48 Commits

Author SHA1 Message Date
hiyouga
b8a827faeb fix #2320
Former-commit-id: 2bc30763e9a40a82484c27b9a472425fdb9b3bd8
2024-01-24 16:19:18 +08:00
hoshi-hiyouga
5159c9719c Update tuner.py
Former-commit-id: 662b9a9dcfadb01a903d3672e277929ec1875ed4
2024-01-21 12:39:38 +08:00
yhyu13
f036b9c7ba Remove manully set use_cache; torch_dtype is not str, save model as bfloat16 used to fail;
Former-commit-id: 9cdbd3bfc8be3f9adc799af8db9a254a47a577a2
2024-01-21 11:12:15 +08:00
hiyouga
b27e91222c format style
Former-commit-id: 638234ceee1b19716e45b6e5f4ea54d9122da4df
2024-01-20 20:15:56 +08:00
hiyouga
2f7684a8ee fix tests
Former-commit-id: f6d6e00337ebef8740d180836dcb18d0e6a3c59a
2024-01-20 19:58:04 +08:00
hiyouga
69e8925249 support longlora for main branch
Former-commit-id: 38af076a75c33da26d641780820694e4b7342d92
2024-01-20 19:25:22 +08:00
hiyouga
6e33982849 Update tuner.py
Former-commit-id: ddd48ce8ab409b4ff206a5c980ba2483988ddc51
2024-01-18 15:06:02 +08:00
hiyouga
4e3bfb799d support function calling
Former-commit-id: d9f1cae35150cce594a7abd96dd2beb811fa33f2
2024-01-18 09:54:23 +08:00
hiyouga
6a954cc075 support export push_to_hub #2183
Former-commit-id: 42859f073434eab0928940e8a9c52f275a2fc93a
2024-01-16 23:59:42 +08:00
hiyouga
69d966eb1f fix #2164
Former-commit-id: 4b2d11ec28130ee6c21dc85614ffcee61a4a5847
2024-01-12 00:27:57 +08:00
hiyouga
6378864390 fix #2161
Former-commit-id: 898ec3696a4d2db48485fb7263f866599437d626
2024-01-11 17:04:13 +08:00
hiyouga
fbe945aba0 improve model export
Former-commit-id: 05ed4e80286d3187fca8c821fdf99279683ed01c
2024-01-09 22:26:24 +08:00
hiyouga
61960189b2 fix #1789
Former-commit-id: 4571068e1e00dc234c9131185fe0924c726add84
2024-01-09 18:31:27 +08:00
hiyouga
6bbcf5ad16 improve model export
Former-commit-id: d2a676c8ba550e1dd7f4e12cb397a32e01831d85
2024-01-05 18:51:49 +08:00
hiyouga
cae66bce3d fix args
Former-commit-id: 65c5b0477c0e62691a1f8790670ba04d7f6d2804
2023-12-28 18:47:19 +08:00
hiyouga
16688b773a fix export format
Former-commit-id: e165354facf7e69f535f9b7d99438f03dbf0293d
2023-12-28 18:40:46 +08:00
hiyouga
d0946f08db fix ppo trainer
Former-commit-id: 5431be42f9c43095d478f2250fac64ef189eb3ad
2023-12-28 18:09:28 +08:00
hiyouga
938c4cb132 fix dpo trainer
Former-commit-id: 074745b1707f98e092749f57041d866c5d55bc04
2023-12-23 01:51:55 +08:00
hiyouga
f0d405f392 support unsloth
Former-commit-id: 7aad0b889d9a316fffd65f32a419078418fc0986
2023-12-23 00:14:33 +08:00
hiyouga
82a79e9fdf fix #1073 #1462 #1735 #1908
Former-commit-id: 31165a9822bd52130b33cd3439f887c26e0679dc
2023-12-20 17:15:40 +08:00
hiyouga
8154b4bdf6 fix #1742
Former-commit-id: 870426ff70c060213ac283b10a9b1f4bf71679ef
2023-12-16 20:50:45 +08:00
hiyouga
4e75ca1222 support dpo-ftx
Former-commit-id: b87c74289d523ef88611b376074199ffd03cf103
2023-12-16 19:21:41 +08:00
hiyouga
7db6fe4754 update tips
Former-commit-id: 3551171d49f0f6aa5f745d80f71939408c9bb3a7
2023-12-15 23:52:50 +08:00
hiyouga
9a88387b91 fix #1770
Former-commit-id: 439a26c27606dc617cfd073ef23256b8f6f7a4fb
2023-12-15 23:50:15 +08:00
hiyouga
7dbc670902 support quantization in export model
