Commit graph

  • 783379670a
    metal : restore original F16 mat-vec multiplication Georgi Gerganov 2023-09-07 15:20:07 +03:00
  • ed92c3d4b2
    metal : put warning in kernel_norm to not combine the loops Georgi Gerganov 2023-09-07 14:59:48 +03:00
  • 5e1c4089d8
    metal : fix kernel_norm Georgi Gerganov 2023-09-07 14:11:21 +03:00
  • 405c8e90a0 PR suggestion Iwan Kawrakow 2023-09-07 11:30:14 +02:00
  • a68e1a5656 Parallel RoPE on metal Iwan Kawrakow 2023-09-05 12:31:18 +02:00
  • c9b57dca57 install only necessary cuda sub packages Alon Faraj 2023-09-07 11:24:01 +03:00
  • 39d1b769f5 cmake : follow recent FTM improvements from Makefile Przemyslaw Pawelczyk 2023-09-06 00:12:36 +02:00
  • 41997549fe make : fix OpenBSD build by exposing newer POSIX definitions Przemyslaw Pawelczyk 2023-09-06 00:21:16 +02:00
  • d007288dd8 make : use BSD-specific FTMs to enable alloca on BSDs Przemyslaw Pawelczyk 2023-09-06 00:18:17 +02:00
  • 09b7e8000e make : enable BSD extensions for DragonFlyBSD to expose RLIMIT_MEMLOCK Przemyslaw Pawelczyk 2023-09-06 00:15:23 +02:00
  • 5e47277633 make : enable Darwin extensions for macOS to expose RLIMIT_MEMLOCK Przemyslaw Pawelczyk 2023-06-28 10:16:41 +02:00
  • 50ff7aa900 Do not use _GNU_SOURCE gratuitously. Przemyslaw Pawelczyk 2023-06-28 10:01:03 +02:00
  • 510851d8de unify and simplify "Copy and pack Cuda runtime" Alon Faraj 2023-09-07 11:18:17 +03:00
  • a65d78bf44 bump actions/checkout to v3 Alon Faraj 2023-09-07 11:15:51 +03:00
  • fec2fb19e4
    ggml : posixify madvise and pagesize (#3037) b1188 Przemysław Pawełczyk 2023-09-07 10:15:06 +02:00
  • c63a1784b9 add freebsd to ci Alon Faraj 2023-09-07 11:14:20 +03:00
  • 2dc96687eb updated lite Concedo 2023-09-07 15:18:24 +08:00
  • 7fb809b94b fixed auto rope scaling (+1 squashed commits) Concedo 2023-09-07 14:19:42 +08:00
  • 01948ddf3c examples : make n_ctx warning work again Cebtenzzre 2023-09-07 00:18:47 -04:00
  • 0bf75b05dc Merge branch 'master' into concedo_experimental Concedo 2023-09-07 10:50:23 +08:00
  • a8673be66e tidy up the new sse stream code Concedo 2023-09-07 10:48:53 +08:00
  • 49089371d1 Merge remote-tracking branch 'kalo/patch-1' into concedo_experimental Concedo 2023-09-07 10:43:19 +08:00
  • f4a6bbd282 convert : check for None explicitly in more places Cebtenzzre 2023-09-06 20:51:51 -04:00
  • 3a26b3c310 gguf: add script for converting falcon 180B jb 2023-09-06 23:04:14 +02:00
  • 6c0c9243bd convert : fix F32 ftype not being saved Cebtenzzre 2023-09-06 17:02:16 -04:00
  • 0c2c9c7545
    fix gradient accumulation bug where the same batch was used for each microstep xaedes 2023-09-06 22:45:36 +02:00
  • de6170d818
    fix gradient accumulation bug where the same batch was used for each microstep xaedes 2023-09-06 21:35:21 +02:00
  • 0393116628
    Merge branch 'master' into finetune-lora xaedes 2023-09-06 20:15:24 +02:00
  • c08fcf5947
    specify default lora rank with '--lora-r N' xaedes 2023-09-06 20:11:22 +02:00
  • 8c2d7e37f9
    improve finetune time measurement xaedes 2023-09-06 18:06:24 +02:00
  • 872cff8570 Merge remote-tracking branch 'origin/master' into cmake-find-package Mason M 2023-09-06 09:07:39 -03:00
  • 178b1850eb
    k-quants : fix zero-weight guard in Q6_K (ref #3040) b1187 Georgi Gerganov 2023-09-06 12:40:57 +03:00
  • 72621682c8
    Merge 759dcdcfdb into ea2c85d5d2 JackJollimore 2023-09-06 06:15:36 -03:00
  • ea2c85d5d2
