41 Commits

Author SHA1 Message Date
hiyouga
c87910ada3 support falcon model #72
Former-commit-id: c136f362c1aa75d3374b151188ba4a55d9313a59
2023-07-05 15:00:06 +08:00
hiyouga
1ce7b5e0f3 update loading logic
Former-commit-id: 4d0fddba213beaa55146b047a78963d1d18185a1
2023-06-28 12:07:16 +08:00
hiyouga
e877b8d55b fix initializing data arguments
Former-commit-id: 18f87c1b25c7d1bbc06ea2260a1473b7f296e0ff
2023-06-27 22:50:23 +08:00
hiyouga
307f5866e9 support save full model, replace BOS token
Former-commit-id: 2e01abfda5706d8913860f52ce3bab98739eae55
2023-06-27 21:40:11 +08:00
hiyouga
8f1d99c926 support prefixes, loading multiple local files
Former-commit-id: cec9760eb890d37b733d8da73d0f3dbf924ca4ef
2023-06-26 15:32:40 +08:00
hiyouga
e4a869dc42 update api
Former-commit-id: f030b09924f0fb07305c244115759ac295e957c7
2023-06-26 13:39:57 +08:00
hiyouga
eeb78bd75c add default template
Former-commit-id: f621f7631a4a9db4a927a6aeb8fefd3a94f14467
2023-06-16 21:12:17 +08:00
hiyouga
c6d56e7109 support loading lora from hub
Former-commit-id: 0574b590ef3c4e317f7e2da25b0e5084dcef42a1
2023-06-16 00:02:17 +08:00
hiyouga
a68808d6d9 support baichuan model
Former-commit-id: 0cee6ad67ffb06f0d7165a0284e39f510a2abc36
2023-06-15 16:02:01 +08:00
hiyouga
dd1e7ed3cf add BOS token in pre-training
Former-commit-id: d668f8b501c367276ef4be372f2eb1753a1b7e86
2023-06-15 01:46:17 +08:00
hiyouga
3419396945 support multiturn training like FastChat
Former-commit-id: b6faf0207d5b637722a1fd45984d27b3ac095fd4
2023-06-14 22:27:39 +08:00
hiyouga
ca90a1e6d9 fix loading valuehead
Former-commit-id: 875e8e23498f6933d657ad154b53611310327e3e
2023-06-13 11:13:06 +08:00
hiyouga
c92bfb158f fix generating args
Former-commit-id: 531a3764d99ab00a0d217ce2ced0347b263dfe68
2023-06-13 01:33:56 +08:00
hiyouga
1fbda5d139 support RM metrics, add generating Args
Former-commit-id: cec6524d6b1be65c5d171a5b3dcaae7818132bc5
2023-06-12 15:48:48 +08:00
BUAADreamer
b1c6ee9cf5 add code for reading from multi files in one directory
Former-commit-id: a2af9df5a99ad529d0a280099b115cde69e02973
2023-06-10 16:27:30 +08:00
BUAADreamer
53727aee3e add code for reading from multi files in one directory
Former-commit-id: 3dd5f9a874d66353bb4379bbe39a89cd425dac3d
2023-06-10 15:53:47 +08:00
hiyouga
587d7a907f tiny fix
Former-commit-id: 2ba5d69c7f6e00e348c88b95331af9a80ede9561
2023-06-07 16:42:31 +08:00
hiyouga
37c6234126 tiny fix
Former-commit-id: edafb977330767b82b6c9591d9ec180046155632
2023-06-07 12:58:14 +08:00
hiyouga
93f2e35035 tiny fix
Former-commit-id: ce43386080fa1535672f4d879ffe6c4360a1ef7d
2023-06-07 12:08:39 +08:00
hiyouga
ce08b4a7ec add prompt template class
Former-commit-id: 909af8f49698a1de3010becf61817cbedecf7879
