Commit Graph

407 Commits

Author SHA1 Message Date
hiyouga b240b6792f fix webui 2023-10-13 16:27:59 +08:00
hiyouga 11bd271364 fix ppo args 2023-10-11 23:40:50 +08:00
hiyouga 2818af0b09 refactor model_dtype, fix PPO trainer 2023-10-11 23:16:01 +08:00
hiyouga 5310e4d182 add averaging in evaluation 2023-10-10 23:16:31 +08:00
hiyouga be420e4179 fix aquila template, repair sft packing mechanism 2023-10-10 18:49:55 +08:00
hiyouga e1dcb8e4dc tiny fix 2023-10-10 17:41:13 +08:00
hiyouga 0a356bc897 fix flash shift short attention 2023-10-09 17:54:48 +08:00
hiyouga 6b24f29c8a fix webui args 2023-10-09 17:13:57 +08:00
hiyouga ab65c3063b fix shift short attention 2023-10-09 17:07:46 +08:00
hiyouga b8dbec086e update webui #1086 2023-10-09 14:50:14 +08:00
hiyouga a683c5b797 fix #1097 2023-10-08 22:29:26 +08:00
hiyouga f9769cff8a add llamafy_qwen.py 2023-10-08 22:05:36 +08:00
hiyouga d11a545463 fix #1068 #1074 2023-09-28 14:39:16 +08:00
hiyouga de19614306 fix bug in packed sft dataset 2023-09-28 01:16:46 +08:00
hiyouga 5d4118b096 tiny fix 2023-09-28 01:03:04 +08:00
hiyouga d2ebd225db tiny fix 2023-09-28 01:02:11 +08:00
hiyouga c902236397 fix #1064 2023-09-28 00:53:29 +08:00
hiyouga b3fbba57eb fix bug in pretraining 2023-09-28 00:45:20 +08:00
hiyouga 84b7486885 fix layer norm dtype 2023-09-28 00:25:55 +08:00
hiyouga b0b0138e1d fix #1026 2023-09-27 22:57:09 +08:00
hiyouga 35fa94723c fix #424 2023-09-27 22:49:43 +08:00
hiyouga a41e12de5e fix #1032 2023-09-27 22:42:16 +08:00
hiyouga 620efe1d8d refactor finetuning Args 2023-09-27 22:28:06 +08:00
hiyouga 90375f600d support LongLoRA 2023-09-27 21:55:50 +08:00
hiyouga 4dd9b4d982 add CMMLU, update eval script 2023-09-23 21:10:17 +08:00
hiyouga f8ff625d76 update evaluate 2023-09-23 11:55:31 +08:00
hiyouga badd2735b5 move file 2023-09-23 11:52:12 +08:00
hiyouga 465ee8119a add MMLU and C-Eval script 2023-09-23 00:34:17 +08:00
hiyouga 5cc7a44784 fix #1000 2023-09-22 15:00:48 +08:00
hiyouga 5f3ab3ddde fix webui 2023-09-21 19:55:38 +08:00
hiyouga dbaef776a1 tiny fix 2023-09-21 19:52:06 +08:00
hiyouga 338b8664ed fix #944 2023-09-21 19:51:02 +08:00
hiyouga ace3f85a72 tiny fix 2023-09-21 15:25:29 +08:00
statelesshz b3e41c6d49 support export model on Ascend NPU 2023-09-20 10:26:02 +08:00
hiyouga 10ab2f8b90 fix webui 2023-09-19 18:35:21 +08:00
hiyouga 7e8655c8b5 fix error info 2023-09-19 18:30:23 +08:00
hiyouga 469f859161 add tests.cal_flops.py 2023-09-16 23:40:41 +08:00
hiyouga 0b5f970c05 fix #913 2023-09-15 20:58:28 +08:00
hiyouga 8632bff811 fix #896 2023-09-14 18:37:34 +08:00
hiyouga 8857e45602 fix #887 2023-09-14 17:56:58 +08:00
mmbwf 30fb721f12
Update utils.py
Fix parameters load error.
