mirror of
https://github.com/ggerganov/llama.cpp.git
synced 2025-01-09 02:01:44 +00:00
9bc6db28d0
* ggml-quants : 1.625 bpw ternary packing for BitNet 1.58b * ggml-quants : faster 1.625 bpw AVX2 vec_dot Not using a lookup table anymore makes it match q4_0 speed. * gguf-py : fix formatting * llama : remove spaces on empty line * ggml-quants : subtract 1 when back in epi8 This makes the 1.625 bpw type go faster than q4_0. Still not the fastest. * ggml-quants : Q2_2 now faster than Q4_K on with AVX2 * ggml-quants : cleanup Q1_3 code formatting * ggml-quants : ARM NEON vec_dot for q2_2 and q1_3 * ggml-quants : use ceiling division when quantizing q1_3 * convert-hf : simplify BitNet pre-quantization This still results in the exact same tensor weights and scales, but it reveals some weirdness in the current algorithm. * convert-hf : allow converting the weird BitNet 1.3B Its FFN size is 5460 which is not convenient. The offending tensors are kept in F16, which makes the final model 5.01 bpw. * bitnet : replace 1.58b with b1.58, as in the paper * ggml-quants : fix build failure on Windows * ggml-quants : attempt to fix Arm 32-bit support * ggml : add some informative comments in q1_3 vec_dot * ggml : add TQ1_0 and TQ2_0 ternary quantization types * ggml : even faster TQ2_0 * ggml : also faster TQ1_0 Same optimization as for TQ2_0 by offsetting the sum instead of the weights. This makes TQ1_0 almost as fast as Q8_0 on AVX2. * ggml : fix build issues in certain environments * ggml : add NEON vec_dot implementation for TQ1_0 and TQ2_0 * ggml : avoid directly using vmlal_high_s8, for 32-bit ARM compat The compiler seems smart enough to use the same instruction even when using vget_high_s8 instead. * ggml : remove q1_3 and q2_2 No more 1.625 bpw and 2.000 bpw, now instead using 1.6875 bpw and 2.0625 bpw with TQ1_0 and TQ2_0, respectively. * llama : remove the separate scale tensors of BitNet b1.58 They won't be needed, since the remaining ternary quant types have built-in scales. * ggml-quants : rename fields of TQ1_0 and TQ2_0 structs for consistency * ggml-quants : allow using vdotq_s32 in TQ2_0 vec_dot Not yet tested on hardware which supports it, might not work or might not even compile. But also it might. It should make the performance better on recent ARM CPUs. * ggml-quants : remove comment about possible format change of TQ2_0 Making it slightly more convenient for AVX512 but less convenient for everything else is not worth the trouble. * gguf-py : Numpy (de)quantization for TQ1_0 and TQ2_0 * ggml-quants : use roundf instead of nearest_int for TQ1_0 and TQ2_0 This does not change anything for ternary models, since their values should never end up being in halfway cases anyway. * convert : allow direct conversion to TQ1_0 and TQ2_0 The token embeddings and output tensors are kept in F16 to allow quantizing them to Q4_K and Q6_K with llama-quantize. * llama : handle fallback for TQ1_0 and TQ2_0 with Q4_0 Q4_0 is not completely symmetric (so not lossless for ternary models), but it should be good enough. * ggml-quants : allow using ARM dot product instructions for TQ1_0 * ggml-quants : deduplicate TQ1_0 and TQ2_0 __ARM_FEATURE_DOTPROD support * ggml : remove unused ggml_mul special case It would otherwise conflict with the more general optimization coming with Mamba-2. * ggml : handle TQ1_0 and TQ2_0 in dequantization-based operators * test-backend-ops : add TQ1_0 and TQ2_0 comments for later Not yet adding uncommented, because some backends like SYCL and Metal do not properly handle unknown types in supports_op for GGML_OP_MUL_MAT. (and Metal also doesn't handle it with GGML_OP_GET_ROWS) Support for TQ1_0 and TQ2_0 for other backends than CPU will be added in follow-up pull requests.
