47 Commits

Author SHA1 Message Date
hiyouga
38755bced7 add template, modify datasets
Former-commit-id: 386f590209e466b51c17a7ac8cee55fc3ce928d7
2023-11-09 15:53:23 +08:00
hoshi-hiyouga
28a9176784 Merge pull request #1436 from lvzii/main
fix tokenizer config changed after pretrain

Former-commit-id: 7ca32d8e69d5a2790b8ac323f6e6d0aea42600e7
2023-11-09 14:30:50 +08:00
hiyouga
f9ebc718d0 support parquet format #1446
Former-commit-id: 3df90b988bd987daa4e3a991c32ae53446481dcd
2023-11-09 14:17:40 +08:00
lvzi
13eb365eb7 fix tokenizer config changed after pretrain
Changing tokenizer's attribute at preprocessing stage will result in saving a wrong tokenizer.
for example, baichuan2

Former-commit-id: 043c316ac8913e10b2274867033f194ea92bfcd6
2023-11-08 15:50:46 +08:00
hiyouga
3d40bdb600 upgrade peft, fix #1088 #1411
Former-commit-id: b2a60905f384ada92618bf21301fe96dac1c10bf
2023-11-07 16:13:36 +08:00
hiyouga
5507014392 fix bug in data loader, support dpo eval
Former-commit-id: b355f6cac99592b66890ccc04e77a9993de0447d
2023-11-03 00:34:26 +08:00
hiyouga
a9db89a025 update data readme (zh)
Former-commit-id: cc8ffa10d877f5893f3940204e5bec6f3266559f
2023-11-02 23:42:49 +08:00
hiyouga
a1b0655457 support sharegpt format, add datasets
Former-commit-id: a8371724130db2fbd7273a480e2acb251e382aec
2023-11-02 23:10:04 +08:00
hiyouga
22b3c913e9 fix #1325
Former-commit-id: 083787dbfe41f58ff59cb16ddde02df98593aef5
2023-11-01 23:38:49 +08:00
hiyouga
fcfcac4858 support dataset cache
Former-commit-id: 3fe7df628db4093d7b3c121ececff60be0aa3a8a
2023-10-26 21:48:45 +08:00
hiyouga
84e27a1c0b remove filter in preprocess
Former-commit-id: 2caf91f824320b226daa4666eda2da7cb853db9c
2023-10-23 23:46:02 +08:00
hiyouga
4930118761 fix #1218
Former-commit-id: 7a11a42dfd414d140cd83b7a74760715d2ae2078
2023-10-19 16:17:41 +08:00
hiyouga
e585c789ce fix #1184
Former-commit-id: af18b0dce7a4ef10b30da069d454010eddd269af
2023-10-14 19:20:11 +08:00
hiyouga
9ef9cb316b fix webui
Former-commit-id: b240b6792fdb734dd77ed54861fdde059feb1855
2023-10-13 16:27:59 +08:00
hiyouga
141937ead6 fix aquila template, repair sft packing mechanism
Former-commit-id: be420e417920211b68f5b86a5ef5426aeaa62bb0
2023-10-10 18:49:55 +08:00
hiyouga
f88088c43d fix bug in packed sft dataset
Former-commit-id: de196143064772db770a45235424b3c911b2e147
2023-09-28 01:16:46 +08:00
hiyouga
8a8ba08bf7 tiny fix
Former-commit-id: d2ebd225dbb922adec99c1eb774c16f5cb973d2c
2023-09-28 01:02:11 +08:00
hiyouga
d2ce9b879b fix bug in pretraining
Former-commit-id: b3fbba57eb8be33a018b5904bdf08d1c95412005
2023-09-28 00:45:20 +08:00
hiyouga
108c31e1fc support LongLoRA
Former-commit-id: 90375f600d5601866836123597fa3ef52008eeef
2023-09-27 21:55:50 +08:00
hiyouga
4581d09fa6 fix #944
Former-commit-id: 338b8664edea5ae65192ac657bb013581245ae15
2023-09-21 19:51:02 +08:00
hiyouga
5377d0bf95 fix #896
Former-commit-id: 8632bff81110b202919e27b33294898f16638c9d
2023-09-14 18:37:34 +08:00
hiyouga
f865d0bd51 fix lora target
Former-commit-id: a51b7c98acc599de5ed2eaeeebe7b184105722c5
2023-09-09 17:04:45 +08:00
hiyouga
f74b980650 fix baichuan templates
Former-commit-id: 85b1f6632a752029dabdaed87c58986deb3a6b1d
