q5h_1 = _mm256_slli_epi16(_mm256_srli_epi16(_mm256_and_si256(hbits, hmask), bit++), 4);
const __m256i q5_1 = _mm256_add_epi8(q5l_1, q5h_1);
hmask = _mm256_slli_epi16(hmask, 1);
const __m256i q8_0 = _mm256_loadu_si256((const __m256i*)q8); q8 += 32;
const __m256i q8_1 = _mm256_loadu_si256((const __m256i*)q8); q8 += 32;
__m256i p16_0 = _mm256_maddubs_epi16(q5_0, q8_0);
__m256i p16_1 = _mm256_maddubs_epi16(q5_1, q8_1);
p16_0 = _mm256_madd_epi16(scale_0, p16_0);
p16_1 = _mm256_madd_epi16(scale_1, p16_1);
sumi = _mm256_add_epi32(sumi, _mm256_add_epi32(p16_0, p16_1));
}
__m256 vd = _mm256_set1_ps(d);
acc = _mm256_fmadd_ps(vd, _mm256_cvtepi32_ps(sumi), acc);
}
*s = hsum_float_8(acc) + summs;
k-quants : add AVX support to dot functions (#1916) * k_quants : add AVX support * k_quants : apply review comments
2023-06-26 16:46:07 +00:00
#elif defined __AVX__
const __m128i m4 = _mm_set1_epi8(0xF);
const __m128i mzero = _mm_setzero_si128();
const __m128i mone = _mm_set1_epi8(1);
const __m128i m2 = _mm_set1_epi8(2);
__m256 acc = _mm256_setzero_ps();
float summs = 0.f;
for (int i = 0; i < nb; ++i) {
ggml : move FP16 <-> FP32 code to ggml-impl.h (#3861) * ggml : move FP16 <-> FP32 stuff to ggml-impl.h ggml-ci * tests : fix ARM build * ggml : explicitly initialize deprecated type traits * ggml : add math.h to ggml-impl.h * ggml : remove duplicate static assert macros * ggml : prefix lookup tables with ggml_ ggml-ci * ggml-impl : move extern "C" to start of file
2023-10-30 17:19:15 +00:00
const float d = y[i].d * GGML_FP16_TO_FP32(x[i].d);
const float dmin = -y[i].d * GGML_FP16_TO_FP32(x[i].dmin);
k-quants : add AVX support to dot functions (#1916) * k_quants : add AVX support * k_quants : apply review comments
2023-06-26 16:46:07 +00:00
const uint8_t * restrict q5 = x[i].qs;
const int8_t * restrict q8 = y[i].qs;
memcpy(utmp, x[i].scales, 12);
utmp[3] = ((utmp[2] >> 4) & kmask2) | (((utmp[1] >> 6) & kmask3) << 4);
const uint32_t uaux = utmp[1] & kmask1;
utmp[1] = (utmp[2] & kmask2) | (((utmp[0] >> 6) & kmask3) << 4);
utmp[2] = uaux;
utmp[0] &= kmask1;
const __m128i utmps = _mm_set_epi32(utmp[3], utmp[2], utmp[1], utmp[0]);
const __m128i scales = _mm_cvtepu8_epi16(utmps);
const __m128i mins = _mm_cvtepu8_epi16(_mm_unpackhi_epi64(utmps, utmps));
const __m128i q8sums_0 = _mm_loadu_si128((const __m128i*)&y[i].bsums[0]);
const __m128i q8sums_1 = _mm_loadu_si128((const __m128i*)&y[i].bsums[8]);
const __m128i q8s = _mm_hadd_epi16(q8sums_0, q8sums_1);
const __m128i prod = _mm_madd_epi16(mins, q8s);
const __m128i hsum = _mm_hadd_epi32(_mm_hadd_epi32(prod, mzero), mzero);
summs += dmin * _mm_extract_epi32(hsum, 0);
const __m128i hbits_0 = _mm_loadu_si128((const __m128i*)&x[i].qh[0]);
const __m128i hbits_1 = _mm_loadu_si128((const __m128i*)&x[i].qh[16]);
__m128i hmask = mone;
__m128i sumi_0 = _mm_setzero_si128();
__m128i sumi_1 = _mm_setzero_si128();
int bit = 0;
__m128i shuffle = _mm_set1_epi16(0x0100);
for (int j = 0; j < QK_K/64; ++j) {
const __m128i scale_0 = _mm_shuffle_epi8(scales, shuffle);
shuffle = _mm_add_epi16(shuffle, m2);
const __m128i scale_1 = _mm_shuffle_epi8(scales, shuffle);
shuffle = _mm_add_epi16(shuffle, m2);
const __m128i q5bits_0 = _mm_loadu_si128((const __m128i*)q5); q5 += 16;
const __m128i q5bits_1 = _mm_loadu_si128((const __m128i*)q5); q5 += 16;
__m128i q5l_0 = _mm_and_si128(q5bits_0, m4);
__m128i q5l_1 = _mm_and_si128(q5bits_1, m4);
__m128i q5h_0 = _mm_slli_epi16(_mm_srli_epi16(_mm_and_si128(hbits_0, hmask), bit), 4);
__m128i q5h_1 = _mm_slli_epi16(_mm_srli_epi16(_mm_and_si128(hbits_1, hmask), bit++), 4);
__m128i q5_0 = _mm_add_epi8(q5l_0, q5h_0);
__m128i q5_1 = _mm_add_epi8(q5l_1, q5h_1);
hmask = _mm_slli_epi16(hmask, 1);
__m128i q8_0 = _mm_loadu_si128((const __m128i*)q8); q8 += 16;
__m128i q8_1 = _mm_loadu_si128((const __m128i*)q8); q8 += 16;
__m128i p16_0 = _mm_maddubs_epi16(q5_0, q8_0);
__m128i p16_1 = _mm_maddubs_epi16(q5_1, q8_1);
p16_0 = _mm_madd_epi16(scale_0, p16_0);
p16_1 = _mm_madd_epi16(scale_0, p16_1);
q5l_0 = _mm_and_si128(_mm_srli_epi16(q5bits_0, 4), m4);
q5l_1 = _mm_and_si128(_mm_srli_epi16(q5bits_1, 4), m4);
q5h_0 = _mm_slli_epi16(_mm_srli_epi16(_mm_and_si128(hbits_0, hmask), bit), 4);
q5h_1 = _mm_slli_epi16(_mm_srli_epi16(_mm_and_si128(hbits_1, hmask), bit++), 4);
q5_0 = _mm_add_epi8(q5l_0, q5h_0);
q5_1 = _mm_add_epi8(q5l_1, q5h_1);
hmask = _mm_slli_epi16(hmask, 1);
q8_0 = _mm_loadu_si128((const __m128i*)q8); q8 += 16;
q8_1 = _mm_loadu_si128((const __m128i*)q8); q8 += 16;
__m128i p16_2 = _mm_maddubs_epi16(q5_0, q8_0);
__m128i p16_3 = _mm_maddubs_epi16(q5_1, q8_1);
p16_2 = _mm_madd_epi16(scale_1, p16_2);
p16_3 = _mm_madd_epi16(scale_1, p16_3);
sumi_0 = _mm_add_epi32(sumi_0, _mm_add_epi32(p16_0, p16_2));
sumi_1 = _mm_add_epi32(sumi_1, _mm_add_epi32(p16_1, p16_3));
}
__m256 vd = _mm256_set1_ps(d);
ggml : workaround for missing _mm256_setr_m128i in GCC < 8 in k_quants.c (#2405)
2023-07-28 18:17:45 +00:00
__m256i sumi = MM256_SET_M128I(sumi_1, sumi_0);
k-quants : add AVX support to dot functions (#1916) * k_quants : add AVX support * k_quants : apply review comments
2023-06-26 16:46:07 +00:00
acc = _mm256_add_ps(_mm256_mul_ps(vd, _mm256_cvtepi32_ps(sumi)), acc);
}
*s = hsum_float_8(acc) + summs;
ggml : add RISC-V Vector Support for K-Quants and improved the existing intrinsics (#3453) * Added RVV intrinsics support for Q8 quantize row and also improved the existing dot product function for risc-v. The RVV intrinsics is added for the following quantize row functions quantize_row_q8_0 quantize_row_q8_1 The following dot product functions have also been optimized by using LMUL = 1/2 instead of LMUL = 1 ggml_vec_dot_q4_0_q8_0 ggml_vec_dot_q4_1_q8_1 ggml_vec_dot_q5_0_q8_0 ggml_vec_dot_q5_1_q8_1 And vector initialization in Q5 by temporary array is also replaced by the vid intrinsics Signed-off-by: Ahmad Tameem <ahmad.tameem@10xengineers.ai> * Added RVV intrinsics support for k_quants This adds RISC-V Vector intrinsics support for the following K_quants functions for both QKK = 256 and QKK = 64 ggml_vec_dot_q2_K_q8_K ggml_vec_dot_q3_K_q8_K ggml_vec_dot_q4_K_q8_K ggml_vec_dot_q5_K_q8_K ggml_vec_dot_q6_K_q8_K Signed-off-by: Ahmad Tameem <ahmad.tameem@10xengineers.ai> --------- Signed-off-by: Ahmad Tameem <ahmad.tameem@10xengineers.ai>
2023-10-03 18:38:19 +00:00
#elif defined __riscv_v_intrinsic
const uint8_t * scales = (const uint8_t*)&utmp[0];
const uint8_t * mins = (const uint8_t*)&utmp[2];
float sumf = 0;
float sums = 0.0;
size_t vl;
for (int i = 0; i < nb; ++i) {
vl = 8;
const uint8_t * restrict q5 = x[i].qs;
const uint8_t * restrict hm = x[i].qh;
const int8_t * restrict q8 = y[i].qs;
ggml : move FP16 <-> FP32 code to ggml-impl.h (#3861) * ggml : move FP16 <-> FP32 stuff to ggml-impl.h ggml-ci * tests : fix ARM build * ggml : explicitly initialize deprecated type traits * ggml : add math.h to ggml-impl.h * ggml : remove duplicate static assert macros * ggml : prefix lookup tables with ggml_ ggml-ci * ggml-impl : move extern "C" to start of file
2023-10-30 17:19:15 +00:00
const float d = GGML_FP16_TO_FP32(x[i].d) * y[i].d;
const float dmin = GGML_FP16_TO_FP32(x[i].dmin) * y[i].d;
ggml : add RISC-V Vector Support for K-Quants and improved the existing intrinsics (#3453) * Added RVV intrinsics support for Q8 quantize row and also improved the existing dot product function for risc-v. The RVV intrinsics is added for the following quantize row functions quantize_row_q8_0 quantize_row_q8_1 The following dot product functions have also been optimized by using LMUL = 1/2 instead of LMUL = 1 ggml_vec_dot_q4_0_q8_0 ggml_vec_dot_q4_1_q8_1 ggml_vec_dot_q5_0_q8_0 ggml_vec_dot_q5_1_q8_1 And vector initialization in Q5 by temporary array is also replaced by the vid intrinsics Signed-off-by: Ahmad Tameem <ahmad.tameem@10xengineers.ai> * Added RVV intrinsics support for k_quants This adds RISC-V Vector intrinsics support for the following K_quants functions for both QKK = 256 and QKK = 64 ggml_vec_dot_q2_K_q8_K ggml_vec_dot_q3_K_q8_K ggml_vec_dot_q4_K_q8_K ggml_vec_dot_q5_K_q8_K ggml_vec_dot_q6_K_q8_K Signed-off-by: Ahmad Tameem <ahmad.tameem@10xengineers.ai> --------- Signed-off-by: Ahmad Tameem <ahmad.tameem@10xengineers.ai>
2023-10-03 18:38:19 +00:00
vint16mf2_t q8sums_0 = __riscv_vlse16_v_i16mf2(y[i].bsums, 4, vl);
vint16mf2_t q8sums_1 = __riscv_vlse16_v_i16mf2(y[i].bsums+1, 4, vl);
vint16mf2_t q8sums = __riscv_vadd_vv_i16mf2(q8sums_0, q8sums_1, vl);
memcpy(utmp, x[i].scales, 12);
utmp[3] = ((utmp[2] >> 4) & kmask2) | (((utmp[1] >> 6) & kmask3) << 4);
const uint32_t uaux = utmp[1] & kmask1;
utmp[1] = (utmp[2] & kmask2) | (((utmp[0] >> 6) & kmask3) << 4);
utmp[2] = uaux;
utmp[0] &= kmask1;
vuint8mf4_t mins8 = __riscv_vle8_v_u8mf4(mins, vl);
vint16mf2_t v_mins = __riscv_vreinterpret_v_u16mf2_i16mf2(__riscv_vzext_vf2_u16mf2(mins8, vl));
vint32m1_t prod = __riscv_vwmul_vv_i32m1(q8sums, v_mins, vl);
vint32m1_t sumi = __riscv_vredsum_vs_i32m1_i32m1(prod, __riscv_vmv_v_x_i32m1(0, 1), vl);
sumf -= dmin * __riscv_vmv_x_s_i32m1_i32(sumi);
vl = 32;
int32_t aux32 = 0;
int is = 0;
uint8_t m = 1;
vint32m1_t vzero = __riscv_vmv_v_x_i32m1(0, 1);
vuint8m1_t vqh = __riscv_vle8_v_u8m1(hm, vl);
for (int j = 0; j < QK_K/64; ++j) {
// load Q5 and Q8
vuint8m1_t q5_x = __riscv_vle8_v_u8m1(q5, vl);
vint8m1_t q8_y1 = __riscv_vle8_v_i8m1(q8, vl);
vint8m1_t q8_y2 = __riscv_vle8_v_i8m1(q8+32, vl);
// compute mask for addition
vint8m1_t q5_a = __riscv_vreinterpret_v_u8m1_i8m1(__riscv_vand_vx_u8m1(q5_x, 0x0F, vl));
vuint8m1_t qh_m1 = __riscv_vand_vx_u8m1(vqh, m, vl);
vbool8_t vmask_1 = __riscv_vmsne_vx_u8m1_b8(qh_m1, 0, vl);
vint8m1_t q5_m1 = __riscv_vadd_vx_i8m1_m(vmask_1, q5_a, 16, vl);
m <<= 1;
vint8m1_t q5_l = __riscv_vreinterpret_v_u8m1_i8m1(__riscv_vsrl_vx_u8m1(q5_x, 0x04, vl));
vuint8m1_t qh_m2 = __riscv_vand_vx_u8m1(vqh, m, vl);
vbool8_t vmask_2 = __riscv_vmsne_vx_u8m1_b8(qh_m2, 0, vl);
vint8m1_t q5_m2 = __riscv_vadd_vx_i8m1_m(vmask_2, q5_l, 16, vl);
m <<= 1;
vint16m2_t v0 = __riscv_vwmul_vv_i16m2(q5_m1, q8_y1, vl);
vint16m2_t v1 = __riscv_vwmul_vv_i16m2(q5_m2, q8_y2, vl);
vint32m4_t vs1 = __riscv_vwmul_vx_i32m4(v0, scales[is++], vl);
vint32m4_t vs2 = __riscv_vwmul_vx_i32m4(v1, scales[is++], vl);
vint32m1_t vacc1 = __riscv_vredsum_vs_i32m4_i32m1(vs1, vzero, vl);
vint32m1_t vacc2 = __riscv_vredsum_vs_i32m4_i32m1(vs2, vzero, vl);
aux32 += __riscv_vmv_x_s_i32m1_i32(vacc1) + __riscv_vmv_x_s_i32m1_i32(vacc2);
q5 += 32; q8 += 64;
}
vfloat32m1_t vaux = __riscv_vfmul_vf_f32m1(__riscv_vfmv_v_f_f32m1(aux32, 1), d, 1);
sums += __riscv_vfmv_f_s_f32m1_f32(vaux);
}
*s = sumf+sums;
ggml : add SOTA 2,3,4,5,6 bit k-quantizations (#1684) * Starting to add k-quantization to ggml I think it is better to have quantization separate from ggml. For now just adding the k-quants there, but it would be better to also factor out the existing ggml quantizations. * Adding Q3_K and Q8_K (de)-quantization * Q3_K now working on CUDA and AVX2/scalar CUDA is not ideal - ~50% slower than Q4_0 for single token prediction, about the same in batch mode (perplexity). CPU single token is ~55 ms (on Ryzen 7950X). * Some improvement for Q3_K on CUDA It is now ~22.5 ms/token on my GPU, so ~30% slower than Q4_0. * Some more CUDA optimizations for Q3_K Single token is now 20.5 ms/token (~20% slower than Q4_0). Perplexity is on par with Q4_0. * Adding Q4_K - scalar, AVX2, CUDA Performance is the same or perhaps very slightly better than Q4_0 on the CPU. On the GPU, single token prediction is ~10% better than Q4_0, batch mode (perplexity is about the same). * Adding Q6_K - scalar, AVX2, CUDA Performance is ~40% lower compared to Q4_K on the CPU. This is to be expected, considering that we are memory bound on the CPU and the 6-bit model is ~44% larger than the 4-bit. On the GPU, single token prediction is ~6% lower than Q4_0, batch mode (perplexity) is even closer (but still slower). * Adding Q5_K - scalar, AVX2, CUDA Performance is ~20% lower compared to Q4_K on the CPU. This is to be expected, considering that we are memory bound on the CPU and the 5-bit model is ~22% larger than the 4-bit. On the GPU, single token prediction is about the same as Q4_0 for both, single token and batch prediction. * Per convention, all QX_K quantizations use Q5_K for output.weight * Adding quantization mixes * Quantization mixes: didn't quite get what I wanted in the last commit * Q4_K dot product for ARM_NEON * Q6_K dot product for ARM_NEON * Q5_K dot product for ARM_NEON * Adding Q3_K dot for ARM_NEON It is 22% slower than Q4_K, despite the smaller model size. On x86_64, where we are memory bound, the Q3_K model is quite a bit faster than Q4_K. * A very slightly faster ARM_NEON Q3_K dot * Adding Q2_K - just CUDA for now Token prediction is pretty good - about 15.5 ms on a RTX 4080. Perplexity is about the same as Q4_K. * Adding scalar and AVX2 Q2_K dot * Adding ARM_NEON Q2_K dot About the same performance as Q4_K. * A slightly faster ARM_NEON Q2_K dot Single token prediction is now ~36 ms on M2 Max. The code is much simpler too. * Fixed bug in Q2_K CUDA dot product kernel Stranegly enough, for the few prompts I tried with the 7B model the responses looked perfectly reasonable. Only realized something is not quite right when I tried the larger models and started getting nonse back. In any case, Q2_K single token evaluation time on an RTX 4080 in a Ryzen7950X box iusing CUDA and model fully loaded on the GPU are ~15.5 ms for 7B, ~25.4 ms for 13B, and ~55.8 ms for 30B. The max number of layers that fit in VRAM for The 65B is 32. With that, we get ~330 ms per token, which is not that much faster than just running on the CPU (~470 ms per token). * Don't print zeros/NaNs when no count histogram has been collected * A 10% faster CUDA vector dot kernel for Q3_K Q3_K is now running at ~18.5 ms / token on CUDA, so the gap to Q4_0 is only 10%. It seems memory acccess pattern is more important for performance than the amount of computation the kernel does. * A slightly daster Q4_K AVX2 dot product For perplexity, where we are less memory bound, time per pass drops by ~5%. Barely measurable difference for single token prediction. * A slightly faster ARM_NEON A4_K dot product * Minor * Fix quantization error test We cannot possibly be expecting rmse < 0.002 for 2- and 3-bit quantization variants. * Fix docker build I have been sloppy with vector reinterpret casts on ARM_NEON. It seems clang is very forgiving in that regard. * Added forgotten ggml.o dependence on k_quants.h to the Makefile * Had unintentionally committed the Makefile with -Ofast enabled * ggml : rename k_quants -> ggml-quants-k, use lowercase in code --------- Co-authored-by: Iwan Kawrakow <iwan.kawrakow@gmail.com> Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
2023-06-05 19:56:18 +00:00
#else
const uint8_t * scales = (const uint8_t*)&utmp[0];
const uint8_t * mins = (const uint8_t*)&utmp[2];
int8_t aux8[QK_K];
int16_t aux16[8];
float sums [8];
int32_t aux32[8];
memset(sums, 0, 8*sizeof(float));
float sumf = 0;
for (int i = 0; i < nb; ++i) {
const uint8_t * restrict q4 = x[i].qs;
const uint8_t * restrict hm = x[i].qh;
const int8_t * restrict q8 = y[i].qs;
memset(aux32, 0, 8*sizeof(int32_t));
int8_t * restrict a = aux8;
uint8_t m = 1;
for (int j = 0; j < QK_K/64; ++j) {
for (int l = 0; l < 32; ++l) a[l] = (int8_t)(q4[l] & 0xF);
for (int l = 0; l < 32; ++l) a[l] += (hm[l] & m ? 16 : 0);
a += 32; m <<= 1;
for (int l = 0; l < 32; ++l) a[l] = (int8_t)(q4[l] >> 4);
for (int l = 0; l < 32; ++l) a[l] += (hm[l] & m ? 16 : 0);
a += 32; m <<= 1;
q4 += 32;
}
memcpy(utmp, x[i].scales, 12);
utmp[3] = ((utmp[2] >> 4) & kmask2) | (((utmp[1] >> 6) & kmask3) << 4);
const uint32_t uaux = utmp[1] & kmask1;
utmp[1] = (utmp[2] & kmask2) | (((utmp[0] >> 6) & kmask3) << 4);
utmp[2] = uaux;
utmp[0] &= kmask1;
int sumi = 0;
for (int j = 0; j < QK_K/16; ++j) sumi += y[i].bsums[j] * mins[j/2];
a = aux8;
int is = 0;
for (int j = 0; j < QK_K/32; ++j) {
int32_t scale = scales[is++];
for (int l = 0; l < 8; ++l) aux16[l] = q8[l] * a[l];
for (int l = 0; l < 8; ++l) aux32[l] += scale * aux16[l];
q8 += 8; a += 8;
for (int l = 0; l < 8; ++l) aux16[l] = q8[l] * a[l];
for (int l = 0; l < 8; ++l) aux32[l] += scale * aux16[l];
q8 += 8; a += 8;
for (int l = 0; l < 8; ++l) aux16[l] = q8[l] * a[l];
for (int l = 0; l < 8; ++l) aux32[l] += scale * aux16[l];
q8 += 8; a += 8;
for (int l = 0; l < 8; ++l) aux16[l] = q8[l] * a[l];
for (int l = 0; l < 8; ++l) aux32[l] += scale * aux16[l];
q8 += 8; a += 8;
}
ggml : move FP16 <-> FP32 code to ggml-impl.h (#3861) * ggml : move FP16 <-> FP32 stuff to ggml-impl.h ggml-ci * tests : fix ARM build * ggml : explicitly initialize deprecated type traits * ggml : add math.h to ggml-impl.h * ggml : remove duplicate static assert macros * ggml : prefix lookup tables with ggml_ ggml-ci * ggml-impl : move extern "C" to start of file
2023-10-30 17:19:15 +00:00
const float d = GGML_FP16_TO_FP32(x[i].d) * y[i].d;
ggml : add SOTA 2,3,4,5,6 bit k-quantizations (#1684) * Starting to add k-quantization to ggml I think it is better to have quantization separate from ggml. For now just adding the k-quants there, but it would be better to also factor out the existing ggml quantizations. * Adding Q3_K and Q8_K (de)-quantization * Q3_K now working on CUDA and AVX2/scalar CUDA is not ideal - ~50% slower than Q4_0 for single token prediction, about the same in batch mode (perplexity). CPU single token is ~55 ms (on Ryzen 7950X). * Some improvement for Q3_K on CUDA It is now ~22.5 ms/token on my GPU, so ~30% slower than Q4_0. * Some more CUDA optimizations for Q3_K Single token is now 20.5 ms/token (~20% slower than Q4_0). Perplexity is on par with Q4_0. * Adding Q4_K - scalar, AVX2, CUDA Performance is the same or perhaps very slightly better than Q4_0 on the CPU. On the GPU, single token prediction is ~10% better than Q4_0, batch mode (perplexity is about the same). * Adding Q6_K - scalar, AVX2, CUDA Performance is ~40% lower compared to Q4_K on the CPU. This is to be expected, considering that we are memory bound on the CPU and the 6-bit model is ~44% larger than the 4-bit. On the GPU, single token prediction is ~6% lower than Q4_0, batch mode (perplexity) is even closer (but still slower). * Adding Q5_K - scalar, AVX2, CUDA Performance is ~20% lower compared to Q4_K on the CPU. This is to be expected, considering that we are memory bound on the CPU and the 5-bit model is ~22% larger than the 4-bit. On the GPU, single token prediction is about the same as Q4_0 for both, single token and batch prediction. * Per convention, all QX_K quantizations use Q5_K for output.weight * Adding quantization mixes * Quantization mixes: didn't quite get what I wanted in the last commit * Q4_K dot product for ARM_NEON * Q6_K dot product for ARM_NEON * Q5_K dot product for ARM_NEON * Adding Q3_K dot for ARM_NEON It is 22% slower than Q4_K, despite the smaller model size. On x86_64, where we are memory bound, the Q3_K model is quite a bit faster than Q4_K. * A very slightly faster ARM_NEON Q3_K dot * Adding Q2_K - just CUDA for now Token prediction is pretty good - about 15.5 ms on a RTX 4080. Perplexity is about the same as Q4_K. * Adding scalar and AVX2 Q2_K dot * Adding ARM_NEON Q2_K dot About the same performance as Q4_K. * A slightly faster ARM_NEON Q2_K dot Single token prediction is now ~36 ms on M2 Max. The code is much simpler too. * Fixed bug in Q2_K CUDA dot product kernel Stranegly enough, for the few prompts I tried with the 7B model the responses looked perfectly reasonable. Only realized something is not quite right when I tried the larger models and started getting nonse back. In any case, Q2_K single token evaluation time on an RTX 4080 in a Ryzen7950X box iusing CUDA and model fully loaded on the GPU are ~15.5 ms for 7B, ~25.4 ms for 13B, and ~55.8 ms for 30B. The max number of layers that fit in VRAM for The 65B is 32. With that, we get ~330 ms per token, which is not that much faster than just running on the CPU (~470 ms per token). * Don't print zeros/NaNs when no count histogram has been collected * A 10% faster CUDA vector dot kernel for Q3_K Q3_K is now running at ~18.5 ms / token on CUDA, so the gap to Q4_0 is only 10%. It seems memory acccess pattern is more important for performance than the amount of computation the kernel does. * A slightly daster Q4_K AVX2 dot product For perplexity, where we are less memory bound, time per pass drops by ~5%. Barely measurable difference for single token prediction. * A slightly faster ARM_NEON A4_K dot product * Minor * Fix quantization error test We cannot possibly be expecting rmse < 0.002 for 2- and 3-bit quantization variants. * Fix docker build I have been sloppy with vector reinterpret casts on ARM_NEON. It seems clang is very forgiving in that regard. * Added forgotten ggml.o dependence on k_quants.h to the Makefile * Had unintentionally committed the Makefile with -Ofast enabled * ggml : rename k_quants -> ggml-quants-k, use lowercase in code --------- Co-authored-by: Iwan Kawrakow <iwan.kawrakow@gmail.com> Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
2023-06-05 19:56:18 +00:00
for (int l = 0; l < 8; ++l) sums[l] += d * aux32[l];
ggml : move FP16 <-> FP32 code to ggml-impl.h (#3861) * ggml : move FP16 <-> FP32 stuff to ggml-impl.h ggml-ci * tests : fix ARM build * ggml : explicitly initialize deprecated type traits * ggml : add math.h to ggml-impl.h * ggml : remove duplicate static assert macros * ggml : prefix lookup tables with ggml_ ggml-ci * ggml-impl : move extern "C" to start of file
2023-10-30 17:19:15 +00:00
const float dmin = GGML_FP16_TO_FP32(x[i].dmin) * y[i].d;
ggml : add SOTA 2,3,4,5,6 bit k-quantizations (#1684) * Starting to add k-quantization to ggml I think it is better to have quantization separate from ggml. For now just adding the k-quants there, but it would be better to also factor out the existing ggml quantizations. * Adding Q3_K and Q8_K (de)-quantization * Q3_K now working on CUDA and AVX2/scalar CUDA is not ideal - ~50% slower than Q4_0 for single token prediction, about the same in batch mode (perplexity). CPU single token is ~55 ms (on Ryzen 7950X). * Some improvement for Q3_K on CUDA It is now ~22.5 ms/token on my GPU, so ~30% slower than Q4_0. * Some more CUDA optimizations for Q3_K Single token is now 20.5 ms/token (~20% slower than Q4_0). Perplexity is on par with Q4_0. * Adding Q4_K - scalar, AVX2, CUDA Performance is the same or perhaps very slightly better than Q4_0 on the CPU. On the GPU, single token prediction is ~10% better than Q4_0, batch mode (perplexity is about the same). * Adding Q6_K - scalar, AVX2, CUDA Performance is ~40% lower compared to Q4_K on the CPU. This is to be expected, considering that we are memory bound on the CPU and the 6-bit model is ~44% larger than the 4-bit. On the GPU, single token prediction is ~6% lower than Q4_0, batch mode (perplexity) is even closer (but still slower). * Adding Q5_K - scalar, AVX2, CUDA Performance is ~20% lower compared to Q4_K on the CPU. This is to be expected, considering that we are memory bound on the CPU and the 5-bit model is ~22% larger than the 4-bit. On the GPU, single token prediction is about the same as Q4_0 for both, single token and batch prediction. * Per convention, all QX_K quantizations use Q5_K for output.weight * Adding quantization mixes * Quantization mixes: didn't quite get what I wanted in the last commit * Q4_K dot product for ARM_NEON * Q6_K dot product for ARM_NEON * Q5_K dot product for ARM_NEON * Adding Q3_K dot for ARM_NEON It is 22% slower than Q4_K, despite the smaller model size. On x86_64, where we are memory bound, the Q3_K model is quite a bit faster than Q4_K. * A very slightly faster ARM_NEON Q3_K dot * Adding Q2_K - just CUDA for now Token prediction is pretty good - about 15.5 ms on a RTX 4080. Perplexity is about the same as Q4_K. * Adding scalar and AVX2 Q2_K dot * Adding ARM_NEON Q2_K dot About the same performance as Q4_K. * A slightly faster ARM_NEON Q2_K dot Single token prediction is now ~36 ms on M2 Max. The code is much simpler too. * Fixed bug in Q2_K CUDA dot product kernel Stranegly enough, for the few prompts I tried with the 7B model the responses looked perfectly reasonable. Only realized something is not quite right when I tried the larger models and started getting nonse back. In any case, Q2_K single token evaluation time on an RTX 4080 in a Ryzen7950X box iusing CUDA and model fully loaded on the GPU are ~15.5 ms for 7B, ~25.4 ms for 13B, and ~55.8 ms for 30B. The max number of layers that fit in VRAM for The 65B is 32. With that, we get ~330 ms per token, which is not that much faster than just running on the CPU (~470 ms per token). * Don't print zeros/NaNs when no count histogram has been collected * A 10% faster CUDA vector dot kernel for Q3_K Q3_K is now running at ~18.5 ms / token on CUDA, so the gap to Q4_0 is only 10%. It seems memory acccess pattern is more important for performance than the amount of computation the kernel does. * A slightly daster Q4_K AVX2 dot product For perplexity, where we are less memory bound, time per pass drops by ~5%. Barely measurable difference for single token prediction. * A slightly faster ARM_NEON A4_K dot product * Minor * Fix quantization error test We cannot possibly be expecting rmse < 0.002 for 2- and 3-bit quantization variants. * Fix docker build I have been sloppy with vector reinterpret casts on ARM_NEON. It seems clang is very forgiving in that regard. * Added forgotten ggml.o dependence on k_quants.h to the Makefile * Had unintentionally committed the Makefile with -Ofast enabled * ggml : rename k_quants -> ggml-quants-k, use lowercase in code --------- Co-authored-by: Iwan Kawrakow <iwan.kawrakow@gmail.com> Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
2023-06-05 19:56:18 +00:00
sumf -= dmin * sumi;
}
for (int l = 0; l < 8; ++l) sumf += sums[l];
*s = sumf;
#endif
}
k-quants : support for super-block size of 64 (#2001) * k_quants: WIP super-blocks with 64 weights * k_quants: WIP super-blocks with 64 weights Q6_K scalar and AVX2 works * k_quants: WIP super-blocks with 64 weights Q4_K scalar and AVX2 works * k_quants: WIP super-blocks with 64 weights Q2_K scalar and AVX2 works. Q2_K is way too slow (it is actually slower than the scalar implementation) * k_quants: WIP super-blocks with 64 weights Q3_K scalar and AVX2 works. * k_quants: WIP super-blocks with 64 weights Q5_K scalar and AVX2 works, and with that all k_quants are done on AVX2 and scalar * k_quants: WIP super-blocks with 64 weights Q6_K working on CUDA. Cannot make it run quite as gast as with super-blocks with 256 weigths: 8% slower on 4080, 20% slower on the 1660 (but there we fit 1 less layer on the GPU because pf the larger model size), so some fraction of these 20% is due to that, * k_quants: WIP super-blocks with 64 weights Q4_K working on CUDA. ~10% slower on GTX-1660, 16% slower on 4080. * k_quants: WIP super-blocks with 64 weights Q2_K working on CUDA. ~3% slower on GTX-1660, 10% slower on 4080. * k_quants: WIP super-blocks with 64 weights Q3_K working on CUDA. * k_quants: WIP super-blocks with 64 weights Q5_K working on CUDA, and with this CUDA is done. * k_quants: WIP super-blocks with 64 weights Q6_K working on ARM_NEON * k_quants: WIP super-blocks with 64 weights Q4_K working on ARM_NEON, but quite a bit slower than 256 weights * k_quants: WIP super-blocks with 64 weights Q2_K working on ARM_NEON, but quite a bit slower than 256 weights * k_quants: WIP super-blocks with 64 weights Q3_K working on ARM_NEON, but quite a bit slower than 256 weights. * k_quants: WIP super-blocks with 64 weights Q5_K working on ARM_NEON, but quite a bit slower than 256 weights. With that, we have full support for ARM_NEON, although performance is not quite there. * k_quants: WIP super-blocks with 64 weights Slightly more efficient Q3_K and Q5_K * k_quants: WIP super-blocks with 64 weights Another small improvement for Q3_K and Q5_K on ARM_NEON * k_quants: WIP super-blocks with 64 weights Yet another speedup for Q5_K on ARM_NEON. We are now within 10% of the QK_K = 256 version. * k_quants: WIP super-blocks with 64 weights * We are able to pass preprocessor macros to the Metal compiler * Q6_K works and is actually slightly more efficient than the QK_K = 256 version (25.2 ms vs 25.8 ms) * k_quants: WIP super-blocks with 64 weights Q4_K works on Metal and is actually slightly faster than QK_K = 256 (21.95 ms vs 24.0 ms). * k_quants: WIP super-blocks with 64 weights Q2_K works on Metal and is very slightly faster than QK_K = 256 (23.8 ms vs 24.2 ms). * k_quants: WIP super-blocks with 64 weights Q3_K works on Metal and is slightly faster than QK_K = 256 (26.6 ms vs 28.3 ms). * k_quants: WIP super-blocks with 64 weights Q5_K works on Metal and is slightly faster than QK_K = 256 (23.7 ms vs 26.3 ms). * k_quants: call them _K, not _k, also on Metal * k_quants: correctly define QK_K in llama.cpp * Fixed bug in q4_K quantization added with the 64-block addition * Simplify via lambda * k_quants: swicth Q3_K to 4-bit scales when QK_K = 64 Otherwise there isn't much benefit from this quantization type. There is some very slight loss in accuracy, but we reduce size by ~7%. E.g., for OpenLLaMA-3B, Q3_K_S perplexity is 8.6131 with 8-bit scales and 8.6352 with 4-bit, while file size decreases from 1.53G to 1.44G. * k_quants: switch Q4_K to 4-bit scales when QK_K = 64 Here the loss in accuracy is greater than for Q3_K, but the Q4_K points still move further to the left on the perplexity vs size curve. * k_quants: forgot to add the Metal changes in last commit * k_quants: change Q5_K to be type 0 when QK_K = 64 Still needs AVX2 implementation * k_quants: AVX2 implementation for new 64-weight Q5_K * k_quants: 10% faster ARM_NEON Q5_K dot product * k_quants: fixed issue caused by merging with master --------- Co-authored-by: Iwan Kawrakow <iwan.kawrakow@gmail.com>
2023-06-26 16:43:07 +00:00
#else
void ggml_vec_dot_q5_K_q8_K(const int n, float * restrict s, const void * restrict vx, const void * restrict vy) {
assert(n % QK_K == 0);
const block_q5_K * restrict x = vx;
const block_q8_K * restrict y = vy;
const int nb = n / QK_K;
#ifdef __ARM_NEON
const uint8x16_t m4b = vdupq_n_u8(0xf);
const uint8x16_t mh = vdupq_n_u8(16);
k-quants : fix build on armv7 (android only) (#2920) * k-quants : fix build on armv7 * ggml : cleanup unused arm32 specific impl * k-quants : avoid some unused vzero / mzero define * ggml-alloc : use 4g for MEASURE_MAX_SIZE in 32-bit arm
2023-09-02 12:23:45 +00:00
const int32x4_t mzero = vdupq_n_s32(0);
k-quants : support for super-block size of 64 (#2001) * k_quants: WIP super-blocks with 64 weights * k_quants: WIP super-blocks with 64 weights Q6_K scalar and AVX2 works * k_quants: WIP super-blocks with 64 weights Q4_K scalar and AVX2 works * k_quants: WIP super-blocks with 64 weights Q2_K scalar and AVX2 works. Q2_K is way too slow (it is actually slower than the scalar implementation) * k_quants: WIP super-blocks with 64 weights Q3_K scalar and AVX2 works. * k_quants: WIP super-blocks with 64 weights Q5_K scalar and AVX2 works, and with that all k_quants are done on AVX2 and scalar * k_quants: WIP super-blocks with 64 weights Q6_K working on CUDA. Cannot make it run quite as gast as with super-blocks with 256 weigths: 8% slower on 4080, 20% slower on the 1660 (but there we fit 1 less layer on the GPU because pf the larger model size), so some fraction of these 20% is due to that, * k_quants: WIP super-blocks with 64 weights Q4_K working on CUDA. ~10% slower on GTX-1660, 16% slower on 4080. * k_quants: WIP super-blocks with 64 weights Q2_K working on CUDA. ~3% slower on GTX-1660, 10% slower on 4080. * k_quants: WIP super-blocks with 64 weights Q3_K working on CUDA. * k_quants: WIP super-blocks with 64 weights Q5_K working on CUDA, and with this CUDA is done. * k_quants: WIP super-blocks with 64 weights Q6_K working on ARM_NEON * k_quants: WIP super-blocks with 64 weights Q4_K working on ARM_NEON, but quite a bit slower than 256 weights * k_quants: WIP super-blocks with 64 weights Q2_K working on ARM_NEON, but quite a bit slower than 256 weights * k_quants: WIP super-blocks with 64 weights Q3_K working on ARM_NEON, but quite a bit slower than 256 weights. * k_quants: WIP super-blocks with 64 weights Q5_K working on ARM_NEON, but quite a bit slower than 256 weights. With that, we have full support for ARM_NEON, although performance is not quite there. * k_quants: WIP super-blocks with 64 weights Slightly more efficient Q3_K and Q5_K * k_quants: WIP super-blocks with 64 weights Another small improvement for Q3_K and Q5_K on ARM_NEON * k_quants: WIP super-blocks with 64 weights Yet another speedup for Q5_K on ARM_NEON. We are now within 10% of the QK_K = 256 version. * k_quants: WIP super-blocks with 64 weights * We are able to pass preprocessor macros to the Metal compiler * Q6_K works and is actually slightly more efficient than the QK_K = 256 version (25.2 ms vs 25.8 ms) * k_quants: WIP super-blocks with 64 weights Q4_K works on Metal and is actually slightly faster than QK_K = 256 (21.95 ms vs 24.0 ms). * k_quants: WIP super-blocks with 64 weights Q2_K works on Metal and is very slightly faster than QK_K = 256 (23.8 ms vs 24.2 ms). * k_quants: WIP super-blocks with 64 weights Q3_K works on Metal and is slightly faster than QK_K = 256 (26.6 ms vs 28.3 ms). * k_quants: WIP super-blocks with 64 weights Q5_K works on Metal and is slightly faster than QK_K = 256 (23.7 ms vs 26.3 ms). * k_quants: call them _K, not _k, also on Metal * k_quants: correctly define QK_K in llama.cpp * Fixed bug in q4_K quantization added with the 64-block addition * Simplify via lambda * k_quants: swicth Q3_K to 4-bit scales when QK_K = 64 Otherwise there isn't much benefit from this quantization type. There is some very slight loss in accuracy, but we reduce size by ~7%. E.g., for OpenLLaMA-3B, Q3_K_S perplexity is 8.6131 with 8-bit scales and 8.6352 with 4-bit, while file size decreases from 1.53G to 1.44G. * k_quants: switch Q4_K to 4-bit scales when QK_K = 64 Here the loss in accuracy is greater than for Q3_K, but the Q4_K points still move further to the left on the perplexity vs size curve. * k_quants: forgot to add the Metal changes in last commit * k_quants: change Q5_K to be type 0 when QK_K = 64 Still needs AVX2 implementation * k_quants: AVX2 implementation for new 64-weight Q5_K * k_quants: 10% faster ARM_NEON Q5_K dot product * k_quants: fixed issue caused by merging with master --------- Co-authored-by: Iwan Kawrakow <iwan.kawrakow@gmail.com>
2023-06-26 16:43:07 +00:00
ggml : sync (im2col, GPU conv, 32-bit arm compat) (#4060) ggml-ci
2023-11-13 14:55:52 +00:00
ggml_int8x16x4_t q5bytes;
ggml_uint8x16x4_t q5h;
k-quants : support for super-block size of 64 (#2001) * k_quants: WIP super-blocks with 64 weights * k_quants: WIP super-blocks with 64 weights Q6_K scalar and AVX2 works * k_quants: WIP super-blocks with 64 weights Q4_K scalar and AVX2 works * k_quants: WIP super-blocks with 64 weights Q2_K scalar and AVX2 works. Q2_K is way too slow (it is actually slower than the scalar implementation) * k_quants: WIP super-blocks with 64 weights Q3_K scalar and AVX2 works. * k_quants: WIP super-blocks with 64 weights Q5_K scalar and AVX2 works, and with that all k_quants are done on AVX2 and scalar * k_quants: WIP super-blocks with 64 weights Q6_K working on CUDA. Cannot make it run quite as gast as with super-blocks with 256 weigths: 8% slower on 4080, 20% slower on the 1660 (but there we fit 1 less layer on the GPU because pf the larger model size), so some fraction of these 20% is due to that, * k_quants: WIP super-blocks with 64 weights Q4_K working on CUDA. ~10% slower on GTX-1660, 16% slower on 4080. * k_quants: WIP super-blocks with 64 weights Q2_K working on CUDA. ~3% slower on GTX-1660, 10% slower on 4080. * k_quants: WIP super-blocks with 64 weights Q3_K working on CUDA. * k_quants: WIP super-blocks with 64 weights Q5_K working on CUDA, and with this CUDA is done. * k_quants: WIP super-blocks with 64 weights Q6_K working on ARM_NEON * k_quants: WIP super-blocks with 64 weights Q4_K working on ARM_NEON, but quite a bit slower than 256 weights * k_quants: WIP super-blocks with 64 weights Q2_K working on ARM_NEON, but quite a bit slower than 256 weights * k_quants: WIP super-blocks with 64 weights Q3_K working on ARM_NEON, but quite a bit slower than 256 weights. * k_quants: WIP super-blocks with 64 weights Q5_K working on ARM_NEON, but quite a bit slower than 256 weights. With that, we have full support for ARM_NEON, although performance is not quite there. * k_quants: WIP super-blocks with 64 weights Slightly more efficient Q3_K and Q5_K * k_quants: WIP super-blocks with 64 weights Another small improvement for Q3_K and Q5_K on ARM_NEON * k_quants: WIP super-blocks with 64 weights Yet another speedup for Q5_K on ARM_NEON. We are now within 10% of the QK_K = 256 version. * k_quants: WIP super-blocks with 64 weights * We are able to pass preprocessor macros to the Metal compiler * Q6_K works and is actually slightly more efficient than the QK_K = 256 version (25.2 ms vs 25.8 ms) * k_quants: WIP super-blocks with 64 weights Q4_K works on Metal and is actually slightly faster than QK_K = 256 (21.95 ms vs 24.0 ms). * k_quants: WIP super-blocks with 64 weights Q2_K works on Metal and is very slightly faster than QK_K = 256 (23.8 ms vs 24.2 ms). * k_quants: WIP super-blocks with 64 weights Q3_K works on Metal and is slightly faster than QK_K = 256 (26.6 ms vs 28.3 ms). * k_quants: WIP super-blocks with 64 weights Q5_K works on Metal and is slightly faster than QK_K = 256 (23.7 ms vs 26.3 ms). * k_quants: call them _K, not _k, also on Metal * k_quants: correctly define QK_K in llama.cpp * Fixed bug in q4_K quantization added with the 64-block addition * Simplify via lambda * k_quants: swicth Q3_K to 4-bit scales when QK_K = 64 Otherwise there isn't much benefit from this quantization type. There is some very slight loss in accuracy, but we reduce size by ~7%. E.g., for OpenLLaMA-3B, Q3_K_S perplexity is 8.6131 with 8-bit scales and 8.6352 with 4-bit, while file size decreases from 1.53G to 1.44G. * k_quants: switch Q4_K to 4-bit scales when QK_K = 64 Here the loss in accuracy is greater than for Q3_K, but the Q4_K points still move further to the left on the perplexity vs size curve. * k_quants: forgot to add the Metal changes in last commit * k_quants: change Q5_K to be type 0 when QK_K = 64 Still needs AVX2 implementation * k_quants: AVX2 implementation for new 64-weight Q5_K * k_quants: 10% faster ARM_NEON Q5_K dot product * k_quants: fixed issue caused by merging with master --------- Co-authored-by: Iwan Kawrakow <iwan.kawrakow@gmail.com>
2023-06-26 16:43:07 +00:00
float sumf = 0;
for (int i = 0; i < nb; ++i) {
const float d = y[i].d * (float)x[i].d;
const int8_t * sc = x[i].scales;
const uint8_t * restrict q5 = x[i].qs;
const uint8_t * restrict qh = x[i].qh;
const int8_t * restrict q8 = y[i].qs;