152 lines
13 KiB
C
152 lines
13 KiB
C
#pragma once
|
|
|
|
#define GGML_COMMON_DECL_C
|
|
#include "ggml-common.h"
|
|
|
|
#include "ggml.h"
|
|
|
|
// GGML internal header
|
|
|
|
#ifdef __cplusplus
|
|
extern "C" {
|
|
#endif
|
|
|
|
// Quantization
|
|
void quantize_row_q4_0_ref(const float * GGML_RESTRICT x, block_q4_0 * GGML_RESTRICT y, int64_t k);
|
|
void quantize_row_q4_1_ref(const float * GGML_RESTRICT x, block_q4_1 * GGML_RESTRICT y, int64_t k);
|
|
void quantize_row_q5_0_ref(const float * GGML_RESTRICT x, block_q5_0 * GGML_RESTRICT y, int64_t k);
|
|
void quantize_row_q5_1_ref(const float * GGML_RESTRICT x, block_q5_1 * GGML_RESTRICT y, int64_t k);
|
|
void quantize_row_q8_0_ref(const float * GGML_RESTRICT x, block_q8_0 * GGML_RESTRICT y, int64_t k);
|
|
void quantize_row_q8_1_ref(const float * GGML_RESTRICT x, block_q8_1 * GGML_RESTRICT y, int64_t k);
|
|
|
|
void quantize_row_q2_K_ref(const float * GGML_RESTRICT x, block_q2_K * GGML_RESTRICT y, int64_t k);
|
|
void quantize_row_q3_K_ref(const float * GGML_RESTRICT x, block_q3_K * GGML_RESTRICT y, int64_t k);
|
|
void quantize_row_q4_K_ref(const float * GGML_RESTRICT x, block_q4_K * GGML_RESTRICT y, int64_t k);
|
|
void quantize_row_q5_K_ref(const float * GGML_RESTRICT x, block_q5_K * GGML_RESTRICT y, int64_t k);
|
|
void quantize_row_q6_K_ref(const float * GGML_RESTRICT x, block_q6_K * GGML_RESTRICT y, int64_t k);
|
|
void quantize_row_q8_K_ref(const float * GGML_RESTRICT x, block_q8_K * GGML_RESTRICT y, int64_t k);
|
|
|
|
void quantize_row_tq1_0_ref(const float * GGML_RESTRICT x, block_tq1_0 * GGML_RESTRICT y, int64_t k);
|
|
void quantize_row_tq2_0_ref(const float * GGML_RESTRICT x, block_tq2_0 * GGML_RESTRICT y, int64_t k);
|
|
|
|
void quantize_row_iq3_xxs_ref(const float * GGML_RESTRICT x, block_iq3_xxs * GGML_RESTRICT y, int64_t k);
|
|
void quantize_row_iq4_nl_ref (const float * GGML_RESTRICT x, block_iq4_nl * GGML_RESTRICT y, int64_t k);
|
|
void quantize_row_iq4_xs_ref (const float * GGML_RESTRICT x, block_iq4_xs * GGML_RESTRICT y, int64_t k);
|
|
void quantize_row_iq3_s_ref (const float * GGML_RESTRICT x, block_iq3_s * GGML_RESTRICT y, int64_t k);
|
|
void quantize_row_iq2_s_ref (const float * GGML_RESTRICT x, block_iq2_s * GGML_RESTRICT y, int64_t k);
|
|
|
|
void quantize_row_q4_0(const float * GGML_RESTRICT x, void * GGML_RESTRICT y, int64_t k);
|
|
void quantize_row_q4_1(const float * GGML_RESTRICT x, void * GGML_RESTRICT y, int64_t k);
|
|
void quantize_row_q5_0(const float * GGML_RESTRICT x, void * GGML_RESTRICT y, int64_t k);
|
|
void quantize_row_q5_1(const float * GGML_RESTRICT x, void * GGML_RESTRICT y, int64_t k);
|
|
void quantize_row_q8_0(const float * GGML_RESTRICT x, void * GGML_RESTRICT y, int64_t k);