2023-09-14 15:38:04 +08:00
hiyouga 7ba57d5b14 fix ppo save model 2023-09-12 16:25:29 +08:00
hiyouga d4be857e23 fix #762 #814 2023-09-12 16:10:10 +08:00
hiyouga 3b306478d4 tiny fix 2023-09-11 18:27:08 +08:00
hiyouga ccb3553576 Release v0.1.8 2023-09-11 17:31:34 +08:00
hiyouga 0fbece85a7 update flashattn, fix ppo save model 2023-09-11 17:25:36 +08:00
hiyouga b218c271ed remove PeftTrainer 2023-09-10 22:23:23 +08:00
hiyouga d8aa1404be support FlashAttention2 2023-09-10 20:43:56 +08:00
hiyouga 815b92e698 fix #850 2023-09-10 14:22:03 +08:00
hiyouga a51b7c98ac fix lora target 2023-09-09 17:04:45 +08:00
hiyouga bca1a247bc support lora target auto find 2023-09-09 15:38:37 +08:00
hiyouga d8d82ca281 fix chatglm2 tokenizer 2023-09-09 13:50:29 +08:00
hiyouga 90bd085ae4 fix bug in DPO data collator 2023-09-08 20:45:07 +08:00
hiyouga b34797a845 fix #761 2023-09-08 20:22:18 +08:00
hiyouga 8ea32e4046 change to right-padding, update reward score #803 2023-09-08 20:04:31 +08:00
hiyouga 8aaaa132d4 fix chatglm template 2023-09-08 14:45:58 +08:00
hiyouga f5351c18e1 update requirements 2023-09-07 19:26:25 +08:00
hiyouga 5a9970dbef fix #818 2023-09-07 19:19:53 +08:00
hiyouga ed1c2c5557 add deepspeed check in PPO training 2023-09-07 19:12:40 +08:00
hiyouga e2bf7c3bad fix #809 2023-09-07 19:04:32 +08:00
hiyouga 85b1f6632a fix baichuan templates 2023-09-07 18:54:14 +08:00
hiyouga 0531886e1f update baichuan2 template 2023-09-06 21:43:06 +08:00
hiyouga 62ce65c628 add Baichuan2 models 2023-09-06 18:36:04 +08:00
Q a19fc2ebf7
fix utils.py bug 2023-09-05 10:38:01 +08:00
hiyouga 370bdb6e43 fix #763 2023-09-01 23:13:05 +08:00
hiyouga a9d1fb72f7 refactor dataset_attr, add eos in pt, fix #757 2023-09-01 19:00:45 +08:00
codemayq 0bcc489c42 update llama2 template 2023-08-30 16:23:56 +08:00
codemayq f7fdc088d4 add dataset stage check 2023-08-30 16:23:08 +08:00
codingma 9b4d16d040
Merge pull request #651 from hiyouga/feature-dataset_stage
add dataset stage
2023-08-28 16:03:45 +08:00
codemayq 24e68d29f2 add text format dataset preview in webui 2023-08-24 19:45:36 +08:00
codemayq ba94c8729d add stage in DatasetAttr 2023-08-23 20:54:53 +08:00
hiyouga 2de1a7610a fix import error 2023-08-23 20:45:03 +08:00
hiyouga 57146c101f fix #649 2023-08-23 20:21:15 +08:00
codemayq c0e4d1e81b add dataset stage and filter dataset when stage chosen in webui 2023-08-23 18:54:23 +08:00
hiyouga 1c702ad538 fix webui 2023-08-23 11:03:35 +08:00
codemayq a7cc6c4140 fix quantization bit is "" 2023-08-23 10:08:17 +08:00
codemayq ec2047b064 fix quantization is "" 2023-08-23 10:04:03 +08:00
hiyouga 4318347d3f update template 2023-08-22 19:46:09 +08:00
hiyouga 5235b15c91 fix #617 2023-08-21 18:16:11 +08:00
hiyouga 02d69b6fde fix #608 2023-08-21 17:49:36 +08:00
hiyouga 0a3f698425 fix baichuan template for training #597 #616 2023-08-21 17:41:51 +08:00
hiyouga 5c052836a0 fix #595 2023-08-20 16:40:00 +08:00
beat4ocean 7b45de6b9f fix KeyError: 'lang' bug 2023-08-20 15:32:36 +08:00
hiyouga 0676497104 fix ppo trainer #551 2023-08-20 14:07:11 +08:00
hiyouga 9c9009f49f Release v0.1.7 2023-08-18 17:21:27 +08:00
hiyouga d75e377b0f tiny fix 2023-08-18 13:07:35 +08:00
hiyouga 53e33418d0 support ppo score norm (trl 0.5.1.dev required) 2023-08-18 12:02:42 +08:00
hiyouga 9020524418 fix PPO trainer #551 , update readme 2023-08-18 11:43:10 +08:00
hiyouga 58f13e22da update training resuming 2023-08-18 01:41:17 +08:00
hoshi-hiyouga 7252903245
Merge branch 'main' into main 2023-08-18 01:37:23 +08:00
hiyouga d125218cde support bf16 ppo #551 2023-08-18 00:40:32 +08:00
hiyouga 9f4c2adc9a fix ChatGLM2 ppo #527 #528 2023-08-18 00:34:59 +08:00
hiyouga be21fc83f9 fix generation bug #532 2023-08-17 22:21:34 +08:00
hiyouga b0ed0dec5e fix streaming in pt stage #548 #549 2023-08-17 17:59:26 +08:00
hiyouga 892fd39373 fix baichuan and intern template 2023-08-17 01:27:20 +08:00
