23 Commits

Author	SHA1	Message	Date
theblackcat102	a1e1445de9	[fix] evaluation dataset is incorrect in reward-model/trainer.py	2023-01-08 03:42:40 +00:00
theblackcat102	116045915e	[fix] syntax error and some typing issue in py38	2023-01-08 02:14:49 +00:00
theblackcat102	9eb401c633	[fix] resolve conflict from main	2023-01-08 01:32:39 +00:00
Szymon Ożóg	d0942a3256	Added option to use cosine scheduler	2023-01-07 16:36:38 +01:00
theblackcat102	3625f39948	[feature] Add GPTJ synthetic dataset, fix reference removal regex for webgpt	2023-01-07 01:36:27 +00:00
theblackcat102	325c97857c	Merge pull request #313 from bth5032/bth5032/78-blackcat-trainer ... Bth5032/78 blackcat trainer	2023-01-05 10:50:04 +08:00
Bobak Hashemi	da79aa04a0	Cleaned up default argument logic.	2023-01-03 21:45:16 -05:00
Bobak Hashemi	4569bcf354	fixed linting	2023-01-03 20:47:33 -05:00
Bobak Hashemi	568a42066a	FP32 Training Works	2023-01-03 00:53:07 -05:00
Bobak Hashemi	34ab948ade	testing rankgen integration into instructor trainer	2023-01-01 23:30:12 -05:00
Gareth Davidson	7000e10bc0	apply pre-commit rules	2023-01-02 00:01:45 +00:00
theblackcat102	fe99b46f2e	[fix] pre-commit update	2023-01-01 11:43:15 +00:00
theblackcat102	168e9ca6b3	[feature] added summary quality rater	2023-01-01 10:19:25 +00:00
theblackcat102	4b7f1f25a1	[fix] Use official split for eval	2023-01-01 03:07:40 +00:00
theblackcat102	a5a2625e2d	[merge] most of the bugs should be fixed. #77	2023-01-01 02:55:54 +00:00
theblackcat102	e27a3eb3c7	[fix] Tidy up todo and trainer comments	2023-01-01 02:22:57 +00:00
theblackcat102	ba336fb087	[fix] fix freeze top N layers	2022-12-31 17:43:27 +00:00
theblackcat102	918b7b7ec0	[feature] Add galactica training config	2023-01-01 01:25:53 +08:00
theblackcat102	f3c299757d	[feature] added configs argument for parameters training and recording	2022-12-31 17:02:46 +00:00
theblackcat102	3a10f1024a	[fix] Fix truncation in collate fn	2022-12-31 09:27:09 +00:00
theblackcat102	b2ef4695a0	[fix] Fix missing accuracy and eval loss	2022-12-31 03:47:54 +00:00
theblackcat102	bcd5c52b3b	[feature] working trainer code	2022-12-31 03:02:10 +00:00
theblackcat102	ad98a28241	[feature] add rank dataset for webgpt and human feedback summary	2022-12-30 17:25:50 +00:00