Commit Graph

  • 72f1a651e2
    Allow converting models from local directories Remixer Dec 2024-11-24 22:46:36 +0400
  • 38e5801edb loading codestral works but no tinference Goekdeniz-Guelmez 2024-11-24 16:26:45 +0100
  • 0f135396ae
    Generation refactor: part 2 (#1099) Awni Hannun 2024-11-23 11:47:06 -0800
  • 83f7c8e3b5 version Awni Hannun 2024-11-23 11:43:09 -0800
  • 5a3f01b081 fix test + faster min p + test Awni Hannun 2024-11-23 11:03:00 -0800
  • f82e49aad9 some cleanup, warnings, tests Awni Hannun 2024-11-22 16:51:22 -0800
  • 9986787303 nit Awni Hannun 2024-11-13 14:00:17 -0800
  • 566af15c34 fixes Awni Hannun 2024-11-07 19:43:42 -0800
  • 431988721f unify with stream_generate Awni Hannun 2024-11-07 17:21:15 -0800
  • 004eb4cc9d
    Tencent HunYuan MOE model (#1100) Awni Hannun 2024-11-23 11:06:26 -0800
  • 0593aaea89 default trust remote code for tokenizer, allow system prompt to be configurable Awni Hannun 2024-11-23 09:19:08 -0800
  • 516d0e3af0 Allow loading from diffusers ckpt Angelos Katharopoulos 2024-11-22 20:52:50 -0800
  • a6ddc27a4e removing last checkpoint file Goekdeniz-Guelmez 2024-11-21 22:33:56 +0100
  • 57b1717cf5 inference fixed Goekdeniz-Guelmez 2024-11-21 22:25:58 +0100
  • 117ffd3909 removing some files Goekdeniz-Guelmez 2024-11-21 22:05:42 +0100
  • e22b2dbf27 Fixed streaming generation and got rid of generating gibberish, but is still a litle slow: 0.222 tokens-per-sec Goekdeniz-Guelmez 2024-11-21 22:01:28 +0100
  • e4eae973e8
    Merge branch 'ml-explore:main' into adding-support-for-mamba2 Gökdeniz Gülmez 2024-11-21 21:06:45 +0100
  • 042280ce50
    Fix format (#1115) Angelos Katharopoulos 2024-11-20 16:15:53 -0800
  • ef656ff4e1 Fix format Angelos Katharopoulos 2024-11-20 16:11:42 -0800
  • 60c7b80350
    Pass seed to sd img2img (#1114) Valentin Roussellet 2024-11-20 15:21:52 -0800
  • 1b1a540df0
    seed Valentin Roussellet 2024-11-20 15:17:27 -0800
  • 04c18832ab add to_buffer at the end of data_iter sakares 2024-11-20 17:36:28 +0800
  • bd6d910ca3
    [MLX LM] Fix f-string formatting in memory warning message (#1105) Alban Lecocq 2024-11-13 15:14:03 +0100
  • 29a842ad6a [MLX LM] Fix f-string formatting in memory warning message alban 2024-11-13 15:11:38 +0100
  • 1d851069ea nits Goekdeniz-Guelmez 2024-11-10 17:21:18 +0100
  • 1a6688384d imopemented multi Token inputs, but still generating Gibberish Goekdeniz-Guelmez 2024-11-10 17:19:00 +0100
  • 2f95b361a8 removed the custom Mamba2Cache adn updated the existing MambaCache but still only one input Token and outputs gibberish Goekdeniz-Guelmez 2024-11-10 16:57:03 +0100
  • 49d3f188f8
    Merge branch 'ml-explore:main' into adding-support-for-mamba2 Gökdeniz Gülmez 2024-11-10 16:36:02 +0100
  • 3a499f9735 fixed inference slowness but it cant handle multible Token inputs and is generateing gibberish Goekdeniz-Guelmez 2024-11-10 16:35:07 +0100
  • 4ddbb988ce Update documentation Chime Ogbuji 2024-11-10 10:10:04 -0500
  • 791727fa1c Merge remote-tracking branch 'origin/completion_only' into completion_only Chime Ogbuji 2024-11-10 09:54:49 -0500