const uint8x8_t qhbits = vld1_u8(qh);
ggml : sync (im2col, GPU conv, 32-bit arm compat) (#4060) ggml-ci
2023-11-13 14:55:52 +00:00
const ggml_uint8x16x2_t q5bits = ggml_vld1q_u8_x2(q5);
const ggml_int8x16x4_t q8bytes = ggml_vld1q_s8_x4(q8);
k-quants : support for super-block size of 64 (#2001) * k_quants: WIP super-blocks with 64 weights * k_quants: WIP super-blocks with 64 weights Q6_K scalar and AVX2 works * k_quants: WIP super-blocks with 64 weights Q4_K scalar and AVX2 works * k_quants: WIP super-blocks with 64 weights Q2_K scalar and AVX2 works. Q2_K is way too slow (it is actually slower than the scalar implementation) * k_quants: WIP super-blocks with 64 weights Q3_K scalar and AVX2 works. * k_quants: WIP super-blocks with 64 weights Q5_K scalar and AVX2 works, and with that all k_quants are done on AVX2 and scalar * k_quants: WIP super-blocks with 64 weights Q6_K working on CUDA. Cannot make it run quite as gast as with super-blocks with 256 weigths: 8% slower on 4080, 20% slower on the 1660 (but there we fit 1 less layer on the GPU because pf the larger model size), so some fraction of these 20% is due to that, * k_quants: WIP super-blocks with 64 weights Q4_K working on CUDA. ~10% slower on GTX-1660, 16% slower on 4080. * k_quants: WIP super-blocks with 64 weights Q2_K working on CUDA. ~3% slower on GTX-1660, 10% slower on 4080. * k_quants: WIP super-blocks with 64 weights Q3_K working on CUDA. * k_quants: WIP super-blocks with 64 weights Q5_K working on CUDA, and with this CUDA is done. * k_quants: WIP super-blocks with 64 weights Q6_K working on ARM_NEON * k_quants: WIP super-blocks with 64 weights Q4_K working on ARM_NEON, but quite a bit slower than 256 weights * k_quants: WIP super-blocks with 64 weights Q2_K working on ARM_NEON, but quite a bit slower than 256 weights * k_quants: WIP super-blocks with 64 weights Q3_K working on ARM_NEON, but quite a bit slower than 256 weights. * k_quants: WIP super-blocks with 64 weights Q5_K working on ARM_NEON, but quite a bit slower than 256 weights. With that, we have full support for ARM_NEON, although performance is not quite there. * k_quants: WIP super-blocks with 64 weights Slightly more efficient Q3_K and Q5_K * k_quants: WIP super-blocks with 64 weights Another small improvement for Q3_K and Q5_K on ARM_NEON * k_quants: WIP super-blocks with 64 weights Yet another speedup for Q5_K on ARM_NEON. We are now within 10% of the QK_K = 256 version. * k_quants: WIP super-blocks with 64 weights * We are able to pass preprocessor macros to the Metal compiler * Q6_K works and is actually slightly more efficient than the QK_K = 256 version (25.2 ms vs 25.8 ms) * k_quants: WIP super-blocks with 64 weights Q4_K works on Metal and is actually slightly faster than QK_K = 256 (21.95 ms vs 24.0 ms). * k_quants: WIP super-blocks with 64 weights Q2_K works on Metal and is very slightly faster than QK_K = 256 (23.8 ms vs 24.2 ms). * k_quants: WIP super-blocks with 64 weights Q3_K works on Metal and is slightly faster than QK_K = 256 (26.6 ms vs 28.3 ms). * k_quants: WIP super-blocks with 64 weights Q5_K works on Metal and is slightly faster than QK_K = 256 (23.7 ms vs 26.3 ms). * k_quants: call them _K, not _k, also on Metal * k_quants: correctly define QK_K in llama.cpp * Fixed bug in q4_K quantization added with the 64-block addition * Simplify via lambda * k_quants: swicth Q3_K to 4-bit scales when QK_K = 64 Otherwise there isn't much benefit from this quantization type. There is some very slight loss in accuracy, but we reduce size by ~7%. E.g., for OpenLLaMA-3B, Q3_K_S perplexity is 8.6131 with 8-bit scales and 8.6352 with 4-bit, while file size decreases from 1.53G to 1.44G. * k_quants: switch Q4_K to 4-bit scales when QK_K = 64 Here the loss in accuracy is greater than for Q3_K, but the Q4_K points still move further to the left on the perplexity vs size curve. * k_quants: forgot to add the Metal changes in last commit * k_quants: change Q5_K to be type 0 when QK_K = 64 Still needs AVX2 implementation * k_quants: AVX2 implementation for new 64-weight Q5_K * k_quants: 10% faster ARM_NEON Q5_K dot product * k_quants: fixed issue caused by merging with master --------- Co-authored-by: Iwan Kawrakow <iwan.kawrakow@gmail.com>
2023-06-26 16:43:07 +00:00
const uint8x16_t htmp = vcombine_u8(qhbits, vshr_n_u8(qhbits, 1));
q5h.val[0] = vbicq_u8(mh, vshlq_n_u8(htmp, 4));
q5h.val[1] = vbicq_u8(mh, vshlq_n_u8(htmp, 2));
q5h.val[2] = vbicq_u8(mh, htmp);
q5h.val[3] = vbicq_u8(mh, vshrq_n_u8(htmp, 2));
q5bytes.val[0] = vsubq_s8(vreinterpretq_s8_u8(vandq_u8(q5bits.val[0], m4b)), vreinterpretq_s8_u8(q5h.val[0]));
q5bytes.val[1] = vsubq_s8(vreinterpretq_s8_u8(vandq_u8(q5bits.val[1], m4b)), vreinterpretq_s8_u8(q5h.val[1]));
q5bytes.val[2] = vsubq_s8(vreinterpretq_s8_u8(vshrq_n_u8(q5bits.val[0], 4)), vreinterpretq_s8_u8(q5h.val[2]));
q5bytes.val[3] = vsubq_s8(vreinterpretq_s8_u8(vshrq_n_u8(q5bits.val[1], 4)), vreinterpretq_s8_u8(q5h.val[3]));
ggml : add ggml_vdotq_s32 alias (#4715) ggml-ci
2023-12-31 09:43:31 +00:00
int32_t sumi1 = sc[0] * vaddvq_s32(ggml_vdotq_s32(mzero, q5bytes.val[0], q8bytes.val[0]));
int32_t sumi2 = sc[1] * vaddvq_s32(ggml_vdotq_s32(mzero, q5bytes.val[1], q8bytes.val[1]));
int32_t sumi3 = sc[2] * vaddvq_s32(ggml_vdotq_s32(mzero, q5bytes.val[2], q8bytes.val[2]));
int32_t sumi4 = sc[3] * vaddvq_s32(ggml_vdotq_s32(mzero, q5bytes.val[3], q8bytes.val[3]));
k-quants : support for super-block size of 64 (#2001) * k_quants: WIP super-blocks with 64 weights * k_quants: WIP super-blocks with 64 weights Q6_K scalar and AVX2 works * k_quants: WIP super-blocks with 64 weights Q4_K scalar and AVX2 works * k_quants: WIP super-blocks with 64 weights Q2_K scalar and AVX2 works. Q2_K is way too slow (it is actually slower than the scalar implementation) * k_quants: WIP super-blocks with 64 weights Q3_K scalar and AVX2 works. * k_quants: WIP super-blocks with 64 weights Q5_K scalar and AVX2 works, and with that all k_quants are done on AVX2 and scalar * k_quants: WIP super-blocks with 64 weights Q6_K working on CUDA. Cannot make it run quite as gast as with super-blocks with 256 weigths: 8% slower on 4080, 20% slower on the 1660 (but there we fit 1 less layer on the GPU because pf the larger model size), so some fraction of these 20% is due to that, * k_quants: WIP super-blocks with 64 weights Q4_K working on CUDA. ~10% slower on GTX-1660, 16% slower on 4080. * k_quants: WIP super-blocks with 64 weights Q2_K working on CUDA. ~3% slower on GTX-1660, 10% slower on 4080. * k_quants: WIP super-blocks with 64 weights Q3_K working on CUDA. * k_quants: WIP super-blocks with 64 weights Q5_K working on CUDA, and with this CUDA is done. * k_quants: WIP super-blocks with 64 weights Q6_K working on ARM_NEON * k_quants: WIP super-blocks with 64 weights Q4_K working on ARM_NEON, but quite a bit slower than 256 weights * k_quants: WIP super-blocks with 64 weights Q2_K working on ARM_NEON, but quite a bit slower than 256 weights * k_quants: WIP super-blocks with 64 weights Q3_K working on ARM_NEON, but quite a bit slower than 256 weights. * k_quants: WIP super-blocks with 64 weights Q5_K working on ARM_NEON, but quite a bit slower than 256 weights. With that, we have full support for ARM_NEON, although performance is not quite there. * k_quants: WIP super-blocks with 64 weights Slightly more efficient Q3_K and Q5_K * k_quants: WIP super-blocks with 64 weights Another small improvement for Q3_K and Q5_K on ARM_NEON * k_quants: WIP super-blocks with 64 weights Yet another speedup for Q5_K on ARM_NEON. We are now within 10% of the QK_K = 256 version. * k_quants: WIP super-blocks with 64 weights * We are able to pass preprocessor macros to the Metal compiler * Q6_K works and is actually slightly more efficient than the QK_K = 256 version (25.2 ms vs 25.8 ms) * k_quants: WIP super-blocks with 64 weights Q4_K works on Metal and is actually slightly faster than QK_K = 256 (21.95 ms vs 24.0 ms). * k_quants: WIP super-blocks with 64 weights Q2_K works on Metal and is very slightly faster than QK_K = 256 (23.8 ms vs 24.2 ms). * k_quants: WIP super-blocks with 64 weights Q3_K works on Metal and is slightly faster than QK_K = 256 (26.6 ms vs 28.3 ms). * k_quants: WIP super-blocks with 64 weights Q5_K works on Metal and is slightly faster than QK_K = 256 (23.7 ms vs 26.3 ms). * k_quants: call them _K, not _k, also on Metal * k_quants: correctly define QK_K in llama.cpp * Fixed bug in q4_K quantization added with the 64-block addition * Simplify via lambda * k_quants: swicth Q3_K to 4-bit scales when QK_K = 64 Otherwise there isn't much benefit from this quantization type. There is some very slight loss in accuracy, but we reduce size by ~7%. E.g., for OpenLLaMA-3B, Q3_K_S perplexity is 8.6131 with 8-bit scales and 8.6352 with 4-bit, while file size decreases from 1.53G to 1.44G. * k_quants: switch Q4_K to 4-bit scales when QK_K = 64 Here the loss in accuracy is greater than for Q3_K, but the Q4_K points still move further to the left on the perplexity vs size curve. * k_quants: forgot to add the Metal changes in last commit * k_quants: change Q5_K to be type 0 when QK_K = 64 Still needs AVX2 implementation * k_quants: AVX2 implementation for new 64-weight Q5_K * k_quants: 10% faster ARM_NEON Q5_K dot product * k_quants: fixed issue caused by merging with master --------- Co-authored-by: Iwan Kawrakow <iwan.kawrakow@gmail.com>
2023-06-26 16:43:07 +00:00
sumf += d * (sumi1 + sumi2 + sumi3 + sumi4);
}
*s = sumf;
#elif defined __AVX2__
const __m256i m4 = _mm256_set1_epi8(0xF);
const __m256i mone = _mm256_set1_epi8(1);
__m256 acc = _mm256_setzero_ps();
for (int i = 0; i < nb; ++i) {
const uint8_t * restrict q5 = x[i].qs;
const int8_t * restrict q8 = y[i].qs;
ggml : move FP16 <-> FP32 code to ggml-impl.h (#3861) * ggml : move FP16 <-> FP32 stuff to ggml-impl.h ggml-ci * tests : fix ARM build * ggml : explicitly initialize deprecated type traits * ggml : add math.h to ggml-impl.h * ggml : remove duplicate static assert macros * ggml : prefix lookup tables with ggml_ ggml-ci * ggml-impl : move extern "C" to start of file
2023-10-30 17:19:15 +00:00
const float d = y[i].d * GGML_FP16_TO_FP32(x[i].d);
k-quants : support for super-block size of 64 (#2001) * k_quants: WIP super-blocks with 64 weights * k_quants: WIP super-blocks with 64 weights Q6_K scalar and AVX2 works * k_quants: WIP super-blocks with 64 weights Q4_K scalar and AVX2 works * k_quants: WIP super-blocks with 64 weights Q2_K scalar and AVX2 works. Q2_K is way too slow (it is actually slower than the scalar implementation) * k_quants: WIP super-blocks with 64 weights Q3_K scalar and AVX2 works. * k_quants: WIP super-blocks with 64 weights Q5_K scalar and AVX2 works, and with that all k_quants are done on AVX2 and scalar * k_quants: WIP super-blocks with 64 weights Q6_K working on CUDA. Cannot make it run quite as gast as with super-blocks with 256 weigths: 8% slower on 4080, 20% slower on the 1660 (but there we fit 1 less layer on the GPU because pf the larger model size), so some fraction of these 20% is due to that, * k_quants: WIP super-blocks with 64 weights Q4_K working on CUDA. ~10% slower on GTX-1660, 16% slower on 4080. * k_quants: WIP super-blocks with 64 weights Q2_K working on CUDA. ~3% slower on GTX-1660, 10% slower on 4080. * k_quants: WIP super-blocks with 64 weights Q3_K working on CUDA. * k_quants: WIP super-blocks with 64 weights Q5_K working on CUDA, and with this CUDA is done. * k_quants: WIP super-blocks with 64 weights Q6_K working on ARM_NEON * k_quants: WIP super-blocks with 64 weights Q4_K working on ARM_NEON, but quite a bit slower than 256 weights * k_quants: WIP super-blocks with 64 weights Q2_K working on ARM_NEON, but quite a bit slower than 256 weights * k_quants: WIP super-blocks with 64 weights Q3_K working on ARM_NEON, but quite a bit slower than 256 weights. * k_quants: WIP super-blocks with 64 weights Q5_K working on ARM_NEON, but quite a bit slower than 256 weights. With that, we have full support for ARM_NEON, although performance is not quite there. * k_quants: WIP super-blocks with 64 weights Slightly more efficient Q3_K and Q5_K * k_quants: WIP super-blocks with 64 weights Another small improvement for Q3_K and Q5_K on ARM_NEON * k_quants: WIP super-blocks with 64 weights Yet another speedup for Q5_K on ARM_NEON. We are now within 10% of the QK_K = 256 version. * k_quants: WIP super-blocks with 64 weights * We are able to pass preprocessor macros to the Metal compiler * Q6_K works and is actually slightly more efficient than the QK_K = 256 version (25.2 ms vs 25.8 ms) * k_quants: WIP super-blocks with 64 weights Q4_K works on Metal and is actually slightly faster than QK_K = 256 (21.95 ms vs 24.0 ms). * k_quants: WIP super-blocks with 64 weights Q2_K works on Metal and is very slightly faster than QK_K = 256 (23.8 ms vs 24.2 ms). * k_quants: WIP super-blocks with 64 weights Q3_K works on Metal and is slightly faster than QK_K = 256 (26.6 ms vs 28.3 ms). * k_quants: WIP super-blocks with 64 weights Q5_K works on Metal and is slightly faster than QK_K = 256 (23.7 ms vs 26.3 ms). * k_quants: call them _K, not _k, also on Metal * k_quants: correctly define QK_K in llama.cpp * Fixed bug in q4_K quantization added with the 64-block addition * Simplify via lambda * k_quants: swicth Q3_K to 4-bit scales when QK_K = 64 Otherwise there isn't much benefit from this quantization type. There is some very slight loss in accuracy, but we reduce size by ~7%. E.g., for OpenLLaMA-3B, Q3_K_S perplexity is 8.6131 with 8-bit scales and 8.6352 with 4-bit, while file size decreases from 1.53G to 1.44G. * k_quants: switch Q4_K to 4-bit scales when QK_K = 64 Here the loss in accuracy is greater than for Q3_K, but the Q4_K points still move further to the left on the perplexity vs size curve. * k_quants: forgot to add the Metal changes in last commit * k_quants: change Q5_K to be type 0 when QK_K = 64 Still needs AVX2 implementation * k_quants: AVX2 implementation for new 64-weight Q5_K * k_quants: 10% faster ARM_NEON Q5_K dot product * k_quants: fixed issue caused by merging with master --------- Co-authored-by: Iwan Kawrakow <iwan.kawrakow@gmail.com>
2023-06-26 16:43:07 +00:00
const __m256i q5bits = _mm256_loadu_si256((const __m256i*)q5);
ggml : workaround for missing _mm256_setr_m128i in GCC < 8 in k_quants.c (#2405)
2023-07-28 18:17:45 +00:00
const __m256i scale_l = MM256_SET_M128I(_mm_set1_epi16(x[i].scales[1]), _mm_set1_epi16(x[i].scales[0]));
const __m256i scale_h = MM256_SET_M128I(_mm_set1_epi16(x[i].scales[3]), _mm_set1_epi16(x[i].scales[2]));
k-quants : support for super-block size of 64 (#2001) * k_quants: WIP super-blocks with 64 weights * k_quants: WIP super-blocks with 64 weights Q6_K scalar and AVX2 works * k_quants: WIP super-blocks with 64 weights Q4_K scalar and AVX2 works * k_quants: WIP super-blocks with 64 weights Q2_K scalar and AVX2 works. Q2_K is way too slow (it is actually slower than the scalar implementation) * k_quants: WIP super-blocks with 64 weights Q3_K scalar and AVX2 works. * k_quants: WIP super-blocks with 64 weights Q5_K scalar and AVX2 works, and with that all k_quants are done on AVX2 and scalar * k_quants: WIP super-blocks with 64 weights Q6_K working on CUDA. Cannot make it run quite as gast as with super-blocks with 256 weigths: 8% slower on 4080, 20% slower on the 1660 (but there we fit 1 less layer on the GPU because pf the larger model size), so some fraction of these 20% is due to that, * k_quants: WIP super-blocks with 64 weights Q4_K working on CUDA. ~10% slower on GTX-1660, 16% slower on 4080. * k_quants: WIP super-blocks with 64 weights Q2_K working on CUDA. ~3% slower on GTX-1660, 10% slower on 4080. * k_quants: WIP super-blocks with 64 weights Q3_K working on CUDA. * k_quants: WIP super-blocks with 64 weights Q5_K working on CUDA, and with this CUDA is done. * k_quants: WIP super-blocks with 64 weights Q6_K working on ARM_NEON * k_quants: WIP super-blocks with 64 weights Q4_K working on ARM_NEON, but quite a bit slower than 256 weights * k_quants: WIP super-blocks with 64 weights Q2_K working on ARM_NEON, but quite a bit slower than 256 weights * k_quants: WIP super-blocks with 64 weights Q3_K working on ARM_NEON, but quite a bit slower than 256 weights. * k_quants: WIP super-blocks with 64 weights Q5_K working on ARM_NEON, but quite a bit slower than 256 weights. With that, we have full support for ARM_NEON, although performance is not quite there. * k_quants: WIP super-blocks with 64 weights Slightly more efficient Q3_K and Q5_K * k_quants: WIP super-blocks with 64 weights Another small improvement for Q3_K and Q5_K on ARM_NEON * k_quants: WIP super-blocks with 64 weights Yet another speedup for Q5_K on ARM_NEON. We are now within 10% of the QK_K = 256 version. * k_quants: WIP super-blocks with 64 weights * We are able to pass preprocessor macros to the Metal compiler * Q6_K works and is actually slightly more efficient than the QK_K = 256 version (25.2 ms vs 25.8 ms) * k_quants: WIP super-blocks with 64 weights Q4_K works on Metal and is actually slightly faster than QK_K = 256 (21.95 ms vs 24.0 ms). * k_quants: WIP super-blocks with 64 weights Q2_K works on Metal and is very slightly faster than QK_K = 256 (23.8 ms vs 24.2 ms). * k_quants: WIP super-blocks with 64 weights Q3_K works on Metal and is slightly faster than QK_K = 256 (26.6 ms vs 28.3 ms). * k_quants: WIP super-blocks with 64 weights Q5_K works on Metal and is slightly faster than QK_K = 256 (23.7 ms vs 26.3 ms). * k_quants: call them _K, not _k, also on Metal * k_quants: correctly define QK_K in llama.cpp * Fixed bug in q4_K quantization added with the 64-block addition * Simplify via lambda * k_quants: swicth Q3_K to 4-bit scales when QK_K = 64 Otherwise there isn't much benefit from this quantization type. There is some very slight loss in accuracy, but we reduce size by ~7%. E.g., for OpenLLaMA-3B, Q3_K_S perplexity is 8.6131 with 8-bit scales and 8.6352 with 4-bit, while file size decreases from 1.53G to 1.44G. * k_quants: switch Q4_K to 4-bit scales when QK_K = 64 Here the loss in accuracy is greater than for Q3_K, but the Q4_K points still move further to the left on the perplexity vs size curve. * k_quants: forgot to add the Metal changes in last commit * k_quants: change Q5_K to be type 0 when QK_K = 64 Still needs AVX2 implementation * k_quants: AVX2 implementation for new 64-weight Q5_K * k_quants: 10% faster ARM_NEON Q5_K dot product * k_quants: fixed issue caused by merging with master --------- Co-authored-by: Iwan Kawrakow <iwan.kawrakow@gmail.com>
2023-06-26 16:43:07 +00:00
int64_t aux64;
memcpy(&aux64, x[i].qh, 8);
const __m128i haux128 = _mm_set_epi64x(aux64 >> 1, aux64);
ggml : workaround for missing _mm256_setr_m128i in GCC < 8 in k_quants.c (#2405)
2023-07-28 18:17:45 +00:00
const __m256i haux256 = MM256_SET_M128I(_mm_srli_epi16(haux128, 2), haux128);
k-quants : support for super-block size of 64 (#2001) * k_quants: WIP super-blocks with 64 weights * k_quants: WIP super-blocks with 64 weights Q6_K scalar and AVX2 works * k_quants: WIP super-blocks with 64 weights Q4_K scalar and AVX2 works * k_quants: WIP super-blocks with 64 weights Q2_K scalar and AVX2 works. Q2_K is way too slow (it is actually slower than the scalar implementation) * k_quants: WIP super-blocks with 64 weights Q3_K scalar and AVX2 works. * k_quants: WIP super-blocks with 64 weights Q5_K scalar and AVX2 works, and with that all k_quants are done on AVX2 and scalar * k_quants: WIP super-blocks with 64 weights Q6_K working on CUDA. Cannot make it run quite as gast as with super-blocks with 256 weigths: 8% slower on 4080, 20% slower on the 1660 (but there we fit 1 less layer on the GPU because pf the larger model size), so some fraction of these 20% is due to that, * k_quants: WIP super-blocks with 64 weights Q4_K working on CUDA. ~10% slower on GTX-1660, 16% slower on 4080. * k_quants: WIP super-blocks with 64 weights Q2_K working on CUDA. ~3% slower on GTX-1660, 10% slower on 4080. * k_quants: WIP super-blocks with 64 weights Q3_K working on CUDA. * k_quants: WIP super-blocks with 64 weights Q5_K working on CUDA, and with this CUDA is done. * k_quants: WIP super-blocks with 64 weights Q6_K working on ARM_NEON * k_quants: WIP super-blocks with 64 weights Q4_K working on ARM_NEON, but quite a bit slower than 256 weights * k_quants: WIP super-blocks with 64 weights Q2_K working on ARM_NEON, but quite a bit slower than 256 weights * k_quants: WIP super-blocks with 64 weights Q3_K working on ARM_NEON, but quite a bit slower than 256 weights. * k_quants: WIP super-blocks with 64 weights Q5_K working on ARM_NEON, but quite a bit slower than 256 weights. With that, we have full support for ARM_NEON, although performance is not quite there. * k_quants: WIP super-blocks with 64 weights Slightly more efficient Q3_K and Q5_K * k_quants: WIP super-blocks with 64 weights Another small improvement for Q3_K and Q5_K on ARM_NEON * k_quants: WIP super-blocks with 64 weights Yet another speedup for Q5_K on ARM_NEON. We are now within 10% of the QK_K = 256 version. * k_quants: WIP super-blocks with 64 weights * We are able to pass preprocessor macros to the Metal compiler * Q6_K works and is actually slightly more efficient than the QK_K = 256 version (25.2 ms vs 25.8 ms) * k_quants: WIP super-blocks with 64 weights Q4_K works on Metal and is actually slightly faster than QK_K = 256 (21.95 ms vs 24.0 ms). * k_quants: WIP super-blocks with 64 weights Q2_K works on Metal and is very slightly faster than QK_K = 256 (23.8 ms vs 24.2 ms). * k_quants: WIP super-blocks with 64 weights Q3_K works on Metal and is slightly faster than QK_K = 256 (26.6 ms vs 28.3 ms). * k_quants: WIP super-blocks with 64 weights Q5_K works on Metal and is slightly faster than QK_K = 256 (23.7 ms vs 26.3 ms). * k_quants: call them _K, not _k, also on Metal * k_quants: correctly define QK_K in llama.cpp * Fixed bug in q4_K quantization added with the 64-block addition * Simplify via lambda * k_quants: swicth Q3_K to 4-bit scales when QK_K = 64 Otherwise there isn't much benefit from this quantization type. There is some very slight loss in accuracy, but we reduce size by ~7%. E.g., for OpenLLaMA-3B, Q3_K_S perplexity is 8.6131 with 8-bit scales and 8.6352 with 4-bit, while file size decreases from 1.53G to 1.44G. * k_quants: switch Q4_K to 4-bit scales when QK_K = 64 Here the loss in accuracy is greater than for Q3_K, but the Q4_K points still move further to the left on the perplexity vs size curve. * k_quants: forgot to add the Metal changes in last commit * k_quants: change Q5_K to be type 0 when QK_K = 64 Still needs AVX2 implementation * k_quants: AVX2 implementation for new 64-weight Q5_K * k_quants: 10% faster ARM_NEON Q5_K dot product * k_quants: fixed issue caused by merging with master --------- Co-authored-by: Iwan Kawrakow <iwan.kawrakow@gmail.com>
2023-06-26 16:43:07 +00:00
const __m256i q5h_0 = _mm256_slli_epi16(_mm256_andnot_si256(haux256, mone), 4);
const __m256i q5h_1 = _mm256_slli_epi16(_mm256_andnot_si256(_mm256_srli_epi16(haux256, 4), mone), 4);
const __m256i q5l_0 = _mm256_and_si256(q5bits, m4);
const __m256i q5l_1 = _mm256_and_si256(_mm256_srli_epi16(q5bits, 4), m4);
const __m256i q8_0 = _mm256_loadu_si256((const __m256i*)(q8+ 0));
const __m256i q8_1 = _mm256_loadu_si256((const __m256i*)(q8+32));
const __m256i p16_0 = _mm256_madd_epi16(scale_l, _mm256_maddubs_epi16(q5l_0, q8_0));
const __m256i p16_1 = _mm256_madd_epi16(scale_h, _mm256_maddubs_epi16(q5l_1, q8_1));
const __m256i s16_0 = _mm256_madd_epi16(scale_l, _mm256_maddubs_epi16(q5h_0, q8_0));
const __m256i s16_1 = _mm256_madd_epi16(scale_h, _mm256_maddubs_epi16(q5h_1, q8_1));
const __m256i dot = _mm256_sub_epi32(_mm256_add_epi32(p16_0, p16_1), _mm256_add_epi32(s16_0, s16_1));
acc = _mm256_fmadd_ps(_mm256_set1_ps(d), _mm256_cvtepi32_ps(dot), acc);
}
*s = hsum_float_8(acc);
k_quants : add AVX support to dot functions with QK_K as 64 (#2339) * add AVX to ggml_vec_dot_q2_K_q8_K() * add AVX to ggml_vec_dot_q3_K_q8_K() * add AVX to ggml_vec_dot_q4_K_q8_K() * add AVX to ggml_vec_dot_q5_K_q8_K() * add AVX to ggml_vec_dot_q6_K_q8_K() * refactor AVX code in ggml_vec_dot_q6_K_q8_K()
2023-07-25 12:13:41 +00:00
#elif defined __AVX__
const __m128i m4 = _mm_set1_epi8(0xF);
const __m128i mone = _mm_set1_epi8(1);
__m256 acc = _mm256_setzero_ps();
for (int i = 0; i < nb; ++i) {
const uint8_t * restrict q5 = x[i].qs;
const int8_t * restrict q8 = y[i].qs;
ggml : move FP16 <-> FP32 code to ggml-impl.h (#3861) * ggml : move FP16 <-> FP32 stuff to ggml-impl.h ggml-ci * tests : fix ARM build * ggml : explicitly initialize deprecated type traits * ggml : add math.h to ggml-impl.h * ggml : remove duplicate static assert macros * ggml : prefix lookup tables with ggml_ ggml-ci * ggml-impl : move extern "C" to start of file
2023-10-30 17:19:15 +00:00
const float d = y[i].d * GGML_FP16_TO_FP32(x[i].d);
k_quants : add AVX support to dot functions with QK_K as 64 (#2339) * add AVX to ggml_vec_dot_q2_K_q8_K() * add AVX to ggml_vec_dot_q3_K_q8_K() * add AVX to ggml_vec_dot_q4_K_q8_K() * add AVX to ggml_vec_dot_q5_K_q8_K() * add AVX to ggml_vec_dot_q6_K_q8_K() * refactor AVX code in ggml_vec_dot_q6_K_q8_K()
2023-07-25 12:13:41 +00:00
const __m256i q5bits = _mm256_loadu_si256((const __m256i*)q5);
const __m128i scale_0 = _mm_set1_epi16(x[i].scales[0]);
const __m128i scale_1 = _mm_set1_epi16(x[i].scales[1]);
const __m128i scale_2 = _mm_set1_epi16(x[i].scales[2]);
const __m128i scale_3 = _mm_set1_epi16(x[i].scales[3]);
int64_t aux64;
memcpy(&aux64, x[i].qh, 8);
const __m128i haux128_0 = _mm_set_epi64x(aux64 >> 1, aux64);
const __m128i haux128_1 = _mm_srli_epi16(haux128_0, 2);
const __m128i q5h_0 = _mm_slli_epi16(_mm_andnot_si128(haux128_0, mone), 4);
const __m128i q5h_1 = _mm_slli_epi16(_mm_andnot_si128(haux128_1, mone), 4);
const __m128i q5h_2 = _mm_slli_epi16(_mm_andnot_si128(_mm_srli_epi16(haux128_0, 4), mone), 4);
const __m128i q5h_3 = _mm_slli_epi16(_mm_andnot_si128(_mm_srli_epi16(haux128_1, 4), mone), 4);
const __m128i q5l_0 = _mm_and_si128(_mm256_extractf128_si256(q5bits, 0), m4);
const __m128i q5l_1 = _mm_and_si128(_mm256_extractf128_si256(q5bits, 1), m4);
const __m128i q5l_2 = _mm_and_si128(_mm_srli_epi16(_mm256_extractf128_si256(q5bits, 0), 4), m4);
const __m128i q5l_3 = _mm_and_si128(_mm_srli_epi16(_mm256_extractf128_si256(q5bits, 1), 4), m4);
const __m256i q8_0 = _mm256_loadu_si256((const __m256i*)(q8+ 0));
const __m256i q8_1 = _mm256_loadu_si256((const __m256i*)(q8+32));
const __m128i p16_0 = _mm_madd_epi16(scale_0, _mm_maddubs_epi16(q5l_0, _mm256_extractf128_si256(q8_0, 0)));
const __m128i p16_1 = _mm_madd_epi16(scale_1, _mm_maddubs_epi16(q5l_1, _mm256_extractf128_si256(q8_0, 1)));
const __m128i p16_2 = _mm_madd_epi16(scale_2, _mm_maddubs_epi16(q5l_2, _mm256_extractf128_si256(q8_1, 0)));
const __m128i p16_3 = _mm_madd_epi16(scale_3, _mm_maddubs_epi16(q5l_3, _mm256_extractf128_si256(q8_1, 1)));
const __m128i s16_0 = _mm_madd_epi16(scale_0, _mm_maddubs_epi16(q5h_0, _mm256_extractf128_si256(q8_0, 0)));
const __m128i s16_1 = _mm_madd_epi16(scale_1, _mm_maddubs_epi16(q5h_1, _mm256_extractf128_si256(q8_0, 1)));
const __m128i s16_2 = _mm_madd_epi16(scale_2, _mm_maddubs_epi16(q5h_2, _mm256_extractf128_si256(q8_1, 0)));
const __m128i s16_3 = _mm_madd_epi16(scale_3, _mm_maddubs_epi16(q5h_3, _mm256_extractf128_si256(q8_1, 1)));
const __m128i dot_0 = _mm_sub_epi32(_mm_add_epi32(p16_0, p16_2), _mm_add_epi32(s16_0, s16_2));
const __m128i dot_1 = _mm_sub_epi32(_mm_add_epi32(p16_1, p16_3), _mm_add_epi32(s16_1, s16_3));
ggml : workaround for missing _mm256_setr_m128i in GCC < 8 in k_quants.c (#2405)
2023-07-28 18:17:45 +00:00
acc = _mm256_add_ps(_mm256_mul_ps(_mm256_set1_ps(d), _mm256_cvtepi32_ps(MM256_SET_M128I(dot_1, dot_0))), acc);
k_quants : add AVX support to dot functions with QK_K as 64 (#2339) * add AVX to ggml_vec_dot_q2_K_q8_K() * add AVX to ggml_vec_dot_q3_K_q8_K() * add AVX to ggml_vec_dot_q4_K_q8_K() * add AVX to ggml_vec_dot_q5_K_q8_K() * add AVX to ggml_vec_dot_q6_K_q8_K() * refactor AVX code in ggml_vec_dot_q6_K_q8_K()
2023-07-25 12:13:41 +00:00
}
*s = hsum_float_8(acc);
ggml : add RISC-V Vector Support for K-Quants and improved the existing intrinsics (#3453) * Added RVV intrinsics support for Q8 quantize row and also improved the existing dot product function for risc-v. The RVV intrinsics is added for the following quantize row functions quantize_row_q8_0 quantize_row_q8_1 The following dot product functions have also been optimized by using LMUL = 1/2 instead of LMUL = 1 ggml_vec_dot_q4_0_q8_0 ggml_vec_dot_q4_1_q8_1 ggml_vec_dot_q5_0_q8_0 ggml_vec_dot_q5_1_q8_1 And vector initialization in Q5 by temporary array is also replaced by the vid intrinsics Signed-off-by: Ahmad Tameem <ahmad.tameem@10xengineers.ai> * Added RVV intrinsics support for k_quants This adds RISC-V Vector intrinsics support for the following K_quants functions for both QKK = 256 and QKK = 64 ggml_vec_dot_q2_K_q8_K ggml_vec_dot_q3_K_q8_K ggml_vec_dot_q4_K_q8_K ggml_vec_dot_q5_K_q8_K ggml_vec_dot_q6_K_q8_K Signed-off-by: Ahmad Tameem <ahmad.tameem@10xengineers.ai> --------- Signed-off-by: Ahmad Tameem <ahmad.tameem@10xengineers.ai>
2023-10-03 18:38:19 +00:00
#elif defined __riscv_v_intrinsic
float sumf = 0;
for (int i = 0; i < nb; ++i) {
const float d = y[i].d * (float)x[i].d;
const int8_t * sc = x[i].scales;
const uint8_t * restrict q5 = x[i].qs;
const uint8_t * restrict qh = x[i].qh;
const int8_t * restrict q8 = y[i].qs;
vint32m1_t vzero = __riscv_vmv_v_x_i32m1(0, 1);
// load qh
vuint8mf4_t qh_x1 = __riscv_vle8_v_u8mf4(qh, 8);
vuint8mf2_t qh_x2 = __riscv_vlmul_ext_v_u8mf4_u8mf2(__riscv_vsrl_vx_u8mf4(qh_x1, 1, 8));
size_t vl = 16;
// combine both qh_1 and qh_2
vuint8mf2_t qh_x = __riscv_vslideup_vx_u8mf2(__riscv_vlmul_ext_v_u8mf4_u8mf2(qh_x1), qh_x2, vl/2, vl);
vuint8mf2_t qh_h0 = __riscv_vand_vx_u8mf2(__riscv_vnot_v_u8mf2(__riscv_vsll_vx_u8mf2(qh_x, 0x4, vl), vl), 16, vl);
vuint8mf2_t qh_h1 = __riscv_vand_vx_u8mf2(__riscv_vnot_v_u8mf2(__riscv_vsll_vx_u8mf2(qh_x, 0x2, vl), vl), 16, vl);
vuint8mf2_t qh_h2 = __riscv_vand_vx_u8mf2(__riscv_vnot_v_u8mf2(qh_x, vl), 16, vl);
vuint8mf2_t qh_h3 = __riscv_vand_vx_u8mf2(__riscv_vnot_v_u8mf2(__riscv_vsrl_vx_u8mf2(qh_x, 0x4, vl), vl), 16, vl);
vint8mf2_t qh_0 = __riscv_vreinterpret_v_u8mf2_i8mf2(qh_h0);
vint8mf2_t qh_1 = __riscv_vreinterpret_v_u8mf2_i8mf2(qh_h1);
vint8mf2_t qh_2 = __riscv_vreinterpret_v_u8mf2_i8mf2(qh_h2);
vint8mf2_t qh_3 = __riscv_vreinterpret_v_u8mf2_i8mf2(qh_h3);
// load q5
vuint8mf2_t q5_x1 = __riscv_vle8_v_u8mf2(q5, vl);
vuint8mf2_t q5_x2 = __riscv_vle8_v_u8mf2(q5+16, vl);
vint8mf2_t q5s_0 = __riscv_vreinterpret_v_u8mf2_i8mf2(__riscv_vand_vx_u8mf2(q5_x1, 0xF, vl));
vint8mf2_t q5s_1 = __riscv_vreinterpret_v_u8mf2_i8mf2(__riscv_vand_vx_u8mf2(q5_x2, 0xF, vl));
vint8mf2_t q5s_2 = __riscv_vreinterpret_v_u8mf2_i8mf2(__riscv_vsrl_vx_u8mf2(q5_x1, 0x4, vl));
vint8mf2_t q5s_3 = __riscv_vreinterpret_v_u8mf2_i8mf2(__riscv_vsrl_vx_u8mf2(q5_x2, 0x4, vl));
vint8mf2_t q5_0 = __riscv_vsub_vv_i8mf2(q5s_0, qh_0, vl);
vint8mf2_t q5_1 = __riscv_vsub_vv_i8mf2(q5s_1, qh_1, vl);
vint8mf2_t q5_2 = __riscv_vsub_vv_i8mf2(q5s_2, qh_2, vl);
vint8mf2_t q5_3 = __riscv_vsub_vv_i8mf2(q5s_3, qh_3, vl);
// load Q8 and multiply it with Q5
vint16m1_t p0 = __riscv_vwmul_vv_i16m1(q5_0, __riscv_vle8_v_i8mf2(q8, vl), vl);
vint16m1_t p1 = __riscv_vwmul_vv_i16m1(q5_1, __riscv_vle8_v_i8mf2(q8+16, vl), vl);
vint16m1_t p2 = __riscv_vwmul_vv_i16m1(q5_2, __riscv_vle8_v_i8mf2(q8+32, vl), vl);
vint16m1_t p3 = __riscv_vwmul_vv_i16m1(q5_3, __riscv_vle8_v_i8mf2(q8+48, vl), vl);
vint32m1_t vs_0 = __riscv_vwredsum_vs_i16m1_i32m1(p0, vzero, vl);
vint32m1_t vs_1 = __riscv_vwredsum_vs_i16m1_i32m1(p1, vzero, vl);
vint32m1_t vs_2 = __riscv_vwredsum_vs_i16m1_i32m1(p2, vzero, vl);
vint32m1_t vs_3 = __riscv_vwredsum_vs_i16m1_i32m1(p3, vzero, vl);
int32_t sumi1 = sc[0] * __riscv_vmv_x_s_i32m1_i32(vs_0);
int32_t sumi2 = sc[1] * __riscv_vmv_x_s_i32m1_i32(vs_1);
int32_t sumi3 = sc[2] * __riscv_vmv_x_s_i32m1_i32(vs_2);
int32_t sumi4 = sc[3] * __riscv_vmv_x_s_i32m1_i32(vs_3);
sumf += d * (sumi1 + sumi2 + sumi3 + sumi4);
}
*s = sumf;
k-quants : support for super-block size of 64 (#2001) * k_quants: WIP super-blocks with 64 weights * k_quants: WIP super-blocks with 64 weights Q6_K scalar and AVX2 works * k_quants: WIP super-blocks with 64 weights Q4_K scalar and AVX2 works * k_quants: WIP super-blocks with 64 weights Q2_K scalar and AVX2 works. Q2_K is way too slow (it is actually slower than the scalar implementation) * k_quants: WIP super-blocks with 64 weights Q3_K scalar and AVX2 works. * k_quants: WIP super-blocks with 64 weights Q5_K scalar and AVX2 works, and with that all k_quants are done on AVX2 and scalar * k_quants: WIP super-blocks with 64 weights Q6_K working on CUDA. Cannot make it run quite as gast as with super-blocks with 256 weigths: 8% slower on 4080, 20% slower on the 1660 (but there we fit 1 less layer on the GPU because pf the larger model size), so some fraction of these 20% is due to that, * k_quants: WIP super-blocks with 64 weights Q4_K working on CUDA. ~10% slower on GTX-1660, 16% slower on 4080. * k_quants: WIP super-blocks with 64 weights Q2_K working on CUDA. ~3% slower on GTX-1660, 10% slower on 4080. * k_quants: WIP super-blocks with 64 weights Q3_K working on CUDA. * k_quants: WIP super-blocks with 64 weights Q5_K working on CUDA, and with this CUDA is done. * k_quants: WIP super-blocks with 64 weights Q6_K working on ARM_NEON * k_quants: WIP super-blocks with 64 weights Q4_K working on ARM_NEON, but quite a bit slower than 256 weights * k_quants: WIP super-blocks with 64 weights Q2_K working on ARM_NEON, but quite a bit slower than 256 weights * k_quants: WIP super-blocks with 64 weights Q3_K working on ARM_NEON, but quite a bit slower than 256 weights. * k_quants: WIP super-blocks with 64 weights Q5_K working on ARM_NEON, but quite a bit slower than 256 weights. With that, we have full support for ARM_NEON, although performance is not quite there. * k_quants: WIP super-blocks with 64 weights Slightly more efficient Q3_K and Q5_K * k_quants: WIP super-blocks with 64 weights Another small improvement for Q3_K and Q5_K on ARM_NEON * k_quants: WIP super-blocks with 64 weights Yet another speedup for Q5_K on ARM_NEON. We are now within 10% of the QK_K = 256 version. * k_quants: WIP super-blocks with 64 weights * We are able to pass preprocessor macros to the Metal compiler * Q6_K works and is actually slightly more efficient than the QK_K = 256 version (25.2 ms vs 25.8 ms) * k_quants: WIP super-blocks with 64 weights Q4_K works on Metal and is actually slightly faster than QK_K = 256 (21.95 ms vs 24.0 ms). * k_quants: WIP super-blocks with 64 weights Q2_K works on Metal and is very slightly faster than QK_K = 256 (23.8 ms vs 24.2 ms). * k_quants: WIP super-blocks with 64 weights Q3_K works on Metal and is slightly faster than QK_K = 256 (26.6 ms vs 28.3 ms). * k_quants: WIP super-blocks with 64 weights Q5_K works on Metal and is slightly faster than QK_K = 256 (23.7 ms vs 26.3 ms). * k_quants: call them _K, not _k, also on Metal * k_quants: correctly define QK_K in llama.cpp * Fixed bug in q4_K quantization added with the 64-block addition * Simplify via lambda * k_quants: swicth Q3_K to 4-bit scales when QK_K = 64 Otherwise there isn't much benefit from this quantization type. There is some very slight loss in accuracy, but we reduce size by ~7%. E.g., for OpenLLaMA-3B, Q3_K_S perplexity is 8.6131 with 8-bit scales and 8.6352 with 4-bit, while file size decreases from 1.53G to 1.44G. * k_quants: switch Q4_K to 4-bit scales when QK_K = 64 Here the loss in accuracy is greater than for Q3_K, but the Q4_K points still move further to the left on the perplexity vs size curve. * k_quants: forgot to add the Metal changes in last commit * k_quants: change Q5_K to be type 0 when QK_K = 64 Still needs AVX2 implementation * k_quants: AVX2 implementation for new 64-weight Q5_K * k_quants: 10% faster ARM_NEON Q5_K dot product * k_quants: fixed issue caused by merging with master --------- Co-authored-by: Iwan Kawrakow <iwan.kawrakow@gmail.com>
2023-06-26 16:43:07 +00:00
#else
Fix scalar version of Q5_K when QK_K = 64 (#2362) Co-authored-by: Iwan Kawrakow <iwan.kawrakow@gmail.com>
2023-07-24 09:55:02 +00:00
int8_t aux8[QK_K];
k-quants : support for super-block size of 64 (#2001) * k_quants: WIP super-blocks with 64 weights * k_quants: WIP super-blocks with 64 weights Q6_K scalar and AVX2 works * k_quants: WIP super-blocks with 64 weights Q4_K scalar and AVX2 works * k_quants: WIP super-blocks with 64 weights Q2_K scalar and AVX2 works. Q2_K is way too slow (it is actually slower than the scalar implementation) * k_quants: WIP super-blocks with 64 weights Q3_K scalar and AVX2 works. * k_quants: WIP super-blocks with 64 weights Q5_K scalar and AVX2 works, and with that all k_quants are done on AVX2 and scalar * k_quants: WIP super-blocks with 64 weights Q6_K working on CUDA. Cannot make it run quite as gast as with super-blocks with 256 weigths: 8% slower on 4080, 20% slower on the 1660 (but there we fit 1 less layer on the GPU because pf the larger model size), so some fraction of these 20% is due to that, * k_quants: WIP super-blocks with 64 weights Q4_K working on CUDA. ~10% slower on GTX-1660, 16% slower on 4080. * k_quants: WIP super-blocks with 64 weights Q2_K working on CUDA. ~3% slower on GTX-1660, 10% slower on 4080. * k_quants: WIP super-blocks with 64 weights Q3_K working on CUDA. * k_quants: WIP super-blocks with 64 weights Q5_K working on CUDA, and with this CUDA is done. * k_quants: WIP super-blocks with 64 weights Q6_K working on ARM_NEON * k_quants: WIP super-blocks with 64 weights Q4_K working on ARM_NEON, but quite a bit slower than 256 weights * k_quants: WIP super-blocks with 64 weights Q2_K working on ARM_NEON, but quite a bit slower than 256 weights * k_quants: WIP super-blocks with 64 weights Q3_K working on ARM_NEON, but quite a bit slower than 256 weights. * k_quants: WIP super-blocks with 64 weights Q5_K working on ARM_NEON, but quite a bit slower than 256 weights. With that, we have full support for ARM_NEON, although performance is not quite there. * k_quants: WIP super-blocks with 64 weights Slightly more efficient Q3_K and Q5_K * k_quants: WIP super-blocks with 64 weights Another small improvement for Q3_K and Q5_K on ARM_NEON * k_quants: WIP super-blocks with 64 weights Yet another speedup for Q5_K on ARM_NEON. We are now within 10% of the QK_K = 256 version. * k_quants: WIP super-blocks with 64 weights * We are able to pass preprocessor macros to the Metal compiler * Q6_K works and is actually slightly more efficient than the QK_K = 256 version (25.2 ms vs 25.8 ms) * k_quants: WIP super-blocks with 64 weights Q4_K works on Metal and is actually slightly faster than QK_K = 256 (21.95 ms vs 24.0 ms). * k_quants: WIP super-blocks with 64 weights Q2_K works on Metal and is very slightly faster than QK_K = 256 (23.8 ms vs 24.2 ms). * k_quants: WIP super-blocks with 64 weights Q3_K works on Metal and is slightly faster than QK_K = 256 (26.6 ms vs 28.3 ms). * k_quants: WIP super-blocks with 64 weights Q5_K works on Metal and is slightly faster than QK_K = 256 (23.7 ms vs 26.3 ms). * k_quants: call them _K, not _k, also on Metal * k_quants: correctly define QK_K in llama.cpp * Fixed bug in q4_K quantization added with the 64-block addition * Simplify via lambda * k_quants: swicth Q3_K to 4-bit scales when QK_K = 64 Otherwise there isn't much benefit from this quantization type. There is some very slight loss in accuracy, but we reduce size by ~7%. E.g., for OpenLLaMA-3B, Q3_K_S perplexity is 8.6131 with 8-bit scales and 8.6352 with 4-bit, while file size decreases from 1.53G to 1.44G. * k_quants: switch Q4_K to 4-bit scales when QK_K = 64 Here the loss in accuracy is greater than for Q3_K, but the Q4_K points still move further to the left on the perplexity vs size curve. * k_quants: forgot to add the Metal changes in last commit * k_quants: change Q5_K to be type 0 when QK_K = 64 Still needs AVX2 implementation * k_quants: AVX2 implementation for new 64-weight Q5_K * k_quants: 10% faster ARM_NEON Q5_K dot product * k_quants: fixed issue caused by merging with master --------- Co-authored-by: Iwan Kawrakow <iwan.kawrakow@gmail.com>