|
|
void quantize_row_q8_1(const float * GGML_RESTRICT x, void * GGML_RESTRICT y, int64_t k);
|
|
|
|
void quantize_row_q2_K(const float * GGML_RESTRICT x, void * GGML_RESTRICT y, int64_t k);
|
|
void quantize_row_q3_K(const float * GGML_RESTRICT x, void * GGML_RESTRICT y, int64_t k);
|
|
void quantize_row_q4_K(const float * GGML_RESTRICT x, void * GGML_RESTRICT y, int64_t k);
|
|
void quantize_row_q5_K(const float * GGML_RESTRICT x, void * GGML_RESTRICT y, int64_t k);
|
|
void quantize_row_q6_K(const float * GGML_RESTRICT x, void * GGML_RESTRICT y, int64_t k);
|
|
void quantize_row_q8_K(const float * GGML_RESTRICT x, void * GGML_RESTRICT y, int64_t k);
|
|
|
|
void quantize_row_tq1_0(const float * GGML_RESTRICT x, void * GGML_RESTRICT y, int64_t k);
|
|
void quantize_row_tq2_0(const float * GGML_RESTRICT x, void * GGML_RESTRICT y, int64_t k);
|
|
|
|
void quantize_row_iq3_xxs(const float * GGML_RESTRICT x, void * GGML_RESTRICT y, int64_t k);
|
|
void quantize_row_iq4_nl (const float * GGML_RESTRICT x, void * GGML_RESTRICT y, int64_t k);
|
|
void quantize_row_iq4_xs (const float * GGML_RESTRICT x, void * GGML_RESTRICT y, int64_t k);
|
|
void quantize_row_iq3_s (const float * GGML_RESTRICT x, void * GGML_RESTRICT y, int64_t k);
|
|
void quantize_row_iq2_s (const float * GGML_RESTRICT x, void * GGML_RESTRICT y, int64_t k);
|
|
|
|
// Dequantization
|
|
void dequantize_row_q4_0(const block_q4_0 * GGML_RESTRICT x, float * GGML_RESTRICT y, int64_t k);
|
|
void dequantize_row_q4_1(const block_q4_1 * GGML_RESTRICT x, float * GGML_RESTRICT y, int64_t k);
|
|
void dequantize_row_q5_0(const block_q5_0 * GGML_RESTRICT x, float * GGML_RESTRICT y, int64_t k);
|
|
void dequantize_row_q5_1(const block_q5_1 * GGML_RESTRICT x, float * GGML_RESTRICT y, int64_t k);
|
|
void dequantize_row_q8_0(const block_q8_0 * GGML_RESTRICT x, float * GGML_RESTRICT y, int64_t k);
|
|
//void dequantize_row_q8_1(const block_q8_1 * GGML_RESTRICT x, float * GGML_RESTRICT y, int64_t k);
|
|
|
|
void dequantize_row_q2_K(const block_q2_K * GGML_RESTRICT x, float * GGML_RESTRICT y, int64_t k);
|
|
void dequantize_row_q3_K(const block_q3_K * GGML_RESTRICT x, float * GGML_RESTRICT y, int64_t k);
|
|
void dequantize_row_q4_K(const block_q4_K * GGML_RESTRICT x, float * GGML_RESTRICT y, int64_t k);
|
|
void dequantize_row_q5_K(const block_q5_K * GGML_RESTRICT x, float * GGML_RESTRICT y, int64_t k);
|
|
void dequantize_row_q6_K(const block_q6_K * GGML_RESTRICT x, float * GGML_RESTRICT y, int64_t k);
|
|
void dequantize_row_q8_K(const block_q8_K * GGML_RESTRICT x, float * GGML_RESTRICT y, int64_t k);
|
|
|
|
void dequantize_row_tq1_0(const block_tq1_0 * GGML_RESTRICT x, float * GGML_RESTRICT y, int64_t k);
