vbatts/llama.cpp
1
0
Fork 0
Code Issues Pull requests Projects Releases Packages Wiki Activity Actions

Commit graph

  • 56551bc11f
         readme : add notice about upcoming breaking change Georgi Gerganov 2023-05-08 22:52:18 +03:00
  • 0e48eb6f6b
         ggml : uniform 5th bit extraction Georgi Gerganov 2023-05-08 22:18:15 +03:00
  • 948d124837
         AVX implementations (#1370) Stephan Walter 2023-05-08 19:14:06 +00:00
  • d155f0f865
         scripts : add script for measuring the time per token Georgi Gerganov 2023-05-08 22:06:54 +03:00
  • de51f6faa5 AVX implementations Stephan Walter 2023-05-06 19:37:48 +02:00
  • 8fbf7777ce
         ggml : fix Q5_0 quantization Georgi Gerganov 2023-05-07 20:26:02 +03:00
  • 60f62bbc85
         ggml : minor formatting Georgi Gerganov 2023-05-07 20:00:01 +03:00
  • 7cdc08a5d1
         ggml : remove Q4_2 mode Georgi Gerganov 2023-05-07 18:26:59 +03:00
  • b47bd2877f
         ggml : update cuBLAS + normalize variable names Georgi Gerganov 2023-05-07 18:23:59 +03:00
  • c216656990
         ggml : fix Q4_1 quantization Georgi Gerganov 2023-05-07 18:07:11 +03:00
  • 4991499a5a
         ggml : remove WASM SIMD bit shuffling + remove vzip for ARM 32-bit Georgi Gerganov 2023-05-05 17:23:41 +03:00
  • ba953d6e21
         ggml : simplify scalar dot Georgi Gerganov 2023-05-05 17:12:58 +03:00
  • c7af9042b3
         ggml : remove Q5_1 bit shuffling (ARM NEON + scalar) Georgi Gerganov 2023-05-05 17:09:11 +03:00
  • 39bb8e7d19
         ggml : 2x faster scalar implementations Georgi Gerganov 2023-05-04 23:31:35 +03:00
  • 796f8ae261
         ggml : remove Q5_0 bit shuffling (ARM NEON) Georgi Gerganov 2023-05-04 22:55:10 +03:00
  • a6a1d96c91
         ggml : remove Q4_2 bit shuffling (WIP, BROKEN) Georgi Gerganov 2023-05-04 22:07:40 +03:00
  • 086cfea11f
         ggml : nibbles_from_floats() + bytes_from_nibbles() (ARM NEON) Georgi Gerganov 2023-05-04 21:51:42 +03:00
  • edb6c8bb66
         ggml : remove Q4_1 bit shuffling (ARM NEON + reference) Georgi Gerganov 2023-05-04 20:53:14 +03:00
  • a546dc6d60
         ggml : remove Q4_0 bit shufling (ARM NEON) Georgi Gerganov 2023-05-03 23:13:37 +03:00
  • 6ca682b19d
         ggml : swap vDSP_vsub args as per documentation Georgi Gerganov 2023-05-08 21:15:41 +03:00
  • 9c3fe4eb76
         swap arguments to vDSP_vdiv call xaedes 2023-05-08 20:09:52 +02:00
  • cafbb785fa
         swap arguments to vDSP_vdiv call xaedes 2023-05-08 20:09:52 +02:00
  • 006db8e0bb fixup! More GPU threads for CUDA kernels JohannesGaessler 2023-05-08 19:45:02 +02:00
  • fe60904eef
         readme : add TOC and Pygmalion instructions (#1359) AlpinDale 2023-05-08 21:03:30 +04:30
  • 6cc42deda5
         ggml : fix nullptr derefs in GGML_OP_CONT and GGML_OP_RESHAPE back Georgi Gerganov 2023-05-08 18:50:04 +03:00
  • 78af3e92c9
         ggml : fix compiler warnings + cosmetic changes Georgi Gerganov 2023-05-08 18:37:17 +03:00
  • 0d72207ac3
         c++ in baby-llama example xaedes 2023-05-08 16:56:41 +02:00
  • 003ba2fb43
         llama : fix hparams shadow (#1367) master-003ba2f Pavol Rusnak 2023-05-08 16:48:21 +02:00
  • f9a6364912
         llama : require first token to be BOS (#1303) master-f9a6364 Georgi Gerganov 2023-05-08 17:41:54 +03:00
  • 0e94ea65c6
         perplexity : add clarifying comments Georgi Gerganov 2023-05-08 17:41:26 +03:00
  • dea9c9359a
         c++ in baby-llama example xaedes 2023-05-08 16:40:31 +02:00
  • cdf40a9bea
         readme : update perplexity values after BOS fix Georgi Gerganov 2023-05-08 17:37:49 +03:00
  • 10632d820c
         llama: fix shadowing Pavol Rusnak 2023-05-08 15:44:59 +02:00
