45 Commits

Author SHA1 Message Date
hiyouga
25182c4779 fix Baichuan-13B
Former-commit-id: 6d9d826b3246349454c68f4d13b862da4de986e2
2023-07-13 23:08:45 +08:00
Jinghuan Shang
c0078c755f Fix typo in common.py
lastest -> latest

Former-commit-id: a0a82a56765ffbdc2aeaa8174835494459ac914a
2023-07-11 18:03:53 -04:00
hiyouga
102f4f425c fix sft encode
Former-commit-id: 2369a96a3200593421ae9afb06e08e2ac8010bb2
2023-07-11 19:50:33 +08:00
hiyouga
2e75897fc1 fix compute dtype
Former-commit-id: 5aadbb22730d19570b039462c91df443dbb34b5f
2023-07-05 15:13:00 +08:00
hiyouga
6b5a085ddf support falcon model #72
Former-commit-id: 72cc3ff0e6de641073de1159196319705f8efe85
2023-07-05 15:00:06 +08:00
hiyouga
11f8c31101 update loading logic
Former-commit-id: f1da17bb0deeb39a29da4dc208951d1ad69bb8ba
2023-06-28 12:07:16 +08:00
hiyouga
4f2c204f13 fix initializing data arguments
Former-commit-id: e6b83c8b87cb93358086121a6f9ccaba5dfa7497
2023-06-27 22:50:23 +08:00
hiyouga
9342c6411b support save full model, replace BOS token
Former-commit-id: 32e56c290802ba971c08f471b94a33daec85671a
2023-06-27 21:40:11 +08:00
hiyouga
b6968a6940 support prefixes, loading multiple local files
Former-commit-id: 6672e09836ed0103693a381ece010377bd0ef4f8
2023-06-26 15:32:40 +08:00
hiyouga
d97da03cd5 update api
Former-commit-id: a90db46e336a657d5fcf480986bfc68c77ad416b
2023-06-26 13:39:57 +08:00
hiyouga
b9e225bc20 add default template
Former-commit-id: c64fb6b83fdbedd62073417213f0215207ff1311
2023-06-16 21:12:17 +08:00
hiyouga
6d1e733311 support loading lora from hub
Former-commit-id: 0b34c962bc3368dca62b18ad6c27a0293c3affa5
2023-06-16 00:02:17 +08:00
hiyouga
fa2c840610 support baichuan model
Former-commit-id: d683042fbcb2ee43b9823262d0a65b64f4cb54cb
2023-06-15 16:02:01 +08:00
hiyouga
11df2ab717 add BOS token in pre-training
Former-commit-id: c57cf5d4a46c57c6f698e5cfd0fd59cce703094d
2023-06-15 01:46:17 +08:00
hiyouga
11bace2e93 support multiturn training like FastChat
Former-commit-id: 629cafb1a09924e82d7ea1f9fba318d3f5593196
2023-06-14 22:27:39 +08:00
hiyouga
febe41a481 fix loading valuehead
Former-commit-id: 7872375d7a0c1d8826206631f6717a91ec49f1b3
2023-06-13 11:13:06 +08:00
hiyouga
fdfc22196c fix generating args
Former-commit-id: 52805a8441bd7b324bd89489de60f18f103c8e4c
2023-06-13 01:33:56 +08:00
hiyouga
0da1b7d9ab support RM metrics, add generating Args
Former-commit-id: c461c6190bc124e98dde7f3cf96a59ce40b26fb0
2023-06-12 15:48:48 +08:00
BUAADreamer
a976cba730 add code for reading from multi files in one directory
Former-commit-id: 9b80cf08b9f0d4aee896b228fb76399e9a7c9d8b
2023-06-10 16:27:30 +08:00
BUAADreamer
2012cb5cbc add code for reading from multi files in one directory
Former-commit-id: b7ebb83a96619e5111b0faa9da9d0feb8d9cdff0
2023-06-10 15:53:47 +08:00
hiyouga
6978c1625a tiny fix
Former-commit-id: c9c795f9c7cd2228410a12af4ec10d3b59be87db
2023-06-07 16:42:31 +08:00
hiyouga
0092e863c1 tiny fix
Former-commit-id: 4a9bc72d90b65db80b375cd141484abfbb0dcf0d