  • 01e330d6bb Add input masking for fine-tuning in documentation Chime Ogbuji 2024-11-10 09:54:32 -0500
  • 800b60239c save checkpoint Goekdeniz-Guelmez 2024-11-10 14:36:26 +0100
  • 98114b92fa
    Merge branch 'ml-explore:main' into sets_of_hf_datasets Chime Ogbuji 2024-11-09 12:52:38 -0500
  • 3080102b80
    Merge branch 'ml-explore:main' into completion_only Chime Ogbuji 2024-11-09 12:52:30 -0500
  • c219da4a1b Merge branch 'main' into flux/pip madroid 2024-11-09 11:15:26 +0800
  • 1e07660184
    FLUX: save train config (#1049) madroid 2024-11-09 09:15:19 +0800
  • f9936f77da
    Merge branch 'ml-explore:main' into sets_of_hf_datasets Chime Ogbuji 2024-11-08 18:47:27 -0500
  • 7f89ace55e
    Merge branch 'ml-explore:main' into completion_only Chime Ogbuji 2024-11-08 18:46:38 -0500
  • 53569da120 format str Awni Hannun 2024-11-08 15:20:49 -0800
  • 0e8b7339af fix Awni Hannun 2024-11-08 15:17:37 -0800
  • 189be53988 hunyuan Awni Hannun 2024-11-08 15:13:03 -0800
  • 9e4ce397f0 Typo Angelos Katharopoulos 2024-11-08 13:03:03 -0800
  • ebf314bdcc Nits Angelos Katharopoulos 2024-11-08 13:01:51 -0800
  • 657b4cc0aa
    [MLX LM] Sampler refactor + a few improvements (#1094) Awni Hannun 2024-11-07 16:15:24 -0800
  • 230215a50d FLUX: dreambooth add main() def madroid 2024-11-07 15:20:58 +0800
  • 39fd6d272f FLUX: fix pre-commit lints madroid 2024-11-07 12:51:22 +0800
  • 1c43a83280 FLUX: add setup config madroid 2024-11-07 12:45:56 +0800
  • e61849a003 FLUX: move cli to mlx_flux dir madroid 2024-11-07 12:35:49 +0800
  • 83c92c2a11 FLUX: rename flux to mlx_flux madroid 2024-11-07 10:32:53 +0800
  • bfa6c2932e Fix Chime Ogbuji 2024-11-06 20:29:12 -0500
  • 960ed79a6b Update sublist search and calculation of input id length Chime Ogbuji 2024-11-06 13:57:59 -0500
  • 90e2da881c Minor fix Chime Ogbuji 2024-11-06 12:58:00 -0500
  • e45ce38f86 Add ability to fetch raw prompt and completion text from completion datasets Chime Ogbuji 2024-11-06 12:53:54 -0500
  • 3c76a253db Fix variable reference Chime Ogbuji 2024-11-06 12:33:49 -0500
  • 906f972d36 save push Goekdeniz-Guelmez 2024-11-06 16:35:46 +0100
  • 6b209c6d3e fix eos handling in stream generate Awni Hannun 2024-11-06 06:48:05 -0800
  • c9994f80e6 fix stream generate Awni Hannun 2024-11-05 17:18:00 -0800
  • f5cd03c64d fix stream Awni Hannun 2024-11-05 17:04:46 -0800
  • 0be87b3c53 refactor sampler/processor and a few improvements Awni Hannun 2024-11-05 17:01:21 -0800
  • 3783156072 starting Awni Hannun 2024-10-27 10:02:45 -0700
  • 4b88c33a26 Updates CL lora tuner with input masking that uses default_loss (and iterate_batches) by default. Chime Ogbuji 2024-11-05 19:10:01 -0500
  • e0d66f5479 Merge remote-tracking branch 'origin/completion_only' into completion_only Chime Ogbuji 2024-11-05 15:26:08 -0500
  • 5579b48974 Minor documentation update Chime Ogbuji 2024-11-05 15:25:45 -0500
  • 603dab57be
    Merge branch 'ml-explore:main' into completion_only Chime Ogbuji 2024-11-05 15:18:05 -0500
  • b7b3332dc5 Replace iterate_input_masked_batches with iterate_delineated_batches, an updated attempt to better sync with iterate_batches logic Chime Ogbuji 2024-11-05 15:17:23 -0500