  • 2f2eff6e13 the dark gods have been sated, and redpajama is integrated... but at what cost? Concedo 2023-05-08 20:58:00 +08:00
  • afa160d5cc fix typo in default model path Andras Salamon 2023-05-08 13:04:15 +01:00
  • 95078cc554
         convert: add ability to convert safetensors files (#1276) ubik2 2023-05-08 04:54:26 -07:00
  • d0199b3bc3 fixup! More GPU threads for CUDA kernels JohannesGaessler 2023-05-08 12:56:32 +02:00
  • d8c36c91f8
         Check for safetensors files first, and only use PyTorch versions when safetensors aren't available ubik2 2023-05-08 00:56:12 -07:00
  • b9904c3093 up ver Concedo 2023-05-08 11:13:16 +08:00
  • 1083876a1b Merge branch 'master' into concedo_experimental Concedo 2023-05-08 11:12:42 +08:00
  • 4244d97c38
         Merge bcc7f8b6bf into 1f48b0abcf MagisterLuddite 2023-05-08 11:11:30 +08:00
  • 89d70886a4 added support for setting custom context size at load time (memory allocation) Concedo 2023-05-08 11:11:25 +08:00
  • 1f48b0abcf
         Documented CUDA reproducibility, added warning (#1346) master-1f48b0a Johannes Gäßler 2023-05-08 02:42:01 +02:00
  • 1ecbece752
         disable slow tests grad0 and opt to avoid exceeding timeouts xaedes 2023-05-08 02:29:36 +02:00
  • f5301061b6
         remove busy loop that was used as sleep for slower sinus wave generation xaedes 2023-05-08 01:12:37 +02:00
  • 4997bc5819
         reduce number of test-grad0 iterations xaedes 2023-05-08 00:57:41 +02:00
  • 2936dd60a4
         remove trailing whitespace xaedes 2023-05-08 00:04:54 +02:00
  • 7c8768f819
         add missing include for strcmp, etc xaedes 2023-05-07 23:42:44 +02:00
  • 660836f0ff
         fix call to ggml_set_name xaedes 2023-05-07 23:39:57 +02:00
  • 9dd8e405fb
         rename print functions in baby-llama example xaedes 2023-05-07 22:43:23 +02:00
  • 47ad186628
         revert disabling of threading for rms_norm and norm xaedes 2023-05-07 21:55:25 +02:00
  • 5d9fed7e7f
         remove shape annotations in llama_eval_internal xaedes 2023-05-07 21:45:21 +02:00
  • d20ba6f6e6
         update static assert of GGML_OP_COUNT xaedes 2023-05-07 21:42:42 +02:00
  • e643fa1619
         smaller default values for baby llama model parameters xaedes 2023-05-07 21:31:55 +02:00
  • 10184b0675
         Added TOC and Pygmalion to README AlpinDale 2023-05-07 23:55:00 +04:30
  • ee565f34e3
         Merge branch 'master' into train-example xaedes 2023-05-07 21:24:12 +02:00
  • 4764842120
         change name of GGML_OP_ADD_AT to GGML_OP_ACC xaedes 2023-05-07 21:14:57 +02:00
  • e0de09d77e
         shorten code using a variable xaedes 2023-05-07 19:48:38 +02:00
  • 49d6daa11e
         vastly improve training results xaedes 2023-05-07 19:46:05 +02:00
  • 93201abdb7
         add trainable lora-only model with all big matrices C split into A,B with A*B=C xaedes 2023-05-07 19:44:51 +02:00
  • 8d8de07a4e fixup! More GPU threads for CUDA kernels JohannesGaessler 2023-05-07 18:34:04 +02:00
  • 070cbcc1bd
         occupanct function Henri Vasserman 2023-05-07 18:10:56 +03:00
  • a3296d50aa
         Merge 'origin/master' into hipblas Henri Vasserman 2023-05-07 18:06:04 +03:00
  • 50148408b5 More GPU threads for CUDA kernels JohannesGaessler 2023-05-06 12:17:45 +02:00
  • e1295513a4
         CI: add Windows CLBlast and OpenBLAS builds (#1277) master-e129551 Henri Vasserman 2023-05-07 14:20:09 +03:00
  • 62beded0e7 Merge branch 'master' into concedo_experimental Concedo 2023-05-07 19:10:01 +08:00
  • f12c6f6ec8 ggml : delete unused function, packNibbles_256 katsu560 2023-05-07 20:04:49 +09:00