    convert-llama-ggml-to-gguf: Try to handle files older than GGJTv3 (#3023) Kerfuffle 2023-09-06 02:49:11 -06:00
  • 69de44d441 Improve some informational output KerfuffleV2 2023-09-06 01:52:42 -06:00
  • b57bd3e14d docker : add git to full-cuda.Dockerfile main-cuda.Dockerfile hongbo.mo 2023-09-06 15:27:07 +08:00
  • 378fd908ee
    readme : fix typo Ikko Eltociear Ashimine 2023-09-06 12:41:34 +09:00
  • d4c6d2c314
    adding train-text-from-scratch to flake.nix takov751 2023-09-06 04:09:52 +01:00
  • ebc972c1b6 updated path to match default folder name for 7B model in readme mtwebb 2023-09-05 19:43:38 -07:00
  • 3d0d55f295 train : use std::lround instead of add + cast (clang-tidy) Cebtenzzre 2023-09-05 19:47:27 -04:00
  • 4c703231cf save-load-state : remove a redundant c_str() call (clang-tidy) Cebtenzzre 2023-09-05 19:44:58 -04:00
  • 6a9d1bd023 quantize : do not use 'else' after 'return' (clang-tidy) Cebtenzzre 2023-09-05 19:43:43 -04:00
  • b3d79bd346 grammar-parser : reserve space in vector (clang-tidy) Cebtenzzre 2023-09-05 19:43:01 -04:00
  • 22ff1408b5 do not throw unhandled exceptions (clang-tidy) Cebtenzzre 2023-09-05 18:58:21 -04:00
  • 87b90f505d train : remove inaccurate my_llama_hparams::operator== (clang-tidy) Cebtenzzre 2023-09-05 18:51:58 -04:00
  • 992b4ee1f3 metal : use sysconf(_SC_PAGESIZE) instead of getpagesize() derived from BSD Przemyslaw Pawelczyk 2023-09-06 01:38:40 +02:00
  • 2c716c0441 ggml : use sysconf(_SC_PAGESIZE) instead of getpagesize() derived from BSD Przemyslaw Pawelczyk 2023-09-06 01:37:55 +02:00
  • 77f9bd7c23 llama : use posix_madvise() instead of madvise() derived from BSD Przemyslaw Pawelczyk 2023-06-28 10:09:17 +02:00
  • f75865610d ggml-alloc : remove useless assignment (clang-tidy) Cebtenzzre 2023-09-05 18:49:07 -04:00
  • 49a700ccff server : make token_with_probs non-const (clang-tidy) Cebtenzzre 2023-09-05 18:46:19 -04:00
  • 0b0fe663f8 prefer const reference parameters (clang-tidy) Cebtenzzre 2023-09-05 18:42:19 -04:00
  • aae2be0f08 fix some unidiomatic conditions (clang-tidy) Cebtenzzre 2023-09-05 18:38:30 -04:00
  • 688aae46e5 .clang-tidy : update checks Cebtenzzre 2023-09-05 18:02:08 -04:00
  • 1dd3fd9323 llama : fix some -Wunused warnings Cebtenzzre 2023-09-05 18:36:03 -04:00
  • ceee3a6c22 build : silence misleading -Warray-bounds warning Cebtenzzre 2023-09-05 18:27:14 -04:00
  • 9912b9efc8
    build : add LLAMA_METAL_NDEBUG flag (#3033) b1185 Cebtenzzre 2023-09-05 18:21:10 -04:00
  • 0f844e3897 make : fix CPPFLAGS Cebtenzzre 2023-09-05 18:14:31 -04:00
  • 9d1a41a966
    Merge pull request #1 from ggerganov/master Marc 2023-09-05 17:06:24 -05:00
  • c8cc7f1a19
    Proposed streaming improvements kalomaze 2023-09-05 16:46:30 -05:00
  • a2ad58e3cb remove dynamic deps from kompute build Aaron Miller 2023-09-05 13:42:27 -07:00
  • c78072d6fa make : improve test target Cebtenzzre 2023-09-05 14:42:28 -04:00
  • d3f298d717 build : add LLAMA_METAL_NDEBUG flag Cebtenzzre 2023-09-05 15:15:14 -04:00
  • 9e2023156e
    make : use new flag variables for recent changes (#3019) b1184 Cebtenzzre 2023-09-05 15:12:00 -04:00
  • de2fe892af
    examples : replace fprintf to stdout with printf (#3017) b1183 Cebtenzzre 2023-09-05 15:10:27 -04:00
  • a7f5c74795 magic.md : fix missing final newline Cebtenzzre 2023-09-05 14:57:53 -04:00
  • b48994bcc1 cmake : fix trailing whitespace Cebtenzzre 2023-09-05 14:47:56 -04:00
