root/llama.cpp
1
0
Fork 0
mirror of https://github.com/ggerganov/llama.cpp.git synced 2025-01-09 02:01:44 +00:00
Code Issues Actions 9 Packages Projects Releases Wiki Activity

Commit Graph

  • 1176f37198 Fix handling of "invalid property" when creating OpenCL command queue Maarten ter Huurne 2023-05-22 23:55:33 +0200
  • 2e6cd4b025
         OpenCL Token Generation Acceleration (#1459) master-2e6cd4b 0cc4m 2023-05-22 23:33:24 +0200
  • bf1f02ddc0
         chat-persistent.sh : use bracket expressions in grep Senemu 2023-05-22 21:17:19 +0000
  • 046def2d9a
         Merge branch 'master' of github.com:biw/llama.cpp into added-disable-tty Ben Williams 2023-05-22 14:03:36 -0700
  • cb28080aef
         Small compiler warning fixes Henri Vasserman 2023-05-22 23:14:15 +0300
  • 4dfd4fe1eb Restore default platform + device selection by id behavior 0cc4m 2023-05-22 21:51:39 +0200
  • e1ee2810ea
         change to fprintf Henri Vasserman 2023-05-22 22:18:02 +0300
  • 6d40cc3a44
         remove trailing whitespace xaedes 2023-05-22 20:56:35 +0200
  • d3acbf644e
         simplify code xaedes 2023-05-22 20:53:57 +0200
  • ee9aaaaebc Add conversion from FP32 quants to FP16 quants model Jason0214 2023-05-23 01:20:11 +0800
  • 4a55951464 Only copy f16/f32 buffer if not already on GPU 0cc4m 2023-05-22 18:46:51 +0200
  • 0651679302
         save checkpoint only when it was trained xaedes 2023-05-22 16:56:28 +0200
  • cc440bd438
         fix bug in get_samples which corrupted training targets xaedes 2023-05-22 16:55:52 +0200
  • b763d6f1f2
         remove unused functions xaedes 2023-05-22 16:54:21 +0200
  • 7894e85788 fixed a bug in previous klite Concedo 2023-05-22 21:54:24 +0800
  • a05da31fe7 updated embedded lite Concedo 2023-05-22 20:58:54 +0800
  • 47e41fa8ce Add means to exit interactive mode changhz 2023-05-22 08:28:25 -0400
  • b78ceb1a2e
         merge-hf-and-lora-to-hf.py FNsi 2023-05-22 19:31:17 +0800
  • 29995194e3
         merge-hf-and-lora-to-hf.py FNsi 2023-05-22 19:29:24 +0800
  • 1fd5d10b07
         Update merge-hf-and-lora-to-hf.py FNsi 2023-05-22 19:27:45 +0800
  • 39287b06da
         Update merge-hf-and-lora-to-hf.py FNsi 2023-05-22 19:26:00 +0800
  • e20e302e87 Merge branch 'master' into concedo_experimental Concedo 2023-05-22 17:05:34 +0800
  • b9f06a7670 mavx only for windows by default, let them eat march native. Concedo 2023-05-22 16:48:55 +0800
  • 981d5ba866 Merge remote-tracking branch 'occam/opencl-dev' into concedo_experimental Concedo 2023-05-22 16:16:48 +0800
  • 169a26d15f removed unused build targets Concedo 2023-05-22 13:53:10 +0800
  • b6a30489a7
         merge-hf-and-lora-to-hf.py FNsi 2023-05-22 13:08:59 +0800
  • 587308a202 fixed some build errors on linux, changed icon resolution, added more error printing Concedo 2023-05-22 12:18:42 +0800
  • 9d058c2096 avoid sending finalize op to thread pool if it does nothing zrm 2023-05-21 18:11:03 -0400
  • 0d23f8ce8d disable mmap prefetch/readahead for NUMA systems zrm 2023-05-21 16:33:10 -0400
  • 42d9b4cfc2
         store optimizer state in training checkpoint and add learning schedule xaedes 2023-05-21 21:36:04 +0200
  • 37c69435f0
         print suppressed newline tokens as string "\n" xaedes 2023-05-21 21:17:46 +0200