Former-commit-id: 3524aa1e58da94ab00e9a2024952ea1b4119b2af
2023-12-15 23:44:50 +08:00
hiyouga
bd03307bbd refactor adapter hparam
Former-commit-id: 0716f5e470afffd2df5a815712b552a4b4797153
2023-12-15 20:53:11 +08:00
hiyouga
027caabbb6 fix ppo trainer save logic
Former-commit-id: d3dccd0693ede18a99f04780f2fd6e3a89810405
2023-12-04 19:00:19 +08:00
hiyouga
6493558c3b fix bug
Former-commit-id: 8b681ee273c28813c599d9d55b2a3540c8ac257d
2023-12-03 21:40:40 +08:00
hiyouga
64eead3fb1 ppo support rm server
Former-commit-id: 747db4017291b0eb91946f57011bb31659056037
2023-12-03 21:38:51 +08:00
hiyouga
1cb390b9b2 implement rm server #1543
Former-commit-id: 7df4f3ab206fddb462f6ed865eaf04234fd72ed6
2023-12-03 20:52:54 +08:00
hiyouga
3d291a82d3 fix #1597
Former-commit-id: 327d7f7efe1fefe4bf4646c07fc4917a42c13383
2023-11-30 21:47:06 +08:00
hiyouga
ba6d290d0b fix #1668
Former-commit-id: 1585962eb7ed042890d4c56422aae749c669dda8
2023-11-30 21:02:00 +08:00
hiyouga
ecfc7d1b50 fix #1658
Former-commit-id: 77d1b14fc2d9703d15bbd879f67df037db9fbb28
2023-11-28 20:57:24 +08:00
hiyouga
ae1048db6d fix #1659
Former-commit-id: 475a3fa0f4c09d4cfd55ec66271a6d3c9eb5f4d2
2023-11-28 20:52:28 +08:00
hiyouga
b015ac35d8 support export size setting
Former-commit-id: 859a6ea9425a09d7263f6436d05102df8129c248
2023-11-26 18:34:09 +08:00
hiyouga
4966bd7911 support GPTQ tuning #729 #1481 #1545 , fix chatglm template #1453 #1480 #1569
Former-commit-id: 9ea93801459b0d271d21a2d730c44abae9106c51
2023-11-20 22:52:11 +08:00
hiyouga
f06c4c8f7a update ppo trainer
Former-commit-id: 5021062493ed63ad1f6133cfb543e4e7f528d2cc
2023-11-20 21:39:15 +08:00
hoshi-hiyouga
d72f123851 Merge pull request #1553 from hannlp/hans
Change the default argument settings for PPO training

Former-commit-id: 48211e3799a16de946360930d3d92f5a40e9d12d
2023-11-20 20:32:55 +08:00
hiyouga
682d81caa9 fix #1567
Former-commit-id: 99a3f06377d2886c4000ce7e3583b12ca965534d
2023-11-20 18:46:36 +08:00
hiyouga
a53afb27eb fix #1263
Former-commit-id: 065bfaeed490a4e03fb48a5adc0b8af4d835a767
2023-11-19 16:05:18 +08:00
hiyouga
48d6d925f7 fix #1558
Former-commit-id: 1740131d63d32aefc0370441baf4716ddb5ebcfe
2023-11-19 14:15:47 +08:00
Yuchen Han
a419122179 Update workflow.py
Former-commit-id: eeb5249d0b6ce0816e1fa47afc3a853c7b267cbf
2023-11-17 00:16:27 -08:00
hiyouga
0ed0b8f9c5 fix bug in freeze tuning
Former-commit-id: ff52b1779c909819d0aef83d3f7ea663199cbe54
2023-11-16 14:25:11 +08:00
hiyouga
678052a7ef fix rlhf callback
Former-commit-id: 1817ffc86fe3463ea91e9359c0e3611979a9d53e
2023-11-16 03:26:19 +08:00
hiyouga
b71da932eb fix bug in PPO training
Former-commit-id: 856522a3df4bb9ddfaaa137119eceb9574873950
2023-11-16 02:32:54 +08:00
hiyouga
eb5a852dd5 fix import bug
Former-commit-id: 35b91ea34caade45dd51813b94da5177b852aa4c
2023-11-16 02:27:03 +08:00
hiyouga
f441932bd1 support full-parameter PPO
Former-commit-id: ce783036001397a20b0b4c5da2fea6d0c03389d2
2023-11-16 02:08:04 +08:00
hiyouga
06a4820836 disentangle model from tuner and rename modules
Former-commit-id: 4736344eb1595ee023a50d49e8118f4eee46305f
2023-11-15 16:29:09 +08:00