|
|
void dequantize_row_tq2_0(const block_tq2_0 * GGML_RESTRICT x, float * GGML_RESTRICT y, int64_t k);
|
|
|
|
void dequantize_row_iq2_xxs(const block_iq2_xxs * GGML_RESTRICT x, float * GGML_RESTRICT y, int64_t k);
|
|
void dequantize_row_iq2_xs (const block_iq2_xs * GGML_RESTRICT x, float * GGML_RESTRICT y, int64_t k);
|
|
void dequantize_row_iq2_s (const block_iq2_s * GGML_RESTRICT x, float * GGML_RESTRICT y, int64_t k);
|
|
void dequantize_row_iq3_xxs(const block_iq3_xxs * GGML_RESTRICT x, float * GGML_RESTRICT y, int64_t k);
|
|
void dequantize_row_iq1_s (const block_iq1_s * GGML_RESTRICT x, float * GGML_RESTRICT y, int64_t k);
|
|
void dequantize_row_iq1_m (const block_iq1_m * GGML_RESTRICT x, float * GGML_RESTRICT y, int64_t k);
|
|
void dequantize_row_iq4_nl (const block_iq4_nl * GGML_RESTRICT x, float * GGML_RESTRICT y, int64_t k);
|
|
void dequantize_row_iq4_xs (const block_iq4_xs * GGML_RESTRICT x, float * GGML_RESTRICT y, int64_t k);
|
|
void dequantize_row_iq3_s (const block_iq3_s * GGML_RESTRICT x, float * GGML_RESTRICT y, int64_t k);
|
|
|
|
// Dot product
|
|
void ggml_vec_dot_q4_0_q8_0(int n, float * GGML_RESTRICT s, size_t bs, const void * GGML_RESTRICT vx, size_t bx, const void * GGML_RESTRICT vy, size_t by, int nrc);
|
|
void ggml_vec_dot_q4_1_q8_1(int n, float * GGML_RESTRICT s, size_t bs, const void * GGML_RESTRICT vx, size_t bx, const void * GGML_RESTRICT vy, size_t by, int nrc);
|
|
void ggml_vec_dot_q5_0_q8_0(int n, float * GGML_RESTRICT s, size_t bs, const void * GGML_RESTRICT vx, size_t bx, const void * GGML_RESTRICT vy, size_t by, int nrc);
|
|
void ggml_vec_dot_q5_1_q8_1(int n, float * GGML_RESTRICT s, size_t bs, const void * GGML_RESTRICT vx, size_t bx, const void * GGML_RESTRICT vy, size_t by, int nrc);
|
|
void ggml_vec_dot_q8_0_q8_0(int n, float * GGML_RESTRICT s, size_t bs, const void * GGML_RESTRICT vx, size_t bx, const void * GGML_RESTRICT vy, size_t by, int nrc);
|
|
|
|
void ggml_vec_dot_q2_K_q8_K(int n, float * GGML_RESTRICT s, size_t bs, const void * GGML_RESTRICT vx, size_t bx, const void * GGML_RESTRICT vy, size_t by, int nrc);
|
|
void ggml_vec_dot_q3_K_q8_K(int n, float * GGML_RESTRICT s, size_t bs, const void * GGML_RESTRICT vx, size_t bx, const void * GGML_RESTRICT vy, size_t by, int nrc);
|
|
void ggml_vec_dot_q4_K_q8_K(int n, float * GGML_RESTRICT s, size_t bs, const void * GGML_RESTRICT vx, size_t bx, const void * GGML_RESTRICT vy, size_t by, int nrc);
|
|
void ggml_vec_dot_q5_K_q8_K(int n, float * GGML_RESTRICT s, size_t bs, const void * GGML_RESTRICT vx, size_t bx, const void * GGML_RESTRICT vy, size_t by, int nrc);
|
|