2023-06-26 16:43:07 +00:00
int16_t aux16[16];
float sums [8];
memset(sums, 0, 8*sizeof(float));
float sumf = 0;
for (int i = 0; i < nb; ++i) {
const uint8_t * restrict q4 = x[i].qs;
const uint8_t * restrict hm = x[i].qh;
const int8_t * restrict q8 = y[i].qs;
Fix scalar version of Q5_K when QK_K = 64 (#2362) Co-authored-by: Iwan Kawrakow <iwan.kawrakow@gmail.com>
2023-07-24 09:55:02 +00:00
int8_t * restrict a = aux8;
k-quants : support for super-block size of 64 (#2001) * k_quants: WIP super-blocks with 64 weights * k_quants: WIP super-blocks with 64 weights Q6_K scalar and AVX2 works * k_quants: WIP super-blocks with 64 weights Q4_K scalar and AVX2 works * k_quants: WIP super-blocks with 64 weights Q2_K scalar and AVX2 works. Q2_K is way too slow (it is actually slower than the scalar implementation) * k_quants: WIP super-blocks with 64 weights Q3_K scalar and AVX2 works. * k_quants: WIP super-blocks with 64 weights Q5_K scalar and AVX2 works, and with that all k_quants are done on AVX2 and scalar * k_quants: WIP super-blocks with 64 weights Q6_K working on CUDA. Cannot make it run quite as gast as with super-blocks with 256 weigths: 8% slower on 4080, 20% slower on the 1660 (but there we fit 1 less layer on the GPU because pf the larger model size), so some fraction of these 20% is due to that, * k_quants: WIP super-blocks with 64 weights Q4_K working on CUDA. ~10% slower on GTX-1660, 16% slower on 4080. * k_quants: WIP super-blocks with 64 weights Q2_K working on CUDA. ~3% slower on GTX-1660, 10% slower on 4080. * k_quants: WIP super-blocks with 64 weights Q3_K working on CUDA. * k_quants: WIP super-blocks with 64 weights Q5_K working on CUDA, and with this CUDA is done. * k_quants: WIP super-blocks with 64 weights Q6_K working on ARM_NEON * k_quants: WIP super-blocks with 64 weights Q4_K working on ARM_NEON, but quite a bit slower than 256 weights * k_quants: WIP super-blocks with 64 weights Q2_K working on ARM_NEON, but quite a bit slower than 256 weights * k_quants: WIP super-blocks with 64 weights Q3_K working on ARM_NEON, but quite a bit slower than 256 weights. * k_quants: WIP super-blocks with 64 weights Q5_K working on ARM_NEON, but quite a bit slower than 256 weights. With that, we have full support for ARM_NEON, although performance is not quite there. * k_quants: WIP super-blocks with 64 weights Slightly more efficient Q3_K and Q5_K * k_quants: WIP super-blocks with 64 weights Another small improvement for Q3_K and Q5_K on ARM_NEON * k_quants: WIP super-blocks with 64 weights Yet another speedup for Q5_K on ARM_NEON. We are now within 10% of the QK_K = 256 version. * k_quants: WIP super-blocks with 64 weights * We are able to pass preprocessor macros to the Metal compiler * Q6_K works and is actually slightly more efficient than the QK_K = 256 version (25.2 ms vs 25.8 ms) * k_quants: WIP super-blocks with 64 weights Q4_K works on Metal and is actually slightly faster than QK_K = 256 (21.95 ms vs 24.0 ms). * k_quants: WIP super-blocks with 64 weights Q2_K works on Metal and is very slightly faster than QK_K = 256 (23.8 ms vs 24.2 ms). * k_quants: WIP super-blocks with 64 weights Q3_K works on Metal and is slightly faster than QK_K = 256 (26.6 ms vs 28.3 ms). * k_quants: WIP super-blocks with 64 weights Q5_K works on Metal and is slightly faster than QK_K = 256 (23.7 ms vs 26.3 ms). * k_quants: call them _K, not _k, also on Metal * k_quants: correctly define QK_K in llama.cpp * Fixed bug in q4_K quantization added with the 64-block addition * Simplify via lambda * k_quants: swicth Q3_K to 4-bit scales when QK_K = 64 Otherwise there isn't much benefit from this quantization type. There is some very slight loss in accuracy, but we reduce size by ~7%. E.g., for OpenLLaMA-3B, Q3_K_S perplexity is 8.6131 with 8-bit scales and 8.6352 with 4-bit, while file size decreases from 1.53G to 1.44G. * k_quants: switch Q4_K to 4-bit scales when QK_K = 64 Here the loss in accuracy is greater than for Q3_K, but the Q4_K points still move further to the left on the perplexity vs size curve. * k_quants: forgot to add the Metal changes in last commit * k_quants: change Q5_K to be type 0 when QK_K = 64 Still needs AVX2 implementation * k_quants: AVX2 implementation for new 64-weight Q5_K * k_quants: 10% faster ARM_NEON Q5_K dot product * k_quants: fixed issue caused by merging with master --------- Co-authored-by: Iwan Kawrakow <iwan.kawrakow@gmail.com>
2023-06-26 16:43:07 +00:00
for (int l = 0; l < 32; ++l) {
a[l+ 0] = q4[l] & 0xF;
a[l+32] = q4[l] >> 4;
}
for (int is = 0; is < 8; ++is) {
uint8_t m = 1 << is;
for (int l = 0; l < 8; ++l) a[8*is + l] -= (hm[l] & m ? 0 : 16);
}
ggml : move FP16 <-> FP32 code to ggml-impl.h (#3861) * ggml : move FP16 <-> FP32 stuff to ggml-impl.h ggml-ci * tests : fix ARM build * ggml : explicitly initialize deprecated type traits * ggml : add math.h to ggml-impl.h * ggml : remove duplicate static assert macros * ggml : prefix lookup tables with ggml_ ggml-ci * ggml-impl : move extern "C" to start of file
2023-10-30 17:19:15 +00:00
const float d = y[i].d * GGML_FP16_TO_FP32(x[i].d);
k-quants : support for super-block size of 64 (#2001) * k_quants: WIP super-blocks with 64 weights * k_quants: WIP super-blocks with 64 weights Q6_K scalar and AVX2 works * k_quants: WIP super-blocks with 64 weights Q4_K scalar and AVX2 works * k_quants: WIP super-blocks with 64 weights Q2_K scalar and AVX2 works. Q2_K is way too slow (it is actually slower than the scalar implementation) * k_quants: WIP super-blocks with 64 weights Q3_K scalar and AVX2 works. * k_quants: WIP super-blocks with 64 weights Q5_K scalar and AVX2 works, and with that all k_quants are done on AVX2 and scalar * k_quants: WIP super-blocks with 64 weights Q6_K working on CUDA. Cannot make it run quite as gast as with super-blocks with 256 weigths: 8% slower on 4080, 20% slower on the 1660 (but there we fit 1 less layer on the GPU because pf the larger model size), so some fraction of these 20% is due to that, * k_quants: WIP super-blocks with 64 weights Q4_K working on CUDA. ~10% slower on GTX-1660, 16% slower on 4080. * k_quants: WIP super-blocks with 64 weights Q2_K working on CUDA. ~3% slower on GTX-1660, 10% slower on 4080. * k_quants: WIP super-blocks with 64 weights Q3_K working on CUDA. * k_quants: WIP super-blocks with 64 weights Q5_K working on CUDA, and with this CUDA is done. * k_quants: WIP super-blocks with 64 weights Q6_K working on ARM_NEON * k_quants: WIP super-blocks with 64 weights Q4_K working on ARM_NEON, but quite a bit slower than 256 weights * k_quants: WIP super-blocks with 64 weights Q2_K working on ARM_NEON, but quite a bit slower than 256 weights * k_quants: WIP super-blocks with 64 weights Q3_K working on ARM_NEON, but quite a bit slower than 256 weights. * k_quants: WIP super-blocks with 64 weights Q5_K working on ARM_NEON, but quite a bit slower than 256 weights. With that, we have full support for ARM_NEON, although performance is not quite there. * k_quants: WIP super-blocks with 64 weights Slightly more efficient Q3_K and Q5_K * k_quants: WIP super-blocks with 64 weights Another small improvement for Q3_K and Q5_K on ARM_NEON * k_quants: WIP super-blocks with 64 weights Yet another speedup for Q5_K on ARM_NEON. We are now within 10% of the QK_K = 256 version. * k_quants: WIP super-blocks with 64 weights * We are able to pass preprocessor macros to the Metal compiler * Q6_K works and is actually slightly more efficient than the QK_K = 256 version (25.2 ms vs 25.8 ms) * k_quants: WIP super-blocks with 64 weights Q4_K works on Metal and is actually slightly faster than QK_K = 256 (21.95 ms vs 24.0 ms). * k_quants: WIP super-blocks with 64 weights Q2_K works on Metal and is very slightly faster than QK_K = 256 (23.8 ms vs 24.2 ms). * k_quants: WIP super-blocks with 64 weights Q3_K works on Metal and is slightly faster than QK_K = 256 (26.6 ms vs 28.3 ms). * k_quants: WIP super-blocks with 64 weights Q5_K works on Metal and is slightly faster than QK_K = 256 (23.7 ms vs 26.3 ms). * k_quants: call them _K, not _k, also on Metal * k_quants: correctly define QK_K in llama.cpp * Fixed bug in q4_K quantization added with the 64-block addition * Simplify via lambda * k_quants: swicth Q3_K to 4-bit scales when QK_K = 64 Otherwise there isn't much benefit from this quantization type. There is some very slight loss in accuracy, but we reduce size by ~7%. E.g., for OpenLLaMA-3B, Q3_K_S perplexity is 8.6131 with 8-bit scales and 8.6352 with 4-bit, while file size decreases from 1.53G to 1.44G. * k_quants: switch Q4_K to 4-bit scales when QK_K = 64 Here the loss in accuracy is greater than for Q3_K, but the Q4_K points still move further to the left on the perplexity vs size curve. * k_quants: forgot to add the Metal changes in last commit * k_quants: change Q5_K to be type 0 when QK_K = 64 Still needs AVX2 implementation * k_quants: AVX2 implementation for new 64-weight Q5_K * k_quants: 10% faster ARM_NEON Q5_K dot product * k_quants: fixed issue caused by merging with master --------- Co-authored-by: Iwan Kawrakow <iwan.kawrakow@gmail.com>
2023-06-26 16:43:07 +00:00
const int8_t * restrict sc = x[i].scales;
for (int j = 0; j < QK_K/16; ++j) {
const float dl = d * sc[j];
for (int l = 0; l < 16; ++l) aux16[l] = q8[l] * a[l];
for (int l = 0; l < 8; ++l) sums[l] += dl * (aux16[l] + aux16[8+l]);
q8 += 16; a += 16;
}
}
for (int l = 0; l < 8; ++l) sumf += sums[l];
*s = sumf;
#endif
}
#endif
ggml : add SOTA 2,3,4,5,6 bit k-quantizations (#1684) * Starting to add k-quantization to ggml I think it is better to have quantization separate from ggml. For now just adding the k-quants there, but it would be better to also factor out the existing ggml quantizations. * Adding Q3_K and Q8_K (de)-quantization * Q3_K now working on CUDA and AVX2/scalar CUDA is not ideal - ~50% slower than Q4_0 for single token prediction, about the same in batch mode (perplexity). CPU single token is ~55 ms (on Ryzen 7950X). * Some improvement for Q3_K on CUDA It is now ~22.5 ms/token on my GPU, so ~30% slower than Q4_0. * Some more CUDA optimizations for Q3_K Single token is now 20.5 ms/token (~20% slower than Q4_0). Perplexity is on par with Q4_0. * Adding Q4_K - scalar, AVX2, CUDA Performance is the same or perhaps very slightly better than Q4_0 on the CPU. On the GPU, single token prediction is ~10% better than Q4_0, batch mode (perplexity is about the same). * Adding Q6_K - scalar, AVX2, CUDA Performance is ~40% lower compared to Q4_K on the CPU. This is to be expected, considering that we are memory bound on the CPU and the 6-bit model is ~44% larger than the 4-bit. On the GPU, single token prediction is ~6% lower than Q4_0, batch mode (perplexity) is even closer (but still slower). * Adding Q5_K - scalar, AVX2, CUDA Performance is ~20% lower compared to Q4_K on the CPU. This is to be expected, considering that we are memory bound on the CPU and the 5-bit model is ~22% larger than the 4-bit. On the GPU, single token prediction is about the same as Q4_0 for both, single token and batch prediction. * Per convention, all QX_K quantizations use Q5_K for output.weight * Adding quantization mixes * Quantization mixes: didn't quite get what I wanted in the last commit * Q4_K dot product for ARM_NEON * Q6_K dot product for ARM_NEON * Q5_K dot product for ARM_NEON * Adding Q3_K dot for ARM_NEON It is 22% slower than Q4_K, despite the smaller model size. On x86_64, where we are memory bound, the Q3_K model is quite a bit faster than Q4_K. * A very slightly faster ARM_NEON Q3_K dot * Adding Q2_K - just CUDA for now Token prediction is pretty good - about 15.5 ms on a RTX 4080. Perplexity is about the same as Q4_K. * Adding scalar and AVX2 Q2_K dot * Adding ARM_NEON Q2_K dot About the same performance as Q4_K. * A slightly faster ARM_NEON Q2_K dot Single token prediction is now ~36 ms on M2 Max. The code is much simpler too. * Fixed bug in Q2_K CUDA dot product kernel Stranegly enough, for the few prompts I tried with the 7B model the responses looked perfectly reasonable. Only realized something is not quite right when I tried the larger models and started getting nonse back. In any case, Q2_K single token evaluation time on an RTX 4080 in a Ryzen7950X box iusing CUDA and model fully loaded on the GPU are ~15.5 ms for 7B, ~25.4 ms for 13B, and ~55.8 ms for 30B. The max number of layers that fit in VRAM for The 65B is 32. With that, we get ~330 ms per token, which is not that much faster than just running on the CPU (~470 ms per token). * Don't print zeros/NaNs when no count histogram has been collected * A 10% faster CUDA vector dot kernel for Q3_K Q3_K is now running at ~18.5 ms / token on CUDA, so the gap to Q4_0 is only 10%. It seems memory acccess pattern is more important for performance than the amount of computation the kernel does. * A slightly daster Q4_K AVX2 dot product For perplexity, where we are less memory bound, time per pass drops by ~5%. Barely measurable difference for single token prediction. * A slightly faster ARM_NEON A4_K dot product * Minor * Fix quantization error test We cannot possibly be expecting rmse < 0.002 for 2- and 3-bit quantization variants. * Fix docker build I have been sloppy with vector reinterpret casts on ARM_NEON. It seems clang is very forgiving in that regard. * Added forgotten ggml.o dependence on k_quants.h to the Makefile * Had unintentionally committed the Makefile with -Ofast enabled * ggml : rename k_quants -> ggml-quants-k, use lowercase in code --------- Co-authored-by: Iwan Kawrakow <iwan.kawrakow@gmail.com> Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
2023-06-05 19:56:18 +00:00
k-quants : support for super-block size of 64 (#2001) * k_quants: WIP super-blocks with 64 weights * k_quants: WIP super-blocks with 64 weights Q6_K scalar and AVX2 works * k_quants: WIP super-blocks with 64 weights Q4_K scalar and AVX2 works * k_quants: WIP super-blocks with 64 weights Q2_K scalar and AVX2 works. Q2_K is way too slow (it is actually slower than the scalar implementation) * k_quants: WIP super-blocks with 64 weights Q3_K scalar and AVX2 works. * k_quants: WIP super-blocks with 64 weights Q5_K scalar and AVX2 works, and with that all k_quants are done on AVX2 and scalar * k_quants: WIP super-blocks with 64 weights Q6_K working on CUDA. Cannot make it run quite as gast as with super-blocks with 256 weigths: 8% slower on 4080, 20% slower on the 1660 (but there we fit 1 less layer on the GPU because pf the larger model size), so some fraction of these 20% is due to that, * k_quants: WIP super-blocks with 64 weights Q4_K working on CUDA. ~10% slower on GTX-1660, 16% slower on 4080. * k_quants: WIP super-blocks with 64 weights Q2_K working on CUDA. ~3% slower on GTX-1660, 10% slower on 4080. * k_quants: WIP super-blocks with 64 weights Q3_K working on CUDA. * k_quants: WIP super-blocks with 64 weights Q5_K working on CUDA, and with this CUDA is done. * k_quants: WIP super-blocks with 64 weights Q6_K working on ARM_NEON * k_quants: WIP super-blocks with 64 weights Q4_K working on ARM_NEON, but quite a bit slower than 256 weights * k_quants: WIP super-blocks with 64 weights Q2_K working on ARM_NEON, but quite a bit slower than 256 weights * k_quants: WIP super-blocks with 64 weights Q3_K working on ARM_NEON, but quite a bit slower than 256 weights. * k_quants: WIP super-blocks with 64 weights Q5_K working on ARM_NEON, but quite a bit slower than 256 weights. With that, we have full support for ARM_NEON, although performance is not quite there. * k_quants: WIP super-blocks with 64 weights Slightly more efficient Q3_K and Q5_K * k_quants: WIP super-blocks with 64 weights Another small improvement for Q3_K and Q5_K on ARM_NEON * k_quants: WIP super-blocks with 64 weights Yet another speedup for Q5_K on ARM_NEON. We are now within 10% of the QK_K = 256 version. * k_quants: WIP super-blocks with 64 weights * We are able to pass preprocessor macros to the Metal compiler * Q6_K works and is actually slightly more efficient than the QK_K = 256 version (25.2 ms vs 25.8 ms) * k_quants: WIP super-blocks with 64 weights Q4_K works on Metal and is actually slightly faster than QK_K = 256 (21.95 ms vs 24.0 ms). * k_quants: WIP super-blocks with 64 weights Q2_K works on Metal and is very slightly faster than QK_K = 256 (23.8 ms vs 24.2 ms). * k_quants: WIP super-blocks with 64 weights Q3_K works on Metal and is slightly faster than QK_K = 256 (26.6 ms vs 28.3 ms). * k_quants: WIP super-blocks with 64 weights Q5_K works on Metal and is slightly faster than QK_K = 256 (23.7 ms vs 26.3 ms). * k_quants: call them _K, not _k, also on Metal * k_quants: correctly define QK_K in llama.cpp * Fixed bug in q4_K quantization added with the 64-block addition * Simplify via lambda * k_quants: swicth Q3_K to 4-bit scales when QK_K = 64 Otherwise there isn't much benefit from this quantization type. There is some very slight loss in accuracy, but we reduce size by ~7%. E.g., for OpenLLaMA-3B, Q3_K_S perplexity is 8.6131 with 8-bit scales and 8.6352 with 4-bit, while file size decreases from 1.53G to 1.44G. * k_quants: switch Q4_K to 4-bit scales when QK_K = 64 Here the loss in accuracy is greater than for Q3_K, but the Q4_K points still move further to the left on the perplexity vs size curve. * k_quants: forgot to add the Metal changes in last commit * k_quants: change Q5_K to be type 0 when QK_K = 64 Still needs AVX2 implementation * k_quants: AVX2 implementation for new 64-weight Q5_K * k_quants: 10% faster ARM_NEON Q5_K dot product * k_quants: fixed issue caused by merging with master --------- Co-authored-by: Iwan Kawrakow <iwan.kawrakow@gmail.com>
2023-06-26 16:43:07 +00:00
#if QK_K == 256
k-quants : allow to optionally disable at compile time (#1734) * k-quants : put behind optional compile flag LLAMA_K_QUANTS * build : enable k-quants by default
2023-06-07 07:59:52 +00:00
void ggml_vec_dot_q6_K_q8_K(const int n, float * restrict s, const void * restrict vx, const void * restrict vy) {
ggml : add SOTA 2,3,4,5,6 bit k-quantizations (#1684) * Starting to add k-quantization to ggml I think it is better to have quantization separate from ggml. For now just adding the k-quants there, but it would be better to also factor out the existing ggml quantizations. * Adding Q3_K and Q8_K (de)-quantization * Q3_K now working on CUDA and AVX2/scalar CUDA is not ideal - ~50% slower than Q4_0 for single token prediction, about the same in batch mode (perplexity). CPU single token is ~55 ms (on Ryzen 7950X). * Some improvement for Q3_K on CUDA It is now ~22.5 ms/token on my GPU, so ~30% slower than Q4_0. * Some more CUDA optimizations for Q3_K Single token is now 20.5 ms/token (~20% slower than Q4_0). Perplexity is on par with Q4_0. * Adding Q4_K - scalar, AVX2, CUDA Performance is the same or perhaps very slightly better than Q4_0 on the CPU. On the GPU, single token prediction is ~10% better than Q4_0, batch mode (perplexity is about the same). * Adding Q6_K - scalar, AVX2, CUDA Performance is ~40% lower compared to Q4_K on the CPU. This is to be expected, considering that we are memory bound on the CPU and the 6-bit model is ~44% larger than the 4-bit. On the GPU, single token prediction is ~6% lower than Q4_0, batch mode (perplexity) is even closer (but still slower). * Adding Q5_K - scalar, AVX2, CUDA Performance is ~20% lower compared to Q4_K on the CPU. This is to be expected, considering that we are memory bound on the CPU and the 5-bit model is ~22% larger than the 4-bit. On the GPU, single token prediction is about the same as Q4_0 for both, single token and batch prediction. * Per convention, all QX_K quantizations use Q5_K for output.weight * Adding quantization mixes * Quantization mixes: didn't quite get what I wanted in the last commit * Q4_K dot product for ARM_NEON * Q6_K dot product for ARM_NEON * Q5_K dot product for ARM_NEON * Adding Q3_K dot for ARM_NEON It is 22% slower than Q4_K, despite the smaller model size. On x86_64, where we are memory bound, the Q3_K model is quite a bit faster than Q4_K. * A very slightly faster ARM_NEON Q3_K dot * Adding Q2_K - just CUDA for now Token prediction is pretty good - about 15.5 ms on a RTX 4080. Perplexity is about the same as Q4_K. * Adding scalar and AVX2 Q2_K dot * Adding ARM_NEON Q2_K dot About the same performance as Q4_K. * A slightly faster ARM_NEON Q2_K dot Single token prediction is now ~36 ms on M2 Max. The code is much simpler too. * Fixed bug in Q2_K CUDA dot product kernel Stranegly enough, for the few prompts I tried with the 7B model the responses looked perfectly reasonable. Only realized something is not quite right when I tried the larger models and started getting nonse back. In any case, Q2_K single token evaluation time on an RTX 4080 in a Ryzen7950X box iusing CUDA and model fully loaded on the GPU are ~15.5 ms for 7B, ~25.4 ms for 13B, and ~55.8 ms for 30B. The max number of layers that fit in VRAM for The 65B is 32. With that, we get ~330 ms per token, which is not that much faster than just running on the CPU (~470 ms per token). * Don't print zeros/NaNs when no count histogram has been collected * A 10% faster CUDA vector dot kernel for Q3_K Q3_K is now running at ~18.5 ms / token on CUDA, so the gap to Q4_0 is only 10%. It seems memory acccess pattern is more important for performance than the amount of computation the kernel does. * A slightly daster Q4_K AVX2 dot product For perplexity, where we are less memory bound, time per pass drops by ~5%. Barely measurable difference for single token prediction. * A slightly faster ARM_NEON A4_K dot product * Minor * Fix quantization error test We cannot possibly be expecting rmse < 0.002 for 2- and 3-bit quantization variants. * Fix docker build I have been sloppy with vector reinterpret casts on ARM_NEON. It seems clang is very forgiving in that regard. * Added forgotten ggml.o dependence on k_quants.h to the Makefile * Had unintentionally committed the Makefile with -Ofast enabled * ggml : rename k_quants -> ggml-quants-k, use lowercase in code --------- Co-authored-by: Iwan Kawrakow <iwan.kawrakow@gmail.com> Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
2023-06-05 19:56:18 +00:00
assert(n % QK_K == 0);
k-quants : allow to optionally disable at compile time (#1734) * k-quants : put behind optional compile flag LLAMA_K_QUANTS * build : enable k-quants by default
2023-06-07 07:59:52 +00:00
const block_q6_K * restrict x = vx;
const block_q8_K * restrict y = vy;
ggml : add SOTA 2,3,4,5,6 bit k-quantizations (#1684) * Starting to add k-quantization to ggml I think it is better to have quantization separate from ggml. For now just adding the k-quants there, but it would be better to also factor out the existing ggml quantizations. * Adding Q3_K and Q8_K (de)-quantization * Q3_K now working on CUDA and AVX2/scalar CUDA is not ideal - ~50% slower than Q4_0 for single token prediction, about the same in batch mode (perplexity). CPU single token is ~55 ms (on Ryzen 7950X). * Some improvement for Q3_K on CUDA It is now ~22.5 ms/token on my GPU, so ~30% slower than Q4_0. * Some more CUDA optimizations for Q3_K Single token is now 20.5 ms/token (~20% slower than Q4_0). Perplexity is on par with Q4_0. * Adding Q4_K - scalar, AVX2, CUDA Performance is the same or perhaps very slightly better than Q4_0 on the CPU. On the GPU, single token prediction is ~10% better than Q4_0, batch mode (perplexity is about the same). * Adding Q6_K - scalar, AVX2, CUDA Performance is ~40% lower compared to Q4_K on the CPU. This is to be expected, considering that we are memory bound on the CPU and the 6-bit model is ~44% larger than the 4-bit. On the GPU, single token prediction is ~6% lower than Q4_0, batch mode (perplexity) is even closer (but still slower). * Adding Q5_K - scalar, AVX2, CUDA Performance is ~20% lower compared to Q4_K on the CPU. This is to be expected, considering that we are memory bound on the CPU and the 5-bit model is ~22% larger than the 4-bit. On the GPU, single token prediction is about the same as Q4_0 for both, single token and batch prediction. * Per convention, all QX_K quantizations use Q5_K for output.weight * Adding quantization mixes * Quantization mixes: didn't quite get what I wanted in the last commit * Q4_K dot product for ARM_NEON * Q6_K dot product for ARM_NEON * Q5_K dot product for ARM_NEON * Adding Q3_K dot for ARM_NEON It is 22% slower than Q4_K, despite the smaller model size. On x86_64, where we are memory bound, the Q3_K model is quite a bit faster than Q4_K. * A very slightly faster ARM_NEON Q3_K dot * Adding Q2_K - just CUDA for now Token prediction is pretty good - about 15.5 ms on a RTX 4080. Perplexity is about the same as Q4_K. * Adding scalar and AVX2 Q2_K dot * Adding ARM_NEON Q2_K dot About the same performance as Q4_K. * A slightly faster ARM_NEON Q2_K dot Single token prediction is now ~36 ms on M2 Max. The code is much simpler too. * Fixed bug in Q2_K CUDA dot product kernel Stranegly enough, for the few prompts I tried with the 7B model the responses looked perfectly reasonable. Only realized something is not quite right when I tried the larger models and started getting nonse back. In any case, Q2_K single token evaluation time on an RTX 4080 in a Ryzen7950X box iusing CUDA and model fully loaded on the GPU are ~15.5 ms for 7B, ~25.4 ms for 13B, and ~55.8 ms for 30B. The max number of layers that fit in VRAM for The 65B is 32. With that, we get ~330 ms per token, which is not that much faster than just running on the CPU (~470 ms per token). * Don't print zeros/NaNs when no count histogram has been collected * A 10% faster CUDA vector dot kernel for Q3_K Q3_K is now running at ~18.5 ms / token on CUDA, so the gap to Q4_0 is only 10%. It seems memory acccess pattern is more important for performance than the amount of computation the kernel does. * A slightly daster Q4_K AVX2 dot product For perplexity, where we are less memory bound, time per pass drops by ~5%. Barely measurable difference for single token prediction. * A slightly faster ARM_NEON A4_K dot product * Minor * Fix quantization error test We cannot possibly be expecting rmse < 0.002 for 2- and 3-bit quantization variants. * Fix docker build I have been sloppy with vector reinterpret casts on ARM_NEON. It seems clang is very forgiving in that regard. * Added forgotten ggml.o dependence on k_quants.h to the Makefile * Had unintentionally committed the Makefile with -Ofast enabled * ggml : rename k_quants -> ggml-quants-k, use lowercase in code --------- Co-authored-by: Iwan Kawrakow <iwan.kawrakow@gmail.com> Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
2023-06-05 19:56:18 +00:00
const int nb = n / QK_K;
#ifdef __ARM_NEON
float sum = 0;
const uint8x16_t m4b = vdupq_n_u8(0xF);
const int32x4_t vzero = vdupq_n_s32(0);
//const int8x16_t m32s = vdupq_n_s8(32);
const uint8x16_t mone = vdupq_n_u8(3);
ggml : sync (im2col, GPU conv, 32-bit arm compat) (#4060) ggml-ci
2023-11-13 14:55:52 +00:00
ggml_int8x16x4_t q6bytes;
ggml_uint8x16x4_t q6h;
ggml : add SOTA 2,3,4,5,6 bit k-quantizations (#1684) * Starting to add k-quantization to ggml I think it is better to have quantization separate from ggml. For now just adding the k-quants there, but it would be better to also factor out the existing ggml quantizations. * Adding Q3_K and Q8_K (de)-quantization * Q3_K now working on CUDA and AVX2/scalar CUDA is not ideal - ~50% slower than Q4_0 for single token prediction, about the same in batch mode (perplexity). CPU single token is ~55 ms (on Ryzen 7950X). * Some improvement for Q3_K on CUDA It is now ~22.5 ms/token on my GPU, so ~30% slower than Q4_0. * Some more CUDA optimizations for Q3_K Single token is now 20.5 ms/token (~20% slower than Q4_0). Perplexity is on par with Q4_0. * Adding Q4_K - scalar, AVX2, CUDA Performance is the same or perhaps very slightly better than Q4_0 on the CPU. On the GPU, single token prediction is ~10% better than Q4_0, batch mode (perplexity is about the same). * Adding Q6_K - scalar, AVX2, CUDA Performance is ~40% lower compared to Q4_K on the CPU. This is to be expected, considering that we are memory bound on the CPU and the 6-bit model is ~44% larger than the 4-bit. On the GPU, single token prediction is ~6% lower than Q4_0, batch mode (perplexity) is even closer (but still slower). * Adding Q5_K - scalar, AVX2, CUDA Performance is ~20% lower compared to Q4_K on the CPU. This is to be expected, considering that we are memory bound on the CPU and the 5-bit model is ~22% larger than the 4-bit. On the GPU, single token prediction is about the same as Q4_0 for both, single token and batch prediction. * Per convention, all QX_K quantizations use Q5_K for output.weight * Adding quantization mixes * Quantization mixes: didn't quite get what I wanted in the last commit * Q4_K dot product for ARM_NEON * Q6_K dot product for ARM_NEON * Q5_K dot product for ARM_NEON * Adding Q3_K dot for ARM_NEON It is 22% slower than Q4_K, despite the smaller model size. On x86_64, where we are memory bound, the Q3_K model is quite a bit faster than Q4_K. * A very slightly faster ARM_NEON Q3_K dot * Adding Q2_K - just CUDA for now Token prediction is pretty good - about 15.5 ms on a RTX 4080. Perplexity is about the same as Q4_K. * Adding scalar and AVX2 Q2_K dot * Adding ARM_NEON Q2_K dot About the same performance as Q4_K. * A slightly faster ARM_NEON Q2_K dot Single token prediction is now ~36 ms on M2 Max. The code is much simpler too. * Fixed bug in Q2_K CUDA dot product kernel Stranegly enough, for the few prompts I tried with the 7B model the responses looked perfectly reasonable. Only realized something is not quite right when I tried the larger models and started getting nonse back. In any case, Q2_K single token evaluation time on an RTX 4080 in a Ryzen7950X box iusing CUDA and model fully loaded on the GPU are ~15.5 ms for 7B, ~25.4 ms for 13B, and ~55.8 ms for 30B. The max number of layers that fit in VRAM for The 65B is 32. With that, we get ~330 ms per token, which is not that much faster than just running on the CPU (~470 ms per token). * Don't print zeros/NaNs when no count histogram has been collected * A 10% faster CUDA vector dot kernel for Q3_K Q3_K is now running at ~18.5 ms / token on CUDA, so the gap to Q4_0 is only 10%. It seems memory acccess pattern is more important for performance than the amount of computation the kernel does. * A slightly daster Q4_K AVX2 dot product For perplexity, where we are less memory bound, time per pass drops by ~5%. Barely measurable difference for single token prediction. * A slightly faster ARM_NEON A4_K dot product * Minor * Fix quantization error test We cannot possibly be expecting rmse < 0.002 for 2- and 3-bit quantization variants. * Fix docker build I have been sloppy with vector reinterpret casts on ARM_NEON. It seems clang is very forgiving in that regard. * Added forgotten ggml.o dependence on k_quants.h to the Makefile * Had unintentionally committed the Makefile with -Ofast enabled * ggml : rename k_quants -> ggml-quants-k, use lowercase in code --------- Co-authored-by: Iwan Kawrakow <iwan.kawrakow@gmail.com> Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
2023-06-05 19:56:18 +00:00
for (int i = 0; i < nb; ++i) {
ggml : move FP16 <-> FP32 code to ggml-impl.h (#3861) * ggml : move FP16 <-> FP32 stuff to ggml-impl.h ggml-ci * tests : fix ARM build * ggml : explicitly initialize deprecated type traits * ggml : add math.h to ggml-impl.h * ggml : remove duplicate static assert macros * ggml : prefix lookup tables with ggml_ ggml-ci * ggml-impl : move extern "C" to start of file
2023-10-30 17:19:15 +00:00
const float d_all = GGML_FP16_TO_FP32(x[i].d);
ggml : add SOTA 2,3,4,5,6 bit k-quantizations (#1684) * Starting to add k-quantization to ggml I think it is better to have quantization separate from ggml. For now just adding the k-quants there, but it would be better to also factor out the existing ggml quantizations. * Adding Q3_K and Q8_K (de)-quantization * Q3_K now working on CUDA and AVX2/scalar CUDA is not ideal - ~50% slower than Q4_0 for single token prediction, about the same in batch mode (perplexity). CPU single token is ~55 ms (on Ryzen 7950X). * Some improvement for Q3_K on CUDA It is now ~22.5 ms/token on my GPU, so ~30% slower than Q4_0. * Some more CUDA optimizations for Q3_K Single token is now 20.5 ms/token (~20% slower than Q4_0). Perplexity is on par with Q4_0. * Adding Q4_K - scalar, AVX2, CUDA Performance is the same or perhaps very slightly better than Q4_0 on the CPU. On the GPU, single token prediction is ~10% better than Q4_0, batch mode (perplexity is about the same). * Adding Q6_K - scalar, AVX2, CUDA Performance is ~40% lower compared to Q4_K on the CPU. This is to be expected, considering that we are memory bound on the CPU and the 6-bit model is ~44% larger than the 4-bit. On the GPU, single token prediction is ~6% lower than Q4_0, batch mode (perplexity) is even closer (but still slower). * Adding Q5_K - scalar, AVX2, CUDA Performance is ~20% lower compared to Q4_K on the CPU. This is to be expected, considering that we are memory bound on the CPU and the 5-bit model is ~22% larger than the 4-bit. On the GPU, single token prediction is about the same as Q4_0 for both, single token and batch prediction. * Per convention, all QX_K quantizations use Q5_K for output.weight * Adding quantization mixes * Quantization mixes: didn't quite get what I wanted in the last commit * Q4_K dot product for ARM_NEON * Q6_K dot product for ARM_NEON * Q5_K dot product for ARM_NEON * Adding Q3_K dot for ARM_NEON It is 22% slower than Q4_K, despite the smaller model size. On x86_64, where we are memory bound, the Q3_K model is quite a bit faster than Q4_K. * A very slightly faster ARM_NEON Q3_K dot * Adding Q2_K - just CUDA for now Token prediction is pretty good - about 15.5 ms on a RTX 4080. Perplexity is about the same as Q4_K. * Adding scalar and AVX2 Q2_K dot * Adding ARM_NEON Q2_K dot About the same performance as Q4_K. * A slightly faster ARM_NEON Q2_K dot Single token prediction is now ~36 ms on M2 Max. The code is much simpler too. * Fixed bug in Q2_K CUDA dot product kernel Stranegly enough, for the few prompts I tried with the 7B model the responses looked perfectly reasonable. Only realized something is not quite right when I tried the larger models and started getting nonse back. In any case, Q2_K single token evaluation time on an RTX 4080 in a Ryzen7950X box iusing CUDA and model fully loaded on the GPU are ~15.5 ms for 7B, ~25.4 ms for 13B, and ~55.8 ms for 30B. The max number of layers that fit in VRAM for The 65B is 32. With that, we get ~330 ms per token, which is not that much faster than just running on the CPU (~470 ms per token). * Don't print zeros/NaNs when no count histogram has been collected * A 10% faster CUDA vector dot kernel for Q3_K Q3_K is now running at ~18.5 ms / token on CUDA, so the gap to Q4_0 is only 10%. It seems memory acccess pattern is more important for performance than the amount of computation the kernel does. * A slightly daster Q4_K AVX2 dot product For perplexity, where we are less memory bound, time per pass drops by ~5%. Barely measurable difference for single token prediction. * A slightly faster ARM_NEON A4_K dot product * Minor * Fix quantization error test We cannot possibly be expecting rmse < 0.002 for 2- and 3-bit quantization variants. * Fix docker build I have been sloppy with vector reinterpret casts on ARM_NEON. It seems clang is very forgiving in that regard. * Added forgotten ggml.o dependence on k_quants.h to the Makefile * Had unintentionally committed the Makefile with -Ofast enabled * ggml : rename k_quants -> ggml-quants-k, use lowercase in code --------- Co-authored-by: Iwan Kawrakow <iwan.kawrakow@gmail.com> Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
2023-06-05 19:56:18 +00:00
const uint8_t * restrict q6 = x[i].ql;
const uint8_t * restrict qh = x[i].qh;
const int8_t * restrict q8 = y[i].qs;
const int8_t * restrict scale = x[i].scales;
ggml : sync (im2col, GPU conv, 32-bit arm compat) (#4060) ggml-ci
2023-11-13 14:55:52 +00:00
const ggml_int16x8x2_t q8sums = ggml_vld1q_s16_x2(y[i].bsums);
