Commit Graph

  • a7e414687e update create_dataset Goekdeniz-Guelmez 2025-02-04 10:45:23 +0100
  • 7b0141455e better create_dataset Goekdeniz-Guelmez 2025-02-04 10:43:00 +0100
  • 9ece9aea02
    Merge branch 'ml-explore:main' into adding-ppo-training Gökdeniz Gülmez 2025-02-04 10:33:47 +0100
  • bd1a42ec2f adding args into dataset handling Goekdeniz-Guelmez 2025-02-04 10:22:34 +0100
  • 7173840283 first succesfull training run Goekdeniz-Guelmez 2025-02-04 09:18:45 +0100
  • 21d0ab6e8a
    fix deepseek sharding (#1242) Awni Hannun 2025-02-03 16:59:50 -0800
  • 3b3c20f508 fix deepseek sharding Awni Hannun 2025-02-03 16:54:58 -0800
  • 0989c073b0
    Optimizations for mamba1 (#1213) Gökdeniz Gülmez 2025-02-03 22:36:08 +0100
  • c43fc7ce59 nits + even faster Awni Hannun 2025-02-03 12:53:11 -0800
  • ca32424043 updates Goekdeniz-Guelmez 2025-02-03 21:57:26 +0100
  • 54e295ea80 fix name funcs Goekdeniz-Guelmez 2025-02-03 19:56:11 +0100
  • 06f9c29c94 print func name Goekdeniz-Guelmez 2025-02-03 19:47:40 +0100
  • 40bca770ae fixes Goekdeniz-Guelmez 2025-02-03 19:43:49 +0100
  • 05d921b788 optims Goekdeniz-Guelmez 2025-02-03 19:37:05 +0100
  • d9924d08d1
    Fix no validation in lora (#1241) Awni Hannun 2025-02-03 09:55:24 -0800
  • 10da47af36 Fix no validation in lora Awni Hannun 2025-02-03 06:52:53 -0800
  • 1d9e4802f0 first working prototype, will try training out at home Goekdeniz-Guelmez 2025-02-03 12:05:29 +0100
  • 23d75cd7ad starting fist training test run Goekdeniz-Guelmez 2025-02-03 10:08:28 +0100
  • 41ff5364d7 Merge branch 'adding-GRPO-training' of https://github.com/Goekdeniz-Guelmez/mlx-examples into adding-GRPO-training Goekdeniz-Guelmez 2025-02-03 09:19:00 +0100
  • a3ed632422 dataset wrapper done Goekdeniz-Guelmez 2025-02-03 09:13:17 +0100
  • 2c96da5155
    Merge branch 'ml-explore:main' into adding-orpo-training Gökdeniz Gülmez 2025-02-03 09:07:52 +0100
  • 94debd5ba8
    Merge branch 'ml-explore:main' into adding-ppo-training Gökdeniz Gülmez 2025-02-03 09:07:37 +0100
  • 734d6f4a69
    Merge branch 'ml-explore:main' into adding-GRPO-training Gökdeniz Gülmez 2025-02-03 09:07:20 +0100
  • d034ca369e adding function for R1 Goekdeniz-Guelmez 2025-02-03 08:26:42 +0100
  • 9c2ef38d4d
    only download local shard (#1240) Awni Hannun 2025-02-02 13:58:44 -0800
  • 2ac502176d only download local shard Awni Hannun 2025-02-02 12:56:46 -0800
  • fbb51f651a small fix Goekdeniz-Guelmez 2025-02-01 16:08:52 +0100
  • a03d434bb9 clean up Goekdeniz-Guelmez 2025-01-31 21:37:15 +0100
  • 541677aa7f cleaning up Goekdeniz-Guelmez 2025-01-31 21:36:24 +0100
  • 5998272ec2 cleaning up some namings Goekdeniz-Guelmez 2025-01-31 21:27:59 +0100
  • 243c9621d9 update lora.py Goekdeniz-Guelmez 2025-01-31 21:10:44 +0100
  • aa7a11c753 updates Goekdeniz-Guelmez 2025-01-31 17:38:01 +0100
  • b379359385 small fix Goekdeniz-Guelmez 2025-01-31 17:19:55 +0100
  • 595125ad4e updates Goekdeniz-Guelmez 2025-01-31 17:19:05 +0100
  • 3ad6405298
    Merge branch 'ml-explore:main' into adding-ppo-training Gökdeniz Gülmez 2025-01-31 17:04:10 +0100