hiyouga d9e62711a3 fix generation 2023-08-16 22:39:54 +08:00
hiyouga 7407d9daa1 fix system prompt 2023-08-16 01:35:52 +08:00
hiyouga 273135f595 fix baichuan template #481 2023-08-15 11:38:21 +08:00
hiyouga af6c011fcb fix ChatGLM RLHF 2023-08-15 11:19:20 +08:00
hiyouga 80b4053602 alert pad_token source 2023-08-15 00:07:56 +08:00
hiyouga 9d0f6214b6 update webui 2023-08-14 22:45:26 +08:00
hoshi-hiyouga adb0f186e9
Merge pull request #511 from hiyouga/feature-autoTemplate
add template match and stage in webui
2023-08-14 22:44:04 +08:00
codemayq 0bf892ff1a auto match template when change model_name 2023-08-14 20:56:05 +08:00
codemayq 79c68e5527 add template match and stage in webui 2023-08-14 20:42:59 +08:00
hiyouga d019956808 fix ChatGLM lm_head #494 2023-08-14 14:14:48 +08:00
hiyouga 20a29297b1 fix bug in webui 2023-08-14 11:38:42 +08:00
hiyouga ca08e5efd3 fix webui cache 2023-08-14 11:37:01 +08:00
hiyouga ec94274ca1 web UI integrating RLHF 2023-08-14 10:48:47 +08:00
hiyouga 2f2fd55d81 fix #480 2023-08-14 00:23:56 +08:00
hiyouga d69b1388e6 fix webui 2023-08-12 23:52:07 +08:00
hiyouga 9dc6a296e3 tiny fix 2023-08-12 22:02:43 +08:00
hiyouga 8545c11c45 fix rope scaling 2023-08-12 22:00:01 +08:00
hiyouga 1836c020c5 update readme 2023-08-12 21:00:11 +08:00
hiyouga fa940c17b8 support rope scaling, fix #475 #476 #478 2023-08-12 20:46:27 +08:00
codemayq 6bc8e9866d add sft script preview in webui 2023-08-12 13:53:55 +08:00
hiyouga dd51c24203 fix unusual output of 8bit models #278 #391 2023-08-12 00:25:29 +08:00
hiyouga a48cb0d474 Release v0.1.6 2023-08-11 23:25:57 +08:00
hiyouga d3844e97e3 add defaults 2023-08-11 13:56:26 +08:00
hiyouga d59f938959 fix stop word in baichuan template 2023-08-11 13:51:46 +08:00
hiyouga 9c6dd10514 fix baichuan template 2023-08-11 13:45:47 +08:00
hiyouga 3ec4351cfd support DPO training (2305.18290) 2023-08-11 03:02:53 +08:00
hoshi-hiyouga 685dae4eff
Merge pull request #451 from jovialchen/main
huggingface login for projects must login while running
2023-08-10 17:25:38 +08:00
hiyouga ad6e7c76c7 fix webui val size 2023-08-10 15:20:44 +08:00
jiongxuc 3e000c2b60 huggingface login for projects must login while running 2023-08-10 14:57:12 +08:00
hiyouga eb6e571cb7 fix template 2023-08-09 23:14:27 +08:00
hiyouga ac29f4d5f0 fix template 2023-08-09 23:10:20 +08:00
hiyouga d86ea314a1 support val set in streaming mode 2023-08-09 23:00:26 +08:00
hiyouga 572ea3bafb fix tokenizer 2023-08-09 17:52:15 +08:00
niuba 2ec68d3398 add last_checkpoint support 2023-08-09 16:39:27 +08:00
hiyouga df946e6949 fix sft trainer 2023-08-09 16:35:03 +08:00
hiyouga 39cd8b6989 fix rm #420, fix template #426, fix #423 2023-08-09 16:23:31 +08:00
hoshi-hiyouga 2d90685358
fix llama2 template 2023-08-09 00:58:27 +08:00
hoshi-hiyouga 32fa5e8d70
fix tokenizer 2023-08-09 00:54:54 +08:00
hiyouga 3a720aac66 update webui 2023-08-09 00:26:11 +08:00
hiyouga eecc4b2131 fix tokenizer #417 2023-08-08 23:59:41 +08:00
hiyouga caa0eda27d fix bug 2023-08-08 21:28:28 +08:00
hiyouga 4b841a6b35 fix bug 2023-08-08 17:55:55 +08:00
hiyouga a9980617f5 fix chatml template #408 2023-08-08 17:44:39 +08:00
hiyouga 5453b93db0 update args spec 2023-08-07 15:23:35 +08:00
hiyouga 20cf27976f update readme 2023-08-07 15:02:02 +08:00
hiyouga 081345baca fix #376 2023-08-07 13:58:59 +08:00
hiyouga 220175ab24 update trainer 2023-08-07 13:34:35 +08:00
hiyouga e21ae01356 fix qwen eos token 2023-08-06 13:31:17 +08:00
hiyouga 7f18d2a335 fix qwen tokenizer #361 2023-08-05 17:06:05 +08:00
hiyouga 1afa51c2fa fix template for tiktoken 2023-08-05 13:42:42 +08:00
hiyouga 53d95725c5 remove redundant code 2023-08-05 00:27:27 +08:00
hiyouga c183b3551d fix template 2023-08-05 00:25:00 +08:00
hiyouga e4a15f863c fix llama2 template 2023-08-05 00:07:54 +08:00
hiyouga d87c8fd8ab fix bos and eos token 2023-08-04 23:55:57 +08:00
hiyouga 8172ad1b5e fix encode 2023-08-04 23:27:55 +08:00