ggml : add SOTA 2,3,4,5,6 bit k-quantizations (#1684) * Starting to add k-quantization to ggml I think it is better to have quantization separate from ggml. For now just adding the k-quants there, but it would be better to also factor out the existing ggml quantizations. * Adding Q3_K and Q8_K (de)-quantization * Q3_K now working on CUDA and AVX2/scalar CUDA is not ideal - ~50% slower than Q4_0 for single token prediction, about the same in batch mode (perplexity). CPU single token is ~55 ms (on Ryzen 7950X). * Some improvement for Q3_K on CUDA It is now ~22.5 ms/token on my GPU, so ~30% slower than Q4_0. * Some more CUDA optimizations for Q3_K Single token is now 20.5 ms/token (~20% slower than Q4_0). Perplexity is on par with Q4_0. * Adding Q4_K - scalar, AVX2, CUDA Performance is the same or perhaps very slightly better than Q4_0 on the CPU. On the GPU, single token prediction is ~10% better than Q4_0, batch mode (perplexity is about the same). * Adding Q6_K - scalar, AVX2, CUDA Performance is ~40% lower compared to Q4_K on the CPU. This is to be expected, considering that we are memory bound on the CPU and the 6-bit model is ~44% larger than the 4-bit. On the GPU, single token prediction is ~6% lower than Q4_0, batch mode (perplexity) is even closer (but still slower). * Adding Q5_K - scalar, AVX2, CUDA Performance is ~20% lower compared to Q4_K on the CPU. This is to be expected, considering that we are memory bound on the CPU and the 5-bit model is ~22% larger than the 4-bit. On the GPU, single token prediction is about the same as Q4_0 for both, single token and batch prediction. * Per convention, all QX_K quantizations use Q5_K for output.weight * Adding quantization mixes * Quantization mixes: didn't quite get what I wanted in the last commit * Q4_K dot product for ARM_NEON * Q6_K dot product for ARM_NEON * Q5_K dot product for ARM_NEON * Adding Q3_K dot for ARM_NEON It is 22% slower than Q4_K, despite the smaller model size. On x86_64, where we are memory bound, the Q3_K model is quite a bit faster than Q4_K. * A very slightly faster ARM_NEON Q3_K dot * Adding Q2_K - just CUDA for now Token prediction is pretty good - about 15.5 ms on a RTX 4080. Perplexity is about the same as Q4_K. * Adding scalar and AVX2 Q2_K dot * Adding ARM_NEON Q2_K dot About the same performance as Q4_K. * A slightly faster ARM_NEON Q2_K dot Single token prediction is now ~36 ms on M2 Max. The code is much simpler too. * Fixed bug in Q2_K CUDA dot product kernel Stranegly enough, for the few prompts I tried with the 7B model the responses looked perfectly reasonable. Only realized something is not quite right when I tried the larger models and started getting nonse back. In any case, Q2_K single token evaluation time on an RTX 4080 in a Ryzen7950X box iusing CUDA and model fully loaded on the GPU are ~15.5 ms for 7B, ~25.4 ms for 13B, and ~55.8 ms for 30B. The max number of layers that fit in VRAM for The 65B is 32. With that, we get ~330 ms per token, which is not that much faster than just running on the CPU (~470 ms per token). * Don't print zeros/NaNs when no count histogram has been collected * A 10% faster CUDA vector dot kernel for Q3_K Q3_K is now running at ~18.5 ms / token on CUDA, so the gap to Q4_0 is only 10%. It seems memory acccess pattern is more important for performance than the amount of computation the kernel does. * A slightly daster Q4_K AVX2 dot product For perplexity, where we are less memory bound, time per pass drops by ~5%. Barely measurable difference for single token prediction. * A slightly faster ARM_NEON A4_K dot product * Minor * Fix quantization error test We cannot possibly be expecting rmse < 0.002 for 2- and 3-bit quantization variants. * Fix docker build I have been sloppy with vector reinterpret casts on ARM_NEON. It seems clang is very forgiving in that regard. * Added forgotten ggml.o dependence on k_quants.h to the Makefile * Had unintentionally committed the Makefile with -Ofast enabled * ggml : rename k_quants -> ggml-quants-k, use lowercase in code --------- Co-authored-by: Iwan Kawrakow <iwan.kawrakow@gmail.com> Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
2023-06-05 19:56:18 +00:00
const int8x16_t scales = vld1q_s8(scale);
cuda : fix jetson compile error (#4560) * fix old jetson compile error * Update Makefile * update jetson detect and cuda version detect * update cuda marco define * update makefile and cuda,fix some issue * Update README.md Co-authored-by: Georgi Gerganov <ggerganov@gmail.com> * Update Makefile * Update README.md --------- Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
2023-12-22 15:11:12 +00:00
const ggml_int16x8x2_t q6scales = {{vmovl_s8(vget_low_s8(scales)), vmovl_s8(vget_high_s8(scales))}};
ggml : add SOTA 2,3,4,5,6 bit k-quantizations (#1684) * Starting to add k-quantization to ggml I think it is better to have quantization separate from ggml. For now just adding the k-quants there, but it would be better to also factor out the existing ggml quantizations. * Adding Q3_K and Q8_K (de)-quantization * Q3_K now working on CUDA and AVX2/scalar CUDA is not ideal - ~50% slower than Q4_0 for single token prediction, about the same in batch mode (perplexity). CPU single token is ~55 ms (on Ryzen 7950X). * Some improvement for Q3_K on CUDA It is now ~22.5 ms/token on my GPU, so ~30% slower than Q4_0. * Some more CUDA optimizations for Q3_K Single token is now 20.5 ms/token (~20% slower than Q4_0). Perplexity is on par with Q4_0. * Adding Q4_K - scalar, AVX2, CUDA Performance is the same or perhaps very slightly better than Q4_0 on the CPU. On the GPU, single token prediction is ~10% better than Q4_0, batch mode (perplexity is about the same). * Adding Q6_K - scalar, AVX2, CUDA Performance is ~40% lower compared to Q4_K on the CPU. This is to be expected, considering that we are memory bound on the CPU and the 6-bit model is ~44% larger than the 4-bit. On the GPU, single token prediction is ~6% lower than Q4_0, batch mode (perplexity) is even closer (but still slower). * Adding Q5_K - scalar, AVX2, CUDA Performance is ~20% lower compared to Q4_K on the CPU. This is to be expected, considering that we are memory bound on the CPU and the 5-bit model is ~22% larger than the 4-bit. On the GPU, single token prediction is about the same as Q4_0 for both, single token and batch prediction. * Per convention, all QX_K quantizations use Q5_K for output.weight * Adding quantization mixes * Quantization mixes: didn't quite get what I wanted in the last commit * Q4_K dot product for ARM_NEON * Q6_K dot product for ARM_NEON * Q5_K dot product for ARM_NEON * Adding Q3_K dot for ARM_NEON It is 22% slower than Q4_K, despite the smaller model size. On x86_64, where we are memory bound, the Q3_K model is quite a bit faster than Q4_K. * A very slightly faster ARM_NEON Q3_K dot * Adding Q2_K - just CUDA for now Token prediction is pretty good - about 15.5 ms on a RTX 4080. Perplexity is about the same as Q4_K. * Adding scalar and AVX2 Q2_K dot * Adding ARM_NEON Q2_K dot About the same performance as Q4_K. * A slightly faster ARM_NEON Q2_K dot Single token prediction is now ~36 ms on M2 Max. The code is much simpler too. * Fixed bug in Q2_K CUDA dot product kernel Stranegly enough, for the few prompts I tried with the 7B model the responses looked perfectly reasonable. Only realized something is not quite right when I tried the larger models and started getting nonse back. In any case, Q2_K single token evaluation time on an RTX 4080 in a Ryzen7950X box iusing CUDA and model fully loaded on the GPU are ~15.5 ms for 7B, ~25.4 ms for 13B, and ~55.8 ms for 30B. The max number of layers that fit in VRAM for The 65B is 32. With that, we get ~330 ms per token, which is not that much faster than just running on the CPU (~470 ms per token). * Don't print zeros/NaNs when no count histogram has been collected * A 10% faster CUDA vector dot kernel for Q3_K Q3_K is now running at ~18.5 ms / token on CUDA, so the gap to Q4_0 is only 10%. It seems memory acccess pattern is more important for performance than the amount of computation the kernel does. * A slightly daster Q4_K AVX2 dot product For perplexity, where we are less memory bound, time per pass drops by ~5%. Barely measurable difference for single token prediction. * A slightly faster ARM_NEON A4_K dot product * Minor * Fix quantization error test We cannot possibly be expecting rmse < 0.002 for 2- and 3-bit quantization variants. * Fix docker build I have been sloppy with vector reinterpret casts on ARM_NEON. It seems clang is very forgiving in that regard. * Added forgotten ggml.o dependence on k_quants.h to the Makefile * Had unintentionally committed the Makefile with -Ofast enabled * ggml : rename k_quants -> ggml-quants-k, use lowercase in code --------- Co-authored-by: Iwan Kawrakow <iwan.kawrakow@gmail.com> Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
2023-06-05 19:56:18 +00:00
const int32x4_t prod = vaddq_s32(vaddq_s32(vmull_s16(vget_low_s16 (q8sums.val[0]), vget_low_s16 (q6scales.val[0])),
vmull_s16(vget_high_s16(q8sums.val[0]), vget_high_s16(q6scales.val[0]))),
vaddq_s32(vmull_s16(vget_low_s16 (q8sums.val[1]), vget_low_s16 (q6scales.val[1])),
vmull_s16(vget_high_s16(q8sums.val[1]), vget_high_s16(q6scales.val[1]))));
int32_t isum_mins = vaddvq_s32(prod);
int32_t isum = 0;
for (int j = 0; j < QK_K/128; ++j) {
ggml : sync (im2col, GPU conv, 32-bit arm compat) (#4060) ggml-ci
2023-11-13 14:55:52 +00:00
ggml_uint8x16x2_t qhbits = ggml_vld1q_u8_x2(qh); qh += 32;
ggml_uint8x16x4_t q6bits = ggml_vld1q_u8_x4(q6); q6 += 64;
ggml_int8x16x4_t q8bytes = ggml_vld1q_s8_x4(q8); q8 += 64;
ggml : add SOTA 2,3,4,5,6 bit k-quantizations (#1684) * Starting to add k-quantization to ggml I think it is better to have quantization separate from ggml. For now just adding the k-quants there, but it would be better to also factor out the existing ggml quantizations. * Adding Q3_K and Q8_K (de)-quantization * Q3_K now working on CUDA and AVX2/scalar CUDA is not ideal - ~50% slower than Q4_0 for single token prediction, about the same in batch mode (perplexity). CPU single token is ~55 ms (on Ryzen 7950X). * Some improvement for Q3_K on CUDA It is now ~22.5 ms/token on my GPU, so ~30% slower than Q4_0. * Some more CUDA optimizations for Q3_K Single token is now 20.5 ms/token (~20% slower than Q4_0). Perplexity is on par with Q4_0. * Adding Q4_K - scalar, AVX2, CUDA Performance is the same or perhaps very slightly better than Q4_0 on the CPU. On the GPU, single token prediction is ~10% better than Q4_0, batch mode (perplexity is about the same). * Adding Q6_K - scalar, AVX2, CUDA Performance is ~40% lower compared to Q4_K on the CPU. This is to be expected, considering that we are memory bound on the CPU and the 6-bit model is ~44% larger than the 4-bit. On the GPU, single token prediction is ~6% lower than Q4_0, batch mode (perplexity) is even closer (but still slower). * Adding Q5_K - scalar, AVX2, CUDA Performance is ~20% lower compared to Q4_K on the CPU. This is to be expected, considering that we are memory bound on the CPU and the 5-bit model is ~22% larger than the 4-bit. On the GPU, single token prediction is about the same as Q4_0 for both, single token and batch prediction. * Per convention, all QX_K quantizations use Q5_K for output.weight * Adding quantization mixes * Quantization mixes: didn't quite get what I wanted in the last commit * Q4_K dot product for ARM_NEON * Q6_K dot product for ARM_NEON * Q5_K dot product for ARM_NEON * Adding Q3_K dot for ARM_NEON It is 22% slower than Q4_K, despite the smaller model size. On x86_64, where we are memory bound, the Q3_K model is quite a bit faster than Q4_K. * A very slightly faster ARM_NEON Q3_K dot * Adding Q2_K - just CUDA for now Token prediction is pretty good - about 15.5 ms on a RTX 4080. Perplexity is about the same as Q4_K. * Adding scalar and AVX2 Q2_K dot * Adding ARM_NEON Q2_K dot About the same performance as Q4_K. * A slightly faster ARM_NEON Q2_K dot Single token prediction is now ~36 ms on M2 Max. The code is much simpler too. * Fixed bug in Q2_K CUDA dot product kernel Stranegly enough, for the few prompts I tried with the 7B model the responses looked perfectly reasonable. Only realized something is not quite right when I tried the larger models and started getting nonse back. In any case, Q2_K single token evaluation time on an RTX 4080 in a Ryzen7950X box iusing CUDA and model fully loaded on the GPU are ~15.5 ms for 7B, ~25.4 ms for 13B, and ~55.8 ms for 30B. The max number of layers that fit in VRAM for The 65B is 32. With that, we get ~330 ms per token, which is not that much faster than just running on the CPU (~470 ms per token). * Don't print zeros/NaNs when no count histogram has been collected * A 10% faster CUDA vector dot kernel for Q3_K Q3_K is now running at ~18.5 ms / token on CUDA, so the gap to Q4_0 is only 10%. It seems memory acccess pattern is more important for performance than the amount of computation the kernel does. * A slightly daster Q4_K AVX2 dot product For perplexity, where we are less memory bound, time per pass drops by ~5%. Barely measurable difference for single token prediction. * A slightly faster ARM_NEON A4_K dot product * Minor * Fix quantization error test We cannot possibly be expecting rmse < 0.002 for 2- and 3-bit quantization variants. * Fix docker build I have been sloppy with vector reinterpret casts on ARM_NEON. It seems clang is very forgiving in that regard. * Added forgotten ggml.o dependence on k_quants.h to the Makefile * Had unintentionally committed the Makefile with -Ofast enabled * ggml : rename k_quants -> ggml-quants-k, use lowercase in code --------- Co-authored-by: Iwan Kawrakow <iwan.kawrakow@gmail.com> Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
2023-06-05 19:56:18 +00:00
q6h.val[0] = vshlq_n_u8(vandq_u8(mone, qhbits.val[0]), 4);
q6h.val[1] = vshlq_n_u8(vandq_u8(mone, qhbits.val[1]), 4);
uint8x16_t shifted = vshrq_n_u8(qhbits.val[0], 2);
q6h.val[2] = vshlq_n_u8(vandq_u8(mone, shifted), 4);
shifted = vshrq_n_u8(qhbits.val[1], 2);
q6h.val[3] = vshlq_n_u8(vandq_u8(mone, shifted), 4);
//q6bytes.val[0] = vsubq_s8(vreinterpretq_s8_u8(vorrq_u8(vandq_u8(q6bits.val[0], m4b), q6h.val[0])), m32s);
//q6bytes.val[1] = vsubq_s8(vreinterpretq_s8_u8(vorrq_u8(vandq_u8(q6bits.val[1], m4b), q6h.val[1])), m32s);
//q6bytes.val[2] = vsubq_s8(vreinterpretq_s8_u8(vorrq_u8(vandq_u8(q6bits.val[2], m4b), q6h.val[2])), m32s);
//q6bytes.val[3] = vsubq_s8(vreinterpretq_s8_u8(vorrq_u8(vandq_u8(q6bits.val[3], m4b), q6h.val[3])), m32s);
q6bytes.val[0] = vreinterpretq_s8_u8(vorrq_u8(vandq_u8(q6bits.val[0], m4b), q6h.val[0]));
q6bytes.val[1] = vreinterpretq_s8_u8(vorrq_u8(vandq_u8(q6bits.val[1], m4b), q6h.val[1]));
q6bytes.val[2] = vreinterpretq_s8_u8(vorrq_u8(vandq_u8(q6bits.val[2], m4b), q6h.val[2]));
q6bytes.val[3] = vreinterpretq_s8_u8(vorrq_u8(vandq_u8(q6bits.val[3], m4b), q6h.val[3]));
ggml : add ggml_vdotq_s32 alias (#4715) ggml-ci
2023-12-31 09:43:31 +00:00
isum += vaddvq_s32(ggml_vdotq_s32(vzero, q6bytes.val[0], q8bytes.val[0])) * scale[0] +
vaddvq_s32(ggml_vdotq_s32(vzero, q6bytes.val[1], q8bytes.val[1])) * scale[1] +
vaddvq_s32(ggml_vdotq_s32(vzero, q6bytes.val[2], q8bytes.val[2])) * scale[2] +
vaddvq_s32(ggml_vdotq_s32(vzero, q6bytes.val[3], q8bytes.val[3])) * scale[3];
ggml : add SOTA 2,3,4,5,6 bit k-quantizations (#1684) * Starting to add k-quantization to ggml I think it is better to have quantization separate from ggml. For now just adding the k-quants there, but it would be better to also factor out the existing ggml quantizations. * Adding Q3_K and Q8_K (de)-quantization * Q3_K now working on CUDA and AVX2/scalar CUDA is not ideal - ~50% slower than Q4_0 for single token prediction, about the same in batch mode (perplexity). CPU single token is ~55 ms (on Ryzen 7950X). * Some improvement for Q3_K on CUDA It is now ~22.5 ms/token on my GPU, so ~30% slower than Q4_0. * Some more CUDA optimizations for Q3_K Single token is now 20.5 ms/token (~20% slower than Q4_0). Perplexity is on par with Q4_0. * Adding Q4_K - scalar, AVX2, CUDA Performance is the same or perhaps very slightly better than Q4_0 on the CPU. On the GPU, single token prediction is ~10% better than Q4_0, batch mode (perplexity is about the same). * Adding Q6_K - scalar, AVX2, CUDA Performance is ~40% lower compared to Q4_K on the CPU. This is to be expected, considering that we are memory bound on the CPU and the 6-bit model is ~44% larger than the 4-bit. On the GPU, single token prediction is ~6% lower than Q4_0, batch mode (perplexity) is even closer (but still slower). * Adding Q5_K - scalar, AVX2, CUDA Performance is ~20% lower compared to Q4_K on the CPU. This is to be expected, considering that we are memory bound on the CPU and the 5-bit model is ~22% larger than the 4-bit. On the GPU, single token prediction is about the same as Q4_0 for both, single token and batch prediction. * Per convention, all QX_K quantizations use Q5_K for output.weight * Adding quantization mixes * Quantization mixes: didn't quite get what I wanted in the last commit * Q4_K dot product for ARM_NEON * Q6_K dot product for ARM_NEON * Q5_K dot product for ARM_NEON * Adding Q3_K dot for ARM_NEON It is 22% slower than Q4_K, despite the smaller model size. On x86_64, where we are memory bound, the Q3_K model is quite a bit faster than Q4_K. * A very slightly faster ARM_NEON Q3_K dot * Adding Q2_K - just CUDA for now Token prediction is pretty good - about 15.5 ms on a RTX 4080. Perplexity is about the same as Q4_K. * Adding scalar and AVX2 Q2_K dot * Adding ARM_NEON Q2_K dot About the same performance as Q4_K. * A slightly faster ARM_NEON Q2_K dot Single token prediction is now ~36 ms on M2 Max. The code is much simpler too. * Fixed bug in Q2_K CUDA dot product kernel Stranegly enough, for the few prompts I tried with the 7B model the responses looked perfectly reasonable. Only realized something is not quite right when I tried the larger models and started getting nonse back. In any case, Q2_K single token evaluation time on an RTX 4080 in a Ryzen7950X box iusing CUDA and model fully loaded on the GPU are ~15.5 ms for 7B, ~25.4 ms for 13B, and ~55.8 ms for 30B. The max number of layers that fit in VRAM for The 65B is 32. With that, we get ~330 ms per token, which is not that much faster than just running on the CPU (~470 ms per token). * Don't print zeros/NaNs when no count histogram has been collected * A 10% faster CUDA vector dot kernel for Q3_K Q3_K is now running at ~18.5 ms / token on CUDA, so the gap to Q4_0 is only 10%. It seems memory acccess pattern is more important for performance than the amount of computation the kernel does. * A slightly daster Q4_K AVX2 dot product For perplexity, where we are less memory bound, time per pass drops by ~5%. Barely measurable difference for single token prediction. * A slightly faster ARM_NEON A4_K dot product * Minor * Fix quantization error test We cannot possibly be expecting rmse < 0.002 for 2- and 3-bit quantization variants. * Fix docker build I have been sloppy with vector reinterpret casts on ARM_NEON. It seems clang is very forgiving in that regard. * Added forgotten ggml.o dependence on k_quants.h to the Makefile * Had unintentionally committed the Makefile with -Ofast enabled * ggml : rename k_quants -> ggml-quants-k, use lowercase in code --------- Co-authored-by: Iwan Kawrakow <iwan.kawrakow@gmail.com> Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
2023-06-05 19:56:18 +00:00
ggml : fix dot product for ARM (#4630) ggml-ci
2023-12-27 09:02:13 +00:00
scale += 4;
ggml : add SOTA 2,3,4,5,6 bit k-quantizations (#1684) * Starting to add k-quantization to ggml I think it is better to have quantization separate from ggml. For now just adding the k-quants there, but it would be better to also factor out the existing ggml quantizations. * Adding Q3_K and Q8_K (de)-quantization * Q3_K now working on CUDA and AVX2/scalar CUDA is not ideal - ~50% slower than Q4_0 for single token prediction, about the same in batch mode (perplexity). CPU single token is ~55 ms (on Ryzen 7950X). * Some improvement for Q3_K on CUDA It is now ~22.5 ms/token on my GPU, so ~30% slower than Q4_0. * Some more CUDA optimizations for Q3_K Single token is now 20.5 ms/token (~20% slower than Q4_0). Perplexity is on par with Q4_0. * Adding Q4_K - scalar, AVX2, CUDA Performance is the same or perhaps very slightly better than Q4_0 on the CPU. On the GPU, single token prediction is ~10% better than Q4_0, batch mode (perplexity is about the same). * Adding Q6_K - scalar, AVX2, CUDA Performance is ~40% lower compared to Q4_K on the CPU. This is to be expected, considering that we are memory bound on the CPU and the 6-bit model is ~44% larger than the 4-bit. On the GPU, single token prediction is ~6% lower than Q4_0, batch mode (perplexity) is even closer (but still slower). * Adding Q5_K - scalar, AVX2, CUDA Performance is ~20% lower compared to Q4_K on the CPU. This is to be expected, considering that we are memory bound on the CPU and the 5-bit model is ~22% larger than the 4-bit. On the GPU, single token prediction is about the same as Q4_0 for both, single token and batch prediction. * Per convention, all QX_K quantizations use Q5_K for output.weight * Adding quantization mixes * Quantization mixes: didn't quite get what I wanted in the last commit * Q4_K dot product for ARM_NEON * Q6_K dot product for ARM_NEON * Q5_K dot product for ARM_NEON * Adding Q3_K dot for ARM_NEON It is 22% slower than Q4_K, despite the smaller model size. On x86_64, where we are memory bound, the Q3_K model is quite a bit faster than Q4_K. * A very slightly faster ARM_NEON Q3_K dot * Adding Q2_K - just CUDA for now Token prediction is pretty good - about 15.5 ms on a RTX 4080. Perplexity is about the same as Q4_K. * Adding scalar and AVX2 Q2_K dot * Adding ARM_NEON Q2_K dot About the same performance as Q4_K. * A slightly faster ARM_NEON Q2_K dot Single token prediction is now ~36 ms on M2 Max. The code is much simpler too. * Fixed bug in Q2_K CUDA dot product kernel Stranegly enough, for the few prompts I tried with the 7B model the responses looked perfectly reasonable. Only realized something is not quite right when I tried the larger models and started getting nonse back. In any case, Q2_K single token evaluation time on an RTX 4080 in a Ryzen7950X box iusing CUDA and model fully loaded on the GPU are ~15.5 ms for 7B, ~25.4 ms for 13B, and ~55.8 ms for 30B. The max number of layers that fit in VRAM for The 65B is 32. With that, we get ~330 ms per token, which is not that much faster than just running on the CPU (~470 ms per token). * Don't print zeros/NaNs when no count histogram has been collected * A 10% faster CUDA vector dot kernel for Q3_K Q3_K is now running at ~18.5 ms / token on CUDA, so the gap to Q4_0 is only 10%. It seems memory acccess pattern is more important for performance than the amount of computation the kernel does. * A slightly daster Q4_K AVX2 dot product For perplexity, where we are less memory bound, time per pass drops by ~5%. Barely measurable difference for single token prediction. * A slightly faster ARM_NEON A4_K dot product * Minor * Fix quantization error test We cannot possibly be expecting rmse < 0.002 for 2- and 3-bit quantization variants. * Fix docker build I have been sloppy with vector reinterpret casts on ARM_NEON. It seems clang is very forgiving in that regard. * Added forgotten ggml.o dependence on k_quants.h to the Makefile * Had unintentionally committed the Makefile with -Ofast enabled * ggml : rename k_quants -> ggml-quants-k, use lowercase in code --------- Co-authored-by: Iwan Kawrakow <iwan.kawrakow@gmail.com> Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
2023-06-05 19:56:18 +00:00
ggml : sync (im2col, GPU conv, 32-bit arm compat) (#4060) ggml-ci
2023-11-13 14:55:52 +00:00
q8bytes = ggml_vld1q_s8_x4(q8); q8 += 64;
ggml : add SOTA 2,3,4,5,6 bit k-quantizations (#1684) * Starting to add k-quantization to ggml I think it is better to have quantization separate from ggml. For now just adding the k-quants there, but it would be better to also factor out the existing ggml quantizations. * Adding Q3_K and Q8_K (de)-quantization * Q3_K now working on CUDA and AVX2/scalar CUDA is not ideal - ~50% slower than Q4_0 for single token prediction, about the same in batch mode (perplexity). CPU single token is ~55 ms (on Ryzen 7950X). * Some improvement for Q3_K on CUDA It is now ~22.5 ms/token on my GPU, so ~30% slower than Q4_0. * Some more CUDA optimizations for Q3_K Single token is now 20.5 ms/token (~20% slower than Q4_0). Perplexity is on par with Q4_0. * Adding Q4_K - scalar, AVX2, CUDA Performance is the same or perhaps very slightly better than Q4_0 on the CPU. On the GPU, single token prediction is ~10% better than Q4_0, batch mode (perplexity is about the same). * Adding Q6_K - scalar, AVX2, CUDA Performance is ~40% lower compared to Q4_K on the CPU. This is to be expected, considering that we are memory bound on the CPU and the 6-bit model is ~44% larger than the 4-bit. On the GPU, single token prediction is ~6% lower than Q4_0, batch mode (perplexity) is even closer (but still slower). * Adding Q5_K - scalar, AVX2, CUDA Performance is ~20% lower compared to Q4_K on the CPU. This is to be expected, considering that we are memory bound on the CPU and the 5-bit model is ~22% larger than the 4-bit. On the GPU, single token prediction is about the same as Q4_0 for both, single token and batch prediction. * Per convention, all QX_K quantizations use Q5_K for output.weight * Adding quantization mixes * Quantization mixes: didn't quite get what I wanted in the last commit * Q4_K dot product for ARM_NEON * Q6_K dot product for ARM_NEON * Q5_K dot product for ARM_NEON * Adding Q3_K dot for ARM_NEON It is 22% slower than Q4_K, despite the smaller model size. On x86_64, where we are memory bound, the Q3_K model is quite a bit faster than Q4_K. * A very slightly faster ARM_NEON Q3_K dot * Adding Q2_K - just CUDA for now Token prediction is pretty good - about 15.5 ms on a RTX 4080. Perplexity is about the same as Q4_K. * Adding scalar and AVX2 Q2_K dot * Adding ARM_NEON Q2_K dot About the same performance as Q4_K. * A slightly faster ARM_NEON Q2_K dot Single token prediction is now ~36 ms on M2 Max. The code is much simpler too. * Fixed bug in Q2_K CUDA dot product kernel Stranegly enough, for the few prompts I tried with the 7B model the responses looked perfectly reasonable. Only realized something is not quite right when I tried the larger models and started getting nonse back. In any case, Q2_K single token evaluation time on an RTX 4080 in a Ryzen7950X box iusing CUDA and model fully loaded on the GPU are ~15.5 ms for 7B, ~25.4 ms for 13B, and ~55.8 ms for 30B. The max number of layers that fit in VRAM for The 65B is 32. With that, we get ~330 ms per token, which is not that much faster than just running on the CPU (~470 ms per token). * Don't print zeros/NaNs when no count histogram has been collected * A 10% faster CUDA vector dot kernel for Q3_K Q3_K is now running at ~18.5 ms / token on CUDA, so the gap to Q4_0 is only 10%. It seems memory acccess pattern is more important for performance than the amount of computation the kernel does. * A slightly daster Q4_K AVX2 dot product For perplexity, where we are less memory bound, time per pass drops by ~5%. Barely measurable difference for single token prediction. * A slightly faster ARM_NEON A4_K dot product * Minor * Fix quantization error test We cannot possibly be expecting rmse < 0.002 for 2- and 3-bit quantization variants. * Fix docker build I have been sloppy with vector reinterpret casts on ARM_NEON. It seems clang is very forgiving in that regard. * Added forgotten ggml.o dependence on k_quants.h to the Makefile * Had unintentionally committed the Makefile with -Ofast enabled * ggml : rename k_quants -> ggml-quants-k, use lowercase in code --------- Co-authored-by: Iwan Kawrakow <iwan.kawrakow@gmail.com> Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
2023-06-05 19:56:18 +00:00
shifted = vshrq_n_u8(qhbits.val[0], 4);
q6h.val[0] = vshlq_n_u8(vandq_u8(mone, shifted), 4);
shifted = vshrq_n_u8(qhbits.val[1], 4);
q6h.val[1] = vshlq_n_u8(vandq_u8(mone, shifted), 4);
shifted = vshrq_n_u8(qhbits.val[0], 6);
q6h.val[2] = vshlq_n_u8(vandq_u8(mone, shifted), 4);
shifted = vshrq_n_u8(qhbits.val[1], 6);
q6h.val[3] = vshlq_n_u8(vandq_u8(mone, shifted), 4);
//q6bytes.val[0] = vsubq_s8(vreinterpretq_s8_u8(vorrq_u8(vshrq_n_u8(q6bits.val[0], 4), q6h.val[0])), m32s);
//q6bytes.val[1] = vsubq_s8(vreinterpretq_s8_u8(vorrq_u8(vshrq_n_u8(q6bits.val[1], 4), q6h.val[1])), m32s);
//q6bytes.val[2] = vsubq_s8(vreinterpretq_s8_u8(vorrq_u8(vshrq_n_u8(q6bits.val[2], 4), q6h.val[2])), m32s);
//q6bytes.val[3] = vsubq_s8(vreinterpretq_s8_u8(vorrq_u8(vshrq_n_u8(q6bits.val[3], 4), q6h.val[3])), m32s);
q6bytes.val[0] = vreinterpretq_s8_u8(vorrq_u8(vshrq_n_u8(q6bits.val[0], 4), q6h.val[0]));
q6bytes.val[1] = vreinterpretq_s8_u8(vorrq_u8(vshrq_n_u8(q6bits.val[1], 4), q6h.val[1]));
q6bytes.val[2] = vreinterpretq_s8_u8(vorrq_u8(vshrq_n_u8(q6bits.val[2], 4), q6h.val[2]));
q6bytes.val[3] = vreinterpretq_s8_u8(vorrq_u8(vshrq_n_u8(q6bits.val[3], 4), q6h.val[3]));
ggml : add ggml_vdotq_s32 alias (#4715) ggml-ci
2023-12-31 09:43:31 +00:00
isum += vaddvq_s32(ggml_vdotq_s32(vzero, q6bytes.val[0], q8bytes.val[0])) * scale[0] +
vaddvq_s32(ggml_vdotq_s32(vzero, q6bytes.val[1], q8bytes.val[1])) * scale[1] +
vaddvq_s32(ggml_vdotq_s32(vzero, q6bytes.val[2], q8bytes.val[2])) * scale[2] +
vaddvq_s32(ggml_vdotq_s32(vzero, q6bytes.val[3], q8bytes.val[3])) * scale[3];
ggml : add SOTA 2,3,4,5,6 bit k-quantizations (#1684) * Starting to add k-quantization to ggml I think it is better to have quantization separate from ggml. For now just adding the k-quants there, but it would be better to also factor out the existing ggml quantizations. * Adding Q3_K and Q8_K (de)-quantization * Q3_K now working on CUDA and AVX2/scalar CUDA is not ideal - ~50% slower than Q4_0 for single token prediction, about the same in batch mode (perplexity). CPU single token is ~55 ms (on Ryzen 7950X). * Some improvement for Q3_K on CUDA It is now ~22.5 ms/token on my GPU, so ~30% slower than Q4_0. * Some more CUDA optimizations for Q3_K Single token is now 20.5 ms/token (~20% slower than Q4_0). Perplexity is on par with Q4_0. * Adding Q4_K - scalar, AVX2, CUDA Performance is the same or perhaps very slightly better than Q4_0 on the CPU. On the GPU, single token prediction is ~10% better than Q4_0, batch mode (perplexity is about the same). * Adding Q6_K - scalar, AVX2, CUDA Performance is ~40% lower compared to Q4_K on the CPU. This is to be expected, considering that we are memory bound on the CPU and the 6-bit model is ~44% larger than the 4-bit. On the GPU, single token prediction is ~6% lower than Q4_0, batch mode (perplexity) is even closer (but still slower). * Adding Q5_K - scalar, AVX2, CUDA Performance is ~20% lower compared to Q4_K on the CPU. This is to be expected, considering that we are memory bound on the CPU and the 5-bit model is ~22% larger than the 4-bit. On the GPU, single token prediction is about the same as Q4_0 for both, single token and batch prediction. * Per convention, all QX_K quantizations use Q5_K for output.weight * Adding quantization mixes * Quantization mixes: didn't quite get what I wanted in the last commit * Q4_K dot product for ARM_NEON * Q6_K dot product for ARM_NEON * Q5_K dot product for ARM_NEON * Adding Q3_K dot for ARM_NEON It is 22% slower than Q4_K, despite the smaller model size. On x86_64, where we are memory bound, the Q3_K model is quite a bit faster than Q4_K. * A very slightly faster ARM_NEON Q3_K dot * Adding Q2_K - just CUDA for now Token prediction is pretty good - about 15.5 ms on a RTX 4080. Perplexity is about the same as Q4_K. * Adding scalar and AVX2 Q2_K dot * Adding ARM_NEON Q2_K dot About the same performance as Q4_K. * A slightly faster ARM_NEON Q2_K dot Single token prediction is now ~36 ms on M2 Max. The code is much simpler too. * Fixed bug in Q2_K CUDA dot product kernel Stranegly enough, for the few prompts I tried with the 7B model the responses looked perfectly reasonable. Only realized something is not quite right when I tried the larger models and started getting nonse back. In any case, Q2_K single token evaluation time on an RTX 4080 in a Ryzen7950X box iusing CUDA and model fully loaded on the GPU are ~15.5 ms for 7B, ~25.4 ms for 13B, and ~55.8 ms for 30B. The max number of layers that fit in VRAM for The 65B is 32. With that, we get ~330 ms per token, which is not that much faster than just running on the CPU (~470 ms per token). * Don't print zeros/NaNs when no count histogram has been collected * A 10% faster CUDA vector dot kernel for Q3_K Q3_K is now running at ~18.5 ms / token on CUDA, so the gap to Q4_0 is only 10%. It seems memory acccess pattern is more important for performance than the amount of computation the kernel does. * A slightly daster Q4_K AVX2 dot product For perplexity, where we are less memory bound, time per pass drops by ~5%. Barely measurable difference for single token prediction. * A slightly faster ARM_NEON A4_K dot product * Minor * Fix quantization error test We cannot possibly be expecting rmse < 0.002 for 2- and 3-bit quantization variants. * Fix docker build I have been sloppy with vector reinterpret casts on ARM_NEON. It seems clang is very forgiving in that regard. * Added forgotten ggml.o dependence on k_quants.h to the Makefile * Had unintentionally committed the Makefile with -Ofast enabled * ggml : rename k_quants -> ggml-quants-k, use lowercase in code --------- Co-authored-by: Iwan Kawrakow <iwan.kawrakow@gmail.com> Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
2023-06-05 19:56:18 +00:00
scale += 4;
}
//sum += isum * d_all * y[i].d;
sum += d_all * y[i].d * (isum - 32 * isum_mins);
}
*s = sum;
#elif defined __AVX2__
const __m256i m4 = _mm256_set1_epi8(0xF);
const __m256i m2 = _mm256_set1_epi8(3);
const __m256i m32s = _mm256_set1_epi8(32);
__m256 acc = _mm256_setzero_ps();
for (int i = 0; i < nb; ++i) {
ggml : move FP16 <-> FP32 code to ggml-impl.h (#3861) * ggml : move FP16 <-> FP32 stuff to ggml-impl.h ggml-ci * tests : fix ARM build * ggml : explicitly initialize deprecated type traits * ggml : add math.h to ggml-impl.h * ggml : remove duplicate static assert macros * ggml : prefix lookup tables with ggml_ ggml-ci * ggml-impl : move extern "C" to start of file
2023-10-30 17:19:15 +00:00
const float d = y[i].d * GGML_FP16_TO_FP32(x[i].d);
ggml : add SOTA 2,3,4,5,6 bit k-quantizations (#1684) * Starting to add k-quantization to ggml I think it is better to have quantization separate from ggml. For now just adding the k-quants there, but it would be better to also factor out the existing ggml quantizations. * Adding Q3_K and Q8_K (de)-quantization * Q3_K now working on CUDA and AVX2/scalar CUDA is not ideal - ~50% slower than Q4_0 for single token prediction, about the same in batch mode (perplexity). CPU single token is ~55 ms (on Ryzen 7950X). * Some improvement for Q3_K on CUDA It is now ~22.5 ms/token on my GPU, so ~30% slower than Q4_0. * Some more CUDA optimizations for Q3_K Single token is now 20.5 ms/token (~20% slower than Q4_0). Perplexity is on par with Q4_0. * Adding Q4_K - scalar, AVX2, CUDA Performance is the same or perhaps very slightly better than Q4_0 on the CPU. On the GPU, single token prediction is ~10% better than Q4_0, batch mode (perplexity is about the same). * Adding Q6_K - scalar, AVX2, CUDA Performance is ~40% lower compared to Q4_K on the CPU. This is to be expected, considering that we are memory bound on the CPU and the 6-bit model is ~44% larger than the 4-bit. On the GPU, single token prediction is ~6% lower than Q4_0, batch mode (perplexity) is even closer (but still slower). * Adding Q5_K - scalar, AVX2, CUDA Performance is ~20% lower compared to Q4_K on the CPU. This is to be expected, considering that we are memory bound on the CPU and the 5-bit model is ~22% larger than the 4-bit. On the GPU, single token prediction is about the same as Q4_0 for both, single token and batch prediction. * Per convention, all QX_K quantizations use Q5_K for output.weight * Adding quantization mixes * Quantization mixes: didn't quite get what I wanted in the last commit * Q4_K dot product for ARM_NEON * Q6_K dot product for ARM_NEON * Q5_K dot product for ARM_NEON * Adding Q3_K dot for ARM_NEON It is 22% slower than Q4_K, despite the smaller model size. On x86_64, where we are memory bound, the Q3_K model is quite a bit faster than Q4_K. * A very slightly faster ARM_NEON Q3_K dot * Adding Q2_K - just CUDA for now Token prediction is pretty good - about 15.5 ms on a RTX 4080. Perplexity is about the same as Q4_K. * Adding scalar and AVX2 Q2_K dot * Adding ARM_NEON Q2_K dot About the same performance as Q4_K. * A slightly faster ARM_NEON Q2_K dot Single token prediction is now ~36 ms on M2 Max. The code is much simpler too. * Fixed bug in Q2_K CUDA dot product kernel Stranegly enough, for the few prompts I tried with the 7B model the responses looked perfectly reasonable. Only realized something is not quite right when I tried the larger models and started getting nonse back. In any case, Q2_K single token evaluation time on an RTX 4080 in a Ryzen7950X box iusing CUDA and model fully loaded on the GPU are ~15.5 ms for 7B, ~25.4 ms for 13B, and ~55.8 ms for 30B. The max number of layers that fit in VRAM for The 65B is 32. With that, we get ~330 ms per token, which is not that much faster than just running on the CPU (~470 ms per token). * Don't print zeros/NaNs when no count histogram has been collected * A 10% faster CUDA vector dot kernel for Q3_K Q3_K is now running at ~18.5 ms / token on CUDA, so the gap to Q4_0 is only 10%. It seems memory acccess pattern is more important for performance than the amount of computation the kernel does. * A slightly daster Q4_K AVX2 dot product For perplexity, where we are less memory bound, time per pass drops by ~5%. Barely measurable difference for single token prediction. * A slightly faster ARM_NEON A4_K dot product * Minor * Fix quantization error test We cannot possibly be expecting rmse < 0.002 for 2- and 3-bit quantization variants. * Fix docker build I have been sloppy with vector reinterpret casts on ARM_NEON. It seems clang is very forgiving in that regard. * Added forgotten ggml.o dependence on k_quants.h to the Makefile * Had unintentionally committed the Makefile with -Ofast enabled * ggml : rename k_quants -> ggml-quants-k, use lowercase in code --------- Co-authored-by: Iwan Kawrakow <iwan.kawrakow@gmail.com> Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
2023-06-05 19:56:18 +00:00
const uint8_t * restrict q4 = x[i].ql;
const uint8_t * restrict qh = x[i].qh;
const int8_t * restrict q8 = y[i].qs;
const __m128i scales = _mm_loadu_si128((const __m128i*)x[i].scales);
__m256i sumi = _mm256_setzero_si256();
int is = 0;
for (int j = 0; j < QK_K/128; ++j) {
const __m128i scale_0 = _mm_shuffle_epi8(scales, get_scale_shuffle(is + 0));
const __m128i scale_1 = _mm_shuffle_epi8(scales, get_scale_shuffle(is + 1));
const __m128i scale_2 = _mm_shuffle_epi8(scales, get_scale_shuffle(is + 2));
const __m128i scale_3 = _mm_shuffle_epi8(scales, get_scale_shuffle(is + 3));
is += 4;
const __m256i q4bits1 = _mm256_loadu_si256((const __m256i*)q4); q4 += 32;
const __m256i q4bits2 = _mm256_loadu_si256((const __m256i*)q4); q4 += 32;
const __m256i q4bitsH = _mm256_loadu_si256((const __m256i*)qh); qh += 32;
const __m256i q4h_0 = _mm256_slli_epi16(_mm256_and_si256(q4bitsH, m2), 4);
const __m256i q4h_1 = _mm256_slli_epi16(_mm256_and_si256(_mm256_srli_epi16(q4bitsH, 2), m2), 4);
const __m256i q4h_2 = _mm256_slli_epi16(_mm256_and_si256(_mm256_srli_epi16(q4bitsH, 4), m2), 4);
const __m256i q4h_3 = _mm256_slli_epi16(_mm256_and_si256(_mm256_srli_epi16(q4bitsH, 6), m2), 4);
const __m256i q4_0 = _mm256_or_si256(_mm256_and_si256(q4bits1, m4), q4h_0);
const __m256i q4_1 = _mm256_or_si256(_mm256_and_si256(q4bits2, m4), q4h_1);
const __m256i q4_2 = _mm256_or_si256(_mm256_and_si256(_mm256_srli_epi16(q4bits1, 4), m4), q4h_2);
const __m256i q4_3 = _mm256_or_si256(_mm256_and_si256(_mm256_srli_epi16(q4bits2, 4), m4), q4h_3);