void ggml_vec_dot_q6_K_q8_K(int n, float * GGML_RESTRICT s, size_t bs, const void * GGML_RESTRICT vx, size_t bx, const void * GGML_RESTRICT vy, size_t by, int nrc);
|
|
|
|
void ggml_vec_dot_tq1_0_q8_K(int n, float * GGML_RESTRICT s, size_t bs, const void * GGML_RESTRICT vx, size_t bx, const void * GGML_RESTRICT vy, size_t by, int nrc);
|
|
void ggml_vec_dot_tq2_0_q8_K(int n, float * GGML_RESTRICT s, size_t bs, const void * GGML_RESTRICT vx, size_t bx, const void * GGML_RESTRICT vy, size_t by, int nrc);
|
|
|
|
void ggml_vec_dot_iq2_xxs_q8_K(int n, float * GGML_RESTRICT s, size_t bs, const void * GGML_RESTRICT vx, size_t bx, const void * GGML_RESTRICT vy, size_t by, int nrc);
|
|
void ggml_vec_dot_iq2_xs_q8_K (int n, float * GGML_RESTRICT s, size_t bs, const void * GGML_RESTRICT vx, size_t bx, const void * GGML_RESTRICT vy, size_t by, int nrc);
|
|
void ggml_vec_dot_iq2_s_q8_K (int n, float * GGML_RESTRICT s, size_t bs, const void * GGML_RESTRICT vx, size_t bx, const void * GGML_RESTRICT vy, size_t by, int nrc);
|
|
void ggml_vec_dot_iq3_xxs_q8_K(int n, float * GGML_RESTRICT s, size_t bs, const void * GGML_RESTRICT vx, size_t bx, const void * GGML_RESTRICT vy, size_t by, int nrc);
|
|
void ggml_vec_dot_iq1_s_q8_K (int n, float * GGML_RESTRICT s, size_t bs, const void * GGML_RESTRICT vx, size_t bx, const void * GGML_RESTRICT vy, size_t by, int nrc);
|
|
void ggml_vec_dot_iq1_m_q8_K (int n, float * GGML_RESTRICT s, size_t bs, const void * GGML_RESTRICT vx, size_t bx, const void * GGML_RESTRICT vy, size_t by, int nrc);
|
|
void ggml_vec_dot_iq4_nl_q8_0 (int n, float * GGML_RESTRICT s, size_t bs, const void * GGML_RESTRICT vx, size_t bx, const void * GGML_RESTRICT vy, size_t by, int nrc);
|
|
void ggml_vec_dot_iq4_xs_q8_K (int n, float * GGML_RESTRICT s, size_t bs, const void * GGML_RESTRICT vx, size_t bx, const void * GGML_RESTRICT vy, size_t by, int nrc);
|
|
void ggml_vec_dot_iq3_s_q8_K (int n, float * GGML_RESTRICT s, size_t bs, const void * GGML_RESTRICT vx, size_t bx, const void * GGML_RESTRICT vy, size_t by, int nrc);
|
|
|
|
// Quantization utilizing an importance matrix (a.k.a. "Activation aWare Quantization")
|
|
size_t quantize_iq2_xxs(const float * GGML_RESTRICT src, void * GGML_RESTRICT dst, int64_t nrows, int64_t n_per_row, const float * imatrix);
|
|
size_t quantize_iq2_xs (const float * GGML_RESTRICT src, void * GGML_RESTRICT dst, int64_t nrows, int64_t n_per_row, const float * imatrix);
|
|
size_t quantize_iq2_s (const float * GGML_RESTRICT src, void * GGML_RESTRICT dst, int64_t nrows, int64_t n_per_row, const float * imatrix);
|
|
size_t quantize_iq3_xxs(const float * GGML_RESTRICT src, void * GGML_RESTRICT dst, int64_t nrows, int64_t n_per_row, const float * imatrix);
|
|