  • 93eb8f7752
         add forward function without using cache, for more performant training xaedes 2023-05-21 21:14:49 +0200
  • 2afd218479
         fix bug in llama_sample_token_mirostat_v2 xaedes 2023-05-21 21:12:10 +0200
  • ec1783c3e0
         add ggml_opt_context, so that we can properly resume training xaedes 2023-05-21 15:16:07 +0200
  • 6fc5f17e21 detect NUMA systems and pin work threads to nodes (linux) zrm 2023-05-21 14:09:52 -0400
  • 7e4ea5beff
         examples : add server example with REST API (#1443) master-7e4ea5b Steward Garcia 2023-05-21 11:51:18 -0600
  • 2257f9f691 Remove trailing space Howard Su 2023-05-21 23:03:36 +0800
  • fea84c3cf5 fix for stupid msvc compiler Concedo 2023-05-21 22:41:33 +0800
  • 80f1faac87 format fix Howard Su 2023-05-21 22:31:19 +0800
  • 006d5707e8 Support V3 format upgrade Howard Su 2023-05-21 22:14:27 +0800
  • 7780e4f479
         make : .PHONY clean (#1553) master-7780e4f Stefan Sydow 2023-05-21 16:03:44 +0200
  • b16c085c49
         examples : fix benchmark-matmult Georgi Gerganov 2023-05-21 16:56:33 +0300
  • f0f6824994
         fix make clean Stefan Sydow 2023-05-21 15:50:54 +0200
  • 60e0c67874 fix compile errors on cuda Concedo 2023-05-21 21:13:17 +0800
  • 1eee9255e7
         add missing default parameters for adam optimizer xaedes 2023-05-21 15:03:51 +0200
  • 33528f5b1d fix for cublas Concedo 2023-05-21 21:03:36 +0800
  • 994be9a4db fix for cublas Concedo 2023-05-21 21:02:21 +0800
  • 57c2f4f909
         fix random weight initialization scale xaedes 2023-05-21 12:18:47 +0200
  • 96514971dd
         use inplace operations in cross_entropy_loss xaedes 2023-05-21 12:17:57 +0200
  • 24127ebf98 updated lite, fixed some encoding issues Concedo 2023-05-21 17:29:00 +0800
  • 265db9834e
         ggml : output 3d sizes in ggml_graph_dump_dot() master-265db98 Georgi Gerganov 2023-05-21 11:56:23 +0300
  • 10cbc311e3 Support more data types Howard Su 2023-05-18 09:49:25 +0800
  • d521d09380 Support Q4_1 Howard Su 2023-05-17 23:42:17 +0800
  • b8d69650dc Upgrade v1 format to v2 by leveraging quantize Howard Su 2023-05-17 23:39:39 +0800
  • 18e9dd87da Explicitely set GEMM type 0cc4m 2023-05-21 08:34:17 +0200
  • b6b39960c0 Use compile args for preprocessing constants 0cc4m 2023-05-21 08:17:17 +0200
  • a1657d0233 Add OpenCL compile options 0cc4m 2023-05-19 21:18:57 +0200
  • e41a7ae40c Fix convert_row_f16 kernel issue 0cc4m 2023-05-18 08:05:19 +0200
  • 457eff920e Deduplicate dequant kernels 0cc4m 2023-05-18 07:35:40 +0200
  • 42e1a2ba3d Fix tensor load to device 0cc4m 2023-05-16 18:49:49 +0200
  • cda2d488f9 Fix error in convert f16 to f32 kernel call 0cc4m 2023-05-16 13:05:33 +0200
  • 915d0d1168 Generate dequant_mul_mat kernels from simple templates 0cc4m 2023-05-16 07:42:01 +0200
  • 1968380373 Fix CMakeLists.txt 0cc4m 2023-05-15 19:51:23 +0200
  • cb588e2aa4 Add remaining dequant_mul_mat functions 0cc4m 2023-05-14 22:19:54 +0200
  • 8c7a7cea2e Fix dequant_mul_mat kernel 0cc4m 2023-05-14 21:26:07 +0200
  • 5f610c90bf Fix bugs in dequant_mul_mat code 0cc4m 2023-05-14 21:14:05 +0200
  • 17e53dbb7e Refactor OpenCL code to work more like the CUDA code, add missing functions 0cc4m 2023-05-14 17:01:46 +0200