const __m256i q8_0 = _mm256_loadu_si256((const __m256i*)q8); q8 += 32;
const __m256i q8_1 = _mm256_loadu_si256((const __m256i*)q8); q8 += 32;
const __m256i q8_2 = _mm256_loadu_si256((const __m256i*)q8); q8 += 32;
const __m256i q8_3 = _mm256_loadu_si256((const __m256i*)q8); q8 += 32;
__m256i q8s_0 = _mm256_maddubs_epi16(m32s, q8_0);
__m256i q8s_1 = _mm256_maddubs_epi16(m32s, q8_1);
__m256i q8s_2 = _mm256_maddubs_epi16(m32s, q8_2);
__m256i q8s_3 = _mm256_maddubs_epi16(m32s, q8_3);
__m256i p16_0 = _mm256_maddubs_epi16(q4_0, q8_0);
__m256i p16_1 = _mm256_maddubs_epi16(q4_1, q8_1);
__m256i p16_2 = _mm256_maddubs_epi16(q4_2, q8_2);
__m256i p16_3 = _mm256_maddubs_epi16(q4_3, q8_3);
p16_0 = _mm256_sub_epi16(p16_0, q8s_0);
p16_1 = _mm256_sub_epi16(p16_1, q8s_1);
p16_2 = _mm256_sub_epi16(p16_2, q8s_2);
p16_3 = _mm256_sub_epi16(p16_3, q8s_3);
p16_0 = _mm256_madd_epi16(_mm256_cvtepi8_epi16(scale_0), p16_0);
p16_1 = _mm256_madd_epi16(_mm256_cvtepi8_epi16(scale_1), p16_1);
p16_2 = _mm256_madd_epi16(_mm256_cvtepi8_epi16(scale_2), p16_2);
p16_3 = _mm256_madd_epi16(_mm256_cvtepi8_epi16(scale_3), p16_3);
sumi = _mm256_add_epi32(sumi, _mm256_add_epi32(p16_0, p16_1));
sumi = _mm256_add_epi32(sumi, _mm256_add_epi32(p16_2, p16_3));
}
acc = _mm256_fmadd_ps(_mm256_broadcast_ss(&d), _mm256_cvtepi32_ps(sumi), acc);
}
*s = hsum_float_8(acc);
k-quants : add AVX support to dot functions (#1916) * k_quants : add AVX support * k_quants : apply review comments
2023-06-26 16:46:07 +00:00
#elif defined __AVX__
const __m128i m4 = _mm_set1_epi8(0xF);
const __m128i m3 = _mm_set1_epi8(3);
const __m128i m32s = _mm_set1_epi8(32);
const __m128i m2 = _mm_set1_epi8(2);
__m256 acc = _mm256_setzero_ps();
for (int i = 0; i < nb; ++i) {
ggml : move FP16 <-> FP32 code to ggml-impl.h (#3861) * ggml : move FP16 <-> FP32 stuff to ggml-impl.h ggml-ci * tests : fix ARM build * ggml : explicitly initialize deprecated type traits * ggml : add math.h to ggml-impl.h * ggml : remove duplicate static assert macros * ggml : prefix lookup tables with ggml_ ggml-ci * ggml-impl : move extern "C" to start of file
2023-10-30 17:19:15 +00:00
const float d = y[i].d * GGML_FP16_TO_FP32(x[i].d);
k-quants : add AVX support to dot functions (#1916) * k_quants : add AVX support * k_quants : apply review comments
2023-06-26 16:46:07 +00:00
const uint8_t * restrict q4 = x[i].ql;
const uint8_t * restrict qh = x[i].qh;
const int8_t * restrict q8 = y[i].qs;
const __m128i scales = _mm_loadu_si128((const __m128i*)x[i].scales);
__m128i sumi_0 = _mm_setzero_si128();
__m128i sumi_1 = _mm_setzero_si128();
__m128i shuffle = _mm_set_epi64x(0x0101010101010101, 0x0000000000000000);
for (int j = 0; j < QK_K/128; ++j) {
const __m128i q4bitsH_0 = _mm_loadu_si128((const __m128i*)qh); qh += 16;
const __m128i q4bitsH_1 = _mm_loadu_si128((const __m128i*)qh); qh += 16;
const __m128i q4h_0 = _mm_slli_epi16(_mm_and_si128(q4bitsH_0, m3), 4);
const __m128i q4h_1 = _mm_slli_epi16(_mm_and_si128(q4bitsH_1, m3), 4);
const __m128i q4h_2 = _mm_slli_epi16(_mm_and_si128(_mm_srli_epi16(q4bitsH_0, 2), m3), 4);
const __m128i q4h_3 = _mm_slli_epi16(_mm_and_si128(_mm_srli_epi16(q4bitsH_1, 2), m3), 4);
const __m128i q4h_4 = _mm_slli_epi16(_mm_and_si128(_mm_srli_epi16(q4bitsH_0, 4), m3), 4);
const __m128i q4h_5 = _mm_slli_epi16(_mm_and_si128(_mm_srli_epi16(q4bitsH_1, 4), m3), 4);
const __m128i q4h_6 = _mm_slli_epi16(_mm_and_si128(_mm_srli_epi16(q4bitsH_0, 6), m3), 4);
const __m128i q4h_7 = _mm_slli_epi16(_mm_and_si128(_mm_srli_epi16(q4bitsH_1, 6), m3), 4);
const __m128i q4bits1_0 = _mm_loadu_si128((const __m128i*)q4); q4 += 16;
const __m128i q4bits1_1 = _mm_loadu_si128((const __m128i*)q4); q4 += 16;
const __m128i q4bits2_0 = _mm_loadu_si128((const __m128i*)q4); q4 += 16;
const __m128i q4bits2_1 = _mm_loadu_si128((const __m128i*)q4); q4 += 16;
const __m128i q4_0 = _mm_or_si128(_mm_and_si128(q4bits1_0, m4), q4h_0);
const __m128i q4_1 = _mm_or_si128(_mm_and_si128(q4bits1_1, m4), q4h_1);
const __m128i q4_2 = _mm_or_si128(_mm_and_si128(q4bits2_0, m4), q4h_2);
const __m128i q4_3 = _mm_or_si128(_mm_and_si128(q4bits2_1, m4), q4h_3);
const __m128i q4_4 = _mm_or_si128(_mm_and_si128(_mm_srli_epi16(q4bits1_0, 4), m4), q4h_4);
const __m128i q4_5 = _mm_or_si128(_mm_and_si128(_mm_srli_epi16(q4bits1_1, 4), m4), q4h_5);
const __m128i q4_6 = _mm_or_si128(_mm_and_si128(_mm_srli_epi16(q4bits2_0, 4), m4), q4h_6);
const __m128i q4_7 = _mm_or_si128(_mm_and_si128(_mm_srli_epi16(q4bits2_1, 4), m4), q4h_7);
const __m128i q8_0 = _mm_loadu_si128((const __m128i*)q8); q8 += 16;
const __m128i q8_1 = _mm_loadu_si128((const __m128i*)q8); q8 += 16;
const __m128i q8_2 = _mm_loadu_si128((const __m128i*)q8); q8 += 16;
const __m128i q8_3 = _mm_loadu_si128((const __m128i*)q8); q8 += 16;
const __m128i q8_4 = _mm_loadu_si128((const __m128i*)q8); q8 += 16;
const __m128i q8_5 = _mm_loadu_si128((const __m128i*)q8); q8 += 16;
const __m128i q8_6 = _mm_loadu_si128((const __m128i*)q8); q8 += 16;
const __m128i q8_7 = _mm_loadu_si128((const __m128i*)q8); q8 += 16;
__m128i q8s_0 = _mm_maddubs_epi16(m32s, q8_0);
__m128i q8s_1 = _mm_maddubs_epi16(m32s, q8_1);
__m128i q8s_2 = _mm_maddubs_epi16(m32s, q8_2);
__m128i q8s_3 = _mm_maddubs_epi16(m32s, q8_3);
__m128i q8s_4 = _mm_maddubs_epi16(m32s, q8_4);
__m128i q8s_5 = _mm_maddubs_epi16(m32s, q8_5);
__m128i q8s_6 = _mm_maddubs_epi16(m32s, q8_6);
__m128i q8s_7 = _mm_maddubs_epi16(m32s, q8_7);
__m128i p16_0 = _mm_maddubs_epi16(q4_0, q8_0);
__m128i p16_1 = _mm_maddubs_epi16(q4_1, q8_1);
__m128i p16_2 = _mm_maddubs_epi16(q4_2, q8_2);
__m128i p16_3 = _mm_maddubs_epi16(q4_3, q8_3);
__m128i p16_4 = _mm_maddubs_epi16(q4_4, q8_4);
__m128i p16_5 = _mm_maddubs_epi16(q4_5, q8_5);
__m128i p16_6 = _mm_maddubs_epi16(q4_6, q8_6);
__m128i p16_7 = _mm_maddubs_epi16(q4_7, q8_7);
p16_0 = _mm_sub_epi16(p16_0, q8s_0);
p16_1 = _mm_sub_epi16(p16_1, q8s_1);
p16_2 = _mm_sub_epi16(p16_2, q8s_2);
p16_3 = _mm_sub_epi16(p16_3, q8s_3);
p16_4 = _mm_sub_epi16(p16_4, q8s_4);
p16_5 = _mm_sub_epi16(p16_5, q8s_5);
p16_6 = _mm_sub_epi16(p16_6, q8s_6);
p16_7 = _mm_sub_epi16(p16_7, q8s_7);
const __m128i scale_0 = _mm_shuffle_epi8(scales, shuffle);
shuffle = _mm_add_epi8(shuffle, m2);
const __m128i scale_1 = _mm_shuffle_epi8(scales, shuffle);
shuffle = _mm_add_epi8(shuffle, m2);
const __m128i scale_2 = _mm_shuffle_epi8(scales, shuffle);
shuffle = _mm_add_epi8(shuffle, m2);
const __m128i scale_3 = _mm_shuffle_epi8(scales, shuffle);
shuffle = _mm_add_epi8(shuffle, m2);
p16_0 = _mm_madd_epi16(_mm_cvtepi8_epi16(scale_0), p16_0);
p16_1 = _mm_madd_epi16(_mm_cvtepi8_epi16(_mm_unpackhi_epi64(scale_0, scale_0)), p16_1);
p16_2 = _mm_madd_epi16(_mm_cvtepi8_epi16(scale_1), p16_2);
p16_3 = _mm_madd_epi16(_mm_cvtepi8_epi16(_mm_unpackhi_epi64(scale_1, scale_1)), p16_3);
p16_4 = _mm_madd_epi16(_mm_cvtepi8_epi16(scale_2), p16_4);
p16_5 = _mm_madd_epi16(_mm_cvtepi8_epi16(_mm_unpackhi_epi64(scale_2, scale_2)), p16_5);
p16_6 = _mm_madd_epi16(_mm_cvtepi8_epi16(scale_3), p16_6);
p16_7 = _mm_madd_epi16(_mm_cvtepi8_epi16(_mm_unpackhi_epi64(scale_3, scale_3)), p16_7);
sumi_0 = _mm_add_epi32(sumi_0, _mm_add_epi32(p16_0, p16_2));
sumi_1 = _mm_add_epi32(sumi_1, _mm_add_epi32(p16_1, p16_3));
sumi_0 = _mm_add_epi32(sumi_0, _mm_add_epi32(p16_4, p16_6));
sumi_1 = _mm_add_epi32(sumi_1, _mm_add_epi32(p16_5, p16_7));
}
ggml : workaround for missing _mm256_setr_m128i in GCC < 8 in k_quants.c (#2405)
2023-07-28 18:17:45 +00:00
__m256i sumi = MM256_SET_M128I(sumi_1, sumi_0);
k-quants : add AVX support to dot functions (#1916) * k_quants : add AVX support * k_quants : apply review comments
2023-06-26 16:46:07 +00:00
acc = _mm256_add_ps(_mm256_mul_ps(_mm256_broadcast_ss(&d), _mm256_cvtepi32_ps(sumi)), acc);
}
*s = hsum_float_8(acc);
ggml : add RISC-V Vector Support for K-Quants and improved the existing intrinsics (#3453) * Added RVV intrinsics support for Q8 quantize row and also improved the existing dot product function for risc-v. The RVV intrinsics is added for the following quantize row functions quantize_row_q8_0 quantize_row_q8_1 The following dot product functions have also been optimized by using LMUL = 1/2 instead of LMUL = 1 ggml_vec_dot_q4_0_q8_0 ggml_vec_dot_q4_1_q8_1 ggml_vec_dot_q5_0_q8_0 ggml_vec_dot_q5_1_q8_1 And vector initialization in Q5 by temporary array is also replaced by the vid intrinsics Signed-off-by: Ahmad Tameem <ahmad.tameem@10xengineers.ai> * Added RVV intrinsics support for k_quants This adds RISC-V Vector intrinsics support for the following K_quants functions for both QKK = 256 and QKK = 64 ggml_vec_dot_q2_K_q8_K ggml_vec_dot_q3_K_q8_K ggml_vec_dot_q4_K_q8_K ggml_vec_dot_q5_K_q8_K ggml_vec_dot_q6_K_q8_K Signed-off-by: Ahmad Tameem <ahmad.tameem@10xengineers.ai> --------- Signed-off-by: Ahmad Tameem <ahmad.tameem@10xengineers.ai>
2023-10-03 18:38:19 +00:00
#elif defined __riscv_v_intrinsic
float sumf = 0;
for (int i = 0; i < nb; ++i) {
ggml : move FP16 <-> FP32 code to ggml-impl.h (#3861) * ggml : move FP16 <-> FP32 stuff to ggml-impl.h ggml-ci * tests : fix ARM build * ggml : explicitly initialize deprecated type traits * ggml : add math.h to ggml-impl.h * ggml : remove duplicate static assert macros * ggml : prefix lookup tables with ggml_ ggml-ci * ggml-impl : move extern "C" to start of file
2023-10-30 17:19:15 +00:00
const float d = GGML_FP16_TO_FP32(x[i].d) * y[i].d;
ggml : add RISC-V Vector Support for K-Quants and improved the existing intrinsics (#3453) * Added RVV intrinsics support for Q8 quantize row and also improved the existing dot product function for risc-v. The RVV intrinsics is added for the following quantize row functions quantize_row_q8_0 quantize_row_q8_1 The following dot product functions have also been optimized by using LMUL = 1/2 instead of LMUL = 1 ggml_vec_dot_q4_0_q8_0 ggml_vec_dot_q4_1_q8_1 ggml_vec_dot_q5_0_q8_0 ggml_vec_dot_q5_1_q8_1 And vector initialization in Q5 by temporary array is also replaced by the vid intrinsics Signed-off-by: Ahmad Tameem <ahmad.tameem@10xengineers.ai> * Added RVV intrinsics support for k_quants This adds RISC-V Vector intrinsics support for the following K_quants functions for both QKK = 256 and QKK = 64 ggml_vec_dot_q2_K_q8_K ggml_vec_dot_q3_K_q8_K ggml_vec_dot_q4_K_q8_K ggml_vec_dot_q5_K_q8_K ggml_vec_dot_q6_K_q8_K Signed-off-by: Ahmad Tameem <ahmad.tameem@10xengineers.ai> --------- Signed-off-by: Ahmad Tameem <ahmad.tameem@10xengineers.ai>
2023-10-03 18:38:19 +00:00
const uint8_t * restrict q6 = x[i].ql;
const uint8_t * restrict qh = x[i].qh;
const int8_t * restrict q8 = y[i].qs;
const int8_t * restrict scale = x[i].scales;
size_t vl;
vint32m1_t vzero = __riscv_vmv_v_x_i32m1(0, 1);
int sum_t = 0;
int is = 0;
for (int j = 0; j < QK_K/128; ++j) {
vl = 32;
// load qh
vuint8m1_t qh_x = __riscv_vle8_v_u8m1(qh, vl);
// load Q6
vuint8m1_t q6_0 = __riscv_vle8_v_u8m1(q6, vl);
vuint8m1_t q6_1 = __riscv_vle8_v_u8m1(q6+32, vl);
vuint8m1_t q6a_0 = __riscv_vand_vx_u8m1(q6_0, 0x0F, vl);
vuint8m1_t q6a_1 = __riscv_vand_vx_u8m1(q6_1, 0x0F, vl);
vuint8m1_t q6s_0 = __riscv_vsrl_vx_u8m1(q6_0, 0x04, vl);
vuint8m1_t q6s_1 = __riscv_vsrl_vx_u8m1(q6_1, 0x04, vl);
vuint8m1_t qh_0 = __riscv_vand_vx_u8m1(qh_x, 0x03, vl);
vuint8m1_t qh_1 = __riscv_vand_vx_u8m1(__riscv_vsrl_vx_u8m1(qh_x, 0x2, vl), 0x03 , vl);
vuint8m1_t qh_2 = __riscv_vand_vx_u8m1(__riscv_vsrl_vx_u8m1(qh_x, 0x4, vl), 0x03 , vl);
vuint8m1_t qh_3 = __riscv_vand_vx_u8m1(__riscv_vsrl_vx_u8m1(qh_x, 0x6, vl), 0x03 , vl);
vuint8m1_t qhi_0 = __riscv_vor_vv_u8m1(q6a_0, __riscv_vsll_vx_u8m1(qh_0, 0x04, vl), vl);
vuint8m1_t qhi_1 = __riscv_vor_vv_u8m1(q6a_1, __riscv_vsll_vx_u8m1(qh_1, 0x04, vl), vl);
vuint8m1_t qhi_2 = __riscv_vor_vv_u8m1(q6s_0, __riscv_vsll_vx_u8m1(qh_2, 0x04, vl), vl);
vuint8m1_t qhi_3 = __riscv_vor_vv_u8m1(q6s_1, __riscv_vsll_vx_u8m1(qh_3, 0x04, vl), vl);
vint8m1_t a_0 = __riscv_vsub_vx_i8m1(__riscv_vreinterpret_v_u8m1_i8m1(qhi_0), 32, vl);
vint8m1_t a_1 = __riscv_vsub_vx_i8m1(__riscv_vreinterpret_v_u8m1_i8m1(qhi_1), 32, vl);
vint8m1_t a_2 = __riscv_vsub_vx_i8m1(__riscv_vreinterpret_v_u8m1_i8m1(qhi_2), 32, vl);
vint8m1_t a_3 = __riscv_vsub_vx_i8m1(__riscv_vreinterpret_v_u8m1_i8m1(qhi_3), 32, vl);
// load Q8 and take product
vint16m2_t va_q_0 = __riscv_vwmul_vv_i16m2(a_0, __riscv_vle8_v_i8m1(q8, vl), vl);
vint16m2_t va_q_1 = __riscv_vwmul_vv_i16m2(a_1, __riscv_vle8_v_i8m1(q8+32, vl), vl);
vint16m2_t va_q_2 = __riscv_vwmul_vv_i16m2(a_2, __riscv_vle8_v_i8m1(q8+64, vl), vl);
vint16m2_t va_q_3 = __riscv_vwmul_vv_i16m2(a_3, __riscv_vle8_v_i8m1(q8+96, vl), vl);
vl = 16;
vint32m2_t vaux_0 = __riscv_vwmul_vx_i32m2(__riscv_vget_v_i16m2_i16m1(va_q_0, 0), scale[is+0], vl);
vint32m2_t vaux_1 = __riscv_vwmul_vx_i32m2(__riscv_vget_v_i16m2_i16m1(va_q_0, 1), scale[is+1], vl);
vint32m2_t vaux_2 = __riscv_vwmul_vx_i32m2(__riscv_vget_v_i16m2_i16m1(va_q_1, 0), scale[is+2], vl);
vint32m2_t vaux_3 = __riscv_vwmul_vx_i32m2(__riscv_vget_v_i16m2_i16m1(va_q_1, 1), scale[is+3], vl);
vint32m2_t vaux_4 = __riscv_vwmul_vx_i32m2(__riscv_vget_v_i16m2_i16m1(va_q_2, 0), scale[is+4], vl);
vint32m2_t vaux_5 = __riscv_vwmul_vx_i32m2(__riscv_vget_v_i16m2_i16m1(va_q_2, 1), scale[is+5], vl);
vint32m2_t vaux_6 = __riscv_vwmul_vx_i32m2(__riscv_vget_v_i16m2_i16m1(va_q_3, 0), scale[is+6], vl);
vint32m2_t vaux_7 = __riscv_vwmul_vx_i32m2(__riscv_vget_v_i16m2_i16m1(va_q_3, 1), scale[is+7], vl);
vint32m1_t isum0 = __riscv_vredsum_vs_i32m2_i32m1(__riscv_vadd_vv_i32m2(vaux_0, vaux_1, vl), vzero, vl);
vint32m1_t isum1 = __riscv_vredsum_vs_i32m2_i32m1(__riscv_vadd_vv_i32m2(vaux_2, vaux_3, vl), isum0, vl);
vint32m1_t isum2 = __riscv_vredsum_vs_i32m2_i32m1(__riscv_vadd_vv_i32m2(vaux_4, vaux_5, vl), isum1, vl);
vint32m1_t isum3 = __riscv_vredsum_vs_i32m2_i32m1(__riscv_vadd_vv_i32m2(vaux_6, vaux_7, vl), isum2, vl);
sum_t += __riscv_vmv_x_s_i32m1_i32(isum3);
q6 += 64; qh += 32; q8 += 128; is=8;
}
sumf += d * sum_t;
}
*s = sumf;
ggml : add SOTA 2,3,4,5,6 bit k-quantizations (#1684) * Starting to add k-quantization to ggml I think it is better to have quantization separate from ggml. For now just adding the k-quants there, but it would be better to also factor out the existing ggml quantizations. * Adding Q3_K and Q8_K (de)-quantization * Q3_K now working on CUDA and AVX2/scalar CUDA is not ideal - ~50% slower than Q4_0 for single token prediction, about the same in batch mode (perplexity). CPU single token is ~55 ms (on Ryzen 7950X). * Some improvement for Q3_K on CUDA It is now ~22.5 ms/token on my GPU, so ~30% slower than Q4_0. * Some more CUDA optimizations for Q3_K Single token is now 20.5 ms/token (~20% slower than Q4_0). Perplexity is on par with Q4_0. * Adding Q4_K - scalar, AVX2, CUDA Performance is the same or perhaps very slightly better than Q4_0 on the CPU. On the GPU, single token prediction is ~10% better than Q4_0, batch mode (perplexity is about the same). * Adding Q6_K - scalar, AVX2, CUDA Performance is ~40% lower compared to Q4_K on the CPU. This is to be expected, considering that we are memory bound on the CPU and the 6-bit model is ~44% larger than the 4-bit. On the GPU, single token prediction is ~6% lower than Q4_0, batch mode (perplexity) is even closer (but still slower). * Adding Q5_K - scalar, AVX2, CUDA Performance is ~20% lower compared to Q4_K on the CPU. This is to be expected, considering that we are memory bound on the CPU and the 5-bit model is ~22% larger than the 4-bit. On the GPU, single token prediction is about the same as Q4_0 for both, single token and batch prediction. * Per convention, all QX_K quantizations use Q5_K for output.weight * Adding quantization mixes * Quantization mixes: didn't quite get what I wanted in the last commit * Q4_K dot product for ARM_NEON * Q6_K dot product for ARM_NEON * Q5_K dot product for ARM_NEON * Adding Q3_K dot for ARM_NEON It is 22% slower than Q4_K, despite the smaller model size. On x86_64, where we are memory bound, the Q3_K model is quite a bit faster than Q4_K. * A very slightly faster ARM_NEON Q3_K dot * Adding Q2_K - just CUDA for now Token prediction is pretty good - about 15.5 ms on a RTX 4080. Perplexity is about the same as Q4_K. * Adding scalar and AVX2 Q2_K dot * Adding ARM_NEON Q2_K dot About the same performance as Q4_K. * A slightly faster ARM_NEON Q2_K dot Single token prediction is now ~36 ms on M2 Max. The code is much simpler too. * Fixed bug in Q2_K CUDA dot product kernel Stranegly enough, for the few prompts I tried with the 7B model the responses looked perfectly reasonable. Only realized something is not quite right when I tried the larger models and started getting nonse back. In any case, Q2_K single token evaluation time on an RTX 4080 in a Ryzen7950X box iusing CUDA and model fully loaded on the GPU are ~15.5 ms for 7B, ~25.4 ms for 13B, and ~55.8 ms for 30B. The max number of layers that fit in VRAM for The 65B is 32. With that, we get ~330 ms per token, which is not that much faster than just running on the CPU (~470 ms per token). * Don't print zeros/NaNs when no count histogram has been collected * A 10% faster CUDA vector dot kernel for Q3_K Q3_K is now running at ~18.5 ms / token on CUDA, so the gap to Q4_0 is only 10%. It seems memory acccess pattern is more important for performance than the amount of computation the kernel does. * A slightly daster Q4_K AVX2 dot product For perplexity, where we are less memory bound, time per pass drops by ~5%. Barely measurable difference for single token prediction. * A slightly faster ARM_NEON A4_K dot product * Minor * Fix quantization error test We cannot possibly be expecting rmse < 0.002 for 2- and 3-bit quantization variants. * Fix docker build I have been sloppy with vector reinterpret casts on ARM_NEON. It seems clang is very forgiving in that regard. * Added forgotten ggml.o dependence on k_quants.h to the Makefile * Had unintentionally committed the Makefile with -Ofast enabled * ggml : rename k_quants -> ggml-quants-k, use lowercase in code --------- Co-authored-by: Iwan Kawrakow <iwan.kawrakow@gmail.com> Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
2023-06-05 19:56:18 +00:00
#else
int8_t aux8[QK_K];
int16_t aux16[8];
float sums [8];
int32_t aux32[8];
memset(sums, 0, 8*sizeof(float));
float sumf = 0;
for (int i = 0; i < nb; ++i) {
const uint8_t * restrict q4 = x[i].ql;
const uint8_t * restrict qh = x[i].qh;
const int8_t * restrict q8 = y[i].qs;
memset(aux32, 0, 8*sizeof(int32_t));
int8_t * restrict a = aux8;
for (int j = 0; j < QK_K; j += 128) {
for (int l = 0; l < 32; ++l) {
a[l + 0] = (int8_t)((q4[l + 0] & 0xF) | (((qh[l] >> 0) & 3) << 4)) - 32;
a[l + 32] = (int8_t)((q4[l + 32] & 0xF) | (((qh[l] >> 2) & 3) << 4)) - 32;
a[l + 64] = (int8_t)((q4[l + 0] >> 4) | (((qh[l] >> 4) & 3) << 4)) - 32;
a[l + 96] = (int8_t)((q4[l + 32] >> 4) | (((qh[l] >> 6) & 3) << 4)) - 32;
}
a += 128;
q4 += 64;
qh += 32;
}
a = aux8;
int is = 0;
for (int j = 0; j < QK_K/16; ++j) {
int scale = x[i].scales[is++];
for (int l = 0; l < 8; ++l) aux16[l] = q8[l] * a[l];
for (int l = 0; l < 8; ++l) aux32[l] += scale * aux16[l];
q8 += 8; a += 8;
for (int l = 0; l < 8; ++l) aux16[l] = q8[l] * a[l];
for (int l = 0; l < 8; ++l) aux32[l] += scale * aux16[l];
q8 += 8; a += 8;
}
ggml : move FP16 <-> FP32 code to ggml-impl.h (#3861) * ggml : move FP16 <-> FP32 stuff to ggml-impl.h ggml-ci * tests : fix ARM build * ggml : explicitly initialize deprecated type traits * ggml : add math.h to ggml-impl.h * ggml : remove duplicate static assert macros * ggml : prefix lookup tables with ggml_ ggml-ci * ggml-impl : move extern "C" to start of file
2023-10-30 17:19:15 +00:00
const float d = GGML_FP16_TO_FP32(x[i].d) * y[i].d;
ggml : add SOTA 2,3,4,5,6 bit k-quantizations (#1684) * Starting to add k-quantization to ggml I think it is better to have quantization separate from ggml. For now just adding the k-quants there, but it would be better to also factor out the existing ggml quantizations. * Adding Q3_K and Q8_K (de)-quantization * Q3_K now working on CUDA and AVX2/scalar CUDA is not ideal - ~50% slower than Q4_0 for single token prediction, about the same in batch mode (perplexity). CPU single token is ~55 ms (on Ryzen 7950X). * Some improvement for Q3_K on CUDA It is now ~22.5 ms/token on my GPU, so ~30% slower than Q4_0. * Some more CUDA optimizations for Q3_K Single token is now 20.5 ms/token (~20% slower than Q4_0). Perplexity is on par with Q4_0. * Adding Q4_K - scalar, AVX2, CUDA Performance is the same or perhaps very slightly better than Q4_0 on the CPU. On the GPU, single token prediction is ~10% better than Q4_0, batch mode (perplexity is about the same). * Adding Q6_K - scalar, AVX2, CUDA Performance is ~40% lower compared to Q4_K on the CPU. This is to be expected, considering that we are memory bound on the CPU and the 6-bit model is ~44% larger than the 4-bit. On the GPU, single token prediction is ~6% lower than Q4_0, batch mode (perplexity) is even closer (but still slower). * Adding Q5_K - scalar, AVX2, CUDA Performance is ~20% lower compared to Q4_K on the CPU. This is to be expected, considering that we are memory bound on the CPU and the 5-bit model is ~22% larger than the 4-bit. On the GPU, single token prediction is about the same as Q4_0 for both, single token and batch prediction. * Per convention, all QX_K quantizations use Q5_K for output.weight * Adding quantization mixes * Quantization mixes: didn't quite get what I wanted in the last commit * Q4_K dot product for ARM_NEON * Q6_K dot product for ARM_NEON * Q5_K dot product for ARM_NEON * Adding Q3_K dot for ARM_NEON It is 22% slower than Q4_K, despite the smaller model size. On x86_64, where we are memory bound, the Q3_K model is quite a bit faster than Q4_K. * A very slightly faster ARM_NEON Q3_K dot * Adding Q2_K - just CUDA for now Token prediction is pretty good - about 15.5 ms on a RTX 4080. Perplexity is about the same as Q4_K. * Adding scalar and AVX2 Q2_K dot * Adding ARM_NEON Q2_K dot About the same performance as Q4_K. * A slightly faster ARM_NEON Q2_K dot Single token prediction is now ~36 ms on M2 Max. The code is much simpler too. * Fixed bug in Q2_K CUDA dot product kernel Stranegly enough, for the few prompts I tried with the 7B model the responses looked perfectly reasonable. Only realized something is not quite right when I tried the larger models and started getting nonse back. In any case, Q2_K single token evaluation time on an RTX 4080 in a Ryzen7950X box iusing CUDA and model fully loaded on the GPU are ~15.5 ms for 7B, ~25.4 ms for 13B, and ~55.8 ms for 30B. The max number of layers that fit in VRAM for The 65B is 32. With that, we get ~330 ms per token, which is not that much faster than just running on the CPU (~470 ms per token). * Don't print zeros/NaNs when no count histogram has been collected * A 10% faster CUDA vector dot kernel for Q3_K Q3_K is now running at ~18.5 ms / token on CUDA, so the gap to Q4_0 is only 10%. It seems memory acccess pattern is more important for performance than the amount of computation the kernel does. * A slightly daster Q4_K AVX2 dot product For perplexity, where we are less memory bound, time per pass drops by ~5%. Barely measurable difference for single token prediction. * A slightly faster ARM_NEON A4_K dot product * Minor * Fix quantization error test We cannot possibly be expecting rmse < 0.002 for 2- and 3-bit quantization variants. * Fix docker build I have been sloppy with vector reinterpret casts on ARM_NEON. It seems clang is very forgiving in that regard. * Added forgotten ggml.o dependence on k_quants.h to the Makefile * Had unintentionally committed the Makefile with -Ofast enabled * ggml : rename k_quants -> ggml-quants-k, use lowercase in code --------- Co-authored-by: Iwan Kawrakow <iwan.kawrakow@gmail.com> Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
2023-06-05 19:56:18 +00:00
for (int l = 0; l < 8; ++l) sums[l] += d * aux32[l];
}
for (int l = 0; l < 8; ++l) sumf += sums[l];
*s = sumf;
#endif
}
k-quants : support for super-block size of 64 (#2001) * k_quants: WIP super-blocks with 64 weights * k_quants: WIP super-blocks with 64 weights Q6_K scalar and AVX2 works * k_quants: WIP super-blocks with 64 weights Q4_K scalar and AVX2 works * k_quants: WIP super-blocks with 64 weights Q2_K scalar and AVX2 works. Q2_K is way too slow (it is actually slower than the scalar implementation) * k_quants: WIP super-blocks with 64 weights Q3_K scalar and AVX2 works. * k_quants: WIP super-blocks with 64 weights Q5_K scalar and AVX2 works, and with that all k_quants are done on AVX2 and scalar * k_quants: WIP super-blocks with 64 weights Q6_K working on CUDA. Cannot make it run quite as gast as with super-blocks with 256 weigths: 8% slower on 4080, 20% slower on the 1660 (but there we fit 1 less layer on the GPU because pf the larger model size), so some fraction of these 20% is due to that, * k_quants: WIP super-blocks with 64 weights Q4_K working on CUDA. ~10% slower on GTX-1660, 16% slower on 4080. * k_quants: WIP super-blocks with 64 weights Q2_K working on CUDA. ~3% slower on GTX-1660, 10% slower on 4080. * k_quants: WIP super-blocks with 64 weights Q3_K working on CUDA. * k_quants: WIP super-blocks with 64 weights Q5_K working on CUDA, and with this CUDA is done. * k_quants: WIP super-blocks with 64 weights Q6_K working on ARM_NEON * k_quants: WIP super-blocks with 64 weights Q4_K working on ARM_NEON, but quite a bit slower than 256 weights * k_quants: WIP super-blocks with 64 weights Q2_K working on ARM_NEON, but quite a bit slower than 256 weights * k_quants: WIP super-blocks with 64 weights Q3_K working on ARM_NEON, but quite a bit slower than 256 weights. * k_quants: WIP super-blocks with 64 weights Q5_K working on ARM_NEON, but quite a bit slower than 256 weights. With that, we have full support for ARM_NEON, although performance is not quite there. * k_quants: WIP super-blocks with 64 weights Slightly more efficient Q3_K and Q5_K * k_quants: WIP super-blocks with 64 weights Another small improvement for Q3_K and Q5_K on ARM_NEON * k_quants: WIP super-blocks with 64 weights Yet another speedup for Q5_K on ARM_NEON. We are now within 10% of the QK_K = 256 version. * k_quants: WIP super-blocks with 64 weights * We are able to pass preprocessor macros to the Metal compiler * Q6_K works and is actually slightly more efficient than the QK_K = 256 version (25.2 ms vs 25.8 ms) * k_quants: WIP super-blocks with 64 weights Q4_K works on Metal and is actually slightly faster than QK_K = 256 (21.95 ms vs 24.0 ms). * k_quants: WIP super-blocks with 64 weights Q2_K works on Metal and is very slightly faster than QK_K = 256 (23.8 ms vs 24.2 ms). * k_quants: WIP super-blocks with 64 weights Q3_K works on Metal and is slightly faster than QK_K = 256 (26.6 ms vs 28.3 ms). * k_quants: WIP super-blocks with 64 weights Q5_K works on Metal and is slightly faster than QK_K = 256 (23.7 ms vs 26.3 ms). * k_quants: call them _K, not _k, also on Metal * k_quants: correctly define QK_K in llama.cpp * Fixed bug in q4_K quantization added with the 64-block addition * Simplify via lambda * k_quants: swicth Q3_K to 4-bit scales when QK_K = 64 Otherwise there isn't much benefit from this quantization type. There is some very slight loss in accuracy, but we reduce size by ~7%. E.g., for OpenLLaMA-3B, Q3_K_S perplexity is 8.6131 with 8-bit scales and 8.6352 with 4-bit, while file size decreases from 1.53G to 1.44G. * k_quants: switch Q4_K to 4-bit scales when QK_K = 64 Here the loss in accuracy is greater than for Q3_K, but the Q4_K points still move further to the left on the perplexity vs size curve. * k_quants: forgot to add the Metal changes in last commit * k_quants: change Q5_K to be type 0 when QK_K = 64 Still needs AVX2 implementation * k_quants: AVX2 implementation for new 64-weight Q5_K * k_quants: 10% faster ARM_NEON Q5_K dot product * k_quants: fixed issue caused by merging with master --------- Co-authored-by: Iwan Kawrakow <iwan.kawrakow@gmail.com>
2023-06-26 16:43:07 +00:00
#else
void ggml_vec_dot_q6_K_q8_K(const int n, float * restrict s, const void * restrict vx, const void * restrict vy) {
assert(n % QK_K == 0);
const block_q6_K * restrict x = vx;
const block_q8_K * restrict y = vy;
const int nb = n / QK_K;
#ifdef __ARM_NEON
float sum = 0;
const uint8x16_t m4b = vdupq_n_u8(0xF);
const int8x16_t m32s = vdupq_n_s8(32);
k-quants : fix build on armv7 (android only) (#2920) * k-quants : fix build on armv7 * ggml : cleanup unused arm32 specific impl * k-quants : avoid some unused vzero / mzero define * ggml-alloc : use 4g for MEASURE_MAX_SIZE in 32-bit arm
2023-09-02 12:23:45 +00:00
const int32x4_t vzero = vdupq_n_s32(0);
k-quants : support for super-block size of 64 (#2001) * k_quants: WIP super-blocks with 64 weights * k_quants: WIP super-blocks with 64 weights Q6_K scalar and AVX2 works * k_quants: WIP super-blocks with 64 weights Q4_K scalar and AVX2 works * k_quants: WIP super-blocks with 64 weights Q2_K scalar and AVX2 works. Q2_K is way too slow (it is actually slower than the scalar implementation) * k_quants: WIP super-blocks with 64 weights Q3_K scalar and AVX2 works. * k_quants: WIP super-blocks with 64 weights Q5_K scalar and AVX2 works, and with that all k_quants are done on AVX2 and scalar * k_quants: WIP super-blocks with 64 weights Q6_K working on CUDA. Cannot make it run quite as gast as with super-blocks with 256 weigths: 8% slower on 4080, 20% slower on the 1660 (but there we fit 1 less layer on the GPU because pf the larger model size), so some fraction of these 20% is due to that, * k_quants: WIP super-blocks with 64 weights Q4_K working on CUDA. ~10% slower on GTX-1660, 16% slower on 4080. * k_quants: WIP super-blocks with 64 weights Q2_K working on CUDA. ~3% slower on GTX-1660, 10% slower on 4080. * k_quants: WIP super-blocks with 64 weights Q3_K working on CUDA. * k_quants: WIP super-blocks with 64 weights Q5_K working on CUDA, and with this CUDA is done. * k_quants: WIP super-blocks with 64 weights Q6_K working on ARM_NEON * k_quants: WIP super-blocks with 64 weights Q4_K working on ARM_NEON, but quite a bit slower than 256 weights * k_quants: WIP super-blocks with 64 weights Q2_K working on ARM_NEON, but quite a bit slower than 256 weights * k_quants: WIP super-blocks with 64 weights Q3_K working on ARM_NEON, but quite a bit slower than 256 weights. * k_quants: WIP super-blocks with 64 weights Q5_K working on ARM_NEON, but quite a bit slower than 256 weights. With that, we have full support for ARM_NEON, although performance is not quite there. * k_quants: WIP super-blocks with 64 weights Slightly more efficient Q3_K and Q5_K * k_quants: WIP super-blocks with 64 weights Another small improvement for Q3_K and Q5_K on ARM_NEON * k_quants: WIP super-blocks with 64 weights Yet another speedup for Q5_K on ARM_NEON. We are now within 10% of the QK_K = 256 version. * k_quants: WIP super-blocks with 64 weights * We are able to pass preprocessor macros to the Metal compiler * Q6_K works and is actually slightly more efficient than the QK_K = 256 version (25.2 ms vs 25.8 ms) * k_quants: WIP super-blocks with 64 weights Q4_K works on Metal and is actually slightly faster than QK_K = 256 (21.95 ms vs 24.0 ms). * k_quants: WIP super-blocks with 64 weights Q2_K works on Metal and is very slightly faster than QK_K = 256 (23.8 ms vs 24.2 ms). * k_quants: WIP super-blocks with 64 weights Q3_K works on Metal and is slightly faster than QK_K = 256 (26.6 ms vs 28.3 ms). * k_quants: WIP super-blocks with 64 weights Q5_K works on Metal and is slightly faster than QK_K = 256 (23.7 ms vs 26.3 ms). * k_quants: call them _K, not _k, also on Metal * k_quants: correctly define QK_K in llama.cpp * Fixed bug in q4_K quantization added with the 64-block addition * Simplify via lambda * k_quants: swicth Q3_K to 4-bit scales when QK_K = 64 Otherwise there isn't much benefit from this quantization type. There is some very slight loss in accuracy, but we reduce size by ~7%. E.g., for OpenLLaMA-3B, Q3_K_S perplexity is 8.6131 with 8-bit scales and 8.6352 with 4-bit, while file size decreases from 1.53G to 1.44G. * k_quants: switch Q4_K to 4-bit scales when QK_K = 64 Here the loss in accuracy is greater than for Q3_K, but the Q4_K points still move further to the left on the perplexity vs size curve. * k_quants: forgot to add the Metal changes in last commit * k_quants: change Q5_K to be type 0 when QK_K = 64 Still needs AVX2 implementation * k_quants: AVX2 implementation for new 64-weight Q5_K * k_quants: 10% faster ARM_NEON Q5_K dot product * k_quants: fixed issue caused by merging with master --------- Co-authored-by: Iwan Kawrakow <iwan.kawrakow@gmail.com>
2023-06-26 16:43:07 +00:00
const uint8x16_t mone = vdupq_n_u8(3);
ggml : sync (im2col, GPU conv, 32-bit arm compat) (#4060) ggml-ci
2023-11-13 14:55:52 +00:00
ggml_int8x16x4_t q6bytes;
ggml_uint8x16x4_t q6h;
k-quants : support for super-block size of 64 (#2001) * k_quants: WIP super-blocks with 64 weights * k_quants: WIP super-blocks with 64 weights Q6_K scalar and AVX2 works * k_quants: WIP super-blocks with 64 weights Q4_K scalar and AVX2 works * k_quants: WIP super-blocks with 64 weights Q2_K scalar and AVX2 works. Q2_K is way too slow (it is actually slower than the scalar implementation) * k_quants: WIP super-blocks with 64 weights Q3_K scalar and AVX2 works. * k_quants: WIP super-blocks with 64 weights Q5_K scalar and AVX2 works, and with that all k_quants are done on AVX2 and scalar * k_quants: WIP super-blocks with 64 weights Q6_K working on CUDA. Cannot make it run quite as gast as with super-blocks with 256 weigths: 8% slower on 4080, 20% slower on the 1660 (but there we fit 1 less layer on the GPU because pf the larger model size), so some fraction of these 20% is due to that, * k_quants: WIP super-blocks with 64 weights Q4_K working on CUDA. ~10% slower on GTX-1660, 16% slower on 4080. * k_quants: WIP super-blocks with 64 weights Q2_K working on CUDA. ~3% slower on GTX-1660, 10% slower on 4080. * k_quants: WIP super-blocks with 64 weights Q3_K working on CUDA. * k_quants: WIP super-blocks with 64 weights Q5_K working on CUDA, and with this CUDA is done. * k_quants: WIP super-blocks with 64 weights Q6_K working on ARM_NEON * k_quants: WIP super-blocks with 64 weights Q4_K working on ARM_NEON, but quite a bit slower than 256 weights * k_quants: WIP super-blocks with 64 weights Q2_K working on ARM_NEON, but quite a bit slower than 256 weights * k_quants: WIP super-blocks with 64 weights Q3_K working on ARM_NEON, but quite a bit slower than 256 weights. * k_quants: WIP super-blocks with 64 weights Q5_K working on ARM_NEON, but quite a bit slower than 256 weights. With that, we have full support for ARM_NEON, although performance is not quite there. * k_quants: WIP super-blocks with 64 weights Slightly more efficient Q3_K and Q5_K * k_quants: WIP super-blocks with 64 weights Another small improvement for Q3_K and Q5_K on ARM_NEON * k_quants: WIP super-blocks with 64 weights Yet another speedup for Q5_K on ARM_NEON. We are now within 10% of the QK_K = 256 version. * k_quants: WIP super-blocks with 64 weights * We are able to pass preprocessor macros to the Metal compiler * Q6_K works and is actually slightly more efficient than the QK_K = 256 version (25.2 ms vs 25.8 ms) * k_quants: WIP super-blocks with 64 weights Q4_K works on Metal and is actually slightly faster than QK_K = 256 (21.95 ms vs 24.0 ms). * k_quants: WIP super-blocks with 64 weights Q2_K works on Metal and is very slightly faster than QK_K = 256 (23.8 ms vs 24.2 ms). * k_quants: WIP super-blocks with 64 weights Q3_K works on Metal and is slightly faster than QK_K = 256 (26.6 ms vs 28.3 ms). * k_quants: WIP super-blocks with 64 weights Q5_K works on Metal and is slightly faster than QK_K = 256 (23.7 ms vs 26.3 ms). * k_quants: call them _K, not _k, also on Metal * k_quants: correctly define QK_K in llama.cpp * Fixed bug in q4_K quantization added with the 64-block addition * Simplify via lambda * k_quants: swicth Q3_K to 4-bit scales when QK_K = 64 Otherwise there isn't much benefit from this quantization type. There is some very slight loss in accuracy, but we reduce size by ~7%. E.g., for OpenLLaMA-3B, Q3_K_S perplexity is 8.6131 with 8-bit scales and 8.6352 with 4-bit, while file size decreases from 1.53G to 1.44G. * k_quants: switch Q4_K to 4-bit scales when QK_K = 64 Here the loss in accuracy is greater than for Q3_K, but the Q4_K points still move further to the left on the perplexity vs size curve. * k_quants: forgot to add the Metal changes in last commit * k_quants: change Q5_K to be type 0 when QK_K = 64 Still needs AVX2 implementation * k_quants: AVX2 implementation for new 64-weight Q5_K * k_quants: 10% faster ARM_NEON Q5_K dot product * k_quants: fixed issue caused by merging with master --------- Co-authored-by: Iwan Kawrakow <iwan.kawrakow@gmail.com>
2023-06-26 16:43:07 +00:00
for (int i = 0; i < nb; ++i) {
const float d_all = (float)x[i].d;
const uint8_t * restrict q6 = x[i].ql;
const uint8_t * restrict qh = x[i].qh;
const int8_t * restrict q8 = y[i].qs;
const int8_t * restrict scale = x[i].scales;
int32_t isum = 0;
ggml : sync (im2col, GPU conv, 32-bit arm compat) (#4060) ggml-ci
2023-11-13 14:55:52 +00:00
uint8x16_t qhbits = vld1q_u8(qh);
ggml_uint8x16x2_t q6bits = ggml_vld1q_u8_x2(q6);
ggml_int8x16x4_t q8bytes = ggml_vld1q_s8_x4(q8);