size_t quantize_iq1_s (const float * GGML_RESTRICT src, void * GGML_RESTRICT dst, int64_t nrows, int64_t n_per_row, const float * imatrix);
|
|
size_t quantize_iq1_m (const float * GGML_RESTRICT src, void * GGML_RESTRICT dst, int64_t nrows, int64_t n_per_row, const float * imatrix);
|
|
size_t quantize_iq4_nl (const float * GGML_RESTRICT src, void * GGML_RESTRICT dst, int64_t nrows, int64_t n_per_row, const float * imatrix);
|
|
size_t quantize_iq4_xs (const float * GGML_RESTRICT src, void * GGML_RESTRICT dst, int64_t nrows, int64_t n_per_row, const float * imatrix);
|
|
size_t quantize_iq3_s (const float * GGML_RESTRICT src, void * GGML_RESTRICT dst, int64_t nrows, int64_t n_per_row, const float * imatrix);
|
|
|
|
size_t quantize_tq1_0(const float * GGML_RESTRICT src, void * GGML_RESTRICT dst, int64_t nrows, int64_t n_per_row, const float * imatrix);
|
|
size_t quantize_tq2_0(const float * GGML_RESTRICT src, void * GGML_RESTRICT dst, int64_t nrows, int64_t n_per_row, const float * imatrix);
|
|
|
|
size_t quantize_q2_K(const float * GGML_RESTRICT src, void * GGML_RESTRICT dst, int64_t nrows, int64_t n_per_row, const float * imatrix);
|
|
size_t quantize_q3_K(const float * GGML_RESTRICT src, void * GGML_RESTRICT dst, int64_t nrows, int64_t n_per_row, const float * imatrix);
|
|
size_t quantize_q4_K(const float * GGML_RESTRICT src, void * GGML_RESTRICT dst, int64_t nrows, int64_t n_per_row, const float * imatrix);
|
|
size_t quantize_q5_K(const float * GGML_RESTRICT src, void * GGML_RESTRICT dst, int64_t nrows, int64_t n_per_row, const float * imatrix);
|
|
size_t quantize_q6_K(const float * GGML_RESTRICT src, void * GGML_RESTRICT dst, int64_t nrows, int64_t n_per_row, const float * imatrix);
|
|
size_t quantize_q4_0(const float * GGML_RESTRICT src, void * GGML_RESTRICT dst, int64_t nrows, int64_t n_per_row, const float * imatrix);
|
|
size_t quantize_q4_1(const float * GGML_RESTRICT src, void * GGML_RESTRICT dst, int64_t nrows, int64_t n_per_row, const float * imatrix);
|
|
size_t quantize_q5_0(const float * GGML_RESTRICT src, void * GGML_RESTRICT dst, int64_t nrows, int64_t n_per_row, const float * imatrix);
|
|
size_t quantize_q5_1(const float * GGML_RESTRICT src, void * GGML_RESTRICT dst, int64_t nrows, int64_t n_per_row, const float * imatrix);
|
|
size_t quantize_q8_0(const float * GGML_RESTRICT src, void * GGML_RESTRICT dst, int64_t nrows, int64_t n_per_row, const float * imatrix);
|
|
|
|
void iq2xs_init_impl(enum ggml_type type);
|
|
void iq2xs_free_impl(enum ggml_type type);
|
|
void iq3xs_init_impl(int grid_size);
|
|
void iq3xs_free_impl(int grid_size);
|
|
|
|
#if defined(__ARM_FEATURE_SVE)
|
|
extern int ggml_sve_cnt_b;
|
|
#endif
|
|
|
|
#ifdef __cplusplus
|
|
}
|
|
#endif
|