  • a7e3bee4cc Move back to C++ for OpenCL 0cc4m 2023-05-14 17:00:37 +0200
  • 651f50f6ca
         merge-hf-and-lora-to-hf.py FNsi 2023-05-21 11:28:00 +0800
  • 5dbdc65700
         merge-hf-and-lora-to-hf.py FNsi 2023-05-21 11:19:18 +0800
  • 28bec1eb25
         merge-hf-and-lora-to-hf.py FNsi 2023-05-21 11:17:39 +0800
  • d892edcf7d
         Update merge-hf-and-lora-to-hf.py FNsi 2023-05-21 11:14:52 +0800
  • 84d3432a98
         Rename merge-HF-and-lora-to-HF.py to merge-hf-and-lora-to-hf.py FNsi 2023-05-21 11:09:29 +0800
  • e970d41095
         Update and rename merge.py to merge-HF-and-lora-to-HF.py FNsi 2023-05-21 11:09:08 +0800
  • 600ace39c8
         update warp size Henri Vasserman 2023-05-20 23:42:20 +0300
  • b19fefef94
         Forwardcompat Henri Vasserman 2023-05-20 23:28:08 +0300
  • 75e4548821 missed out gpt2 Concedo 2023-05-21 01:44:47 +0800
  • 2ead735f08 initial integration completed Concedo 2023-05-21 01:29:20 +0800
  • d6123f738a Merge commit 'ea600071cb005267e9e8f2629c1e406dd5fde083' into concedo_experimental Concedo 2023-05-21 01:27:27 +0800
  • fab49c685e
         ggml : update WASM SIMD master-fab49c6 Georgi Gerganov 2023-05-20 20:00:41 +0300
  • d418146535 fixed a token decoding bug Concedo 2023-05-21 00:53:20 +0800
  • d1824f1e88 Merge branch 'master' into concedo_experimental Concedo 2023-05-21 00:30:06 +0800
  • 5032e0fd64 trying to fix ggjt v3 Concedo 2023-05-21 00:29:50 +0800
  • c048bcfec4 remove old filever checks (+7 squashed commit) Concedo 2023-05-20 16:47:44 +0800
  • c66115b833
         Merge 'origin/master' into hipblas Henri Vasserman 2023-05-20 18:29:31 +0300
  • b8ee340abe
         feature : support blis and other blas implementation (#1536) master-b8ee340 Zenix 2023-05-20 23:58:31 +0900
  • 9ecb30f959
         OpenCL: Fixes for older devices. (#1435) master-9ecb30f Henri Vasserman 2023-05-20 17:57:39 +0300
  • 6b5a4ab957 Fix: blas changes on ci zenix 2023-05-20 22:03:33 +0900
  • c29378e5a8
         clang-tidi Henri Vasserman 2023-05-20 16:03:25 +0300
  • 29cf5596fe
         llama : define magic numbers as integer constants (#1518) (#1520) master-29cf559 Juuso Alasuutari 2023-05-20 15:58:15 +0300
  • ef17d99f65
         implement AdamW in ggml_opt_adam by adding weight decay parameter (default 0.001f) xaedes 2023-05-20 14:54:40 +0200
  • c69f0cd6e4 Define magic numbers as integer constants (#1518) Juuso Alasuutari 2023-05-20 15:54:03 +0300
  • f4e9ce7998
         enable gradient propagation for inplace add1 and scale operations xaedes 2023-05-20 14:49:19 +0200
  • a6aafdd719
         add ggml_add1_inplace to header xaedes 2023-05-20 14:47:56 +0200
  • 3de84b2606
         ggml : add ggml_clamp() (#1539) master-3de84b2 Georgi Gerganov 2023-05-20 15:34:45 +0300
  • 71ac58ae53
         make clang-tidy happy Henri Vasserman 2023-05-20 15:29:26 +0300
  • ad9ab0e3fe
         editorconfig fixes Henri Vasserman 2023-05-20 15:27:24 +0300
  • 4f97f73db2
         fix indexing issue Henri Vasserman 2023-05-20 15:21:38 +0300
  • affc76edfd
         cuda : loading models directly into VRAM, norm calculation on GPU, broadcasting for ggml_mul (#1483) master-affc76e Johannes Gäßler 2023-05-20 14:19:28 +0200
  • 37f2c6c251 Add forgotten fclose() JohannesGaessler 2023-05-20 14:16:40 +0200