k-quants : support for super-block size of 64 (#2001) * k_quants: WIP super-blocks with 64 weights * k_quants: WIP super-blocks with 64 weights Q6_K scalar and AVX2 works * k_quants: WIP super-blocks with 64 weights Q4_K scalar and AVX2 works * k_quants: WIP super-blocks with 64 weights Q2_K scalar and AVX2 works. Q2_K is way too slow (it is actually slower than the scalar implementation) * k_quants: WIP super-blocks with 64 weights Q3_K scalar and AVX2 works. * k_quants: WIP super-blocks with 64 weights Q5_K scalar and AVX2 works, and with that all k_quants are done on AVX2 and scalar * k_quants: WIP super-blocks with 64 weights Q6_K working on CUDA. Cannot make it run quite as gast as with super-blocks with 256 weigths: 8% slower on 4080, 20% slower on the 1660 (but there we fit 1 less layer on the GPU because pf the larger model size), so some fraction of these 20% is due to that, * k_quants: WIP super-blocks with 64 weights Q4_K working on CUDA. ~10% slower on GTX-1660, 16% slower on 4080. * k_quants: WIP super-blocks with 64 weights Q2_K working on CUDA. ~3% slower on GTX-1660, 10% slower on 4080. * k_quants: WIP super-blocks with 64 weights Q3_K working on CUDA. * k_quants: WIP super-blocks with 64 weights Q5_K working on CUDA, and with this CUDA is done. * k_quants: WIP super-blocks with 64 weights Q6_K working on ARM_NEON * k_quants: WIP super-blocks with 64 weights Q4_K working on ARM_NEON, but quite a bit slower than 256 weights * k_quants: WIP super-blocks with 64 weights Q2_K working on ARM_NEON, but quite a bit slower than 256 weights * k_quants: WIP super-blocks with 64 weights Q3_K working on ARM_NEON, but quite a bit slower than 256 weights. * k_quants: WIP super-blocks with 64 weights Q5_K working on ARM_NEON, but quite a bit slower than 256 weights. With that, we have full support for ARM_NEON, although performance is not quite there. * k_quants: WIP super-blocks with 64 weights Slightly more efficient Q3_K and Q5_K * k_quants: WIP super-blocks with 64 weights Another small improvement for Q3_K and Q5_K on ARM_NEON * k_quants: WIP super-blocks with 64 weights Yet another speedup for Q5_K on ARM_NEON. We are now within 10% of the QK_K = 256 version. * k_quants: WIP super-blocks with 64 weights * We are able to pass preprocessor macros to the Metal compiler * Q6_K works and is actually slightly more efficient than the QK_K = 256 version (25.2 ms vs 25.8 ms) * k_quants: WIP super-blocks with 64 weights Q4_K works on Metal and is actually slightly faster than QK_K = 256 (21.95 ms vs 24.0 ms). * k_quants: WIP super-blocks with 64 weights Q2_K works on Metal and is very slightly faster than QK_K = 256 (23.8 ms vs 24.2 ms). * k_quants: WIP super-blocks with 64 weights Q3_K works on Metal and is slightly faster than QK_K = 256 (26.6 ms vs 28.3 ms). * k_quants: WIP super-blocks with 64 weights Q5_K works on Metal and is slightly faster than QK_K = 256 (23.7 ms vs 26.3 ms). * k_quants: call them _K, not _k, also on Metal * k_quants: correctly define QK_K in llama.cpp * Fixed bug in q4_K quantization added with the 64-block addition * Simplify via lambda * k_quants: swicth Q3_K to 4-bit scales when QK_K = 64 Otherwise there isn't much benefit from this quantization type. There is some very slight loss in accuracy, but we reduce size by ~7%. E.g., for OpenLLaMA-3B, Q3_K_S perplexity is 8.6131 with 8-bit scales and 8.6352 with 4-bit, while file size decreases from 1.53G to 1.44G. * k_quants: switch Q4_K to 4-bit scales when QK_K = 64 Here the loss in accuracy is greater than for Q3_K, but the Q4_K points still move further to the left on the perplexity vs size curve. * k_quants: forgot to add the Metal changes in last commit * k_quants: change Q5_K to be type 0 when QK_K = 64 Still needs AVX2 implementation * k_quants: AVX2 implementation for new 64-weight Q5_K * k_quants: 10% faster ARM_NEON Q5_K dot product * k_quants: fixed issue caused by merging with master --------- Co-authored-by: Iwan Kawrakow <iwan.kawrakow@gmail.com>
2023-06-26 16:43:07 +00:00
q6h.val[0] = vshlq_n_u8(vandq_u8(mone, qhbits), 4);
uint8x16_t shifted = vshrq_n_u8(qhbits, 2);
q6h.val[1] = vshlq_n_u8(vandq_u8(mone, shifted), 4);
shifted = vshrq_n_u8(qhbits, 4);
q6h.val[2] = vshlq_n_u8(vandq_u8(mone, shifted), 4);
shifted = vshrq_n_u8(qhbits, 6);
q6h.val[3] = vshlq_n_u8(vandq_u8(mone, shifted), 4);
q6bytes.val[0] = vsubq_s8(vreinterpretq_s8_u8(vorrq_u8(vandq_u8(q6bits.val[0], m4b), q6h.val[0])), m32s);
q6bytes.val[1] = vsubq_s8(vreinterpretq_s8_u8(vorrq_u8(vandq_u8(q6bits.val[1], m4b), q6h.val[1])), m32s);
q6bytes.val[2] = vsubq_s8(vreinterpretq_s8_u8(vorrq_u8(vshrq_n_u8(q6bits.val[0], 4), q6h.val[2])), m32s);
q6bytes.val[3] = vsubq_s8(vreinterpretq_s8_u8(vorrq_u8(vshrq_n_u8(q6bits.val[1], 4), q6h.val[3])), m32s);
ggml : add ggml_vdotq_s32 alias (#4715) ggml-ci
2023-12-31 09:43:31 +00:00
isum += vaddvq_s32(ggml_vdotq_s32(vzero, q6bytes.val[0], q8bytes.val[0])) * scale[0] +
vaddvq_s32(ggml_vdotq_s32(vzero, q6bytes.val[1], q8bytes.val[1])) * scale[1] +
vaddvq_s32(ggml_vdotq_s32(vzero, q6bytes.val[2], q8bytes.val[2])) * scale[2] +
vaddvq_s32(ggml_vdotq_s32(vzero, q6bytes.val[3], q8bytes.val[3])) * scale[3];
k-quants : support for super-block size of 64 (#2001) * k_quants: WIP super-blocks with 64 weights * k_quants: WIP super-blocks with 64 weights Q6_K scalar and AVX2 works * k_quants: WIP super-blocks with 64 weights Q4_K scalar and AVX2 works * k_quants: WIP super-blocks with 64 weights Q2_K scalar and AVX2 works. Q2_K is way too slow (it is actually slower than the scalar implementation) * k_quants: WIP super-blocks with 64 weights Q3_K scalar and AVX2 works. * k_quants: WIP super-blocks with 64 weights Q5_K scalar and AVX2 works, and with that all k_quants are done on AVX2 and scalar * k_quants: WIP super-blocks with 64 weights Q6_K working on CUDA. Cannot make it run quite as gast as with super-blocks with 256 weigths: 8% slower on 4080, 20% slower on the 1660 (but there we fit 1 less layer on the GPU because pf the larger model size), so some fraction of these 20% is due to that, * k_quants: WIP super-blocks with 64 weights Q4_K working on CUDA. ~10% slower on GTX-1660, 16% slower on 4080. * k_quants: WIP super-blocks with 64 weights Q2_K working on CUDA. ~3% slower on GTX-1660, 10% slower on 4080. * k_quants: WIP super-blocks with 64 weights Q3_K working on CUDA. * k_quants: WIP super-blocks with 64 weights Q5_K working on CUDA, and with this CUDA is done. * k_quants: WIP super-blocks with 64 weights Q6_K working on ARM_NEON * k_quants: WIP super-blocks with 64 weights Q4_K working on ARM_NEON, but quite a bit slower than 256 weights * k_quants: WIP super-blocks with 64 weights Q2_K working on ARM_NEON, but quite a bit slower than 256 weights * k_quants: WIP super-blocks with 64 weights Q3_K working on ARM_NEON, but quite a bit slower than 256 weights. * k_quants: WIP super-blocks with 64 weights Q5_K working on ARM_NEON, but quite a bit slower than 256 weights. With that, we have full support for ARM_NEON, although performance is not quite there. * k_quants: WIP super-blocks with 64 weights Slightly more efficient Q3_K and Q5_K * k_quants: WIP super-blocks with 64 weights Another small improvement for Q3_K and Q5_K on ARM_NEON * k_quants: WIP super-blocks with 64 weights Yet another speedup for Q5_K on ARM_NEON. We are now within 10% of the QK_K = 256 version. * k_quants: WIP super-blocks with 64 weights * We are able to pass preprocessor macros to the Metal compiler * Q6_K works and is actually slightly more efficient than the QK_K = 256 version (25.2 ms vs 25.8 ms) * k_quants: WIP super-blocks with 64 weights Q4_K works on Metal and is actually slightly faster than QK_K = 256 (21.95 ms vs 24.0 ms). * k_quants: WIP super-blocks with 64 weights Q2_K works on Metal and is very slightly faster than QK_K = 256 (23.8 ms vs 24.2 ms). * k_quants: WIP super-blocks with 64 weights Q3_K works on Metal and is slightly faster than QK_K = 256 (26.6 ms vs 28.3 ms). * k_quants: WIP super-blocks with 64 weights Q5_K works on Metal and is slightly faster than QK_K = 256 (23.7 ms vs 26.3 ms). * k_quants: call them _K, not _k, also on Metal * k_quants: correctly define QK_K in llama.cpp * Fixed bug in q4_K quantization added with the 64-block addition * Simplify via lambda * k_quants: swicth Q3_K to 4-bit scales when QK_K = 64 Otherwise there isn't much benefit from this quantization type. There is some very slight loss in accuracy, but we reduce size by ~7%. E.g., for OpenLLaMA-3B, Q3_K_S perplexity is 8.6131 with 8-bit scales and 8.6352 with 4-bit, while file size decreases from 1.53G to 1.44G. * k_quants: switch Q4_K to 4-bit scales when QK_K = 64 Here the loss in accuracy is greater than for Q3_K, but the Q4_K points still move further to the left on the perplexity vs size curve. * k_quants: forgot to add the Metal changes in last commit * k_quants: change Q5_K to be type 0 when QK_K = 64 Still needs AVX2 implementation * k_quants: AVX2 implementation for new 64-weight Q5_K * k_quants: 10% faster ARM_NEON Q5_K dot product * k_quants: fixed issue caused by merging with master --------- Co-authored-by: Iwan Kawrakow <iwan.kawrakow@gmail.com>
2023-06-26 16:43:07 +00:00
sum += isum * d_all * y[i].d;
}
*s = sum;
#elif defined __AVX2__
const __m256i m4 = _mm256_set1_epi8(0xF);
const __m256i m2 = _mm256_set1_epi8(3);
const __m256i m32s = _mm256_set1_epi8(32);
__m256 acc = _mm256_setzero_ps();
for (int i = 0; i < nb; ++i) {
ggml : move FP16 <-> FP32 code to ggml-impl.h (#3861) * ggml : move FP16 <-> FP32 stuff to ggml-impl.h ggml-ci * tests : fix ARM build * ggml : explicitly initialize deprecated type traits * ggml : add math.h to ggml-impl.h * ggml : remove duplicate static assert macros * ggml : prefix lookup tables with ggml_ ggml-ci * ggml-impl : move extern "C" to start of file
2023-10-30 17:19:15 +00:00
const float d = y[i].d * GGML_FP16_TO_FP32(x[i].d);
k-quants : support for super-block size of 64 (#2001) * k_quants: WIP super-blocks with 64 weights * k_quants: WIP super-blocks with 64 weights Q6_K scalar and AVX2 works * k_quants: WIP super-blocks with 64 weights Q4_K scalar and AVX2 works * k_quants: WIP super-blocks with 64 weights Q2_K scalar and AVX2 works. Q2_K is way too slow (it is actually slower than the scalar implementation) * k_quants: WIP super-blocks with 64 weights Q3_K scalar and AVX2 works. * k_quants: WIP super-blocks with 64 weights Q5_K scalar and AVX2 works, and with that all k_quants are done on AVX2 and scalar * k_quants: WIP super-blocks with 64 weights Q6_K working on CUDA. Cannot make it run quite as gast as with super-blocks with 256 weigths: 8% slower on 4080, 20% slower on the 1660 (but there we fit 1 less layer on the GPU because pf the larger model size), so some fraction of these 20% is due to that, * k_quants: WIP super-blocks with 64 weights Q4_K working on CUDA. ~10% slower on GTX-1660, 16% slower on 4080. * k_quants: WIP super-blocks with 64 weights Q2_K working on CUDA. ~3% slower on GTX-1660, 10% slower on 4080. * k_quants: WIP super-blocks with 64 weights Q3_K working on CUDA. * k_quants: WIP super-blocks with 64 weights Q5_K working on CUDA, and with this CUDA is done. * k_quants: WIP super-blocks with 64 weights Q6_K working on ARM_NEON * k_quants: WIP super-blocks with 64 weights Q4_K working on ARM_NEON, but quite a bit slower than 256 weights * k_quants: WIP super-blocks with 64 weights Q2_K working on ARM_NEON, but quite a bit slower than 256 weights * k_quants: WIP super-blocks with 64 weights Q3_K working on ARM_NEON, but quite a bit slower than 256 weights. * k_quants: WIP super-blocks with 64 weights Q5_K working on ARM_NEON, but quite a bit slower than 256 weights. With that, we have full support for ARM_NEON, although performance is not quite there. * k_quants: WIP super-blocks with 64 weights Slightly more efficient Q3_K and Q5_K * k_quants: WIP super-blocks with 64 weights Another small improvement for Q3_K and Q5_K on ARM_NEON * k_quants: WIP super-blocks with 64 weights Yet another speedup for Q5_K on ARM_NEON. We are now within 10% of the QK_K = 256 version. * k_quants: WIP super-blocks with 64 weights * We are able to pass preprocessor macros to the Metal compiler * Q6_K works and is actually slightly more efficient than the QK_K = 256 version (25.2 ms vs 25.8 ms) * k_quants: WIP super-blocks with 64 weights Q4_K works on Metal and is actually slightly faster than QK_K = 256 (21.95 ms vs 24.0 ms). * k_quants: WIP super-blocks with 64 weights Q2_K works on Metal and is very slightly faster than QK_K = 256 (23.8 ms vs 24.2 ms). * k_quants: WIP super-blocks with 64 weights Q3_K works on Metal and is slightly faster than QK_K = 256 (26.6 ms vs 28.3 ms). * k_quants: WIP super-blocks with 64 weights Q5_K works on Metal and is slightly faster than QK_K = 256 (23.7 ms vs 26.3 ms). * k_quants: call them _K, not _k, also on Metal * k_quants: correctly define QK_K in llama.cpp * Fixed bug in q4_K quantization added with the 64-block addition * Simplify via lambda * k_quants: swicth Q3_K to 4-bit scales when QK_K = 64 Otherwise there isn't much benefit from this quantization type. There is some very slight loss in accuracy, but we reduce size by ~7%. E.g., for OpenLLaMA-3B, Q3_K_S perplexity is 8.6131 with 8-bit scales and 8.6352 with 4-bit, while file size decreases from 1.53G to 1.44G. * k_quants: switch Q4_K to 4-bit scales when QK_K = 64 Here the loss in accuracy is greater than for Q3_K, but the Q4_K points still move further to the left on the perplexity vs size curve. * k_quants: forgot to add the Metal changes in last commit * k_quants: change Q5_K to be type 0 when QK_K = 64 Still needs AVX2 implementation * k_quants: AVX2 implementation for new 64-weight Q5_K * k_quants: 10% faster ARM_NEON Q5_K dot product * k_quants: fixed issue caused by merging with master --------- Co-authored-by: Iwan Kawrakow <iwan.kawrakow@gmail.com>
2023-06-26 16:43:07 +00:00
const uint8_t * restrict q4 = x[i].ql;
const uint8_t * restrict qh = x[i].qh;
const int8_t * restrict q8 = y[i].qs;
const __m64 scales_1 = _mm_set1_pi8(x[i].scales[0]);
const __m64 scales_2 = _mm_set1_pi8(x[i].scales[1]);
const __m64 scales_3 = _mm_set1_pi8(x[i].scales[2]);
const __m64 scales_4 = _mm_set1_pi8(x[i].scales[3]);
__m256i sumi = _mm256_setzero_si256();
const __m128i scale_0 = _mm_set_epi64(scales_2, scales_1);
const __m128i scale_1 = _mm_set_epi64(scales_4, scales_3);
const __m256i q4bits1 = _mm256_loadu_si256((const __m256i*)q4);
const __m128i q4bitsH = _mm_loadu_si128((const __m128i*)qh);
ggml : workaround for missing _mm256_setr_m128i in GCC < 8 in k_quants.c (#2405)
2023-07-28 18:17:45 +00:00
const __m256i q4h_0 = _mm256_slli_epi16(_mm256_and_si256(MM256_SET_M128I(_mm_srli_epi16(q4bitsH, 2), q4bitsH), m2), 4);
const __m256i q4h_1 = _mm256_slli_epi16(_mm256_and_si256(MM256_SET_M128I(_mm_srli_epi16(q4bitsH, 6), _mm_srli_epi16(q4bitsH, 4)), m2), 4);
k-quants : support for super-block size of 64 (#2001) * k_quants: WIP super-blocks with 64 weights * k_quants: WIP super-blocks with 64 weights Q6_K scalar and AVX2 works * k_quants: WIP super-blocks with 64 weights Q4_K scalar and AVX2 works * k_quants: WIP super-blocks with 64 weights Q2_K scalar and AVX2 works. Q2_K is way too slow (it is actually slower than the scalar implementation) * k_quants: WIP super-blocks with 64 weights Q3_K scalar and AVX2 works. * k_quants: WIP super-blocks with 64 weights Q5_K scalar and AVX2 works, and with that all k_quants are done on AVX2 and scalar * k_quants: WIP super-blocks with 64 weights Q6_K working on CUDA. Cannot make it run quite as gast as with super-blocks with 256 weigths: 8% slower on 4080, 20% slower on the 1660 (but there we fit 1 less layer on the GPU because pf the larger model size), so some fraction of these 20% is due to that, * k_quants: WIP super-blocks with 64 weights Q4_K working on CUDA. ~10% slower on GTX-1660, 16% slower on 4080. * k_quants: WIP super-blocks with 64 weights Q2_K working on CUDA. ~3% slower on GTX-1660, 10% slower on 4080. * k_quants: WIP super-blocks with 64 weights Q3_K working on CUDA. * k_quants: WIP super-blocks with 64 weights Q5_K working on CUDA, and with this CUDA is done. * k_quants: WIP super-blocks with 64 weights Q6_K working on ARM_NEON * k_quants: WIP super-blocks with 64 weights Q4_K working on ARM_NEON, but quite a bit slower than 256 weights * k_quants: WIP super-blocks with 64 weights Q2_K working on ARM_NEON, but quite a bit slower than 256 weights * k_quants: WIP super-blocks with 64 weights Q3_K working on ARM_NEON, but quite a bit slower than 256 weights. * k_quants: WIP super-blocks with 64 weights Q5_K working on ARM_NEON, but quite a bit slower than 256 weights. With that, we have full support for ARM_NEON, although performance is not quite there. * k_quants: WIP super-blocks with 64 weights Slightly more efficient Q3_K and Q5_K * k_quants: WIP super-blocks with 64 weights Another small improvement for Q3_K and Q5_K on ARM_NEON * k_quants: WIP super-blocks with 64 weights Yet another speedup for Q5_K on ARM_NEON. We are now within 10% of the QK_K = 256 version. * k_quants: WIP super-blocks with 64 weights * We are able to pass preprocessor macros to the Metal compiler * Q6_K works and is actually slightly more efficient than the QK_K = 256 version (25.2 ms vs 25.8 ms) * k_quants: WIP super-blocks with 64 weights Q4_K works on Metal and is actually slightly faster than QK_K = 256 (21.95 ms vs 24.0 ms). * k_quants: WIP super-blocks with 64 weights Q2_K works on Metal and is very slightly faster than QK_K = 256 (23.8 ms vs 24.2 ms). * k_quants: WIP super-blocks with 64 weights Q3_K works on Metal and is slightly faster than QK_K = 256 (26.6 ms vs 28.3 ms). * k_quants: WIP super-blocks with 64 weights Q5_K works on Metal and is slightly faster than QK_K = 256 (23.7 ms vs 26.3 ms). * k_quants: call them _K, not _k, also on Metal * k_quants: correctly define QK_K in llama.cpp * Fixed bug in q4_K quantization added with the 64-block addition * Simplify via lambda * k_quants: swicth Q3_K to 4-bit scales when QK_K = 64 Otherwise there isn't much benefit from this quantization type. There is some very slight loss in accuracy, but we reduce size by ~7%. E.g., for OpenLLaMA-3B, Q3_K_S perplexity is 8.6131 with 8-bit scales and 8.6352 with 4-bit, while file size decreases from 1.53G to 1.44G. * k_quants: switch Q4_K to 4-bit scales when QK_K = 64 Here the loss in accuracy is greater than for Q3_K, but the Q4_K points still move further to the left on the perplexity vs size curve. * k_quants: forgot to add the Metal changes in last commit * k_quants: change Q5_K to be type 0 when QK_K = 64 Still needs AVX2 implementation * k_quants: AVX2 implementation for new 64-weight Q5_K * k_quants: 10% faster ARM_NEON Q5_K dot product * k_quants: fixed issue caused by merging with master --------- Co-authored-by: Iwan Kawrakow <iwan.kawrakow@gmail.com>
2023-06-26 16:43:07 +00:00
const __m256i q4_0 = _mm256_or_si256(_mm256_and_si256(q4bits1, m4), q4h_0);
const __m256i q4_1 = _mm256_or_si256(_mm256_and_si256(_mm256_srli_epi16(q4bits1, 4), m4), q4h_1);
const __m256i q8_0 = _mm256_loadu_si256((const __m256i*)(q8+ 0));
const __m256i q8_1 = _mm256_loadu_si256((const __m256i*)(q8+32));
__m256i q8s_0 = _mm256_maddubs_epi16(m32s, q8_0);
__m256i q8s_1 = _mm256_maddubs_epi16(m32s, q8_1);
__m256i p16_0 = _mm256_maddubs_epi16(q4_0, q8_0);
__m256i p16_1 = _mm256_maddubs_epi16(q4_1, q8_1);
p16_0 = _mm256_sub_epi16(p16_0, q8s_0);
p16_1 = _mm256_sub_epi16(p16_1, q8s_1);
p16_0 = _mm256_madd_epi16(_mm256_cvtepi8_epi16(scale_0), p16_0);
p16_1 = _mm256_madd_epi16(_mm256_cvtepi8_epi16(scale_1), p16_1);
sumi = _mm256_add_epi32(sumi, _mm256_add_epi32(p16_0, p16_1));
acc = _mm256_fmadd_ps(_mm256_broadcast_ss(&d), _mm256_cvtepi32_ps(sumi), acc);
}
*s = hsum_float_8(acc);
k_quants : add AVX support to dot functions with QK_K as 64 (#2339) * add AVX to ggml_vec_dot_q2_K_q8_K() * add AVX to ggml_vec_dot_q3_K_q8_K() * add AVX to ggml_vec_dot_q4_K_q8_K() * add AVX to ggml_vec_dot_q5_K_q8_K() * add AVX to ggml_vec_dot_q6_K_q8_K() * refactor AVX code in ggml_vec_dot_q6_K_q8_K()
2023-07-25 12:13:41 +00:00
#elif defined __AVX__
const __m128i m4 = _mm_set1_epi8(0xF);
const __m128i m2 = _mm_set1_epi8(3);
const __m128i m32s = _mm_set1_epi8(32);
__m256 acc = _mm256_setzero_ps();
for (int i = 0; i < nb; ++i) {
ggml : move FP16 <-> FP32 code to ggml-impl.h (#3861) * ggml : move FP16 <-> FP32 stuff to ggml-impl.h ggml-ci * tests : fix ARM build * ggml : explicitly initialize deprecated type traits * ggml : add math.h to ggml-impl.h * ggml : remove duplicate static assert macros * ggml : prefix lookup tables with ggml_ ggml-ci * ggml-impl : move extern "C" to start of file
2023-10-30 17:19:15 +00:00
const float d = y[i].d * GGML_FP16_TO_FP32(x[i].d);
k_quants : add AVX support to dot functions with QK_K as 64 (#2339) * add AVX to ggml_vec_dot_q2_K_q8_K() * add AVX to ggml_vec_dot_q3_K_q8_K() * add AVX to ggml_vec_dot_q4_K_q8_K() * add AVX to ggml_vec_dot_q5_K_q8_K() * add AVX to ggml_vec_dot_q6_K_q8_K() * refactor AVX code in ggml_vec_dot_q6_K_q8_K()
2023-07-25 12:13:41 +00:00
const uint8_t * restrict q4 = x[i].ql;
const uint8_t * restrict qh = x[i].qh;
const int8_t * restrict q8 = y[i].qs;
const __m64 scales_1 = _mm_set1_pi8(x[i].scales[0]);
const __m64 scales_2 = _mm_set1_pi8(x[i].scales[1]);
const __m64 scales_3 = _mm_set1_pi8(x[i].scales[2]);
const __m64 scales_4 = _mm_set1_pi8(x[i].scales[3]);
__m128i sumi_0 = _mm_setzero_si128();
__m128i sumi_1 = _mm_setzero_si128();
const __m128i scale_0 = _mm_set_epi64(scales_2, scales_1);
const __m128i scale_1 = _mm_set_epi64(scales_4, scales_3);
const __m256i q4bits1 = _mm256_loadu_si256((const __m256i*)q4);
const __m128i q4bitsH = _mm_loadu_si128((const __m128i*)qh);
const __m128i q4h_0 = _mm_slli_epi16(_mm_and_si128(q4bitsH, m2), 4);
const __m128i q4h_1 = _mm_slli_epi16(_mm_and_si128(_mm_srli_epi16(q4bitsH, 2), m2), 4);
const __m128i q4h_2 = _mm_slli_epi16(_mm_and_si128(_mm_srli_epi16(q4bitsH, 4), m2), 4);
const __m128i q4h_3 = _mm_slli_epi16(_mm_and_si128(_mm_srli_epi16(q4bitsH, 6), m2), 4);
const __m128i q4_0 = _mm_or_si128(_mm_and_si128(_mm256_extractf128_si256(q4bits1, 0), m4), q4h_0);
const __m128i q4_1 = _mm_or_si128(_mm_and_si128(_mm256_extractf128_si256(q4bits1, 1), m4), q4h_1);
const __m128i q4_2 = _mm_or_si128(_mm_and_si128(_mm_srli_epi16(_mm256_extractf128_si256(q4bits1, 0), 4), m4), q4h_2);
const __m128i q4_3 = _mm_or_si128(_mm_and_si128(_mm_srli_epi16(_mm256_extractf128_si256(q4bits1, 1), 4), m4), q4h_3);
const __m256i q8_0 = _mm256_loadu_si256((const __m256i*)(q8+ 0));
const __m256i q8_1 = _mm256_loadu_si256((const __m256i*)(q8+32));
__m128i q8s_0 = _mm_maddubs_epi16(m32s, _mm256_extractf128_si256(q8_0, 0));
__m128i q8s_1 = _mm_maddubs_epi16(m32s, _mm256_extractf128_si256(q8_0, 1));
__m128i q8s_2 = _mm_maddubs_epi16(m32s, _mm256_extractf128_si256(q8_1, 0));
__m128i q8s_3 = _mm_maddubs_epi16(m32s, _mm256_extractf128_si256(q8_1, 1));
__m128i p16_0 = _mm_maddubs_epi16(q4_0, _mm256_extractf128_si256(q8_0, 0));
__m128i p16_1 = _mm_maddubs_epi16(q4_1, _mm256_extractf128_si256(q8_0, 1));
__m128i p16_2 = _mm_maddubs_epi16(q4_2, _mm256_extractf128_si256(q8_1, 0));
__m128i p16_3 = _mm_maddubs_epi16(q4_3, _mm256_extractf128_si256(q8_1, 1));
p16_0 = _mm_sub_epi16(p16_0, q8s_0);
p16_1 = _mm_sub_epi16(p16_1, q8s_1);
p16_2 = _mm_sub_epi16(p16_2, q8s_2);
p16_3 = _mm_sub_epi16(p16_3, q8s_3);
p16_0 = _mm_madd_epi16(_mm_cvtepi8_epi16(scale_0), p16_0);
p16_1 = _mm_madd_epi16(_mm_cvtepi8_epi16(_mm_unpackhi_epi64(scale_0, scale_0)), p16_1);
p16_2 = _mm_madd_epi16(_mm_cvtepi8_epi16(scale_1), p16_2);
p16_3 = _mm_madd_epi16(_mm_cvtepi8_epi16(_mm_unpackhi_epi64(scale_1, scale_1)), p16_3);
sumi_0 = _mm_add_epi32(sumi_0, _mm_add_epi32(p16_0, p16_2));
sumi_1 = _mm_add_epi32(sumi_1, _mm_add_epi32(p16_1, p16_3));
ggml : workaround for missing _mm256_setr_m128i in GCC < 8 in k_quants.c (#2405)
2023-07-28 18:17:45 +00:00
acc = _mm256_add_ps(_mm256_mul_ps(_mm256_broadcast_ss(&d), _mm256_cvtepi32_ps(MM256_SET_M128I(sumi_1, sumi_0))), acc);
k_quants : add AVX support to dot functions with QK_K as 64 (#2339) * add AVX to ggml_vec_dot_q2_K_q8_K() * add AVX to ggml_vec_dot_q3_K_q8_K() * add AVX to ggml_vec_dot_q4_K_q8_K() * add AVX to ggml_vec_dot_q5_K_q8_K() * add AVX to ggml_vec_dot_q6_K_q8_K() * refactor AVX code in ggml_vec_dot_q6_K_q8_K()
2023-07-25 12:13:41 +00:00
}
*s = hsum_float_8(acc);
ggml : add RISC-V Vector Support for K-Quants and improved the existing intrinsics (#3453) * Added RVV intrinsics support for Q8 quantize row and also improved the existing dot product function for risc-v. The RVV intrinsics is added for the following quantize row functions quantize_row_q8_0 quantize_row_q8_1 The following dot product functions have also been optimized by using LMUL = 1/2 instead of LMUL = 1 ggml_vec_dot_q4_0_q8_0 ggml_vec_dot_q4_1_q8_1 ggml_vec_dot_q5_0_q8_0 ggml_vec_dot_q5_1_q8_1 And vector initialization in Q5 by temporary array is also replaced by the vid intrinsics Signed-off-by: Ahmad Tameem <ahmad.tameem@10xengineers.ai> * Added RVV intrinsics support for k_quants This adds RISC-V Vector intrinsics support for the following K_quants functions for both QKK = 256 and QKK = 64 ggml_vec_dot_q2_K_q8_K ggml_vec_dot_q3_K_q8_K ggml_vec_dot_q4_K_q8_K ggml_vec_dot_q5_K_q8_K ggml_vec_dot_q6_K_q8_K Signed-off-by: Ahmad Tameem <ahmad.tameem@10xengineers.ai> --------- Signed-off-by: Ahmad Tameem <ahmad.tameem@10xengineers.ai>
2023-10-03 18:38:19 +00:00
#elif defined __riscv_v_intrinsic
float sumf = 0;
for (int i = 0; i < nb; ++i) {
const float d_all = (float)x[i].d;
const uint8_t * restrict q6 = x[i].ql;
const uint8_t * restrict qh = x[i].qh;
const int8_t * restrict q8 = y[i].qs;
const int8_t * restrict scale = x[i].scales;
int32_t isum = 0;
size_t vl = 16;
vint32m1_t vzero = __riscv_vmv_v_x_i32m1(0, 1);
// load Q6
vuint8mf2_t q6_0 = __riscv_vle8_v_u8mf2(q6, vl);
vuint8mf2_t q6_1 = __riscv_vle8_v_u8mf2(q6+16, vl);
// load qh
vuint8mf2_t qh_x = __riscv_vle8_v_u8mf2(qh, vl);
vuint8mf2_t qh0 = __riscv_vsll_vx_u8mf2(__riscv_vand_vx_u8mf2(qh_x, 0x3, vl), 0x4, vl);
qh_x = __riscv_vsrl_vx_u8mf2(qh_x, 0x2, vl);
vuint8mf2_t qh1 = __riscv_vsll_vx_u8mf2(__riscv_vand_vx_u8mf2(qh_x, 0x3, vl), 0x4, vl);
qh_x = __riscv_vsrl_vx_u8mf2(qh_x, 0x2, vl);
vuint8mf2_t qh2 = __riscv_vsll_vx_u8mf2(__riscv_vand_vx_u8mf2(qh_x, 0x3, vl), 0x4, vl);
qh_x = __riscv_vsrl_vx_u8mf2(qh_x, 0x2, vl);
vuint8mf2_t qh3 = __riscv_vsll_vx_u8mf2(__riscv_vand_vx_u8mf2(qh_x, 0x3, vl), 0x4, vl);
vuint8mf2_t q6h_0 = __riscv_vor_vv_u8mf2(__riscv_vand_vx_u8mf2(q6_0, 0xF, vl), qh0, vl);
vuint8mf2_t q6h_1 = __riscv_vor_vv_u8mf2(__riscv_vand_vx_u8mf2(q6_1, 0xF, vl), qh1, vl);
vuint8mf2_t q6h_2 = __riscv_vor_vv_u8mf2(__riscv_vsrl_vx_u8mf2(q6_0, 0x4, vl), qh2, vl);
vuint8mf2_t q6h_3 = __riscv_vor_vv_u8mf2(__riscv_vsrl_vx_u8mf2(q6_1, 0x4, vl), qh3, vl);
vint8mf2_t q6v_0 = __riscv_vsub_vx_i8mf2(__riscv_vreinterpret_v_u8mf2_i8mf2(q6h_0), 32, vl);
vint8mf2_t q6v_1 = __riscv_vsub_vx_i8mf2(__riscv_vreinterpret_v_u8mf2_i8mf2(q6h_1), 32, vl);
vint8mf2_t q6v_2 = __riscv_vsub_vx_i8mf2(__riscv_vreinterpret_v_u8mf2_i8mf2(q6h_2), 32, vl);
vint8mf2_t q6v_3 = __riscv_vsub_vx_i8mf2(__riscv_vreinterpret_v_u8mf2_i8mf2(q6h_3), 32, vl);
// load Q8 and take product
vint16m1_t p0 = __riscv_vwmul_vv_i16m1(q6v_0, __riscv_vle8_v_i8mf2(q8, vl), vl);
vint16m1_t p1 = __riscv_vwmul_vv_i16m1(q6v_1, __riscv_vle8_v_i8mf2(q8+16, vl), vl);
vint16m1_t p2 = __riscv_vwmul_vv_i16m1(q6v_2, __riscv_vle8_v_i8mf2(q8+32, vl), vl);
vint16m1_t p3 = __riscv_vwmul_vv_i16m1(q6v_3, __riscv_vle8_v_i8mf2(q8+48, vl), vl);
vint32m1_t vs_0 = __riscv_vwredsum_vs_i16m1_i32m1(p0, vzero, vl);
vint32m1_t vs_1 = __riscv_vwredsum_vs_i16m1_i32m1(p1, vzero, vl);
vint32m1_t vs_2 = __riscv_vwredsum_vs_i16m1_i32m1(p2, vzero, vl);
vint32m1_t vs_3 = __riscv_vwredsum_vs_i16m1_i32m1(p3, vzero, vl);
isum += __riscv_vmv_x_s_i32m1_i32(vs_0) * scale[0];
isum += __riscv_vmv_x_s_i32m1_i32(vs_1) * scale[1];
isum += __riscv_vmv_x_s_i32m1_i32(vs_2) * scale[2];
isum += __riscv_vmv_x_s_i32m1_i32(vs_3) * scale[3];
sumf += isum * d_all * y[i].d;
}
*s = sumf;
k-quants : support for super-block size of 64 (#2001) * k_quants: WIP super-blocks with 64 weights * k_quants: WIP super-blocks with 64 weights Q6_K scalar and AVX2 works * k_quants: WIP super-blocks with 64 weights Q4_K scalar and AVX2 works * k_quants: WIP super-blocks with 64 weights Q2_K scalar and AVX2 works. Q2_K is way too slow (it is actually slower than the scalar implementation) * k_quants: WIP super-blocks with 64 weights Q3_K scalar and AVX2 works. * k_quants: WIP super-blocks with 64 weights Q5_K scalar and AVX2 works, and with that all k_quants are done on AVX2 and scalar * k_quants: WIP super-blocks with 64 weights Q6_K working on CUDA. Cannot make it run quite as gast as with super-blocks with 256 weigths: 8% slower on 4080, 20% slower on the 1660 (but there we fit 1 less layer on the GPU because pf the larger model size), so some fraction of these 20% is due to that, * k_quants: WIP super-blocks with 64 weights Q4_K working on CUDA. ~10% slower on GTX-1660, 16% slower on 4080. * k_quants: WIP super-blocks with 64 weights Q2_K working on CUDA. ~3% slower on GTX-1660, 10% slower on 4080. * k_quants: WIP super-blocks with 64 weights Q3_K working on CUDA. * k_quants: WIP super-blocks with 64 weights Q5_K working on CUDA, and with this CUDA is done. * k_quants: WIP super-blocks with 64 weights Q6_K working on ARM_NEON * k_quants: WIP super-blocks with 64 weights Q4_K working on ARM_NEON, but quite a bit slower than 256 weights * k_quants: WIP super-blocks with 64 weights Q2_K working on ARM_NEON, but quite a bit slower than 256 weights * k_quants: WIP super-blocks with 64 weights Q3_K working on ARM_NEON, but quite a bit slower than 256 weights. * k_quants: WIP super-blocks with 64 weights Q5_K working on ARM_NEON, but quite a bit slower than 256 weights. With that, we have full support for ARM_NEON, although performance is not quite there. * k_quants: WIP super-blocks with 64 weights Slightly more efficient Q3_K and Q5_K * k_quants: WIP super-blocks with 64 weights Another small improvement for Q3_K and Q5_K on ARM_NEON * k_quants: WIP super-blocks with 64 weights Yet another speedup for Q5_K on ARM_NEON. We are now within 10% of the QK_K = 256 version. * k_quants: WIP super-blocks with 64 weights * We are able to pass preprocessor macros to the Metal compiler * Q6_K works and is actually slightly more efficient than the QK_K = 256 version (25.2 ms vs 25.8 ms) * k_quants: WIP super-blocks with 64 weights Q4_K works on Metal and is actually slightly faster than QK_K = 256 (21.95 ms vs 24.0 ms). * k_quants: WIP super-blocks with 64 weights Q2_K works on Metal and is very slightly faster than QK_K = 256 (23.8 ms vs 24.2 ms). * k_quants: WIP super-blocks with 64 weights Q3_K works on Metal and is slightly faster than QK_K = 256 (26.6 ms vs 28.3 ms). * k_quants: WIP super-blocks with 64 weights Q5_K works on Metal and is slightly faster than QK_K = 256 (23.7 ms vs 26.3 ms). * k_quants: call them _K, not _k, also on Metal * k_quants: correctly define QK_K in llama.cpp * Fixed bug in q4_K quantization added with the 64-block addition * Simplify via lambda * k_quants: swicth Q3_K to 4-bit scales when QK_K = 64 Otherwise there isn't much benefit from this quantization type. There is some very slight loss in accuracy, but we reduce size by ~7%. E.g., for OpenLLaMA-3B, Q3_K_S perplexity is 8.6131 with 8-bit scales and 8.6352 with 4-bit, while file size decreases from 1.53G to 1.44G. * k_quants: switch Q4_K to 4-bit scales when QK_K = 64 Here the loss in accuracy is greater than for Q3_K, but the Q4_K points still move further to the left on the perplexity vs size curve. * k_quants: forgot to add the Metal changes in last commit * k_quants: change Q5_K to be type 0 when QK_K = 64 Still needs AVX2 implementation * k_quants: AVX2 implementation for new 64-weight Q5_K * k_quants: 10% faster ARM_NEON Q5_K dot product * k_quants: fixed issue caused by merging with master --------- Co-authored-by: Iwan Kawrakow <iwan.kawrakow@gmail.com>
2023-06-26 16:43:07 +00:00
#else
int8_t aux8[QK_K];
int16_t aux16[8];
float sums [8];
int32_t aux32[8];
memset(sums, 0, 8*sizeof(float));
float sumf = 0;
for (int i = 0; i < nb; ++i) {
const uint8_t * restrict q4 = x[i].ql;
const uint8_t * restrict qh = x[i].qh;
const int8_t * restrict q8 = y[i].qs;
memset(aux32, 0, 8*sizeof(int32_t));
int8_t * restrict a = aux8;
for (int l = 0; l < 16; ++l) {
a[l+ 0] = (int8_t)((q4[l+ 0] & 0xF) | (((qh[l] >> 0) & 3) << 4)) - 32;
a[l+16] = (int8_t)((q4[l+16] & 0xF) | (((qh[l] >> 2) & 3) << 4)) - 32;
a[l+32] = (int8_t)((q4[l+ 0] >> 4) | (((qh[l] >> 4) & 3) << 4)) - 32;
a[l+48] = (int8_t)((q4[l+16] >> 4) | (((qh[l] >> 6) & 3) << 4)) - 32;
}
int is = 0;
for (int j = 0; j < QK_K/16; ++j) {
int scale = x[i].scales[is++];
for (int l = 0; l < 8; ++l) aux16[l] = q8[l] * a[l];
for (int l = 0; l < 8; ++l) aux32[l] += scale * aux16[l];
q8 += 8; a += 8;
for (int l = 0; l < 8; ++l) aux16[l] = q8[l] * a[l];
for (int l = 0; l < 8; ++l) aux32[l] += scale * aux16[l];
q8 += 8; a += 8;
}
ggml : move FP16 <-> FP32 code to ggml-impl.h (#3861) * ggml : move FP16 <-> FP32 stuff to ggml-impl.h ggml-ci * tests : fix ARM build * ggml : explicitly initialize deprecated type traits * ggml : add math.h to ggml-impl.h * ggml : remove duplicate static assert macros * ggml : prefix lookup tables with ggml_ ggml-ci * ggml-impl : move extern "C" to start of file
2023-10-30 17:19:15 +00:00
const float d = GGML_FP16_TO_FP32(x[i].d) * y[i].d;
k-quants : support for super-block size of 64 (#2001) * k_quants: WIP super-blocks with 64 weights * k_quants: WIP super-blocks with 64 weights Q6_K scalar and AVX2 works * k_quants: WIP super-blocks with 64 weights Q4_K scalar and AVX2 works * k_quants: WIP super-blocks with 64 weights Q2_K scalar and AVX2 works. Q2_K is way too slow (it is actually slower than the scalar implementation) * k_quants: WIP super-blocks with 64 weights Q3_K scalar and AVX2 works. * k_quants: WIP super-blocks with 64 weights Q5_K scalar and AVX2 works, and with that all k_quants are done on AVX2 and scalar * k_quants: WIP super-blocks with 64 weights Q6_K working on CUDA. Cannot make it run quite as gast as with super-blocks with 256 weigths: 8% slower on 4080, 20% slower on the 1660 (but there we fit 1 less layer on the GPU because pf the larger model size), so some fraction of these 20% is due to that, * k_quants: WIP super-blocks with 64 weights Q4_K working on CUDA. ~10% slower on GTX-1660, 16% slower on 4080. * k_quants: WIP super-blocks with 64 weights Q2_K working on CUDA. ~3% slower on GTX-1660, 10% slower on 4080. * k_quants: WIP super-blocks with 64 weights Q3_K working on CUDA. * k_quants: WIP super-blocks with 64 weights Q5_K working on CUDA, and with this CUDA is done. * k_quants: WIP super-blocks with 64 weights Q6_K working on ARM_NEON * k_quants: WIP super-blocks with 64 weights Q4_K working on ARM_NEON, but quite a bit slower than 256 weights * k_quants: WIP super-blocks with 64 weights Q2_K working on ARM_NEON, but quite a bit slower than 256 weights * k_quants: WIP super-blocks with 64 weights Q3_K working on ARM_NEON, but quite a bit slower than 256 weights. * k_quants: WIP super-blocks with 64 weights Q5_K working on ARM_NEON, but quite a bit slower than 256 weights. With that, we have full support for ARM_NEON, although performance is not quite there. * k_quants: WIP super-blocks with 64 weights Slightly more efficient Q3_K and Q5_K * k_quants: WIP super-blocks with 64 weights Another small improvement for Q3_K and Q5_K on ARM_NEON * k_quants: WIP super-blocks with 64 weights Yet another speedup for Q5_K on ARM_NEON. We are now within 10% of the QK_K = 256 version. * k_quants: WIP super-blocks with 64 weights * We are able to pass preprocessor macros to the Metal compiler * Q6_K works and is actually slightly more efficient than the QK_K = 256 version (25.2 ms vs 25.8 ms) * k_quants: WIP super-blocks with 64 weights Q4_K works on Metal and is actually slightly faster than QK_K = 256 (21.95 ms vs 24.0 ms). * k_quants: WIP super-blocks with 64 weights Q2_K works on Metal and is very slightly faster than QK_K = 256 (23.8 ms vs 24.2 ms). * k_quants: WIP super-blocks with 64 weights Q3_K works on Metal and is slightly faster than QK_K = 256 (26.6 ms vs 28.3 ms). * k_quants: WIP super-blocks with 64 weights Q5_K works on Metal and is slightly faster than QK_K = 256 (23.7 ms vs 26.3 ms). * k_quants: call them _K, not _k, also on Metal * k_quants: correctly define QK_K in llama.cpp * Fixed bug in q4_K quantization added with the 64-block addition * Simplify via lambda * k_quants: swicth Q3_K to 4-bit scales when QK_K = 64 Otherwise there isn't much benefit from this quantization type. There is some very slight loss in accuracy, but we reduce size by ~7%. E.g., for OpenLLaMA-3B, Q3_K_S perplexity is 8.6131 with 8-bit scales and 8.6352 with 4-bit, while file size decreases from 1.53G to 1.44G. * k_quants: switch Q4_K to 4-bit scales when QK_K = 64 Here the loss in accuracy is greater than for Q3_K, but the Q4_K points still move further to the left on the perplexity vs size curve. * k_quants: forgot to add the Metal changes in last commit * k_quants: change Q5_K to be type 0 when QK_K = 64 Still needs AVX2 implementation * k_quants: AVX2 implementation for new 64-weight Q5_K * k_quants: 10% faster ARM_NEON Q5_K dot product * k_quants: fixed issue caused by merging with master --------- Co-authored-by: Iwan Kawrakow <iwan.kawrakow@gmail.com>
2023-06-26 16:43:07 +00:00
for (int l = 0; l < 8; ++l) sums[l] += d * aux32[l];
}
for (int l = 0; l < 8; ++l) sumf += sums[l];
*s = sumf;
#endif
}
#endif
SOTA 2-bit quants (#4773) * iq2_xxs: basics * iq2_xxs: scalar and AVX2 dot products Needed to change Q8_K to have quants in the -127...127 range, else the IQ2_XXS AVX implementation becomes very awkward. The alternative would have been to use Q8_0 instead. Perhaps I'll change later, for now this is what we have. * iq2_xxs: ARM_NEON dot product Somehow strangely slow (112 ms/token). * iq2_xxs: WIP Metal Dequantize works, something is still wrong with the dot product. * iq2_xxs: Metal dot product now works We have PP-512 = 475 t/s TG-128 = 47.3 t/s Not the greatest performance, but not complete garbage either. * iq2_xxs: slighty faster dot product TG-128 is now 48.4 t/s * iq2_xxs: slighty faster dot product TG-128 is now 50.9 t/s * iq2_xxs: even faster Metal dot product TG-128 is now 54.1 t/s. Strangely enough, putting the signs lookup table into shared memory has a bigger impact than the grid values being in shared memory. * iq2_xxs: dequantize CUDA kernel - fix conflict with master * iq2_xxs: quantized CUDA dot product (MMVQ) We get TG-128 = 153.1 t/s * iq2_xxs: slightly faster CUDA dot product TG-128 is now at 155.1 t/s. * iq2_xxs: add to llama ftype enum * iq2_xxs: fix MoE on Metal * Fix missing MMQ ops when on hipBLAS I had put the ggml_supports_mmq call at the wrong place. * Fix bug in qequantize_row_iq2_xxs The 0.25f factor was missing. Great detective work by @ggerganov! * Fixing tests * PR suggestion --------- Co-authored-by: Iwan Kawrakow <iwan.kawrakow@gmail.com>
2024-01-08 15:02:32 +00:00
static const int8_t keven_signs_q2xs[1024] = {
1, 1, 1, 1, 1, 1, 1, 1, -1, 1, 1, 1, 1, 1, 1, -1, 1, -1, 1, 1, 1, 1, 1, -1, -1, -1, 1, 1, 1, 1, 1, 1,
1, 1, -1, 1, 1, 1, 1, -1, -1, 1, -1, 1, 1, 1, 1, 1, 1, -1, -1, 1, 1, 1, 1, 1, -1, -1, -1, 1, 1, 1, 1, -1,
1, 1, 1, -1, 1, 1, 1, -1, -1, 1, 1, -1, 1, 1, 1, 1, 1, -1, 1, -1, 1, 1, 1, 1, -1, -1, 1, -1, 1, 1, 1, -1,
1, 1, -1, -1, 1, 1, 1, 1, -1, 1, -1, -1, 1, 1, 1, -1, 1, -1, -1, -1, 1, 1, 1, -1, -1, -1, -1, -1, 1, 1, 1, 1,
1, 1, 1, 1, -1, 1, 1, -1, -1, 1, 1, 1, -1, 1, 1, 1, 1, -1, 1, 1, -1, 1, 1, 1, -1, -1, 1, 1, -1, 1, 1, -1,
1, 1, -1, 1, -1, 1, 1, 1, -1, 1, -1, 1, -1, 1, 1, -1, 1, -1, -1, 1, -1, 1, 1, -1, -1, -1, -1, 1, -1, 1, 1, 1,
1, 1, 1, -1, -1, 1, 1, 1, -1, 1, 1, -1, -1, 1, 1, -1, 1, -1, 1, -1, -1, 1, 1, -1, -1, -1, 1, -1, -1, 1, 1, 1,
1, 1, -1, -1, -1, 1, 1, -1, -1, 1, -1, -1, -1, 1, 1, 1, 1, -1, -1, -1, -1, 1, 1, 1, -1, -1, -1, -1, -1, 1, 1, -1,
1, 1, 1, 1, 1, -1, 1, -1, -1, 1, 1, 1, 1, -1, 1, 1, 1, -1, 1, 1, 1, -1, 1, 1, -1, -1, 1, 1, 1, -1, 1, -1,
1, 1, -1, 1, 1, -1, 1, 1, -1, 1, -1, 1, 1, -1, 1, -1, 1, -1, -1, 1, 1, -1, 1, -1, -1, -1, -1, 1, 1, -1, 1, 1,
1, 1, 1, -1, 1, -1, 1, 1, -1, 1, 1, -1, 1, -1, 1, -1, 1, -1, 1, -1, 1, -1, 1, -1, -1, -1, 1, -1, 1, -1, 1, 1,
1, 1, -1, -1, 1, -1, 1, -1, -1, 1, -1, -1, 1, -1, 1, 1, 1, -1, -1, -1, 1, -1, 1, 1, -1, -1, -1, -1, 1, -1, 1, -1,
1, 1, 1, 1, -1, -1, 1, 1, -1, 1, 1, 1, -1, -1, 1, -1, 1, -1, 1, 1, -1, -1, 1, -1, -1, -1, 1, 1, -1, -1, 1, 1,