  • ed9e81dd58
    Fix rotating kv cache size (#1093) Angelos Katharopoulos 2024-11-05 10:24:24 -0800
  • 6fd1f70f73
    fix spm decoder multi-byte (#1092) Awni Hannun 2024-11-05 06:06:26 -0800
  • 19884e5932 Fix rotating kv cache size Angelos Katharopoulos 2024-11-04 23:12:40 -0800
  • a1fbc52cf2
    Merge branch 'ml-explore:main' into completion_only Chime Ogbuji 2024-11-04 22:00:55 -0500
  • 01faa4a692 fix spm decoder multi-byte Awni Hannun 2024-11-04 16:46:37 -0800
  • 4394633ce0
    mlx_whisper: add support for audio input from stdin (#1012) Anthony Wu 2024-11-04 14:02:13 -0800
  • bd6b08e813 some nits Awni Hannun 2024-11-04 13:59:53 -0800
  • 3b526f0aa1
    Add support for falcon-mamba (#1074) ilyasch2 2024-11-05 00:23:30 +0400
  • 6a8dd0df60 nit Awni Hannun 2024-11-04 12:20:36 -0800
  • 31cb8cac94 nits Awni Hannun 2024-11-04 12:19:20 -0800
  • 82e3338987
    chore(mlx-lm): add max token arg for mlx_lm.chat (#1089) Anchen 2024-11-04 22:06:34 +0800
  • 95fb22449b
    Merge branch 'ml-explore:main' into completion_only Chime Ogbuji 2024-11-04 08:48:26 -0500
  • 739a3e62a0 chore: update the default max token value Anchen 2024-11-04 14:58:17 +0800
  • 78b24a2375 Fix index calculation Chime Ogbuji 2024-11-03 20:36:55 -0500
  • 24f40c3b8d Fix iteration over HF dataset collection Chime Ogbuji 2024-11-03 20:30:47 -0500
  • e477060a00 Fix keyword argument invokation Chime Ogbuji 2024-11-03 20:26:15 -0500
  • 04cf93df55 Fixes to references to hf_datasets Chime Ogbuji 2024-11-03 20:04:15 -0500
  • c72122064a Fixes to config format in documentattion Chime Ogbuji 2024-11-03 20:00:35 -0500
  • 1f6c370690 Updates to LoRA documentation Chime Ogbuji 2024-11-03 19:41:09 -0500
  • 9df7bbbe3a Generalize HF datasets to a collection of HF dataasets via datasets, adds support for custom chat HF datasets (#1088), and fixes (#1087) Chime Ogbuji 2024-11-03 19:11:54 -0500
  • e0e6847d20 chore(mlx-lm): add max token arg for mlx_lm.chat Anchen 2024-11-04 07:14:19 +0800
  • 331148d8ec
    Enable distributed LoRA training (#821) Angelos Katharopoulos 2024-11-02 18:02:31 -0700
  • 29c954f4cb
    fix (#1082) Awni Hannun 2024-11-02 13:51:38 -0700
  • 4212387e97 fix Awni Hannun 2024-11-01 17:34:36 -0700
  • 0f799947d0
    fix (#1079) Awni Hannun 2024-11-01 16:30:32 -0700
  • e510987870
    Clear cache every now and then (#1081) Awni Hannun 2024-11-01 14:15:32 -0700
  • 8cdd9da4a6 don't need user arg anymore Awni Hannun 2024-11-01 13:56:55 -0700
  • c102d528ae clear cache every now and then Awni Hannun 2024-11-01 13:54:05 -0700
  • 8160e0c4e5
    Whisper improvements (#1080) Awni Hannun 2024-11-01 10:52:28 -0700
  • 197546a058 version Awni Hannun 2024-11-01 08:49:38 -0700
  • 292f6235f3 speed up decoder Awni Hannun 2024-11-01 08:44:04 -0700
  • 85ffd2c96a
    Quantized KV Cache (#1075) Alex Barron 2024-10-31 16:59:52 -0700
  • c5e09a1725 Fix the test Angelos Katharopoulos 2024-10-31 16:28:55 -0700
  • 83a7a17f84 support different k and v head dims Alex Barron 2024-10-31 16:24:40 -0700