Commit Graph

56 Commits

Author SHA1 Message Date
hiyouga
7b3c1f29ff fix packing for eager/sdpa attn
Former-commit-id: 6fd6aa4530
2024-07-04 01:52:43 +08:00
hiyouga
3595d98b4c fix #4609
unwrap_model_for_generation(reward_model) is necessary for zero3 training


Former-commit-id: 8845e94f91
2024-07-03 19:45:51 +08:00
hiyouga
104151d558 tiny fix
Former-commit-id: 8b1172b910
2024-07-03 02:31:50 +08:00
hiyouga
c9e9beee4e tiny fix
Former-commit-id: 71cdf8956e
2024-07-02 23:06:13 +08:00
hiyouga
ea2d3f6c18 remove rlhf support for chatglm2&3
Former-commit-id: 821bb6660e
2024-07-02 23:03:17 +08:00
hiyouga
4828bed837 upcast logits
Former-commit-id: c13ae2df19
2024-07-02 22:32:05 +08:00
hiyouga
cc31014002 improve rlhf
Former-commit-id: c47ab6c072
2024-07-02 22:23:08 +08:00
hiyouga
08296f4092 fix ppo callbacks
Former-commit-id: 4c296001c4
2024-07-02 17:34:56 +08:00
hiyouga
2cf03017a0 tiny fix
Former-commit-id: 73280b7dc7
2024-07-01 05:43:17 +08:00
hiyouga
54e786346e add eval acc
Former-commit-id: 1856a08e87
2024-07-01 03:51:20 +08:00
hiyouga
2b006beab1 loose gemma2 attention
Former-commit-id: 2f4b89ace1
2024-06-29 01:42:14 +08:00
hiyouga
d3b7c489f2 add Gemma2 models
Former-commit-id: 6f63050e1b
2024-06-28 01:26:50 +08:00
hiyouga
835f0578c2 refactor pissa, improve llamaboard
Former-commit-id: 8baf3b22b0
2024-06-28 01:04:24 +08:00
hzhaoy
e1751f6398 fix #4579
Former-commit-id: 677c86594e
2024-06-27 13:49:57 +08:00
hiyouga
28e613efd0 fix #4458
Former-commit-id: 8d6cd69ac4
2024-06-26 19:52:35 +08:00
hiyouga
ad0304e147 fix #4379
Former-commit-id: cc016461e6
2024-06-25 02:31:44 +08:00
hiyouga
a225b5a70c tiny fix about badam
Former-commit-id: 095fab58d3
2024-06-25 01:54:53 +08:00
hoshi-hiyouga
fe6ef6400c Merge pull request #4352 from Ledzy/main
[Enhancement] Support ZeRO-3 when using BAdam

Former-commit-id: d0f953bf5b
2024-06-25 01:49:13 +08:00
hiyouga
7be502c5c5 update readme
Former-commit-id: e507e60638
2024-06-24 18:22:12 +08:00
hiyouga
7735456561 fix templates
Former-commit-id: 4cff6a4ad5
2024-06-19 17:44:05 +08:00
Jonery
c779899f7b Cleaner integration.
Former-commit-id: 5c2ff1b749
2024-06-19 12:29:40 +08:00
Jonery
3a5eacb4cf Support distributed BAdam.
Former-commit-id: 0f72aac8c9
2024-06-18 12:27:47 +08:00
Jonery
5d59f6562a Merge remote-tracking branch 'upstream/main'
Former-commit-id: ea1f3ba5e0
2024-06-17 18:44:51 +08:00
Jonery
756566342d adapt for badam with ds zero3
Former-commit-id: 33b4372778
2024-06-17 18:18:10 +08:00
hiyouga
ce4a27a5f7 fix tol
Former-commit-id: 46093b5786
2024-06-16 01:38:44 +08:00
hiyouga
f25b8626bf support pissa
Former-commit-id: 8c1046d78a
2024-06-16 01:08:12 +08:00
hiyouga
c0c6b8075a tiny fix
Former-commit-id: 38b6b0f52e
2024-06-16 01:06:41 +08:00
hiyouga
2946153cea add license
Former-commit-id: d87108daa6
2024-06-15 17:54:33 +08:00
hiyouga
ab66ae8cd2 fix #4295
Former-commit-id: 78589cf90c
2024-06-15 04:34:55 +08:00
hiyouga
a3f4925c2c add test cases
Former-commit-id: b27269bd2b
2024-06-15 04:05:54 +08:00
hiyouga
8fccaf20c5 fix #4221
Former-commit-id: 6baafd4eb3
2024-06-13 02:48:21 +08:00
hiyouga
81ed4d8abf fix #4209
DeepSpeed ZeRO3 has inflight param error when calling model.eval()


Former-commit-id: cf9f2d6c42
2024-06-13 02:25:50 +08:00
hiyouga
833aa324c2 clean code
Former-commit-id: 2ed8270112
2024-06-13 01:58:16 +08:00
hiyouga
5834651c4a fix #4198
Former-commit-id: 89f2bd8c8c
2024-06-11 15:38:38 +08:00
hiyouga
ca9468ff04 tiny fix
Former-commit-id: f8d8690bf4
2024-06-07 05:19:21 +08:00
hiyouga
4f3c89a6eb fix ppo trainer save zero3 model
accelerator.get_state_dict(ds_model) should be called at all ranks


Former-commit-id: 4489d73ac7
2024-06-07 05:14:19 +08:00
hiyouga
f76d427332 fix ppo in trl 0.8.6
Former-commit-id: 2702d7e952
2024-06-07 04:48:29 +08:00
hiyouga
d3196318be fix #4120
Former-commit-id: f9e818d79c
2024-06-07 04:18:05 +08:00
hiyouga
8da149ba40 rename files
Former-commit-id: 74f96efef9
2024-06-07 00:09:06 +08:00
hiyouga
368695483d fix ppo+zero3 #3108
Former-commit-id: 76c61905b2
2024-06-06 23:30:07 +08:00
hiyouga
e0aadd4b34 fix ppo dataset bug #4012
Former-commit-id: 149610c636
2024-06-06 19:03:20 +08:00
hiyouga
e898d8bbc4 update trainers
Former-commit-id: fad2591e31
2024-06-06 18:45:49 +08:00
hiyouga
a16786d8ba fix #4090
Former-commit-id: 67fe822324
2024-06-06 00:50:32 +08:00
hiyouga
6f7b6ae0c3 remove gc warnings in DPO&KTO
Former-commit-id: f9a206509e
2024-06-03 22:53:54 +08:00
hoshi-hiyouga
5d96cf146e Update trainer.py
Former-commit-id: 24499f40dc
2024-06-03 22:08:38 +08:00
enji.zhou
e58aca0602 fix KTO Trainer Sampler
Former-commit-id: 34a2c5087a
2024-06-03 21:32:38 +08:00
Uminosachi
0de4e1e9e2 Set scheduler_specific_kwargs to get_scheduler
Former-commit-id: 14e97dc119
2024-05-31 13:45:39 +09:00
hiyouga
468d0e7ed1 10x generate in ppo w/ zero3
https://github.com/huggingface/trl/pull/1483

Former-commit-id: 65cd8bdbdb
2024-05-29 00:23:23 +08:00
hiyouga
bfac965f9c update dpo, kto trainer
Former-commit-id: 7c8e01bb74
2024-05-29 00:14:29 +08:00
hiyouga
14f6cc2b7c clean kto trainer
Former-commit-id: 900e1ea622
2024-05-28 21:43:26 +08:00