1, 1, -1, 1, -1, -1, 1, -1, -1, 1, -1, 1, -1, -1, 1, 1, 1, -1, -1, 1, -1, -1, 1, 1, -1, -1, -1, 1, -1, -1, 1, -1,
1, 1, 1, -1, -1, -1, 1, -1, -1, 1, 1, -1, -1, -1, 1, 1, 1, -1, 1, -1, -1, -1, 1, 1, -1, -1, 1, -1, -1, -1, 1, -1,
1, 1, -1, -1, -1, -1, 1, 1, -1, 1, -1, -1, -1, -1, 1, -1, 1, -1, -1, -1, -1, -1, 1, -1, -1, -1, -1, -1, -1, -1, 1, 1,
1, 1, 1, 1, 1, 1, -1, -1, -1, 1, 1, 1, 1, 1, -1, 1, 1, -1, 1, 1, 1, 1, -1, 1, -1, -1, 1, 1, 1, 1, -1, -1,
1, 1, -1, 1, 1, 1, -1, 1, -1, 1, -1, 1, 1, 1, -1, -1, 1, -1, -1, 1, 1, 1, -1, -1, -1, -1, -1, 1, 1, 1, -1, 1,
1, 1, 1, -1, 1, 1, -1, 1, -1, 1, 1, -1, 1, 1, -1, -1, 1, -1, 1, -1, 1, 1, -1, -1, -1, -1, 1, -1, 1, 1, -1, 1,
1, 1, -1, -1, 1, 1, -1, -1, -1, 1, -1, -1, 1, 1, -1, 1, 1, -1, -1, -1, 1, 1, -1, 1, -1, -1, -1, -1, 1, 1, -1, -1,
1, 1, 1, 1, -1, 1, -1, 1, -1, 1, 1, 1, -1, 1, -1, -1, 1, -1, 1, 1, -1, 1, -1, -1, -1, -1, 1, 1, -1, 1, -1, 1,
1, 1, -1, 1, -1, 1, -1, -1, -1, 1, -1, 1, -1, 1, -1, 1, 1, -1, -1, 1, -1, 1, -1, 1, -1, -1, -1, 1, -1, 1, -1, -1,
1, 1, 1, -1, -1, 1, -1, -1, -1, 1, 1, -1, -1, 1, -1, 1, 1, -1, 1, -1, -1, 1, -1, 1, -1, -1, 1, -1, -1, 1, -1, -1,
1, 1, -1, -1, -1, 1, -1, 1, -1, 1, -1, -1, -1, 1, -1, -1, 1, -1, -1, -1, -1, 1, -1, -1, -1, -1, -1, -1, -1, 1, -1, 1,
1, 1, 1, 1, 1, -1, -1, 1, -1, 1, 1, 1, 1, -1, -1, -1, 1, -1, 1, 1, 1, -1, -1, -1, -1, -1, 1, 1, 1, -1, -1, 1,
1, 1, -1, 1, 1, -1, -1, -1, -1, 1, -1, 1, 1, -1, -1, 1, 1, -1, -1, 1, 1, -1, -1, 1, -1, -1, -1, 1, 1, -1, -1, -1,
1, 1, 1, -1, 1, -1, -1, -1, -1, 1, 1, -1, 1, -1, -1, 1, 1, -1, 1, -1, 1, -1, -1, 1, -1, -1, 1, -1, 1, -1, -1, -1,
1, 1, -1, -1, 1, -1, -1, 1, -1, 1, -1, -1, 1, -1, -1, -1, 1, -1, -1, -1, 1, -1, -1, -1, -1, -1, -1, -1, 1, -1, -1, 1,
1, 1, 1, 1, -1, -1, -1, -1, -1, 1, 1, 1, -1, -1, -1, 1, 1, -1, 1, 1, -1, -1, -1, 1, -1, -1, 1, 1, -1, -1, -1, -1,
1, 1, -1, 1, -1, -1, -1, 1, -1, 1, -1, 1, -1, -1, -1, -1, 1, -1, -1, 1, -1, -1, -1, -1, -1, -1, -1, 1, -1, -1, -1, 1,
1, 1, 1, -1, -1, -1, -1, 1, -1, 1, 1, -1, -1, -1, -1, -1, 1, -1, 1, -1, -1, -1, -1, -1, -1, -1, 1, -1, -1, -1, -1, 1,
1, 1, -1, -1, -1, -1, -1, -1, -1, 1, -1, -1, -1, -1, -1, 1, 1, -1, -1, -1, -1, -1, -1, 1, -1, -1, -1, -1, -1, -1, -1, -1,
};
void ggml_vec_dot_iq2_xxs_q8_K(const int n, float * restrict s, const void * restrict vx, const void * restrict vy) {
assert(n % QK_K == 0);
const block_iq2_xxs * restrict x = vx;
const block_q8_K * restrict y = vy;
const int nb = n / QK_K;
#if defined(__ARM_NEON)
const uint64_t * signs64 = (const uint64_t *)keven_signs_q2xs;
uint32_t aux32[4];
const uint8_t * aux8 = (const uint8_t *)aux32;
ggml : fix vld1q_s8_x4 32-bit compat (#4828) * ggml : fix vld1q_s8_x4 32-bit compat ggml-ci * ggml : fix 32-bit ARM compat (cont) ggml-ci
2024-01-09 08:42:06 +00:00
ggml_int8x16x4_t q2u;
ggml_int8x16x4_t q2s;
ggml_int8x16x4_t q8b;
SOTA 2-bit quants (#4773) * iq2_xxs: basics * iq2_xxs: scalar and AVX2 dot products Needed to change Q8_K to have quants in the -127...127 range, else the IQ2_XXS AVX implementation becomes very awkward. The alternative would have been to use Q8_0 instead. Perhaps I'll change later, for now this is what we have. * iq2_xxs: ARM_NEON dot product Somehow strangely slow (112 ms/token). * iq2_xxs: WIP Metal Dequantize works, something is still wrong with the dot product. * iq2_xxs: Metal dot product now works We have PP-512 = 475 t/s TG-128 = 47.3 t/s Not the greatest performance, but not complete garbage either. * iq2_xxs: slighty faster dot product TG-128 is now 48.4 t/s * iq2_xxs: slighty faster dot product TG-128 is now 50.9 t/s * iq2_xxs: even faster Metal dot product TG-128 is now 54.1 t/s. Strangely enough, putting the signs lookup table into shared memory has a bigger impact than the grid values being in shared memory. * iq2_xxs: dequantize CUDA kernel - fix conflict with master * iq2_xxs: quantized CUDA dot product (MMVQ) We get TG-128 = 153.1 t/s * iq2_xxs: slightly faster CUDA dot product TG-128 is now at 155.1 t/s. * iq2_xxs: add to llama ftype enum * iq2_xxs: fix MoE on Metal * Fix missing MMQ ops when on hipBLAS I had put the ggml_supports_mmq call at the wrong place. * Fix bug in qequantize_row_iq2_xxs The 0.25f factor was missing. Great detective work by @ggerganov! * Fixing tests * PR suggestion --------- Co-authored-by: Iwan Kawrakow <iwan.kawrakow@gmail.com>
2024-01-08 15:02:32 +00:00
float sumf = 0;
for (int i = 0; i < nb; ++i) {
const float d = GGML_FP16_TO_FP32(x[i].d) * y[i].d;
const uint16_t * restrict q2 = x[i].qs;
const int8_t * restrict q8 = y[i].qs;
float sumf1 = 0, sumf2 = 0;
for (int ib32 = 0; ib32 < QK_K/32; ib32 += 2) {
ggml : fix vld1q_s8_x4 32-bit compat (#4828) * ggml : fix vld1q_s8_x4 32-bit compat ggml-ci * ggml : fix 32-bit ARM compat (cont) ggml-ci
2024-01-09 08:42:06 +00:00
q8b = ggml_vld1q_s8_x4(q8); q8 += 64;
SOTA 2-bit quants (#4773) * iq2_xxs: basics * iq2_xxs: scalar and AVX2 dot products Needed to change Q8_K to have quants in the -127...127 range, else the IQ2_XXS AVX implementation becomes very awkward. The alternative would have been to use Q8_0 instead. Perhaps I'll change later, for now this is what we have. * iq2_xxs: ARM_NEON dot product Somehow strangely slow (112 ms/token). * iq2_xxs: WIP Metal Dequantize works, something is still wrong with the dot product. * iq2_xxs: Metal dot product now works We have PP-512 = 475 t/s TG-128 = 47.3 t/s Not the greatest performance, but not complete garbage either. * iq2_xxs: slighty faster dot product TG-128 is now 48.4 t/s * iq2_xxs: slighty faster dot product TG-128 is now 50.9 t/s * iq2_xxs: even faster Metal dot product TG-128 is now 54.1 t/s. Strangely enough, putting the signs lookup table into shared memory has a bigger impact than the grid values being in shared memory. * iq2_xxs: dequantize CUDA kernel - fix conflict with master * iq2_xxs: quantized CUDA dot product (MMVQ) We get TG-128 = 153.1 t/s * iq2_xxs: slightly faster CUDA dot product TG-128 is now at 155.1 t/s. * iq2_xxs: add to llama ftype enum * iq2_xxs: fix MoE on Metal * Fix missing MMQ ops when on hipBLAS I had put the ggml_supports_mmq call at the wrong place. * Fix bug in qequantize_row_iq2_xxs The 0.25f factor was missing. Great detective work by @ggerganov! * Fixing tests * PR suggestion --------- Co-authored-by: Iwan Kawrakow <iwan.kawrakow@gmail.com>
2024-01-08 15:02:32 +00:00
memcpy(aux32, q2, 4*sizeof(uint32_t)); q2 += 8;
q2u.val[0] = vcombine_s8(vld1_s8((const void *)(iq2xxs_grid + aux8[ 0])), vld1_s8((const void *)(iq2xxs_grid + aux8[ 1])));
q2u.val[1] = vcombine_s8(vld1_s8((const void *)(iq2xxs_grid + aux8[ 2])), vld1_s8((const void *)(iq2xxs_grid + aux8[ 3])));
q2u.val[2] = vcombine_s8(vld1_s8((const void *)(iq2xxs_grid + aux8[ 8])), vld1_s8((const void *)(iq2xxs_grid + aux8[ 9])));
q2u.val[3] = vcombine_s8(vld1_s8((const void *)(iq2xxs_grid + aux8[10])), vld1_s8((const void *)(iq2xxs_grid + aux8[11])));
q2s.val[0] = vcombine_s8(vld1_s8((const void *)(signs64 + ((aux32[1] >> 0) & 127))), vld1_s8((const void *)(signs64 + ((aux32[1] >> 7) & 127))));
q2s.val[1] = vcombine_s8(vld1_s8((const void *)(signs64 + ((aux32[1] >> 14) & 127))), vld1_s8((const void *)(signs64 + ((aux32[1] >> 21) & 127))));
q2s.val[2] = vcombine_s8(vld1_s8((const void *)(signs64 + ((aux32[3] >> 0) & 127))), vld1_s8((const void *)(signs64 + ((aux32[3] >> 7) & 127))));
q2s.val[3] = vcombine_s8(vld1_s8((const void *)(signs64 + ((aux32[3] >> 14) & 127))), vld1_s8((const void *)(signs64 + ((aux32[3] >> 21) & 127))));
q2u.val[0] = vmulq_s8(q2u.val[0], q2s.val[0]);
q2u.val[1] = vmulq_s8(q2u.val[1], q2s.val[1]);
q2u.val[2] = vmulq_s8(q2u.val[2], q2s.val[2]);
q2u.val[3] = vmulq_s8(q2u.val[3], q2s.val[3]);
const int32x4_t p1 = ggml_vdotq_s32(ggml_vdotq_s32(vdupq_n_s32(0), q2u.val[0], q8b.val[0]), q2u.val[1], q8b.val[1]);
const int32x4_t p2 = ggml_vdotq_s32(ggml_vdotq_s32(vdupq_n_s32(0), q2u.val[2], q8b.val[2]), q2u.val[3], q8b.val[3]);
sumf1 += vaddvq_s32(p1) * (0.5f + (aux32[1] >> 28));
sumf2 += vaddvq_s32(p2) * (0.5f + (aux32[3] >> 28));
}
sumf += d*(sumf1 + sumf2);
}
*s = 0.25f * sumf;
#elif defined(__AVX2__)
const uint64_t * signs64 = (const uint64_t *)keven_signs_q2xs;
uint32_t aux32[4];
const uint8_t * aux8 = (const uint8_t *)aux32;
__m256 accumf = _mm256_setzero_ps();
for (int i = 0; i < nb; ++i) {
const float d = GGML_FP16_TO_FP32(x[i].d) * y[i].d;
const uint16_t * restrict q2 = x[i].qs;
const int8_t * restrict q8 = y[i].qs;
__m256i sumi1 = _mm256_setzero_si256();
__m256i sumi2 = _mm256_setzero_si256();
for (int ib32 = 0; ib32 < QK_K/32; ib32 += 2) {
const __m256i q8_1 = _mm256_loadu_si256((const __m256i *)q8); q8 += 32;
const __m256i q8_2 = _mm256_loadu_si256((const __m256i *)q8); q8 += 32;
memcpy(aux32, q2, 4*sizeof(uint32_t)); q2 += 8;
const __m256i q2_1 = _mm256_set_epi64x(iq2xxs_grid[aux8[ 3]], iq2xxs_grid[aux8[ 2]], iq2xxs_grid[aux8[1]], iq2xxs_grid[aux8[0]]);
const __m256i q2_2 = _mm256_set_epi64x(iq2xxs_grid[aux8[11]], iq2xxs_grid[aux8[10]], iq2xxs_grid[aux8[9]], iq2xxs_grid[aux8[8]]);
const __m256i s2_1 = _mm256_set_epi64x(signs64[(aux32[1] >> 21) & 127], signs64[(aux32[1] >> 14) & 127],
signs64[(aux32[1] >> 7) & 127], signs64[(aux32[1] >> 0) & 127]);
const __m256i s2_2 = _mm256_set_epi64x(signs64[(aux32[3] >> 21) & 127], signs64[(aux32[3] >> 14) & 127],
signs64[(aux32[3] >> 7) & 127], signs64[(aux32[3] >> 0) & 127]);
const __m256i q8s_1 = _mm256_sign_epi8(q8_1, s2_1);
const __m256i q8s_2 = _mm256_sign_epi8(q8_2, s2_2);
const __m256i dot1 = _mm256_maddubs_epi16(q2_1, q8s_1);
const __m256i dot2 = _mm256_maddubs_epi16(q2_2, q8s_2);
const uint16_t ls1 = aux32[1] >> 28;
const uint16_t ls2 = aux32[3] >> 28;
const __m256i p1 = _mm256_madd_epi16(dot1, _mm256_set1_epi16(2*ls1+1));
const __m256i p2 = _mm256_madd_epi16(dot2, _mm256_set1_epi16(2*ls2+1));
sumi1 = _mm256_add_epi32(sumi1, p1);
sumi2 = _mm256_add_epi32(sumi2, p2);
}
accumf = _mm256_fmadd_ps(_mm256_set1_ps(d), _mm256_cvtepi32_ps(_mm256_add_epi32(sumi1, sumi2)), accumf);
}
*s = 0.125f * hsum_float_8(accumf);
#else
uint32_t aux32[2];
const uint8_t * aux8 = (const uint8_t *)aux32;
float sumf = 0.f;
for (int i = 0; i < nb; ++i) {
const float d = GGML_FP16_TO_FP32(x[i].d) * y[i].d;
const uint16_t * restrict q2 = x[i].qs;
const int8_t * restrict q8 = y[i].qs;
int32_t bsum = 0;
for (int ib32 = 0; ib32 < QK_K/32; ++ib32) {
memcpy(aux32, q2, 2*sizeof(uint32_t));
q2 += 4;
const uint32_t ls = 2*(aux32[1] >> 28) + 1;
int32_t sumi = 0;
for (int l = 0; l < 4; ++l) {
const uint8_t * grid = (const uint8_t *)(iq2xxs_grid + aux8[l]);
const uint8_t signs = ksigns_iq2xs[(aux32[1] >> 7*l) & 127];
for (int j = 0; j < 8; ++j) {
sumi += grid[j] * q8[j] * (signs & kmask_iq2xs[j] ? -1 : 1);
}
q8 += 8;
}
bsum += sumi * ls;
}
sumf += d * bsum;
}
*s = 0.125f * sumf;
#endif
}
ggml : SOTA 2-bit quants (add IQ2_XS) (#4856) * iq2_xs: basics * iq2_xs: this should have been in the basics * iq2_xs: CUDA and scalar CPU works * iq2_xs: WIP Metal * iq2_xs: Metal now works * iq2_xs: working, but dog slow, ARM_NEON dot product * iq2_xs: better ARM_NEON dot product We are now at 19.5 t/s for TG-128 and 61 t/s for PP-512 when running on the CPU. * iq2_xs: AVX2 dot product - 19.5 t/s * iq2_xs: faster AVX2 dit product 21.4 t/s for TG-128, 59.2 t/s for PP-512. The latter is 2x compared to the previous version. * iq2_xs: had forgotten to delete iq2-data.h * Add llama enum for IQ2_XS --------- Co-authored-by: Iwan Kawrakow <iwan.kawrakow@gmail.com>
2024-01-11 19:39:39 +00:00
void ggml_vec_dot_iq2_xs_q8_K(const int n, float * restrict s, const void * restrict vx, const void * restrict vy) {
assert(n % QK_K == 0);
const block_iq2_xs * restrict x = vx;
const block_q8_K * restrict y = vy;
const int nb = n / QK_K;
#if defined(__ARM_NEON)
const uint64_t * signs64 = (const uint64_t *)keven_signs_q2xs;
ggml : fix 32-bit ARM compat for IQ2_XS (whisper/1758) * ggml : fix 32-bit ARM compat * ggml : fix fix * ggml : fix fix fix
2024-01-12 12:02:30 +00:00
ggml_int8x16x4_t q2u;
ggml_int8x16x4_t q2s;
ggml_int8x16x4_t q8b;
ggml : SOTA 2-bit quants (add IQ2_XS) (#4856) * iq2_xs: basics * iq2_xs: this should have been in the basics * iq2_xs: CUDA and scalar CPU works * iq2_xs: WIP Metal * iq2_xs: Metal now works * iq2_xs: working, but dog slow, ARM_NEON dot product * iq2_xs: better ARM_NEON dot product We are now at 19.5 t/s for TG-128 and 61 t/s for PP-512 when running on the CPU. * iq2_xs: AVX2 dot product - 19.5 t/s * iq2_xs: faster AVX2 dit product 21.4 t/s for TG-128, 59.2 t/s for PP-512. The latter is 2x compared to the previous version. * iq2_xs: had forgotten to delete iq2-data.h * Add llama enum for IQ2_XS --------- Co-authored-by: Iwan Kawrakow <iwan.kawrakow@gmail.com>
2024-01-11 19:39:39 +00:00
int32x4x4_t scales32;
float sumf = 0;
for (int i = 0; i < nb; ++i) {
const float d = GGML_FP16_TO_FP32(x[i].d) * y[i].d;
const uint16_t * restrict q2 = x[i].qs;
const int8_t * restrict q8 = y[i].qs;
const uint8x8_t scales8 = vld1_u8(x[i].scales);
const uint8x8_t scales_l = vand_u8(scales8, vdup_n_u8(0xf));
const uint8x8_t scales_h = vshr_n_u8(scales8, 4);
uint8x16_t scales = vcombine_u8(vzip1_u8(scales_l, scales_h), vzip2_u8(scales_l, scales_h));
scales = vaddq_u8(vshlq_n_u8(scales, 1), vdupq_n_u8(1));
const uint16x8_t scales1 = vmovl_u8(vget_low_u8(scales));
const uint16x8_t scales2 = vmovl_u8(vget_high_u8(scales));
scales32.val[0] = vreinterpretq_s32_u32(vmovl_u16(vget_low_u16(scales1)));
scales32.val[1] = vreinterpretq_s32_u32(vmovl_u16(vget_high_u16(scales1)));
scales32.val[2] = vreinterpretq_s32_u32(vmovl_u16(vget_low_u16(scales2)));
scales32.val[3] = vreinterpretq_s32_u32(vmovl_u16(vget_high_u16(scales2)));
int32x4_t sumi = vdupq_n_s32(0);
for (int ib64 = 0; ib64 < QK_K/64; ++ib64) {
ggml : fix 32-bit ARM compat for IQ2_XS (whisper/1758) * ggml : fix 32-bit ARM compat * ggml : fix fix * ggml : fix fix fix
2024-01-12 12:02:30 +00:00
q8b = ggml_vld1q_s8_x4(q8); q8 += 64;
ggml : SOTA 2-bit quants (add IQ2_XS) (#4856) * iq2_xs: basics * iq2_xs: this should have been in the basics * iq2_xs: CUDA and scalar CPU works * iq2_xs: WIP Metal * iq2_xs: Metal now works * iq2_xs: working, but dog slow, ARM_NEON dot product * iq2_xs: better ARM_NEON dot product We are now at 19.5 t/s for TG-128 and 61 t/s for PP-512 when running on the CPU. * iq2_xs: AVX2 dot product - 19.5 t/s * iq2_xs: faster AVX2 dit product 21.4 t/s for TG-128, 59.2 t/s for PP-512. The latter is 2x compared to the previous version. * iq2_xs: had forgotten to delete iq2-data.h * Add llama enum for IQ2_XS --------- Co-authored-by: Iwan Kawrakow <iwan.kawrakow@gmail.com>
2024-01-11 19:39:39 +00:00
q2u.val[0] = vcombine_s8(vld1_s8((const void *)(iq2xs_grid + (q2[0] & 511))), vld1_s8((const void *)(iq2xs_grid + (q2[1] & 511))));
q2u.val[1] = vcombine_s8(vld1_s8((const void *)(iq2xs_grid + (q2[2] & 511))), vld1_s8((const void *)(iq2xs_grid + (q2[3] & 511))));
q2u.val[2] = vcombine_s8(vld1_s8((const void *)(iq2xs_grid + (q2[4] & 511))), vld1_s8((const void *)(iq2xs_grid + (q2[5] & 511))));
q2u.val[3] = vcombine_s8(vld1_s8((const void *)(iq2xs_grid + (q2[6] & 511))), vld1_s8((const void *)(iq2xs_grid + (q2[7] & 511))));
q2s.val[0] = vcombine_s8(vld1_s8((const void *)(signs64 + (q2[0] >> 9))), vld1_s8((const void *)(signs64 + (q2[1] >> 9))));
q2s.val[1] = vcombine_s8(vld1_s8((const void *)(signs64 + (q2[2] >> 9))), vld1_s8((const void *)(signs64 + (q2[3] >> 9))));
q2s.val[2] = vcombine_s8(vld1_s8((const void *)(signs64 + (q2[4] >> 9))), vld1_s8((const void *)(signs64 + (q2[5] >> 9))));
q2s.val[3] = vcombine_s8(vld1_s8((const void *)(signs64 + (q2[6] >> 9))), vld1_s8((const void *)(signs64 + (q2[7] >> 9))));
q2u.val[0] = vmulq_s8(q2u.val[0], q2s.val[0]);
q2u.val[1] = vmulq_s8(q2u.val[1], q2s.val[1]);
q2u.val[2] = vmulq_s8(q2u.val[2], q2s.val[2]);
q2u.val[3] = vmulq_s8(q2u.val[3], q2s.val[3]);
const int32x4_t p1 = ggml_vdotq_s32(vdupq_n_s32(0), q2u.val[0], q8b.val[0]);
const int32x4_t p2 = ggml_vdotq_s32(vdupq_n_s32(0), q2u.val[1], q8b.val[1]);
const int32x4_t p3 = ggml_vdotq_s32(vdupq_n_s32(0), q2u.val[2], q8b.val[2]);
const int32x4_t p4 = ggml_vdotq_s32(vdupq_n_s32(0), q2u.val[3], q8b.val[3]);
const int32x4_t p = vpaddq_s32(vpaddq_s32(p1, p2), vpaddq_s32(p3, p4));
sumi = vmlaq_s32(sumi, p, scales32.val[ib64]);
q2 += 8;
}
sumf += d*vaddvq_s32(sumi);
}
*s = 0.125f * sumf;
#elif defined(__AVX2__)
const __m128i m4 = _mm_set1_epi8(0xf);
const __m128i m1 = _mm_set1_epi8(1);
Faster AVX2 dot product for IQ2_XS (#5187) * iq2xs: faster AVX2 dot product * iq2xs: small AVX2 imrovement * Speed up computing sign bits in AVX2 iq2_xs dot product --------- Co-authored-by: Iwan Kawrakow <iwan.kawrakow@gmail.com> Co-authored-by: Peter Reid <peter@peterreid.net>
2024-01-30 13:15:07 +00:00
const __m256i m511 = _mm256_set1_epi16(511);
const __m256i mone = _mm256_set1_epi8(1);
ggml : SOTA 2-bit quants (add IQ2_XS) (#4856) * iq2_xs: basics * iq2_xs: this should have been in the basics * iq2_xs: CUDA and scalar CPU works * iq2_xs: WIP Metal * iq2_xs: Metal now works * iq2_xs: working, but dog slow, ARM_NEON dot product * iq2_xs: better ARM_NEON dot product We are now at 19.5 t/s for TG-128 and 61 t/s for PP-512 when running on the CPU. * iq2_xs: AVX2 dot product - 19.5 t/s * iq2_xs: faster AVX2 dit product 21.4 t/s for TG-128, 59.2 t/s for PP-512. The latter is 2x compared to the previous version. * iq2_xs: had forgotten to delete iq2-data.h * Add llama enum for IQ2_XS --------- Co-authored-by: Iwan Kawrakow <iwan.kawrakow@gmail.com>
2024-01-11 19:39:39 +00:00
Faster AVX2 dot product for IQ2_XS (#5187) * iq2xs: faster AVX2 dot product * iq2xs: small AVX2 imrovement * Speed up computing sign bits in AVX2 iq2_xs dot product --------- Co-authored-by: Iwan Kawrakow <iwan.kawrakow@gmail.com> Co-authored-by: Peter Reid <peter@peterreid.net>
2024-01-30 13:15:07 +00:00
static const uint8_t k_bit_helper[32] = {
0x00, 0x80, 0x80, 0x00, 0x80, 0x00, 0x00, 0x80, 0x80, 0x00, 0x00, 0x80, 0x00, 0x80, 0x80, 0x00,
0x00, 0x80, 0x80, 0x00, 0x80, 0x00, 0x00, 0x80, 0x80, 0x00, 0x00, 0x80, 0x00, 0x80, 0x80, 0x00,
};
static const char block_sign_shuffle_mask_1[32] = {
0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x02, 0x02, 0x02, 0x02, 0x02, 0x02, 0x02, 0x02,
0x04, 0x04, 0x04, 0x04, 0x04, 0x04, 0x04, 0x04, 0x06, 0x06, 0x06, 0x06, 0x06, 0x06, 0x06, 0x06,
};
static const char block_sign_shuffle_mask_2[32] = {
0x08, 0x08, 0x08, 0x08, 0x08, 0x08, 0x08, 0x08, 0x0a, 0x0a, 0x0a, 0x0a, 0x0a, 0x0a, 0x0a, 0x0a,
0x0c, 0x0c, 0x0c, 0x0c, 0x0c, 0x0c, 0x0c, 0x0c, 0x0e, 0x0e, 0x0e, 0x0e, 0x0e, 0x0e, 0x0e, 0x0e,
};
static const uint8_t bit_selector_mask_bytes[32] = {
0x01, 0x02, 0x04, 0x08, 0x10, 0x20, 0x40, 0x80, 0x01, 0x02, 0x04, 0x08, 0x10, 0x20, 0x40, 0x80,
0x01, 0x02, 0x04, 0x08, 0x10, 0x20, 0x40, 0x80, 0x01, 0x02, 0x04, 0x08, 0x10, 0x20, 0x40, 0x80,
};
const __m256i bit_helper = _mm256_loadu_si256((const __m256i*)k_bit_helper);
const __m256i bit_selector_mask = _mm256_loadu_si256((const __m256i*)bit_selector_mask_bytes);
const __m256i block_sign_shuffle_1 = _mm256_loadu_si256((const __m256i*)block_sign_shuffle_mask_1);
const __m256i block_sign_shuffle_2 = _mm256_loadu_si256((const __m256i*)block_sign_shuffle_mask_2);
ggml : SOTA 2-bit quants (add IQ2_XS) (#4856) * iq2_xs: basics * iq2_xs: this should have been in the basics * iq2_xs: CUDA and scalar CPU works * iq2_xs: WIP Metal * iq2_xs: Metal now works * iq2_xs: working, but dog slow, ARM_NEON dot product * iq2_xs: better ARM_NEON dot product We are now at 19.5 t/s for TG-128 and 61 t/s for PP-512 when running on the CPU. * iq2_xs: AVX2 dot product - 19.5 t/s * iq2_xs: faster AVX2 dit product 21.4 t/s for TG-128, 59.2 t/s for PP-512. The latter is 2x compared to the previous version. * iq2_xs: had forgotten to delete iq2-data.h * Add llama enum for IQ2_XS --------- Co-authored-by: Iwan Kawrakow <iwan.kawrakow@gmail.com>
2024-01-11 19:39:39 +00:00
uint64_t aux64;
// somewhat hacky, but gives a significant boost in performance
Faster AVX2 dot product for IQ2_XS (#5187) * iq2xs: faster AVX2 dot product * iq2xs: small AVX2 imrovement * Speed up computing sign bits in AVX2 iq2_xs dot product --------- Co-authored-by: Iwan Kawrakow <iwan.kawrakow@gmail.com> Co-authored-by: Peter Reid <peter@peterreid.net>
2024-01-30 13:15:07 +00:00
__m256i aux_gindex;
ggml : SOTA 2-bit quants (add IQ2_XS) (#4856) * iq2_xs: basics * iq2_xs: this should have been in the basics * iq2_xs: CUDA and scalar CPU works * iq2_xs: WIP Metal * iq2_xs: Metal now works * iq2_xs: working, but dog slow, ARM_NEON dot product * iq2_xs: better ARM_NEON dot product We are now at 19.5 t/s for TG-128 and 61 t/s for PP-512 when running on the CPU. * iq2_xs: AVX2 dot product - 19.5 t/s * iq2_xs: faster AVX2 dit product 21.4 t/s for TG-128, 59.2 t/s for PP-512. The latter is 2x compared to the previous version. * iq2_xs: had forgotten to delete iq2-data.h * Add llama enum for IQ2_XS --------- Co-authored-by: Iwan Kawrakow <iwan.kawrakow@gmail.com>
2024-01-11 19:39:39 +00:00
const uint16_t * gindex = (const uint16_t *)&aux_gindex;
__m256 accumf = _mm256_setzero_ps();
for (int i = 0; i < nb; ++i) {
const float d = GGML_FP16_TO_FP32(x[i].d) * y[i].d;
const uint16_t * restrict q2 = x[i].qs;
const int8_t * restrict q8 = y[i].qs;
memcpy(&aux64, x[i].scales, 8);
__m128i stmp = _mm_set1_epi64x(aux64);
stmp = _mm_unpacklo_epi8(_mm_and_si128(stmp, m4), _mm_and_si128(_mm_srli_epi16(stmp, 4), m4));
const __m128i scales = _mm_add_epi8(_mm_slli_epi16(stmp, 1), m1);
__m256i sumi1 = _mm256_setzero_si256();
__m256i sumi2 = _mm256_setzero_si256();
Faster AVX2 dot product for IQ2_XS (#5187) * iq2xs: faster AVX2 dot product * iq2xs: small AVX2 imrovement * Speed up computing sign bits in AVX2 iq2_xs dot product --------- Co-authored-by: Iwan Kawrakow <iwan.kawrakow@gmail.com> Co-authored-by: Peter Reid <peter@peterreid.net>
2024-01-30 13:15:07 +00:00
for (int ib32 = 0; ib32 < QK_K/32; ib32 += 4) {
const __m256i q2_data = _mm256_loadu_si256((const __m256i*)q2); q2 += 16;
aux_gindex = _mm256_and_si256(q2_data, m511);
const __m256i partial_sign_bits = _mm256_srli_epi16(q2_data, 9);
const __m256i partial_sign_bits_upper = _mm256_srli_epi16(q2_data, 13);
const __m256i partial_sign_bits_for_counting = _mm256_xor_si256(partial_sign_bits, partial_sign_bits_upper);
const __m256i odd_bits = _mm256_shuffle_epi8(bit_helper, partial_sign_bits_for_counting);
const __m256i full_sign_bits = _mm256_or_si256(partial_sign_bits, odd_bits);
ggml : SOTA 2-bit quants (add IQ2_XS) (#4856) * iq2_xs: basics * iq2_xs: this should have been in the basics * iq2_xs: CUDA and scalar CPU works * iq2_xs: WIP Metal * iq2_xs: Metal now works * iq2_xs: working, but dog slow, ARM_NEON dot product * iq2_xs: better ARM_NEON dot product We are now at 19.5 t/s for TG-128 and 61 t/s for PP-512 when running on the CPU. * iq2_xs: AVX2 dot product - 19.5 t/s * iq2_xs: faster AVX2 dit product 21.4 t/s for TG-128, 59.2 t/s for PP-512. The latter is 2x compared to the previous version. * iq2_xs: had forgotten to delete iq2-data.h * Add llama enum for IQ2_XS --------- Co-authored-by: Iwan Kawrakow <iwan.kawrakow@gmail.com>
2024-01-11 19:39:39 +00:00
const __m256i q8_1 = _mm256_loadu_si256((const __m256i *)q8); q8 += 32;
const __m256i q8_2 = _mm256_loadu_si256((const __m256i *)q8); q8 += 32;
Faster AVX2 dot product for IQ2_XS (#5187) * iq2xs: faster AVX2 dot product * iq2xs: small AVX2 imrovement * Speed up computing sign bits in AVX2 iq2_xs dot product --------- Co-authored-by: Iwan Kawrakow <iwan.kawrakow@gmail.com> Co-authored-by: Peter Reid <peter@peterreid.net>
2024-01-30 13:15:07 +00:00
const __m256i q8_3 = _mm256_loadu_si256((const __m256i *)q8); q8 += 32;
const __m256i q8_4 = _mm256_loadu_si256((const __m256i *)q8); q8 += 32;
const __m256i q2_1 = _mm256_set_epi64x(iq2xs_grid[gindex[ 3]], iq2xs_grid[gindex[ 2]],
iq2xs_grid[gindex[ 1]], iq2xs_grid[gindex[ 0]]);
const __m256i q2_2 = _mm256_set_epi64x(iq2xs_grid[gindex[ 7]], iq2xs_grid[gindex[ 6]],
iq2xs_grid[gindex[ 5]], iq2xs_grid[gindex[ 4]]);
const __m256i q2_3 = _mm256_set_epi64x(iq2xs_grid[gindex[11]], iq2xs_grid[gindex[10]],
iq2xs_grid[gindex[ 9]], iq2xs_grid[gindex[ 8]]);
const __m256i q2_4 = _mm256_set_epi64x(iq2xs_grid[gindex[15]], iq2xs_grid[gindex[14]],
iq2xs_grid[gindex[13]], iq2xs_grid[gindex[12]]);
const __m128i full_signs_l = _mm256_castsi256_si128(full_sign_bits);
const __m128i full_signs_h = _mm256_extractf128_si256(full_sign_bits, 1);
const __m256i full_signs_1 = _mm256_set_m128i(full_signs_l, full_signs_l);
const __m256i full_signs_2 = _mm256_set_m128i(full_signs_h, full_signs_h);
__m256i signs;
signs = _mm256_shuffle_epi8(full_signs_1, block_sign_shuffle_1);
signs = _mm256_cmpeq_epi8(_mm256_and_si256(signs, bit_selector_mask), bit_selector_mask);
const __m256i q8s_1 = _mm256_sign_epi8(q8_1, _mm256_or_si256(signs, mone));
signs = _mm256_shuffle_epi8(full_signs_1, block_sign_shuffle_2);
signs = _mm256_cmpeq_epi8(_mm256_and_si256(signs, bit_selector_mask), bit_selector_mask);
const __m256i q8s_2 = _mm256_sign_epi8(q8_2, _mm256_or_si256(signs, mone));
signs = _mm256_shuffle_epi8(full_signs_2, block_sign_shuffle_1);
signs = _mm256_cmpeq_epi8(_mm256_and_si256(signs, bit_selector_mask), bit_selector_mask);
const __m256i q8s_3 = _mm256_sign_epi8(q8_3, _mm256_or_si256(signs, mone));
signs = _mm256_shuffle_epi8(full_signs_2, block_sign_shuffle_2);
signs = _mm256_cmpeq_epi8(_mm256_and_si256(signs, bit_selector_mask), bit_selector_mask);
const __m256i q8s_4 = _mm256_sign_epi8(q8_4, _mm256_or_si256(signs, mone));
ggml : SOTA 2-bit quants (add IQ2_XS) (#4856) * iq2_xs: basics * iq2_xs: this should have been in the basics * iq2_xs: CUDA and scalar CPU works * iq2_xs: WIP Metal * iq2_xs: Metal now works * iq2_xs: working, but dog slow, ARM_NEON dot product * iq2_xs: better ARM_NEON dot product We are now at 19.5 t/s for TG-128 and 61 t/s for PP-512 when running on the CPU. * iq2_xs: AVX2 dot product - 19.5 t/s * iq2_xs: faster AVX2 dit product 21.4 t/s for TG-128, 59.2 t/s for PP-512. The latter is 2x compared to the previous version. * iq2_xs: had forgotten to delete iq2-data.h * Add llama enum for IQ2_XS --------- Co-authored-by: Iwan Kawrakow <iwan.kawrakow@gmail.com>
2024-01-11 19:39:39 +00:00
const __m256i dot1 = _mm256_maddubs_epi16(q2_1, q8s_1);
const __m256i dot2 = _mm256_maddubs_epi16(q2_2, q8s_2);
Faster AVX2 dot product for IQ2_XS (#5187) * iq2xs: faster AVX2 dot product * iq2xs: small AVX2 imrovement * Speed up computing sign bits in AVX2 iq2_xs dot product --------- Co-authored-by: Iwan Kawrakow <iwan.kawrakow@gmail.com> Co-authored-by: Peter Reid <peter@peterreid.net>
2024-01-30 13:15:07 +00:00
const __m256i dot3 = _mm256_maddubs_epi16(q2_3, q8s_3);
const __m256i dot4 = _mm256_maddubs_epi16(q2_4, q8s_4);
ggml : SOTA 2-bit quants (add IQ2_XS) (#4856) * iq2_xs: basics * iq2_xs: this should have been in the basics * iq2_xs: CUDA and scalar CPU works * iq2_xs: WIP Metal * iq2_xs: Metal now works * iq2_xs: working, but dog slow, ARM_NEON dot product * iq2_xs: better ARM_NEON dot product We are now at 19.5 t/s for TG-128 and 61 t/s for PP-512 when running on the CPU. * iq2_xs: AVX2 dot product - 19.5 t/s * iq2_xs: faster AVX2 dit product 21.4 t/s for TG-128, 59.2 t/s for PP-512. The latter is 2x compared to the previous version. * iq2_xs: had forgotten to delete iq2-data.h * Add llama enum for IQ2_XS --------- Co-authored-by: Iwan Kawrakow <iwan.kawrakow@gmail.com>
2024-01-11 19:39:39 +00:00
const __m256i sc1 = _mm256_cvtepi8_epi16(_mm_shuffle_epi8(scales, get_scale_shuffle(ib32+0)));
const __m256i sc2 = _mm256_cvtepi8_epi16(_mm_shuffle_epi8(scales, get_scale_shuffle(ib32+1)));
Faster AVX2 dot product for IQ2_XS (#5187) * iq2xs: faster AVX2 dot product * iq2xs: small AVX2 imrovement * Speed up computing sign bits in AVX2 iq2_xs dot product --------- Co-authored-by: Iwan Kawrakow <iwan.kawrakow@gmail.com> Co-authored-by: Peter Reid <peter@peterreid.net>
2024-01-30 13:15:07 +00:00
const __m256i sc3 = _mm256_cvtepi8_epi16(_mm_shuffle_epi8(scales, get_scale_shuffle(ib32+2)));
const __m256i sc4 = _mm256_cvtepi8_epi16(_mm_shuffle_epi8(scales, get_scale_shuffle(ib32+3)));
ggml : SOTA 2-bit quants (add IQ2_XS) (#4856) * iq2_xs: basics * iq2_xs: this should have been in the basics * iq2_xs: CUDA and scalar CPU works * iq2_xs: WIP Metal * iq2_xs: Metal now works * iq2_xs: working, but dog slow, ARM_NEON dot product * iq2_xs: better ARM_NEON dot product We are now at 19.5 t/s for TG-128 and 61 t/s for PP-512 when running on the CPU. * iq2_xs: AVX2 dot product - 19.5 t/s * iq2_xs: faster AVX2 dit product 21.4 t/s for TG-128, 59.2 t/s for PP-512. The latter is 2x compared to the previous version. * iq2_xs: had forgotten to delete iq2-data.h * Add llama enum for IQ2_XS --------- Co-authored-by: Iwan Kawrakow <iwan.kawrakow@gmail.com>
2024-01-11 19:39:39 +00:00
sumi1 = _mm256_add_epi32(sumi1, _mm256_madd_epi16(dot1, sc1));
sumi2 = _mm256_add_epi32(sumi2, _mm256_madd_epi16(dot2, sc2));
Faster AVX2 dot product for IQ2_XS (#5187) * iq2xs: faster AVX2 dot product * iq2xs: small AVX2 imrovement * Speed up computing sign bits in AVX2 iq2_xs dot product --------- Co-authored-by: Iwan Kawrakow <iwan.kawrakow@gmail.com> Co-authored-by: Peter Reid <peter@peterreid.net>
2024-01-30 13:15:07 +00:00
sumi1 = _mm256_add_epi32(sumi1, _mm256_madd_epi16(dot3, sc3));
sumi2 = _mm256_add_epi32(sumi2, _mm256_madd_epi16(dot4, sc4));
ggml : SOTA 2-bit quants (add IQ2_XS) (#4856) * iq2_xs: basics * iq2_xs: this should have been in the basics * iq2_xs: CUDA and scalar CPU works * iq2_xs: WIP Metal * iq2_xs: Metal now works * iq2_xs: working, but dog slow, ARM_NEON dot product * iq2_xs: better ARM_NEON dot product We are now at 19.5 t/s for TG-128 and 61 t/s for PP-512 when running on the CPU. * iq2_xs: AVX2 dot product - 19.5 t/s * iq2_xs: faster AVX2 dit product 21.4 t/s for TG-128, 59.2 t/s for PP-512. The latter is 2x compared to the previous version. * iq2_xs: had forgotten to delete iq2-data.h * Add llama enum for IQ2_XS --------- Co-authored-by: Iwan Kawrakow <iwan.kawrakow@gmail.com>
2024-01-11 19:39:39 +00:00
}
accumf = _mm256_fmadd_ps(_mm256_set1_ps(d), _mm256_cvtepi32_ps(_mm256_add_epi32(sumi1, sumi2)), accumf);
}
*s = 0.125f * hsum_float_8(accumf);
#else
float sumf = 0.f;
for (int i = 0; i < nb; ++i) {
const float d = GGML_FP16_TO_FP32(x[i].d) * y[i].d;
const uint16_t * restrict q2 = x[i].qs;
const uint8_t * restrict sc = x[i].scales;
const int8_t * restrict q8 = y[i].qs;
int32_t bsum = 0;
for (int ib32 = 0; ib32 < QK_K/32; ++ib32) {
const uint16_t ls1 = 2*(sc[ib32] & 0xf) + 1;
const uint16_t ls2 = 2*(sc[ib32] >> 4) + 1;
int32_t sumi = 0;
for (int l = 0; l < 2; ++l) {
const uint8_t * grid = (const uint8_t *)(iq2xs_grid + (q2[l] & 511));
const uint8_t signs = ksigns_iq2xs[q2[l] >> 9];
for (int j = 0; j < 8; ++j) {
sumi += grid[j] * q8[j] * (signs & kmask_iq2xs[j] ? -1 : 1);
}
q8 += 8;
}
bsum += sumi * ls1;
sumi = 0;
for (int l = 2; l < 4; ++l) {
const uint8_t * grid = (const uint8_t *)(iq2xs_grid + (q2[l] & 511));
const uint8_t signs = ksigns_iq2xs[q2[l] >> 9];
for (int j = 0; j < 8; ++j) {
sumi += grid[j] * q8[j] * (signs & kmask_iq2xs[j] ? -1 : 1);
}
q8 += 8;
}
bsum += sumi * ls2;
q2 += 4;
}
sumf += d * bsum;
}
*s = 0.125f * sumf;
#endif
}
2-bit quantizations (#4897) * imatrix: load * imatrix: WIP * imatrix: Add Q2_K quantization * imatrix: also guard against Q2_K_S quantization without importance matrix * imatrix: guard even more against low-bit quantization misuse --------- Co-authored-by: Iwan Kawrakow <iwan.kawrakow@gmail.com>
2024-01-14 07:45:56 +00:00
SOTA 3-bit quants (#5196) * iq3_xxs: quantize/dequantize RMSE seems a bit high-ish at about half-way between q2_K and q3_K, so need to check more. * iq3_xxs: CUDA dequantize works * iq2_xxs: tuning quantization * iq3_xxs: starting to look better PPL on wiki.test.raw LLaMA-v1-7B: 6.4218 LLaMA-v2-7B: 6.3560 Mistral-7B : 6.0717 This is better than Q3_K_XS, with a 5% reduction in quantized model size. * iq3_xxs: CUDA dot product We have PP-512: 5891 t/s TG-128: 143.9 t/s * iq3_xxs: scalar and AVX2 dot products * iq3_xxs: ARM_NEON and Metal Metal performance is decent, ARM_NEON is pathetic * iq3_xxs: slightly better grid points * Faster iq3_xxs and iq2_xs dot products on CUDA * iq3_xxs: add some quant mix * iq3_xxs: fix failing quantization test Dot product still fails. Is this real? * iq3_xxs: hopefully fix ROCm * iq3_xxs: failing tests This time the dot product accuracy did find an actual bug in the AVX2 implementation. * Add IQ3_XXS to test-backend-ops --------- Co-authored-by: Iwan Kawrakow <iwan.kawrakow@gmail.com>
2024-01-30 13:14:12 +00:00
// TODO
void ggml_vec_dot_iq3_xxs_q8_K(const int n, float * restrict s, const void * restrict vx, const void * restrict vy) {
assert(n % QK_K == 0);
const block_iq3_xxs * restrict x = vx;
const block_q8_K * restrict y = vy;
const int nb = n / QK_K;
#if defined(__ARM_NEON)
const uint64_t * signs64 = (const uint64_t *)keven_signs_q2xs;
uint32_t aux32[2];
ggml_int8x16x4_t q3s;
ggml_int8x16x4_t q8b;
float sumf = 0;
for (int i = 0; i < nb; ++i) {
const float d = GGML_FP16_TO_FP32(x[i].d) * y[i].d;
const uint8_t * restrict q3 = x[i].qs;
const uint8_t * restrict gas = x[i].qs + QK_K/4;
const int8_t * restrict q8 = y[i].qs;
float sumf1 = 0, sumf2 = 0;
for (int ib32 = 0; ib32 < QK_K/32; ib32 += 2) {
q8b = ggml_vld1q_s8_x4(q8); q8 += 64;
memcpy(aux32, gas, 2*sizeof(uint32_t)); gas += 2*sizeof(uint32_t);
const uint32x4_t aux32x4_0 = {iq3xxs_grid[q3[ 0]], iq3xxs_grid[q3[ 1]], iq3xxs_grid[q3[ 2]], iq3xxs_grid[q3[ 3]]};
const uint32x4_t aux32x4_1 = {iq3xxs_grid[q3[ 4]], iq3xxs_grid[q3[ 5]], iq3xxs_grid[q3[ 6]], iq3xxs_grid[q3[ 7]]};
const uint32x4_t aux32x4_2 = {iq3xxs_grid[q3[ 8]], iq3xxs_grid[q3[ 9]], iq3xxs_grid[q3[10]], iq3xxs_grid[q3[11]]};
const uint32x4_t aux32x4_3 = {iq3xxs_grid[q3[12]], iq3xxs_grid[q3[13]], iq3xxs_grid[q3[14]], iq3xxs_grid[q3[15]]};
q3 += 16;
q3s.val[0] = vcombine_s8(vld1_s8((const void *)(signs64 + ((aux32[0] >> 0) & 127))), vld1_s8((const void *)(signs64 + ((aux32[0] >> 7) & 127))));
q3s.val[1] = vcombine_s8(vld1_s8((const void *)(signs64 + ((aux32[0] >> 14) & 127))), vld1_s8((const void *)(signs64 + ((aux32[0] >> 21) & 127))));
q3s.val[2] = vcombine_s8(vld1_s8((const void *)(signs64 + ((aux32[1] >> 0) & 127))), vld1_s8((const void *)(signs64 + ((aux32[1] >> 7) & 127))));
q3s.val[3] = vcombine_s8(vld1_s8((const void *)(signs64 + ((aux32[1] >> 14) & 127))), vld1_s8((const void *)(signs64 + ((aux32[1] >> 21) & 127))));
q3s.val[0] = vmulq_s8(q3s.val[0], vreinterpretq_s8_u32(aux32x4_0));
q3s.val[1] = vmulq_s8(q3s.val[1], vreinterpretq_s8_u32(aux32x4_1));
q3s.val[2] = vmulq_s8(q3s.val[2], vreinterpretq_s8_u32(aux32x4_2));
q3s.val[3] = vmulq_s8(q3s.val[3], vreinterpretq_s8_u32(aux32x4_3));
const int32x4_t p1 = ggml_vdotq_s32(ggml_vdotq_s32(vdupq_n_s32(0), q3s.val[0], q8b.val[0]), q3s.val[1], q8b.val[1]);
const int32x4_t p2 = ggml_vdotq_s32(ggml_vdotq_s32(vdupq_n_s32(0), q3s.val[2], q8b.val[2]), q3s.val[3], q8b.val[3]);
sumf1 += vaddvq_s32(p1) * (0.5f + (aux32[0] >> 28));
sumf2 += vaddvq_s32(p2) * (0.5f + (aux32[1] >> 28));
}
sumf += d*(sumf1 + sumf2);
}
*s = 0.5f * sumf;
#elif defined(__AVX2__)
const uint64_t * signs64 = (const uint64_t *)keven_signs_q2xs;
uint32_t aux32[2];
__m256 accumf = _mm256_setzero_ps();
for (int i = 0; i < nb; ++i) {
const float d = GGML_FP16_TO_FP32(x[i].d) * y[i].d;
const uint8_t * restrict q3 = x[i].qs;
const uint8_t * restrict gas = x[i].qs + QK_K/4;
const int8_t * restrict q8 = y[i].qs;
__m256i sumi1 = _mm256_setzero_si256();
__m256i sumi2 = _mm256_setzero_si256();
for (int ib32 = 0; ib32 < QK_K/32; ib32 += 2) {
const __m256i q8_1 = _mm256_loadu_si256((const __m256i *)q8); q8 += 32;
const __m256i q8_2 = _mm256_loadu_si256((const __m256i *)q8); q8 += 32;
const __m256i q2_1 = _mm256_set_epi32(iq3xxs_grid[q3[7]], iq3xxs_grid[q3[6]], iq3xxs_grid[q3[5]], iq3xxs_grid[q3[4]],
iq3xxs_grid[q3[3]], iq3xxs_grid[q3[2]], iq3xxs_grid[q3[1]], iq3xxs_grid[q3[0]]);
q3 += 8;
const __m256i q2_2 = _mm256_set_epi32(iq3xxs_grid[q3[7]], iq3xxs_grid[q3[6]], iq3xxs_grid[q3[5]], iq3xxs_grid[q3[4]],
iq3xxs_grid[q3[3]], iq3xxs_grid[q3[2]], iq3xxs_grid[q3[1]], iq3xxs_grid[q3[0]]);
q3 += 8;
memcpy(aux32, gas, 8); gas += 8;
const __m256i s2_1 = _mm256_set_epi64x(signs64[(aux32[0] >> 21) & 127], signs64[(aux32[0] >> 14) & 127],
signs64[(aux32[0] >> 7) & 127], signs64[(aux32[0] >> 0) & 127]);
const __m256i s2_2 = _mm256_set_epi64x(signs64[(aux32[1] >> 21) & 127], signs64[(aux32[1] >> 14) & 127],
signs64[(aux32[1] >> 7) & 127], signs64[(aux32[1] >> 0) & 127]);
const __m256i q8s_1 = _mm256_sign_epi8(q8_1, s2_1);
const __m256i q8s_2 = _mm256_sign_epi8(q8_2, s2_2);
const __m256i dot1 = _mm256_maddubs_epi16(q2_1, q8s_1);
const __m256i dot2 = _mm256_maddubs_epi16(q2_2, q8s_2);
const uint16_t ls1 = aux32[0] >> 28;
const uint16_t ls2 = aux32[1] >> 28;
const __m256i p1 = _mm256_madd_epi16(dot1, _mm256_set1_epi16(2*ls1+1));
const __m256i p2 = _mm256_madd_epi16(dot2, _mm256_set1_epi16(2*ls2+1));
sumi1 = _mm256_add_epi32(sumi1, p1);
sumi2 = _mm256_add_epi32(sumi2, p2);
}
accumf = _mm256_fmadd_ps(_mm256_set1_ps(d), _mm256_cvtepi32_ps(_mm256_add_epi32(sumi1, sumi2)), accumf);
}
*s = 0.25f * hsum_float_8(accumf);
#else
uint32_t aux32;
float sumf = 0.f;
for (int i = 0; i < nb; ++i) {
const float d = GGML_FP16_TO_FP32(x[i].d) * y[i].d;
const uint8_t * restrict q3 = x[i].qs;
const uint8_t * restrict gas = x[i].qs + QK_K/4;
const int8_t * restrict q8 = y[i].qs;
int32_t bsum = 0;
for (int ib32 = 0; ib32 < QK_K/32; ++ib32) {
memcpy(&aux32, gas, sizeof(uint32_t)); gas += sizeof(uint32_t);
const uint32_t ls = 2*(aux32 >> 28) + 1;
int32_t sumi = 0;
for (int l = 0; l < 4; ++l) {
const uint8_t * grid1 = (const uint8_t *)(iq3xxs_grid + q3[2*l+0]);
const uint8_t * grid2 = (const uint8_t *)(iq3xxs_grid + q3[2*l+1]);
const uint8_t signs = ksigns_iq2xs[(aux32 >> 7*l) & 127];
for (int j = 0; j < 4; ++j) {
sumi += grid1[j] * q8[j+0] * (signs & kmask_iq2xs[j+0] ? -1 : 1);
sumi += grid2[j] * q8[j+4] * (signs & kmask_iq2xs[j+4] ? -1 : 1);
}
q8 += 8;
}
q3 += 8;
bsum += sumi * ls;
}
sumf += d * bsum;
}
*s = 0.25f * sumf;
#endif
}
2-bit quantizations (#4897) * imatrix: load * imatrix: WIP * imatrix: Add Q2_K quantization * imatrix: also guard against Q2_K_S quantization without importance matrix * imatrix: guard even more against low-bit quantization misuse --------- Co-authored-by: Iwan Kawrakow <iwan.kawrakow@gmail.com>
2024-01-14 07:45:56 +00:00
// ================================ IQ2 quantization =============================================
typedef struct {
uint64_t * grid;
int * map;
uint16_t * neighbours;
} iq2_entry_t;
static iq2_entry_t iq2_data[2] = {
{NULL, NULL, NULL},
{NULL, NULL, NULL},
};
static inline int iq2_data_index(int grid_size) {
GGML_ASSERT(grid_size == 256 || grid_size == 512);
return grid_size == 256 ? 0 : 1;
}
static int iq2_compare_func(const void * left, const void * right) {
const int * l = (const int *)left;
const int * r = (const int *)right;
return l[0] < r[0] ? -1 : l[0] > r[0] ? 1 : l[1] < r[1] ? -1 : l[1] > r[1] ? 1 : 0;
}