2023-06-07 12:58:14 +08:00
hiyouga
f2dda11101 tiny fix
Former-commit-id: 7115bee4310888ec2e5f104e8d2c1f7127fb6ce6
2023-06-07 12:08:39 +08:00
hiyouga
8875f565ad add prompt template class
Former-commit-id: 3d7e3a38d00aa5d9664824093043951af8c3f707
2023-06-07 11:55:25 +08:00
hiyouga
701a1d84c6 fix inference, add prompt template
Former-commit-id: 3940e50c71472b210bbc1b01248bf85a191c4065
2023-06-07 10:52:35 +08:00
hiyouga
0b903fed18 recover logging
Former-commit-id: d74014496e4ccda2de4482075a91747854facddd
2023-06-06 21:36:37 +08:00
hiyouga
5e5db11833 support distributed quantized training
Former-commit-id: 74ff23a4f36f859f791f7b4be6f1877edc68f12f
2023-06-06 17:39:41 +08:00
hiyouga
2770a2ee58 tiny fix
Former-commit-id: 3c5da617cdab34c6cae038e3a06d0468ae4c6c86
2023-06-05 15:25:22 +08:00
hiyouga
a96cfbee03 tiny fix
Former-commit-id: 5ce3e0056948aded120b63e365a892f9d8c3c840
2023-06-04 16:35:50 +08:00
hiyouga
363e0da084 tiny fix
Former-commit-id: a98ebf62fb82ffe5aaaea6a1ce3d4c60d23a5728
2023-06-04 12:55:40 +08:00
hiyouga
de447e7aeb support QLoRA
Former-commit-id: d89597e28fe9b91246e58c55eeb9082436940481
2023-06-04 00:08:56 +08:00
hiyouga
d2e80fff76 fix int8 inference
Former-commit-id: d05202943e9634526f96d189288f67852d3d1c40
2023-06-03 23:22:05 +08:00
hiyouga
4e224fac7c fix int8 inference
Former-commit-id: fcf3506bef28504dd679c2210bdc84e5868e05fe
2023-06-03 21:17:47 +08:00
hiyouga
315e2bea67 add ziya prompt template
Former-commit-id: 321e44ac54a91260cf00a4caa1991708814473fc
2023-06-03 19:05:51 +08:00
hiyouga
5389fdacd4 use low_cpu_mem_usage to speed up loading
Former-commit-id: 7891e4c200566a4a47088e93efd1fbebcb46528e
2023-06-03 18:19:01 +08:00
hiyouga
cba25893d3 add logits processor
Former-commit-id: f6f4b1554ae1e8849b437d705ffa34ce7ebd56bb
2023-06-03 16:34:54 +08:00
hiyouga
587d0f5311 alter rewards data type
Former-commit-id: 3eb7eb2d37525da50fe401ab7c59532e6e1ef984
2023-06-02 14:19:51 +08:00
hiyouga
7ef5821cba fix possibly OOM error
Former-commit-id: 0d590dffb41b0e832d9f87d20a23bcd0acd983aa
2023-06-01 23:54:44 +08:00
hiyouga
a74fcc4149 support BLOOM models
Former-commit-id: 1314b6ea39a01aa8ac325e1d875ac013d43aec45
2023-05-31 16:54:06 +08:00
hiyouga
f8d03f3aa9 remove dummy code
Former-commit-id: e6bc89d280945bbf48281107145c40a41d7cbd56
2023-05-30 16:28:00 +08:00
hiyouga
bb6f731461 add pre-training script
Former-commit-id: 935d58de2b3a2eadc4f0bed28c3ad7dee32e9fd5
2023-05-29 21:37:22 +08:00
hiyouga
6f89f64c73 fix checkpoint loading
Former-commit-id: d31aa5c2c0bcb6a4ef4a62e21693548dd9acaae6
2023-05-29 17:43:16 +08:00
hiyouga
e158cd8b32 tiny fix
Former-commit-id: eae79707d31fd8be2cf4bee4d610557bbd49f6e7
2023-05-29 09:42:29 +08:00
hiyouga
a4384e442c use fp16 model, add logcallback
Former-commit-id: bea275d51338b49ce855eec0178e759607265e3d
2023-05-28 21:30:28 +08:00
hiyouga
54574f1dfa Initial commit
Former-commit-id: 5ca8e1d63727e7bcb8cab16542c763c47e48184a
2023-05-28 18:09:04 +08:00