  • 9e39c544eb initial commit Goekdeniz-Guelmez 2025-01-31 17:03:40 +0100
  • a57d553fc1 update Goekdeniz-Guelmez 2025-01-31 16:57:43 +0100
  • 80bcf68956 grpo_trainer shoudl be done Goekdeniz-Guelmez 2025-01-31 16:54:18 +0100
  • 6c58aa995c updates Goekdeniz-Guelmez 2025-01-31 16:27:31 +0100
  • b31d9cbb65 removing is-reference-free argument Goekdeniz-Guelmez 2025-01-31 00:01:43 +0100
  • 93370ff1c3 updates ans fixing the KL div lines Goekdeniz-Guelmez 2025-01-30 23:55:34 +0100
  • 70360d9d0d
    Merge ec06c04f4f into e8afb59de4 Sindhu Satish 2025-01-29 07:42:19 -0800
  • ec06c04f4f revert revision changes and retain qwen2 support Sindhu Satish 2025-01-29 07:37:58 -0800
  • ba6c7d3aba Qwen2 support Sindhu Satish 2025-01-29 07:30:11 -0800
  • b0520e7708 Bug fix - Qwen2 support Sindhu Satish 2025-01-29 06:21:36 -0800
  • b1e573d6e8
    Merge branch 'ml-explore:main' into adding-GRPO-training Gökdeniz Gülmez 2025-01-29 15:07:52 +0100
  • b3d6fc38cd
    Merge branch 'ml-explore:main' into adding-dpo-training Gökdeniz Gülmez 2025-01-29 15:07:37 +0100
  • ceccb4c9e9
    Merge branch 'ml-explore:main' into adding-orpo-training Gökdeniz Gülmez 2025-01-29 15:07:22 +0100
  • 57e10446b0
    Merge branch 'ml-explore:main' into adding-support-for-mamba2 Gökdeniz Gülmez 2025-01-29 15:07:11 +0100
  • a7e35ee748
    Merge branch 'ml-explore:main' into optimizations-for-mamba1 Gökdeniz Gülmez 2025-01-29 15:07:00 +0100
  • 651f9a5cf8
    Merge dd1690df81 into e8afb59de4 Sindhu Satish 2025-01-29 06:01:07 -0800
  • dd1690df81 bug fix Sindhu Satish 2025-01-29 06:00:06 -0800
  • e89a131668 Include revision version for HF models while loading Sindhu Satish 2025-01-29 05:53:18 -0800
  • 5e0ae83487 initial commit, gn Goekdeniz-Guelmez 2025-01-29 00:19:07 +0100
  • e8afb59de4
    better overflow correction (#1229) Awni Hannun 2025-01-28 14:37:30 -0800
  • 001323070a better overflow correction Awni Hannun 2025-01-28 12:27:45 -0800
  • 6a367fa31e update Copyright to this year Goekdeniz-Guelmez 2025-01-28 21:04:03 +0100
  • 0d4f2c4dc0 cleaning up and adding apple copyright to helium modelfile Goekdeniz-Guelmez 2025-01-28 21:02:50 +0100
  • a928bba375 Pass down TrainingArgs instance to iterate_batches function and TrainingCallback methods Addresses #1224 Chime Ogbuji 2025-01-28 14:50:38 -0500
  • 7b29cf0eda
    Merge branch 'ml-explore:main' into optimizations-for-mamba1 Gökdeniz Gülmez 2025-01-28 20:32:33 +0100
  • 7a83077cd7
    chore(mlx-lm): support text type content in messages (#1225) Anchen 2025-01-28 12:13:50 +1100
  • d98bf6f798 nits + format Awni Hannun 2025-01-27 16:43:51 -0800
  • f44a52e2dc
    batched min p and fix spec gen sampling (#1222) Awni Hannun 2025-01-27 15:40:31 -0800
  • 07f3d7d6bb chore: optimize the messagef content processing anchen 2025-01-27 09:37:29 +1100
  • 649d3f82ae fix ACKNOWLEDGMENTS Goekdeniz-Guelmez 2025-01-26 17:05:34 +0100
  • 9e5482ee74
    Merge branch 'main' into adding-dpo-training Gökdeniz Gülmez 2025-01-26 17:01:37 +0100
  • 3642a9df9b update ACKNOWLEDGMENTS Goekdeniz-Guelmez 2025-01-26 17:00:32 +0100
  • 294d189eed
    Merge branch 'main' into adding-orpo-training Gökdeniz Gülmez 2025-01-26 16:59:37 +0100