ggml : add IQ2 to test-backend-ops + refactoring (#4990) * ggml : add IQ2 to test-backend-ops + refactoring ggml-ci * cuda : update supports_op for IQ2 ggml-ci * ci : enable LLAMA_CUBLAS=1 for CUDA nodes ggml-ci * cuda : fix out-of-bounds-access in `mul_mat_vec_q` ggml-ci * tests : avoid creating RNGs for each Q tensor ggml-ci * tests : avoid creating RNGs for each tensor ggml-ci
2024-01-17 16:54:56 +00:00
void iq2xs_init_impl(int grid_size) {
2-bit quantizations (#4897) * imatrix: load * imatrix: WIP * imatrix: Add Q2_K quantization * imatrix: also guard against Q2_K_S quantization without importance matrix * imatrix: guard even more against low-bit quantization misuse --------- Co-authored-by: Iwan Kawrakow <iwan.kawrakow@gmail.com>
2024-01-14 07:45:56 +00:00
const int gindex = iq2_data_index(grid_size);
if (iq2_data[gindex].grid) {
return;
}
static const uint16_t kgrid_256[256] = {
0, 2, 5, 8, 10, 17, 20, 32, 34, 40, 42, 65, 68, 80, 88, 97,
100, 128, 130, 138, 162, 257, 260, 272, 277, 320, 388, 408, 512, 514, 546, 642,
1025, 1028, 1040, 1057, 1060, 1088, 1090, 1096, 1120, 1153, 1156, 1168, 1188, 1280, 1282, 1288,
1312, 1350, 1385, 1408, 1425, 1545, 1552, 1600, 1668, 1700, 2048, 2053, 2056, 2068, 2088, 2113,
2116, 2128, 2130, 2184, 2308, 2368, 2562, 2580, 4097, 4100, 4112, 4129, 4160, 4192, 4228, 4240,
4245, 4352, 4360, 4384, 4432, 4442, 4480, 4644, 4677, 5120, 5128, 5152, 5157, 5193, 5248, 5400,
5474, 5632, 5654, 6145, 6148, 6160, 6208, 6273, 6400, 6405, 6560, 6737, 8192, 8194, 8202, 8260,
8289, 8320, 8322, 8489, 8520, 8704, 8706, 9217, 9220, 9232, 9280, 9302, 9472, 9537, 9572, 9872,
10248, 10272, 10388, 10820, 16385, 16388, 16400, 16408, 16417, 16420, 16448, 16456, 16470, 16480, 16513, 16516,
16528, 16640, 16672, 16737, 16768, 16773, 16897, 16912, 16968, 16982, 17000, 17408, 17416, 17440, 17536, 17561,
17682, 17700, 17920, 18433, 18436, 18448, 18496, 18501, 18688, 18776, 18785, 18818, 19013, 19088, 20480, 20488,
20497, 20505, 20512, 20608, 20616, 20740, 20802, 20900, 21137, 21648, 21650, 21770, 22017, 22100, 22528, 22545,
22553, 22628, 22848, 23048, 24580, 24592, 24640, 24680, 24832, 24917, 25112, 25184, 25600, 25605, 25872, 25874,
25988, 26690, 32768, 32770, 32778, 32833, 32898, 33028, 33048, 33088, 33297, 33793, 33796, 33808, 33813, 33856,
33888, 34048, 34118, 34196, 34313, 34368, 34400, 34818, 35076, 35345, 36868, 36880, 36900, 36928, 37025, 37142,
37248, 37445, 37888, 37922, 37956, 38225, 39041, 39200, 40962, 41040, 41093, 41225, 41472, 42008, 43088, 43268,
};
static const uint16_t kgrid_512[512] = {
0, 2, 5, 8, 10, 17, 20, 22, 25, 32, 34, 37, 40, 65, 68, 70,
73, 80, 82, 85, 88, 97, 100, 128, 130, 133, 136, 145, 148, 153, 160, 257,
260, 262, 265, 272, 274, 277, 280, 282, 289, 292, 320, 322, 325, 328, 337, 340,
352, 360, 385, 388, 400, 512, 514, 517, 520, 529, 532, 544, 577, 580, 592, 597,
640, 650, 1025, 1028, 1030, 1033, 1040, 1042, 1045, 1048, 1057, 1060, 1088, 1090, 1093, 1096,
1105, 1108, 1110, 1120, 1153, 1156, 1168, 1280, 1282, 1285, 1288, 1297, 1300, 1312, 1345, 1348,
1360, 1377, 1408, 1537, 1540, 1552, 1574, 1600, 1602, 1668, 2048, 2050, 2053, 2056, 2058, 2065,
2068, 2080, 2085, 2113, 2116, 2128, 2136, 2176, 2208, 2218, 2305, 2308, 2320, 2368, 2433, 2441,
2560, 2592, 2600, 2710, 2720, 4097, 4100, 4102, 4105, 4112, 4114, 4117, 4120, 4129, 4132, 4160,
4162, 4165, 4168, 4177, 4180, 4192, 4202, 4225, 4228, 4240, 4352, 4354, 4357, 4360, 4369, 4372,
4384, 4417, 4420, 4432, 4480, 4500, 4502, 4609, 4612, 4614, 4624, 4672, 4704, 5120, 5122, 5125,
5128, 5137, 5140, 5152, 5185, 5188, 5193, 5200, 5220, 5248, 5377, 5380, 5392, 5440, 5632, 5652,
5705, 6145, 6148, 6160, 6162, 6208, 6228, 6278, 6400, 6405, 6502, 6737, 6825, 8192, 8194, 8197,
8200, 8202, 8209, 8212, 8224, 8257, 8260, 8272, 8320, 8352, 8449, 8452, 8464, 8512, 8520, 8549,
8704, 8738, 8832, 8872, 9217, 9220, 9232, 9257, 9280, 9472, 9537, 9554, 9625, 9729, 9754, 9894,
10240, 10248, 10250, 10272, 10325, 10376, 10402, 10600, 10640, 10760, 10784, 10882, 10888, 10890, 16385, 16388,
16390, 16393, 16400, 16402, 16405, 16408, 16417, 16420, 16448, 16450, 16453, 16456, 16458, 16465, 16468, 16480,
16485, 16513, 16516, 16528, 16640, 16642, 16645, 16648, 16657, 16660, 16672, 16705, 16708, 16720, 16768, 16773,
16802, 16897, 16900, 16912, 16914, 16937, 16960, 17408, 17410, 17413, 17416, 17425, 17428, 17433, 17440, 17473,
17476, 17488, 17536, 17556, 17665, 17668, 17680, 17700, 17728, 17818, 17920, 17930, 17988, 18000, 18433, 18436,
18448, 18496, 18501, 18516, 18530, 18688, 18705, 18756, 18768, 18793, 18948, 20480, 20482, 20485, 20488, 20497,
20500, 20512, 20520, 20545, 20548, 20560, 20608, 20737, 20740, 20752, 20757, 20800, 20802, 20992, 21060, 21162,
21505, 21508, 21520, 21537, 21568, 21600, 21633, 21665, 21760, 21768, 21888, 21896, 22049, 22120, 22177, 22528,
22548, 22593, 22608, 22681, 22810, 22848, 22850, 23173, 24577, 24580, 24592, 24640, 24660, 24674, 24710, 24745,
24832, 25124, 25162, 25234, 25600, 25622, 25872, 25920, 25925, 26020, 26625, 26730, 26917, 27142, 27220, 27234,
32768, 32770, 32773, 32776, 32785, 32788, 32800, 32810, 32833, 32836, 32848, 32896, 32898, 32936, 32938, 33025,
33028, 33030, 33040, 33088, 33105, 33113, 33280, 33312, 33408, 33410, 33440, 33448, 33793, 33796, 33808, 33810,
33813, 33856, 33888, 33929, 34048, 34116, 34213, 34328, 34410, 34816, 34824, 34853, 34906, 34944, 34946, 34984,
35078, 35362, 35456, 35464, 35478, 35496, 36865, 36868, 36880, 36928, 36950, 36996, 37120, 37154, 37220, 37462,
37513, 37888, 37893, 37956, 37968, 37976, 38185, 38288, 38290, 38465, 38993, 39078, 39241, 39445, 39520, 40960,
40962, 40968, 40970, 40992, 41002, 41120, 41297, 41305, 41382, 41472, 41474, 41480, 41514, 41600, 41632, 42048,
42133, 42597, 42648, 43018, 43040, 43042, 43048, 43168, 43176, 43268, 43396, 43398, 43560, 43562, 43665, 43690,
};
const int kmap_size = 43692;
const int nwant = 2;
const uint16_t * kgrid = grid_size == 256 ? kgrid_256 : kgrid_512;
uint64_t * kgrid_q2xs;
int * kmap_q2xs;
uint16_t * kneighbors_q2xs;
printf("================================================================= %s(grid_size = %d)\n", __func__, grid_size);
uint64_t * the_grid = (uint64_t *)malloc(grid_size*sizeof(uint64_t));
for (int k = 0; k < grid_size; ++k) {
int8_t * pos = (int8_t *)(the_grid + k);
for (int i = 0; i < 8; ++i) {
int l = (kgrid[k] >> 2*i) & 0x3;
pos[i] = 2*l + 1;
}
}
kgrid_q2xs = the_grid;
iq2_data[gindex].grid = the_grid;
kmap_q2xs = (int *)malloc(kmap_size*sizeof(int));
iq2_data[gindex].map = kmap_q2xs;
for (int i = 0; i < kmap_size; ++i) kmap_q2xs[i] = -1;
uint64_t aux64;
uint8_t * aux8 = (uint8_t *)&aux64;
for (int i = 0; i < grid_size; ++i) {
aux64 = kgrid_q2xs[i];
uint16_t index = 0;
for (int k=0; k<8; ++k) {
uint16_t q = (aux8[k] - 1)/2;
index |= (q << 2*k);
}
kmap_q2xs[index] = i;
}
int8_t pos[8];
int * dist2 = (int *)malloc(2*grid_size*sizeof(int));
int num_neighbors = 0, num_not_in_map = 0;
for (int i = 0; i < kmap_size; ++i) {
if (kmap_q2xs[i] >= 0) continue;
++num_not_in_map;
for (int k = 0; k < 8; ++k) {
int l = (i >> 2*k) & 0x3;
pos[k] = 2*l + 1;
}
for (int j = 0; j < grid_size; ++j) {
const int8_t * pg = (const int8_t *)(kgrid_q2xs + j);
int d2 = 0;
for (int k = 0; k < 8; ++k) d2 += (pg[k] - pos[k])*(pg[k] - pos[k]);
dist2[2*j+0] = d2;
dist2[2*j+1] = j;
}
qsort(dist2, grid_size, 2*sizeof(int), iq2_compare_func);
int n = 0; int d2 = dist2[0];
int nhave = 1;
for (int j = 0; j < grid_size; ++j) {
if (dist2[2*j] > d2) {
if (nhave == nwant) break;
d2 = dist2[2*j];
++nhave;
}
++n;
}
num_neighbors += n;
}
printf("%s: %d neighbours in total\n", __func__, num_neighbors);
kneighbors_q2xs = (uint16_t *)malloc((num_neighbors + num_not_in_map)*sizeof(uint16_t));
iq2_data[gindex].neighbours = kneighbors_q2xs;
int counter = 0;
for (int i = 0; i < kmap_size; ++i) {
if (kmap_q2xs[i] >= 0) continue;
for (int k = 0; k < 8; ++k) {
int l = (i >> 2*k) & 0x3;
pos[k] = 2*l + 1;
}
for (int j = 0; j < grid_size; ++j) {
const int8_t * pg = (const int8_t *)(kgrid_q2xs + j);
int d2 = 0;
for (int k = 0; k < 8; ++k) d2 += (pg[k] - pos[k])*(pg[k] - pos[k]);
dist2[2*j+0] = d2;
dist2[2*j+1] = j;
}
qsort(dist2, grid_size, 2*sizeof(int), iq2_compare_func);
kmap_q2xs[i] = -(counter + 1);
int d2 = dist2[0];
uint16_t * start = &kneighbors_q2xs[counter++];
int n = 0, nhave = 1;
for (int j = 0; j < grid_size; ++j) {
if (dist2[2*j] > d2) {
if (nhave == nwant) break;
d2 = dist2[2*j];
++nhave;
}
kneighbors_q2xs[counter++] = dist2[2*j+1];
++n;
}
*start = n;
}
free(dist2);
}
ggml : add IQ2 to test-backend-ops + refactoring (#4990) * ggml : add IQ2 to test-backend-ops + refactoring ggml-ci * cuda : update supports_op for IQ2 ggml-ci * ci : enable LLAMA_CUBLAS=1 for CUDA nodes ggml-ci * cuda : fix out-of-bounds-access in `mul_mat_vec_q` ggml-ci * tests : avoid creating RNGs for each Q tensor ggml-ci * tests : avoid creating RNGs for each tensor ggml-ci
2024-01-17 16:54:56 +00:00
void iq2xs_free_impl(int grid_size) {
2-bit quantizations (#4897) * imatrix: load * imatrix: WIP * imatrix: Add Q2_K quantization * imatrix: also guard against Q2_K_S quantization without importance matrix * imatrix: guard even more against low-bit quantization misuse --------- Co-authored-by: Iwan Kawrakow <iwan.kawrakow@gmail.com>
2024-01-14 07:45:56 +00:00
GGML_ASSERT(grid_size == 256 || grid_size == 512 || grid_size == 1024);
const int gindex = iq2_data_index(grid_size);
if (iq2_data[gindex].grid) {
free(iq2_data[gindex].grid); iq2_data[gindex].grid = NULL;
free(iq2_data[gindex].map); iq2_data[gindex].map = NULL;
free(iq2_data[gindex].neighbours); iq2_data[gindex].neighbours = NULL;
}
}
static int iq2_find_best_neighbour(const uint16_t * restrict neighbours, const uint64_t * restrict grid,
const float * restrict xval, const float * restrict weight, float scale, int8_t * restrict L) {
int num_neighbors = neighbours[0];
GGML_ASSERT(num_neighbors > 0);
float best_d2 = FLT_MAX;
int grid_index = -1;
for (int j = 1; j <= num_neighbors; ++j) {
const int8_t * pg = (const int8_t *)(grid + neighbours[j]);
float d2 = 0;
for (int i = 0; i < 8; ++i) {
float q = pg[i];
float diff = scale*q - xval[i];
d2 += weight[i]*diff*diff;
}
if (d2 < best_d2) {
best_d2 = d2; grid_index = neighbours[j];
}
}
GGML_ASSERT(grid_index >= 0);
const int8_t * pg = (const int8_t *)(grid + grid_index);
for (int i = 0; i < 8; ++i) L[i] = (pg[i] - 1)/2;
return grid_index;
}
static void quantize_row_iq2_xxs_impl(const float * restrict x, void * restrict vy, int n, const float * restrict quant_weights) {
const int gindex = iq2_data_index(256);
const uint64_t * kgrid_q2xs = iq2_data[gindex].grid;
const int * kmap_q2xs = iq2_data[gindex].map;
const uint16_t * kneighbors_q2xs = iq2_data[gindex].neighbours;
ggml : add IQ2 to test-backend-ops + refactoring (#4990) * ggml : add IQ2 to test-backend-ops + refactoring ggml-ci * cuda : update supports_op for IQ2 ggml-ci * ci : enable LLAMA_CUBLAS=1 for CUDA nodes ggml-ci * cuda : fix out-of-bounds-access in `mul_mat_vec_q` ggml-ci * tests : avoid creating RNGs for each Q tensor ggml-ci * tests : avoid creating RNGs for each tensor ggml-ci
2024-01-17 16:54:56 +00:00
GGML_ASSERT(quant_weights && "missing quantization weights");
GGML_ASSERT(kgrid_q2xs && "forgot to call ggml_quantize_init()?");
GGML_ASSERT(kmap_q2xs && "forgot to call ggml_quantize_init()?");
GGML_ASSERT(kneighbors_q2xs && "forgot to call ggml_quantize_init()?");
2-bit quantizations (#4897) * imatrix: load * imatrix: WIP * imatrix: Add Q2_K quantization * imatrix: also guard against Q2_K_S quantization without importance matrix * imatrix: guard even more against low-bit quantization misuse --------- Co-authored-by: Iwan Kawrakow <iwan.kawrakow@gmail.com>
2024-01-14 07:45:56 +00:00
GGML_ASSERT(n%QK_K == 0);
const int kMaxQ = 3;
const int nbl = n/256;
block_iq2_xxs * y = vy;
float scales[QK_K/32];
float weight[32];
float xval[32];
int8_t L[32];
int8_t Laux[32];
float waux[32];
uint8_t block_signs[4];
uint32_t q2[2*(QK_K/32)];
for (int ibl = 0; ibl < nbl; ++ibl) {
y[ibl].d = GGML_FP32_TO_FP16(0.f);
memset(q2, 0, QK_K/4);
float max_scale = 0;
const float * xbl = x + QK_K*ibl;
float sumx2 = 0;
for (int i = 0; i < QK_K; ++i) sumx2 += xbl[i]*xbl[i];
float sigma2 = sumx2/QK_K;
for (int ib = 0; ib < QK_K/32; ++ib) {
const float * xb = xbl + 32*ib;
const float * qw = quant_weights + QK_K*ibl + 32*ib;
for (int i = 0; i < 32; ++i) weight[i] = qw[i] * sqrtf(sigma2 + xb[i]*xb[i]);
for (int i = 0; i < 32; ++i) waux[i] = sqrtf(weight[i]);
for (int k = 0; k < 4; ++k) {
int nflip = 0;
uint8_t s = 0;
for (int i = 0; i < 8; ++i) {
if (xb[8*k + i] >= 0) xval[8*k + i] = xb[8*k + i];
else {
xval[8*k + i] = -xb[8*k + i]; ++nflip; s |= (1 << i);
}
}
if (nflip%2) {
int imin = 0; float min = weight[8*k+imin]*xb[8*k+imin]*xb[8*k+imin];
for (int i = 1; i < 8; ++i) {
float ax = weight[8*k+i]*xb[8*k+i]*xb[8*k+i];
if (ax < min) {
min = ax; imin = i;
}
}
xval[8*k+imin] = -xval[8*k+imin];
s ^= (1 << imin);
}
block_signs[k] = s & 127;
}
float max = xval[0];
for (int i = 1; i < 32; ++i) max = MAX(max, xval[i]);
if (!max) {
scales[ib] = 0;
memset(L, 0, 32);
continue;
}
iq2_xxs: tune quantization (#5320) We get slightly better PPL, and we cut quantization time in nearly half. The trick is to 1st quantize without forcing points onto the E8-lattice. We can then use a narrower search range around the block scale that we got that way. Co-authored-by: Iwan Kawrakow <iwan.kawrakow@gmail.com>
2024-02-05 08:46:06 +00:00
float scale = make_qp_quants(32, kMaxQ+1, xval, (uint8_t*)L, weight);
float eff_max = scale*kMaxQ;
2-bit quantizations (#4897) * imatrix: load * imatrix: WIP * imatrix: Add Q2_K quantization * imatrix: also guard against Q2_K_S quantization without importance matrix * imatrix: guard even more against low-bit quantization misuse --------- Co-authored-by: Iwan Kawrakow <iwan.kawrakow@gmail.com>
2024-01-14 07:45:56 +00:00
float best = 0;
iq2_xxs: tune quantization (#5320) We get slightly better PPL, and we cut quantization time in nearly half. The trick is to 1st quantize without forcing points onto the E8-lattice. We can then use a narrower search range around the block scale that we got that way. Co-authored-by: Iwan Kawrakow <iwan.kawrakow@gmail.com>
2024-02-05 08:46:06 +00:00
for (int is = -6; is <= 6; ++is) {
float id = (2*kMaxQ-1+is*0.1f)/eff_max;
2-bit quantizations (#4897) * imatrix: load * imatrix: WIP * imatrix: Add Q2_K quantization * imatrix: also guard against Q2_K_S quantization without importance matrix * imatrix: guard even more against low-bit quantization misuse --------- Co-authored-by: Iwan Kawrakow <iwan.kawrakow@gmail.com>
2024-01-14 07:45:56 +00:00
float this_scale = 1/id;
for (int k = 0; k < 4; ++k) {
for (int i = 0; i < 8; ++i) {
int l = nearest_int(0.5f*(id*xval[8*k+i]-1));
Laux[8*k+i] = MAX(0, MIN(kMaxQ-1, l));
}
uint16_t u = 0;
for (int i = 0; i < 8; ++i) u |= (Laux[8*k+i] << 2*i);
int grid_index = kmap_q2xs[u];
if (grid_index < 0) {
const uint16_t * neighbours = kneighbors_q2xs - kmap_q2xs[u] - 1;
grid_index = iq2_find_best_neighbour(neighbours, kgrid_q2xs, xval + 8*k, waux + 8*k, this_scale, Laux + 8*k);
}
}
float sumqx = 0, sumq2 = 0;
for (int i = 0; i < 32; ++i) {
float w = weight[i];
float q = 2*Laux[i] + 1;
sumqx += w*xval[i]*q;
sumq2 += w*q*q;
}
if (sumq2 > 0 && sumqx*sumqx > best*sumq2) {
scale = sumqx/sumq2; best = scale*sumqx;
iq2_xxs: tune quantization (#5320) We get slightly better PPL, and we cut quantization time in nearly half. The trick is to 1st quantize without forcing points onto the E8-lattice. We can then use a narrower search range around the block scale that we got that way. Co-authored-by: Iwan Kawrakow <iwan.kawrakow@gmail.com>
2024-02-05 08:46:06 +00:00
memcpy(L, Laux, 32);
2-bit quantizations (#4897) * imatrix: load * imatrix: WIP * imatrix: Add Q2_K quantization * imatrix: also guard against Q2_K_S quantization without importance matrix * imatrix: guard even more against low-bit quantization misuse --------- Co-authored-by: Iwan Kawrakow <iwan.kawrakow@gmail.com>
2024-01-14 07:45:56 +00:00
}
}
iq2_xxs: tune quantization (#5320) We get slightly better PPL, and we cut quantization time in nearly half. The trick is to 1st quantize without forcing points onto the E8-lattice. We can then use a narrower search range around the block scale that we got that way. Co-authored-by: Iwan Kawrakow <iwan.kawrakow@gmail.com>
2024-02-05 08:46:06 +00:00
if (scale > 0) {
2-bit quantizations (#4897) * imatrix: load * imatrix: WIP * imatrix: Add Q2_K quantization * imatrix: also guard against Q2_K_S quantization without importance matrix * imatrix: guard even more against low-bit quantization misuse --------- Co-authored-by: Iwan Kawrakow <iwan.kawrakow@gmail.com>
2024-01-14 07:45:56 +00:00
float id = 1/scale;
for (int k = 0; k < 4; ++k) {
uint16_t u = 0;
for (int i = 0; i < 8; ++i) {
int l = nearest_int(0.5f*(id*xval[8*k+i]-1));
l = MAX(0, MIN(kMaxQ-1, l));
u |= (l << 2*i);
}
int grid_index = kmap_q2xs[u];
if (grid_index < 0) {
const uint16_t * neighbours = kneighbors_q2xs - kmap_q2xs[u] - 1;
grid_index = iq2_find_best_neighbour(neighbours, kgrid_q2xs, xval + 8*k, waux + 8*k, scale, L + 8*k);
}
const int8_t * pg = (const int8_t *)(kgrid_q2xs + grid_index);
for (int i = 0; i < 8; ++i) L[8*k+i] = (pg[i] - 1)/2;
}
float sumqx = 0, sumq2 = 0;
for (int i = 0; i < 32; ++i) {
float w = weight[i];
float q = 2*L[i] + 1;
sumqx += w*xval[i]*q;
sumq2 += w*q*q;
}
if (sumq2 > 0) scale = sumqx/sumq2;
}
if (scale < 0) {
// This should never happen, but just in case, flip scale so that it is positive (we use uint's to encode the scale)
// and correspondingly flip quant signs.
scale = -scale;
for (int k = 0; k < 4; ++k) block_signs[k] = (~block_signs[k]) & 127;
}
for (int k = 0; k < 4; ++k) {
uint16_t u = 0;
for (int i = 0; i < 8; ++i) u |= (L[8*k+i] << 2*i);
int grid_index = kmap_q2xs[u];
if (grid_index < 0) {
printf("Oops: found point %u not on grid:", u);
for (int i = 0; i < 8; ++i) printf(" %d", L[8*k+i]);
printf("\n");
GGML_ASSERT(false);
}
q2[2*ib+0] |= (grid_index << 8*k);
q2[2*ib+1] |= (block_signs[k] << 7*k);
}
GGML_ASSERT(scale >= 0);
scales[ib] = scale;
max_scale = MAX(max_scale, scale);
}
if (!max_scale) {
memset(y[ibl].qs, 0, QK_K/4);
continue;
}
float d = max_scale/31;
y[ibl].d = GGML_FP32_TO_FP16(d);
float id = 1/d;
for (int ib = 0; ib < QK_K/32; ++ib) {
int l = nearest_int(0.5f*(id*scales[ib]-1));
l = MAX(0, MIN(15, l));
q2[2*ib+1] |= ((uint32_t)l << 28);
}
memcpy(y[ibl].qs, q2, QK_K/4);
}
}
static void quantize_row_iq2_xs_impl(const float * restrict x, void * restrict vy, int n, const float * restrict quant_weights) {
const int gindex = iq2_data_index(512);
const uint64_t * kgrid_q2xs = iq2_data[gindex].grid;
const int * kmap_q2xs = iq2_data[gindex].map;
const uint16_t * kneighbors_q2xs = iq2_data[gindex].neighbours;
ggml : add IQ2 to test-backend-ops + refactoring (#4990) * ggml : add IQ2 to test-backend-ops + refactoring ggml-ci * cuda : update supports_op for IQ2 ggml-ci * ci : enable LLAMA_CUBLAS=1 for CUDA nodes ggml-ci * cuda : fix out-of-bounds-access in `mul_mat_vec_q` ggml-ci * tests : avoid creating RNGs for each Q tensor ggml-ci * tests : avoid creating RNGs for each tensor ggml-ci
2024-01-17 16:54:56 +00:00
GGML_ASSERT(quant_weights && "missing quantization weights");
GGML_ASSERT(kmap_q2xs && "forgot to call ggml_quantize_init()?");
GGML_ASSERT(kgrid_q2xs && "forgot to call ggml_quantize_init()?");
GGML_ASSERT(kneighbors_q2xs && "forgot to call ggml_quantize_init()?");
2-bit quantizations (#4897) * imatrix: load * imatrix: WIP * imatrix: Add Q2_K quantization * imatrix: also guard against Q2_K_S quantization without importance matrix * imatrix: guard even more against low-bit quantization misuse --------- Co-authored-by: Iwan Kawrakow <iwan.kawrakow@gmail.com>
2024-01-14 07:45:56 +00:00
GGML_ASSERT(n%QK_K == 0);
const int kMaxQ = 3;
const int nbl = n/256;
block_iq2_xs * y = vy;
float scales[QK_K/16];
float weight[16];
float xval[16];
int8_t L[16];
int8_t Laux[16];
float waux[16];
bool is_on_grid[2];
bool is_on_grid_aux[2];
uint8_t block_signs[2];
uint16_t q2[2*(QK_K/16)];
for (int ibl = 0; ibl < nbl; ++ibl) {
y[ibl].d = GGML_FP32_TO_FP16(0.f);
memset(q2, 0, QK_K/4);
memset(y[ibl].scales, 0, QK_K/32);
float max_scale = 0;
const float * xbl = x + QK_K*ibl;
float sumx2 = 0;
for (int i = 0; i < QK_K; ++i) sumx2 += xbl[i]*xbl[i];
float sigma2 = sumx2/QK_K;
for (int ib = 0; ib < QK_K/16; ++ib) {
const float * xb = xbl + 16*ib;
const float * qw = quant_weights + QK_K*ibl + 16*ib;
for (int i = 0; i < 16; ++i) weight[i] = qw[i] * sqrtf(sigma2 + xb[i]*xb[i]);
for (int i = 0; i < 16; ++i) waux[i] = sqrtf(weight[i]);
for (int k = 0; k < 2; ++k) {
int nflip = 0;
uint8_t s = 0;
for (int i = 0; i < 8; ++i) {
if (xb[8*k + i] >= 0) xval[8*k + i] = xb[8*k + i];
else {
xval[8*k + i] = -xb[8*k + i]; ++nflip; s |= (1 << i);
}
}
if (nflip%2) {
int imin = 0; float min = weight[8*k+imin]*xb[8*k+imin]*xb[8*k+imin];
for (int i = 1; i < 8; ++i) {
float ax = weight[8*k+i]*xb[8*k+i]*xb[8*k+i];
if (ax < min) {
min = ax; imin = i;
}
}
xval[8*k+imin] = -xval[8*k+imin];
s ^= (1 << imin);
}
block_signs[k] = s & 127;
}
float max = xval[0];
for (int i = 1; i < 16; ++i) max = MAX(max, xval[i]);
if (!max) {
scales[ib] = 0;
memset(L, 0, 16);
continue;
}
float best = 0;
float scale = max/(2*kMaxQ-1);
is_on_grid[0] = is_on_grid[1] = true;
for (int is = -9; is <= 9; ++is) {
float id = (2*kMaxQ-1+is*0.1f)/max;
float this_scale = 1/id;
for (int k = 0; k < 2; ++k) {
for (int i = 0; i < 8; ++i) {
int l = nearest_int(0.5f*(id*xval[8*k+i]-1));
Laux[8*k+i] = MAX(0, MIN(kMaxQ-1, l));
}
uint16_t u = 0;
for (int i = 0; i < 8; ++i) u |= (Laux[8*k+i] << 2*i);
int grid_index = kmap_q2xs[u];
is_on_grid_aux[k] = true;
if (grid_index < 0) {
is_on_grid_aux[k] = false;
const uint16_t * neighbours = kneighbors_q2xs - kmap_q2xs[u] - 1;
grid_index = iq2_find_best_neighbour(neighbours, kgrid_q2xs, xval + 8*k, waux + 8*k, this_scale, Laux + 8*k);
}
}
float sumqx = 0, sumq2 = 0;
for (int i = 0; i < 16; ++i) {
float w = weight[i];
float q = 2*Laux[i] + 1;
sumqx += w*xval[i]*q;
sumq2 += w*q*q;
}
if (sumq2 > 0 && sumqx*sumqx > best*sumq2) {
scale = sumqx/sumq2; best = scale*sumqx;
for (int i = 0; i < 16; ++i) L[i] = Laux[i];
for (int k = 0; k < 2; ++k) is_on_grid[k] = is_on_grid_aux[k];
}
}
int n_not_ongrid = 0;
for (int k = 0; k < 2; ++k) if (!is_on_grid[k]) ++n_not_ongrid;
if (n_not_ongrid > 0 && scale > 0) {
float id = 1/scale;
for (int k = 0; k < 2; ++k) {
if (is_on_grid[k]) continue;
uint16_t u = 0;
for (int i = 0; i < 8; ++i) {
int l = nearest_int(0.5f*(id*xval[8*k+i]-1));
l = MAX(0, MIN(kMaxQ-1, l));
u |= (l << 2*i);
L[8*k + i] = l;
}
int grid_index = kmap_q2xs[u];
if (grid_index < 0) {
const uint16_t * neighbours = kneighbors_q2xs - kmap_q2xs[u] - 1;
grid_index = iq2_find_best_neighbour(neighbours, kgrid_q2xs, xval + 8*k, waux + 8*k, scale, L + 8*k);
}
}
float sumqx = 0, sumq2 = 0;
for (int i = 0; i < 16; ++i) {
float w = weight[i];
float q = 2*L[i] + 1;
sumqx += w*xval[i]*q;
sumq2 += w*q*q;
}
if (sumq2 > 0) scale = sumqx/sumq2;
}
if (scale < 0) {
scale = -scale;
for (int k = 0; k < 2; ++k) block_signs[k] = (~block_signs[k]) & 127;
}
for (int k = 0; k < 2; ++k) {
uint16_t u = 0;
for (int i = 0; i < 8; ++i) u |= (L[8*k+i] << 2*i);
int grid_index = kmap_q2xs[u];
if (grid_index < 0) {
printf("Oops: found point %u not on grid:", u);
for (int i = 0; i < 8; ++i) printf(" %d", L[8*k+i]);
printf("\n");
GGML_ASSERT(false);
}
q2[2*ib+k] = grid_index | (block_signs[k] << 9);
}
GGML_ASSERT(scale >= 0);
scales[ib] = scale;
max_scale = MAX(max_scale, scale);
}
if (!max_scale) {
memset(y[ibl].qs, 0, QK_K/4);
continue;
}
float d = max_scale/31;
y[ibl].d = GGML_FP32_TO_FP16(d);
float id = 1/d;
for (int ib = 0; ib < QK_K/16; ++ib) {
int l = nearest_int(0.5f*(id*scales[ib]-1));
l = MAX(0, MIN(15, l));
if (ib%2 == 0) y[ibl].scales[ib/2] = l;
else y[ibl].scales[ib/2] |= (l << 4);
}
memcpy(y[ibl].qs, q2, QK_K/4);
}
}
size_t quantize_iq2_xxs(const float * src, void * dst, int nrow, int n_per_row, int64_t * hist, const float * quant_weights) {
(void)hist;
GGML_ASSERT(n_per_row%QK_K == 0);
int nblock = n_per_row/QK_K;
char * qrow = (char *)dst;
for (int row = 0; row < nrow; ++row) {
quantize_row_iq2_xxs_impl(src, qrow, n_per_row, quant_weights);
src += n_per_row;
qrow += nblock*sizeof(block_iq2_xxs);
}
return nrow * nblock * sizeof(block_iq2_xxs);
}
size_t quantize_iq2_xs(const float * src, void * dst, int nrow, int n_per_row, int64_t * hist, const float * quant_weights) {
(void)hist;
GGML_ASSERT(n_per_row%QK_K == 0);
int nblock = n_per_row/QK_K;
char * qrow = (char *)dst;
for (int row = 0; row < nrow; ++row) {
quantize_row_iq2_xs_impl(src, qrow, n_per_row, quant_weights);
src += n_per_row;
qrow += nblock*sizeof(block_iq2_xs);
}
return nrow * nblock * sizeof(block_iq2_xs);
}
SOTA 3-bit quants (#5196) * iq3_xxs: quantize/dequantize RMSE seems a bit high-ish at about half-way between q2_K and q3_K, so need to check more. * iq3_xxs: CUDA dequantize works * iq2_xxs: tuning quantization * iq3_xxs: starting to look better PPL on wiki.test.raw LLaMA-v1-7B: 6.4218 LLaMA-v2-7B: 6.3560 Mistral-7B : 6.0717 This is better than Q3_K_XS, with a 5% reduction in quantized model size. * iq3_xxs: CUDA dot product We have PP-512: 5891 t/s TG-128: 143.9 t/s * iq3_xxs: scalar and AVX2 dot products * iq3_xxs: ARM_NEON and Metal Metal performance is decent, ARM_NEON is pathetic * iq3_xxs: slightly better grid points * Faster iq3_xxs and iq2_xs dot products on CUDA * iq3_xxs: add some quant mix * iq3_xxs: fix failing quantization test Dot product still fails. Is this real? * iq3_xxs: hopefully fix ROCm * iq3_xxs: failing tests This time the dot product accuracy did find an actual bug in the AVX2 implementation. * Add IQ3_XXS to test-backend-ops --------- Co-authored-by: Iwan Kawrakow <iwan.kawrakow@gmail.com>
2024-01-30 13:14:12 +00:00
//
// ============================================= 3-bit using D4 lattice
//
typedef struct {
uint32_t * grid;
int * map;
uint16_t * neighbours;
} iq3_entry_t;
static iq3_entry_t iq3_data[1] = {
{NULL, NULL, NULL},
};
static inline int iq3_data_index(int grid_size) {
(void)grid_size;
GGML_ASSERT(grid_size == 256);
return 0;
}
static int iq3_compare_func(const void * left, const void * right) {
const int * l = (const int *)left;
const int * r = (const int *)right;
return l[0] < r[0] ? -1 : l[0] > r[0] ? 1 : l[1] < r[1] ? -1 : l[1] > r[1] ? 1 : 0;
}
void iq3xs_init_impl(int grid_size) {
const int gindex = iq3_data_index(grid_size);
if (iq3_data[gindex].grid) {
return;
}
static const uint16_t kgrid_256[256] = {
0, 2, 4, 9, 11, 15, 16, 18, 25, 34, 59, 61, 65, 67, 72, 74,
81, 85, 88, 90, 97, 108, 120, 128, 130, 132, 137, 144, 146, 153, 155, 159,
169, 175, 189, 193, 199, 200, 202, 213, 248, 267, 287, 292, 303, 315, 317, 321,
327, 346, 362, 413, 436, 456, 460, 462, 483, 497, 513, 515, 520, 522, 529, 531,
536, 538, 540, 551, 552, 576, 578, 585, 592, 594, 641, 643, 648, 650, 657, 664,
698, 704, 706, 720, 729, 742, 758, 769, 773, 808, 848, 852, 870, 889, 901, 978,
992, 1024, 1026, 1033, 1035, 1040, 1042, 1046, 1049, 1058, 1089, 1091, 1093, 1096, 1098, 1105,
1112, 1139, 1143, 1144, 1152, 1154, 1161, 1167, 1168, 1170, 1183, 1184, 1197, 1217, 1224, 1228,
1272, 1276, 1309, 1323, 1347, 1367, 1377, 1404, 1473, 1475, 1486, 1509, 1537, 1544, 1546, 1553,
1555, 1576, 1589, 1594, 1600, 1602, 1616, 1625, 1636, 1638, 1665, 1667, 1672, 1685, 1706, 1722,
1737, 1755, 1816, 1831, 1850, 1856, 1862, 1874, 1901, 1932, 1950, 1971, 2011, 2032, 2052, 2063,
2077, 2079, 2091, 2095, 2172, 2192, 2207, 2208, 2224, 2230, 2247, 2277, 2308, 2345, 2356, 2389,
2403, 2424, 2501, 2504, 2506, 2520, 2570, 2593, 2616, 2624, 2630, 2646, 2669, 2700, 2714, 2746,
2754, 2795, 2824, 2835, 2839, 2874, 2882, 2905, 2984, 3028, 3042, 3092, 3108, 3110, 3124, 3153,
3185, 3215, 3252, 3288, 3294, 3364, 3397, 3434, 3483, 3523, 3537, 3587, 3589, 3591, 3592, 3610,
3626, 3670, 3680, 3722, 3749, 3754, 3776, 3789, 3803, 3824, 3857, 3873, 3904, 3906, 3924, 3992,
};
const int kmap_size = 4096;
const int nwant = 2;
const uint16_t * kgrid = kgrid_256;
uint32_t * kgrid_q3xs;
int * kmap_q3xs;
uint16_t * kneighbors_q3xs;
printf("================================================================= %s(grid_size = %d)\n", __func__, grid_size);
uint32_t * the_grid = (uint32_t *)malloc(grid_size*sizeof(uint32_t));
for (int k = 0; k < grid_size; ++k) {
int8_t * pos = (int8_t *)(the_grid + k);
for (int i = 0; i < 4; ++i) {
int l = (kgrid[k] >> 3*i) & 0x7;
pos[i] = 2*l + 1;
}
}
kgrid_q3xs = the_grid;
iq3_data[gindex].grid = the_grid;
kmap_q3xs = (int *)malloc(kmap_size*sizeof(int));
iq3_data[gindex].map = kmap_q3xs;
for (int i = 0; i < kmap_size; ++i) kmap_q3xs[i] = -1;
uint32_t aux32;
uint8_t * aux8 = (uint8_t *)&aux32;
for (int i = 0; i < grid_size; ++i) {
aux32 = kgrid_q3xs[i];
uint16_t index = 0;
for (int k=0; k<4; ++k) {
uint16_t q = (aux8[k] - 1)/2;
index |= (q << 3*k);
}
kmap_q3xs[index] = i;
}
int8_t pos[4];
int * dist2 = (int *)malloc(2*grid_size*sizeof(int));
int num_neighbors = 0, num_not_in_map = 0;
for (int i = 0; i < kmap_size; ++i) {
if (kmap_q3xs[i] >= 0) continue;
++num_not_in_map;
for (int k = 0; k < 4; ++k) {
int l = (i >> 3*k) & 0x7;
pos[k] = 2*l + 1;
}
for (int j = 0; j < grid_size; ++j) {
const int8_t * pg = (const int8_t *)(kgrid_q3xs + j);
int d2 = 0;
for (int k = 0; k < 4; ++k) d2 += (pg[k] - pos[k])*(pg[k] - pos[k]);
dist2[2*j+0] = d2;
dist2[2*j+1] = j;
}
qsort(dist2, grid_size, 2*sizeof(int), iq3_compare_func);
int n = 0; int d2 = dist2[0];
int nhave = 1;
for (int j = 0; j < grid_size; ++j) {
if (dist2[2*j] > d2) {
if (nhave == nwant) break;
d2 = dist2[2*j];
++nhave;
}
++n;
}
num_neighbors += n;
}
printf("%s: %d neighbours in total\n", __func__, num_neighbors);
kneighbors_q3xs = (uint16_t *)malloc((num_neighbors + num_not_in_map)*sizeof(uint16_t));
iq3_data[gindex].neighbours = kneighbors_q3xs;
int counter = 0;
for (int i = 0; i < kmap_size; ++i) {
if (kmap_q3xs[i] >= 0) continue;
for (int k = 0; k < 4; ++k) {
int l = (i >> 3*k) & 0x7;
pos[k] = 2*l + 1;
}
for (int j = 0; j < grid_size; ++j) {
const int8_t * pg = (const int8_t *)(kgrid_q3xs + j);
int d2 = 0;
for (int k = 0; k < 4; ++k) d2 += (pg[k] - pos[k])*(pg[k] - pos[k]);
dist2[2*j+0] = d2;
dist2[2*j+1] = j;
}
qsort(dist2, grid_size, 2*sizeof(int), iq3_compare_func);
kmap_q3xs[i] = -(counter + 1);
int d2 = dist2[0];
uint16_t * start = &kneighbors_q3xs[counter++];
int n = 0, nhave = 1;
for (int j = 0; j < grid_size; ++j) {
if (dist2[2*j] > d2) {
if (nhave == nwant) break;
d2 = dist2[2*j];
++nhave;
}
kneighbors_q3xs[counter++] = dist2[2*j+1];
++n;
}
*start = n;
}
free(dist2);
}
void iq3xs_free_impl(int grid_size) {
GGML_ASSERT(grid_size == 256);
const int gindex = iq3_data_index(grid_size);
if (iq3_data[gindex].grid) {
free(iq3_data[gindex].grid); iq3_data[gindex].grid = NULL;
free(iq3_data[gindex].map); iq3_data[gindex].map = NULL;
free(iq3_data[gindex].neighbours); iq3_data[gindex].neighbours = NULL;
}
}
static int iq3_find_best_neighbour(const uint16_t * restrict neighbours, const uint32_t * restrict grid,
const float * restrict xval, const float * restrict weight, float scale, int8_t * restrict L) {
int num_neighbors = neighbours[0];
GGML_ASSERT(num_neighbors > 0);
float best_d2 = FLT_MAX;
int grid_index = -1;
for (int j = 1; j <= num_neighbors; ++j) {
const int8_t * pg = (const int8_t *)(grid + neighbours[j]);
float d2 = 0;
for (int i = 0; i < 4; ++i) {
float q = pg[i];
float diff = scale*q - xval[i];
d2 += weight[i]*diff*diff;
}
if (d2 < best_d2) {
best_d2 = d2; grid_index = neighbours[j];
}
}
GGML_ASSERT(grid_index >= 0);
const int8_t * pg = (const int8_t *)(grid + grid_index);
for (int i = 0; i < 4; ++i) L[i] = (pg[i] - 1)/2;
return grid_index;
}
static void quantize_row_iq3_xxs_impl(const float * restrict x, void * restrict vy, int n, const float * restrict quant_weights) {
const int gindex = iq3_data_index(256);
const uint32_t * kgrid_q3xs = iq3_data[gindex].grid;
const int * kmap_q3xs = iq3_data[gindex].map;
const uint16_t * kneighbors_q3xs = iq3_data[gindex].neighbours;
//GGML_ASSERT(quant_weights && "missing quantization weights");
GGML_ASSERT(kgrid_q3xs && "forgot to call ggml_quantize_init()?");
GGML_ASSERT(kmap_q3xs && "forgot to call ggml_quantize_init()?");
GGML_ASSERT(kneighbors_q3xs && "forgot to call ggml_quantize_init()?");
GGML_ASSERT(n%QK_K == 0);
const int kMaxQ = 8;
const int nbl = n/256;
block_iq3_xxs * y = vy;
float scales[QK_K/32];
float weight[32];
float xval[32];
int8_t L[32];
int8_t Laux[32];
float waux[32];
bool is_on_grid[8];
bool is_on_grid_aux[8];
uint8_t block_signs[8];
uint8_t q3[3*(QK_K/8)];
uint32_t * scales_and_signs = (uint32_t *)(q3 + QK_K/4);
for (int ibl = 0; ibl < nbl; ++ibl) {
y[ibl].d = GGML_FP32_TO_FP16(0.f);
memset(q3, 0, 3*QK_K/8);
float max_scale = 0;
const float * xbl = x + QK_K*ibl;
float sumx2 = 0;
for (int i = 0; i < QK_K; ++i) sumx2 += xbl[i]*xbl[i];
float sigma2 = sumx2/QK_K;
for (int ib = 0; ib < QK_K/32; ++ib) {
const float * xb = xbl + 32*ib;
if (quant_weights) {
const float * qw = quant_weights + QK_K*ibl + 32*ib;
for (int i = 0; i < 32; ++i) weight[i] = qw[i] * sqrtf(sigma2 + xb[i]*xb[i]);
} else {
for (int i = 0; i < 32; ++i) weight[i] = xb[i]*xb[i];
}
for (int i = 0; i < 32; ++i) waux[i] = sqrtf(weight[i]);
for (int k = 0; k < 4; ++k) {
int nflip = 0;
uint8_t s = 0;
for (int i = 0; i < 8; ++i) {
if (xb[8*k + i] >= 0) xval[8*k + i] = xb[8*k + i];
else {
xval[8*k + i] = -xb[8*k + i]; ++nflip; s |= (1 << i);
}
}
if (nflip%2) {
int imin = 0; float min = weight[8*k+imin]*xb[8*k+imin]*xb[8*k+imin];
for (int i = 1; i < 8; ++i) {
float ax = weight[8*k+i]*xb[8*k+i]*xb[8*k+i];
if (ax < min) {
min = ax; imin = i;
}
}
xval[8*k+imin] = -xval[8*k+imin];
s ^= (1 << imin);
}
block_signs[k] = s & 127;
}
float max = xval[0];
for (int i = 1; i < 32; ++i) max = MAX(max, xval[i]);
if (!max) {
scales[ib] = 0;
memset(L, 0, 32);
continue;
}
float best = 0;
float scale = max/(2*kMaxQ-1);
for (int is = -15; is <= 15; ++is) {
float id = (2*kMaxQ-1+is*0.2f)/max;
float this_scale = 1/id;
for (int k = 0; k < 8; ++k) {
for (int i = 0; i < 4; ++i) {
int l = nearest_int(0.5f*(id*xval[4*k+i]-1));
Laux[4*k+i] = MAX(0, MIN(kMaxQ-1, l));
}
uint16_t u = 0;
for (int i = 0; i < 4; ++i) u |= (Laux[4*k+i] << 3*i);
int grid_index = kmap_q3xs[u];
is_on_grid_aux[k] = true;
if (grid_index < 0) {
is_on_grid_aux[k] = false;
const uint16_t * neighbours = kneighbors_q3xs - kmap_q3xs[u] - 1;
grid_index = iq3_find_best_neighbour(neighbours, kgrid_q3xs, xval + 4*k, waux + 4*k, this_scale, Laux + 4*k);
}
}
float sumqx = 0, sumq2 = 0;
for (int i = 0; i < 32; ++i) {
float w = weight[i];
float q = 2*Laux[i] + 1;
sumqx += w*xval[i]*q;
sumq2 += w*q*q;
}
if (sumq2 > 0 && sumqx*sumqx > best*sumq2) {
scale = sumqx/sumq2; best = scale*sumqx;
for (int i = 0; i < 32; ++i) L[i] = Laux[i];
for (int k = 0; k < 8; ++k) is_on_grid[k] = is_on_grid_aux[k];
}
}
int n_not_ongrid = 0;
for (int k = 0; k < 8; ++k) if (!is_on_grid[k]) ++n_not_ongrid;
if (n_not_ongrid > 0 && scale > 0) {
float id = 1/scale;
for (int k = 0; k < 8; ++k) {
if (is_on_grid[k]) continue;
uint16_t u = 0;
for (int i = 0; i < 4; ++i) {
int l = nearest_int(0.5f*(id*xval[4*k+i]-1));
l = MAX(0, MIN(kMaxQ-1, l));
u |= (l << 3*i);
}
int grid_index = kmap_q3xs[u];
if (grid_index < 0) {
const uint16_t * neighbours = kneighbors_q3xs - kmap_q3xs[u] - 1;
grid_index = iq3_find_best_neighbour(neighbours, kgrid_q3xs, xval + 4*k, waux + 4*k, scale, L + 4*k);
}
const int8_t * pg = (const int8_t *)(kgrid_q3xs + grid_index);
for (int i = 0; i < 4; ++i) L[4*k+i] = (pg[i] - 1)/2;
}
float sumqx = 0, sumq2 = 0;
for (int i = 0; i < 32; ++i) {
float w = weight[i];
float q = 2*L[i] + 1;
sumqx += w*xval[i]*q;
sumq2 += w*q*q;
}
if (sumq2 > 0) scale = sumqx/sumq2;
}
if (scale < 0) {
// This should never happen, but just in case, flip scale so that it is positive (we use uint's to encode the scale)
// and correspondingly flip quant signs.
scale = -scale;
for (int k = 0; k < 4; ++k) block_signs[k] = (~block_signs[k]) & 127;
}
for (int k = 0; k < 8; ++k) {
uint16_t u = 0;
for (int i = 0; i < 4; ++i) u |= (L[4*k+i] << 3*i);
int grid_index = kmap_q3xs[u];
if (grid_index < 0) {
printf("Oops: found point %u not on grid:", u);
for (int i = 0; i < 4; ++i) printf(" %d", L[4*k+i]);
printf("\n");
GGML_ASSERT(false);
}
q3[8*ib+k] = grid_index;
}
scales_and_signs[ib] = block_signs[0] | (block_signs[1] << 7) | (block_signs[2] << 14) | (block_signs[3] << 21);
GGML_ASSERT(scale >= 0);
scales[ib] = scale;
max_scale = MAX(max_scale, scale);
}
if (!max_scale) {
memset(y[ibl].qs, 0, 3*QK_K/8);
continue;
}
float d = max_scale/31;
y[ibl].d = GGML_FP32_TO_FP16(d);
float id = 1/d;
float sumqx = 0, sumq2 = 0;
for (int ib = 0; ib < QK_K/32; ++ib) {
int l = nearest_int(0.5f*(id*scales[ib]-1));
l = MAX(0, MIN(15, l));
scales_and_signs[ib] |= ((uint32_t)l << 28);
if (false) {
const float * xb = xbl + 32*ib;
if (quant_weights) {
const float * qw = quant_weights + QK_K*ibl + 32*ib;
for (int i = 0; i < 32; ++i) weight[i] = qw[i] * sqrtf(sigma2 + xb[i]*xb[i]);
} else {
for (int i = 0; i < 32; ++i) weight[i] = xb[i]*xb[i];
}
const float db = 0.25f * d * (1 + 2*l);
for (int k = 0; k < 8; ++k) {
const int8_t * signs = keven_signs_q2xs + 8*((scales_and_signs[ib] >> 7*(k/2)) & 127) + 4*(k%2);
const float * xk = xb + 4*k;
const float * wk = weight + 4*k;
//const uint8_t * grid = (const uint8_t *)(kgrid_q3xs + q3[8*ib+k]);
const uint8_t * grid = (const uint8_t *)(iq3xxs_grid + q3[8*ib+k]);
float best_mse = 0; int best_index = q3[8*ib+k];
for (int j = 0; j < 4; ++j) {
float diff = db * grid[j] * signs[j] - xk[j];
best_mse += wk[j] * diff * diff;
}
for (int idx = 0; idx < 256; ++idx) {
//grid = (const uint8_t *)(kgrid_q3xs + idx);
grid = (const uint8_t *)(iq3xxs_grid + idx);
float mse = 0;
for (int j = 0; j < 4; ++j) {
float diff = db * grid[j] * signs[j] - xk[j];
mse += wk[j] * diff * diff;
}
if (mse < best_mse) {
best_mse = mse; best_index = idx;
}
}
q3[8*ib+k] = best_index;
//grid = (const uint8_t *)(kgrid_q3xs + best_index);
grid = (const uint8_t *)(iq3xxs_grid + best_index);
for (int j = 0; j < 4; ++j) {
float q = db * grid[j] * signs[j];
sumqx += wk[j] * q * xk[j];
sumq2 += wk[j] * q * q;
}
}
if (sumq2 > 0) y[ibl].d = GGML_FP32_TO_FP16(d*sumqx/sumq2);
}
}
memcpy(y[ibl].qs, q3, 3*QK_K/8);
}
}
size_t quantize_iq3_xxs(const float * src, void * dst, int nrow, int n_per_row, int64_t * hist, const float * quant_weights) {
(void)hist;
GGML_ASSERT(n_per_row%QK_K == 0);
int nblock = n_per_row/QK_K;
char * qrow = (char *)dst;
for (int row = 0; row < nrow; ++row) {
quantize_row_iq3_xxs_impl(src, qrow, n_per_row, quant_weights);
src += n_per_row;
qrow += nblock*sizeof(block_iq3_xxs);
}
return nrow * nblock * sizeof(block_iq3_xxs);
}
void quantize_row_iq3_xxs(const float * restrict x, void * restrict vy, int k) {
assert(k % QK_K == 0);
block_iq3_xxs * restrict y = vy;
quantize_row_iq3_xxs_reference(x, y, k);
}
void quantize_row_iq3_xxs_reference(const float * restrict x, block_iq3_xxs * restrict y, int k) {
assert(k % QK_K == 0);
quantize_row_iq3_xxs_impl(x, y, k, NULL);
}
Reference in New Issue Copy Permalink