  • ab4f1fa8e2 Output correct error message in ggml_new_tensor_impl when out of scratch space. KerfuffleV2 2023-05-07 03:52:37 -06:00
  • 5cc9085353 Works with all characters and control codes + Windows console fixes Danny Daemonic 2023-05-07 02:39:10 -07:00
  • d98869197b
         Merge 'origin/master' into cistuff Henri Vasserman 2023-05-07 12:29:29 +03:00
  • 963b27450c
         change build name to 'clblast' Henri Vasserman 2023-05-07 12:29:20 +03:00
  • 89951748db Add ggml_tensor_required_memory function to calculate how much memory creating a tensor will require. KerfuffleV2 2023-05-07 03:26:58 -06:00
  • 0aefa6ab71
         Merge 'origin/master' into hipblas Henri Vasserman 2023-05-07 12:24:41 +03:00
  • baeb482a94
         Revert to default copy Henri Vasserman 2023-05-07 12:24:12 +03:00
  • 30b2b3d655 There were still struct fields and defines conditionally enabled. KerfuffleV2 2023-05-07 02:25:32 -06:00
  • 7f33230a40
         perplexity : add BOS for each chunk Georgi Gerganov 2023-05-07 11:04:52 +03:00
  • 1b0fd45465
         ggml : Allow usage of CLBlast alongside Accelerate.framework (#1336) master-1b0fd45 swittk 2023-05-07 10:03:23 +07:00
  • 4baa85633a
         Fix build jed/spm-clblast Jed Fox 2023-05-06 21:44:07 -04:00
  • 2dc7fc9491
         + ggml-opencl.c Jed Fox 2023-05-06 21:35:00 -04:00
  • dc61b7c682
         spm: link with CLBlast Jed Fox 2023-05-06 21:34:13 -04:00
  • 0e3d7fd4e9
         Remove default arguments from sampling functions (#1343) Jed Fox 2023-05-06 17:01:47 -04:00
  • e91b83b899
         add GGML_ASSERT to catch ggml_rope and back value errors xaedes 2023-05-07 01:47:14 +02:00
  • 561fbe0d1b
         replace inplace operations for training with copying operations to allow gradient propagation xaedes 2023-05-07 01:33:42 +02:00
  • 956511b248
         fix kv_self gradients for training xaedes 2023-05-07 01:32:46 +02:00
  • 47561de7d8
         add ggml_set(ctx, a, b) to set b in view of a and return modified a xaedes 2023-05-07 01:30:34 +02:00
  • 48bcc4dcf9
         fix backward pass for add_at and change arguments to have same order as in view xaedes 2023-05-07 01:27:11 +02:00
  • 226521a4f1
         optimize loss over multiple samples xaedes 2023-05-07 01:23:51 +02:00
  • 7a5dec24f8
         add square_error_loss and cross_entropy_loss functions xaedes 2023-05-07 01:21:26 +02:00
  • 73fd66e9e5
         fix training get_example_targets xaedes 2023-05-07 01:18:17 +02:00
  • 7523107877 Keep API and context fields the same whether or not GGML_RECOVERABLE_ERRORS is defined. KerfuffleV2 2023-05-06 15:48:57 -06:00
  • 1a6987ad92 First rough draft of recoverable errors feature. KerfuffleV2 2023-05-04 14:42:30 -06:00
  • 3924088512
         Remove default arguments from sampling functions (#1343) master-3924088 Jed Fox 2023-05-06 17:01:47 -04:00
  • 4b18cdf632 Documented CUDA reproducibility, added warning JohannesGaessler 2023-05-06 20:53:47 +02:00
  • 289073a532
         Merge 'origin/master' into hipblas Henri Vasserman 2023-05-06 19:59:41 +03:00
  • 80223d98fd
         add test for ggml_sum_rows gradients xaedes 2023-05-06 18:01:32 +02:00
  • e6186d98a5
         implement ggml_repeat support for rank > 2 tensors xaedes 2023-05-06 18:01:17 +02:00
  • fb9e1dde5c
         Add missing parameters to examples & tests Jed Fox 2023-05-06 11:46:44 -04:00
  • 09236f40b4
         llama license text Henri Vasserman 2023-05-06 18:41:56 +03:00
  • 7a15a8370c
         implement backward pass for ggml_sum_rows, necessary for cross entropy loss xaedes 2023-05-06 17:37:51 +02:00
  • 5724628d31
         add test for ggml_log gradients xaedes 2023-05-06 17:36:21 +02:00