  • de856c7223
    Merge branch 'main' into adding-support-for-mamba2 Gökdeniz Gülmez 2025-01-26 16:58:06 +0100
  • 77faa14ba4
    adding support for kyutai's helium (#1208) Gökdeniz Gülmez 2025-01-26 16:19:07 +0100
  • 5acb03d7bf fixes / nits Awni Hannun 2025-01-26 07:06:21 -0800
  • 557649d8da removing tokenizer and updates Goekdeniz-Guelmez 2025-01-26 15:25:27 +0100
  • 2f2ddd4811 clean up Goekdeniz-Guelmez 2025-01-26 15:17:06 +0100
  • 4d0e52f7c8 more metrics Goekdeniz-Guelmez 2025-01-26 15:09:55 +0100
  • 74ae24b883 chore(mlx-lm): support text type content anchen 2025-01-27 00:00:10 +1100
  • 0ff1289bd9 updates Goekdeniz-Guelmez 2025-01-25 22:03:32 +0100
  • d8e7834345 Removed rejected_rewards handling, Updated batch unpacking to match iterator, Updated batch unpacking to match iterator, Added preference score scaling, Simplified reward calculation, Removed redundant rejected_rewards Goekdeniz-Guelmez 2025-01-25 21:35:37 +0100
  • 86b315fdf9 nits and quality of life improvements Goekdeniz-Guelmez 2025-01-24 22:40:27 +0100
  • 531c3345c6 nits Goekdeniz-Guelmez 2025-01-24 18:13:05 +0100
  • 54fcd8ed63 update DPODataset and added in system field too Goekdeniz-Guelmez 2025-01-24 18:11:56 +0100
  • 09ed837896 updates Goekdeniz-Guelmez 2025-01-24 16:57:18 +0100
  • e3688293ed removing dpo and fixing some stuff for orpo Goekdeniz-Guelmez 2025-01-24 16:09:22 +0100
  • 677b6a5176 batched min p and fix spec gen sampling Awni Hannun 2025-01-23 15:20:21 -0800
  • f787c08585 comments dist-eval Alex Barron 2025-01-23 06:36:31 -0800
  • d5f49d65b9 ordering Alex Barron 2024-12-19 00:08:28 -0800
  • 4385363c0f distributed evaluate Alex Barron 2024-12-18 22:12:08 -0800
  • a1aace4d99
    Add the possibility to cache model instead of loading from disk each time. Haixuan Xavier Tao 2025-01-23 11:55:12 +0100
  • 2462a34194 removig sanitize Goekdeniz-Guelmez 2025-01-22 22:30:15 +0100
  • 0bb001121e niits Goekdeniz-Guelmez 2025-01-22 21:39:29 +0100
  • aefe4ba160 nits Goekdeniz-Guelmez 2025-01-22 21:36:56 +0100
  • e1d549bcd3 nits Goekdeniz-Guelmez 2025-01-22 21:03:21 +0100
  • b0ece88909 nits Goekdeniz-Guelmez 2025-01-22 20:54:20 +0100
  • 44809b3ead
    Merge branch 'ml-explore:main' into optimizations-for-mamba1 Gökdeniz Gülmez 2025-01-22 14:19:22 +0100
  • dd29e74b89
    Merge branch 'ml-explore:main' into adding-support-for-mamba2 Gökdeniz Gülmez 2025-01-22 14:19:06 +0100
  • 520f4c468f
    Merge branch 'ml-explore:main' into adding-support-for-helium Gökdeniz Gülmez 2025-01-22 14:18:51 +0100
  • 4098c3bd2f
    Merge branch 'ml-explore:main' into adding-orpo-training Gökdeniz Gülmez 2025-01-22 14:18:38 +0100
  • 69a8f11f7b
    Merge branch 'ml-explore:main' into adding-dpo-training Gökdeniz Gülmez 2025-01-22 14:18:24 +0100
  • 9a3ddc3e65
    some fixes for pipeline parallel deep seek r1 (#1216) Awni Hannun 2025-01-21 19:40:29 -0800
  • a4b716e65d small optimization Goekdeniz-Guelmez 2025-01-22 00:15:02 +0100
  • df1406735b
    Fix dataset variable name, in datasets.py (#1212) Victor Nogueira 2025-01-21 23:12:43 +0100