  • 645b6a2e47 Rename to convert-llama-ggml-to-gguf.py KerfuffleV2 2023-09-05 11:53:04 -06:00
  • c9c3220c48
    convert: fix convert.py not working with int filename_stem (#3028) Erik Scholz 2023-09-05 19:41:00 +02:00
  • 98f6172281 cmake : fix quoting issues Cebtenzzre 2023-09-05 13:31:11 -04:00
  • 61577f0f9f fixing check and order of generator platform check Eric Sommerlade 2023-09-05 17:55:45 +01:00
  • 86ba3b0d21 added cross-platform compilation for msvc Eric Sommerlade 2023-09-05 17:45:34 +01:00
  • e413621edb convert : remove an obsolete pyright comment Cebtenzzre 2023-09-05 12:05:52 -04:00
  • d3a77329cb
    fix implicit int to string conversion no working on python 3.8.10 Green Sky 2023-09-05 16:08:39 +02:00
  • a0aa620718 Merge branch 'master' into concedo_experimental Concedo 2023-09-05 21:49:24 +08:00
  • 867e7c2255
    Merge branch 'master' into finetune-lora xaedes 2023-09-05 14:48:46 +02:00
  • 39c4b8540d
    Merge branch 'ggerganov:master' into master jameswu2014 2023-09-05 20:11:49 +08:00
  • c8863f3694 feature: add prompts jameswu2014 2023-09-05 20:10:41 +08:00
  • 7cfce5ef0f Moved magic to docs. John Boero 2023-09-05 12:12:49 +01:00
  • 00fe3fdb61 convert-llama-ggmlv3-to-gguf: Try to handle files older than GGJTv3 KerfuffleV2 2023-09-05 03:37:02 -06:00
  • 9fd5aa7ecb fix baichuan1-13b W_pack convert bug jameswu2014 2023-09-05 17:34:36 +08:00
  • d375b8f3aa
    ggml : fix L-BFGS linesearch loop Georgi Gerganov 2023-09-05 12:05:13 +03:00
  • 786e786061
    build : fix compile warnings Georgi Gerganov 2023-09-05 12:02:19 +03:00
  • d59bd97065
    Guard against all weights in a super-block being zero (#3010) b1181 Kawrakow 2023-09-05 09:55:33 +02:00
  • 35938ee3b0
    llama : update logic for number of threads when using BLAS b1180 Georgi Gerganov 2023-09-05 10:46:39 +03:00
  • 2636a8bad4 Added string debugging, removed bos token from end, added mlock. apaz-cli 2023-09-05 01:42:20 -05:00
  • b5efa62504
    speculative : fix typos Georgi Gerganov 2023-09-05 08:55:49 +03:00
  • d9559b78f3
    speculative : be more rewarding upon guessing max drafted tokens Georgi Gerganov 2023-09-05 08:54:08 +03:00
  • 4e6e951399
    Merge branch 'master' into HEAD Georgi Gerganov 2023-09-05 08:50:31 +03:00
  • dddd784c4d
    speculative : improve heuristic impl Georgi Gerganov 2023-09-05 08:49:40 +03:00
  • 921772104b
    speculative : add grammar support (#2991) b1179 Georgi Gerganov 2023-09-05 08:46:17 +03:00
  • cbdc56491d byte_decoder -> byte_encoder simonJJJ 2023-09-05 12:08:25 +08:00
  • 4709349395 make : use new flag variables for recent changes Cebtenzzre 2023-09-04 22:10:31 -04:00
  • a30ae2095c implement new YaRN algorithm Cebtenzzre 2023-09-04 20:08:17 -04:00
  • d07b6aac77
    fix tracking of train_samples and train_tokens xaedes 2023-09-05 02:18:17 +02:00
  • c1c3b0e0c2
    add gradient accumulation xaedes 2023-09-05 01:09:06 +02:00
  • 5e170fe3f7 examples : replace fprintf to stdout with printf Cebtenzzre 2023-09-04 18:47:34 -04:00
  • 9f49b2d520 Removes unnecessary autoreleasing under arc + allows Package.swift to be consumed Marc Terns 2023-09-04 17:41:38 -05:00
  • 2ba85c8609
    py : minor Georgi Gerganov 2023-09-04 22:50:50 +03:00
  • d3afd7131e
    Merge branch 'master' into finetune-lora xaedes 2023-09-04 21:44:05 +02:00
  • 1c1e74ce12 Merge branch 'last-working-old-file-format' Eric Sommerlade 2023-09-04 20:30:09 +01:00