2023-06-07 11:55:25 +08:00
hiyouga
4cd43f018b fix inference, add prompt template
Former-commit-id: 5d021d4ad514974dd9dcc5240871713cf53a87f2
2023-06-07 10:52:35 +08:00
hiyouga
8849bee763 recover logging
Former-commit-id: 13d1f0709c774bb5ec5fb2b4e3c66b2f1226afd2
2023-06-06 21:36:37 +08:00
hiyouga
794c2fd506 support distributed quantized training
Former-commit-id: 4eb17bcf6c8ac51a3ec8cc5459064d1b35c82634
2023-06-06 17:39:41 +08:00
hiyouga
e94cf814ff tiny fix
Former-commit-id: 44298c12355082740857ba650bf44a18d4d3b40d
2023-06-05 15:25:22 +08:00
hiyouga
b982a9df83 tiny fix
Former-commit-id: 38b83533a48e2f5a5817b8fc53a37554c42fd932
2023-06-04 16:35:50 +08:00
hiyouga
95a6f1759b tiny fix
Former-commit-id: eac9921e5cc7be2b686731f28b19983a07009128
2023-06-04 12:55:40 +08:00
hiyouga
4d4636c48e support QLoRA
Former-commit-id: 3b9eee8cd26cfeef945155815175831dec98eb20
2023-06-04 00:08:56 +08:00
hiyouga
08d6079140 fix int8 inference
Former-commit-id: 1bd13d7ca197edaa9a1143b061249b4fa6003b97
2023-06-03 23:22:05 +08:00
hiyouga
4200d5d558 fix int8 inference
Former-commit-id: 0f69a0c19ebe05ba6b2d66b56826b6df100e9f32
2023-06-03 21:17:47 +08:00
hiyouga
342ee89d28 add ziya prompt template
Former-commit-id: de09ee1315759a085e4fcf20e94963293c881aae
2023-06-03 19:05:51 +08:00
hiyouga
c7d71dd8af use low_cpu_mem_usage to speed up loading
Former-commit-id: 771f454ff1deee4929927c58feab7dcd3b854f9c
2023-06-03 18:19:01 +08:00
hiyouga
42c9c8de39 add logits processor
Former-commit-id: dca27b4412e8e41cadcd623582222e1c216db78b
2023-06-03 16:34:54 +08:00
hiyouga
4003ddcc3b alter rewards data type
Former-commit-id: 50d9a20f8103fcfb92a3e2a5e6f0055d27b29d53
2023-06-02 14:19:51 +08:00
hiyouga
7e1be4c21a fix possibly OOM error
Former-commit-id: e6126244c161dc87b0d4d45b8976c02fc9933545
2023-06-01 23:54:44 +08:00
hiyouga
3bfb086399 support BLOOM models
Former-commit-id: 740a5daf5634f70a61b41fa8a31ee4a587fa03f3
2023-05-31 16:54:06 +08:00
hiyouga
ddb456bbcb remove dummy code
Former-commit-id: a72492e6490c44a7edccd572da73c47d6f278cc7
2023-05-30 16:28:00 +08:00
hiyouga
4c7c96e656 add pre-training script
Former-commit-id: 8ff96509fa621054368919988a15b50da1891852
2023-05-29 21:37:22 +08:00
hiyouga
0ab4419b86 fix checkpoint loading
Former-commit-id: c0e5df92d601966444956c65482441bd757fd7a1
2023-05-29 17:43:16 +08:00
hiyouga
03338163c2 tiny fix
Former-commit-id: ce71cc8b6db5d13b87b7d0302f4176c5c76ac4b2
2023-05-29 09:42:29 +08:00
hiyouga
87ba09e035 use fp16 model, add logcallback
Former-commit-id: 0c9fda01e3c61727c939efd9d9398f657a2d69b6
2023-05-28 21:30:28 +08:00
hiyouga
54b8ce7b63 Initial commit
Former-commit-id: 769c6ab56be0c9d26e9289f61ac54a4068d935c1
2023-05-28 18:09:04 +08:00