2023-09-07 18:54:14 +08:00
hiyouga
3b5a9c60b6 fix #763
Former-commit-id: 370bdb6e4309db03e26cad311fb13e5cbb1fc1bf
2023-09-01 23:13:05 +08:00
hiyouga
a4fd976048 refactor dataset_attr, add eos in pt, fix #757
Former-commit-id: a9d1fb72f791ae57a4d12f4e3a7e2abccf6a7077
2023-09-01 19:00:45 +08:00
hiyouga
a46f277477 fix streaming in pt stage #548 #549
Former-commit-id: b0ed0dec5e6788a0344c09a6cc58d1116265fd68
2023-08-17 17:59:26 +08:00
hiyouga
edc15c62fa fix system prompt
Former-commit-id: 7407d9daa16bf6b3cd5002e16b2c53e402d2bc39
2023-08-16 01:35:52 +08:00
hiyouga
79f4ba0d26 Release v0.1.6
Former-commit-id: a48cb0d474ef0648a97387daf5f623498b5e3ee6
2023-08-11 23:25:57 +08:00
hiyouga
21bf79e72b add defaults
Former-commit-id: d3844e97e387b2106a32a576a61318ecec948e23
2023-08-11 13:56:26 +08:00
hiyouga
f1485ab927 fix baichuan template
Former-commit-id: 9c6dd1051417c91074daa7dd6ed6cc53448135ad
2023-08-11 13:45:47 +08:00
hiyouga
abdfa26d06 support DPO training (2305.18290)
Former-commit-id: 3ec4351cfdaf2aefcc7d13345e19d79874ed61d3
2023-08-11 03:02:53 +08:00
hiyouga
6404167ab7 support val set in streaming mode
Former-commit-id: d86ea314a197fd821770d895e988c48d46679047
2023-08-09 23:00:26 +08:00
hiyouga
28a807472b fix rm #420, fix template #426, fix #423
Former-commit-id: 39cd8b6989c9190d213e65467ec41f34ea04c5bc
2023-08-09 16:23:31 +08:00
hiyouga
77aa9853fb fix tokenizer #417
Former-commit-id: eecc4b2131e88b38fcd2659b52799a2f6459822f
2023-08-08 23:59:41 +08:00
hiyouga
9e49438c41 fix bug
Former-commit-id: caa0eda27dbb6cd198b2c2c244edae468417b77d
2023-08-08 21:28:28 +08:00
hiyouga
c796c542c8 fix chatml template #408
Former-commit-id: a9980617f5c6e3356b672c8635696b2f2e308a5e
2023-08-08 17:44:39 +08:00
hiyouga
733b395822 update readme
Former-commit-id: 20cf27976f24db2667955a8007e0ce2baa35fc82
2023-08-07 15:02:02 +08:00
hiyouga
65369ecf48 fix bos and eos token
Former-commit-id: d87c8fd8ab84c9f58c0b1f3fb4ad0adf98b25715
2023-08-04 23:55:57 +08:00
hiyouga
dbb284b5a2 fix encode
Former-commit-id: 8172ad1b5e3fa0b224d761ce6069d0db4397da2d
2023-08-04 23:27:55 +08:00
hiyouga
ea045b0e5b support chatml safe encoding
Former-commit-id: b4852f94065a11c8cd00ffa7e71ac0e0b2bf477a
2023-08-04 23:14:28 +08:00
hiyouga
b32ed1d7be support interleave probs
Former-commit-id: 69744c17e8180e0ad549b57d575454724b820d01
2023-08-04 21:27:35 +08:00
hiyouga
de407b59ea fix bug in preprocessing
Former-commit-id: 968ce0dcce6bfef582ce37aea6566a65f5aac811
2023-08-02 01:10:28 +08:00
hiyouga
43762f96e6 fix #296
Former-commit-id: e3f80774c46625dbeb85adeb5443450a41fb7ba2
2023-08-01 18:43:53 +08:00
hiyouga
e80b75b560 support streaming data, fix #284 #274 #268
Former-commit-id: 0411a4b3e122e7907441bc7a64b004948741a620
2023-07-31 23:33:00 +08:00
hiyouga
f769c2d3fc update web UI, support rm predict #210
Former-commit-id: ed0e186a134de816d6a9278f4e47baa6250a52d1
2023-07-21 13:27:27 +08:00
hiyouga
64b4f71673 simplify code
Former-commit-id: 67a27730744b71795b10260d050501bfe2329c26
2023-07-20 15:08:57 +08:00
hiyouga
a696148d6b modity code structure
Former-commit-id: f75137661358f9070bc70c341dfa2cc5fd69cf94
2023-07-15 16:54:28 +08:00