23 Commits

Author	SHA1	Message	Date
Kohaku-Blueleaf	0af44f0262	Add option for output dir (#84)	2023-03-19 22:01:24 -07:00
Kohaku-Blueleaf	450206caaf	Fix torch.compile call on windows (#81) ... * Windows not support compile * Fix code style	2023-03-19 20:16:02 -07:00
Eric Wang	cfad895aa1	mask prompt in loss	2023-03-19 15:53:21 -07:00
Kakigōri Maker	9dab7ba438	add multi-gpu support (ddp) (#54) ... * add multi-gpu support (ddp) * Update finetune.py	2023-03-17 22:27:33 -07:00
Eric Wang	f7044049ab	dataset cleaning, visualizations	2023-03-17 15:04:25 -07:00
Eric Wang	35029da078	Validation set	2023-03-16 15:05:17 -07:00
Eric Wang	5f6614e6fc	Catch outdated installs	2023-03-16 12:11:47 -07:00
andreas.echavez	1862976b33	Update alpaca-lora to use transformers main branch	2023-03-16 12:11:29 -07:00
Eric Wang	2fa1c66388	repair tokenization logic, again	2023-03-15 23:58:44 -07:00
Eric Wang	024dde7dab	Revert "fix <eos> tokenization" ... This reverts commit 6b69ea8665.	2023-03-15 22:52:54 -07:00
Eric Wang	6b69ea8665	fix <eos> tokenization	2023-03-15 18:21:06 -07:00
Eric Wang	a2607faff0	fix finetuning code :(	2023-03-14 21:45:12 -07:00
Eric Wang	d714a73e8c	Update README.md with new checkpoint details	2023-03-14 21:33:12 -07:00
Eric Wang	ec98533876	Update README.md; clean up hyperparameters	2023-03-14 16:30:38 -07:00
Eric Wang	46ddd2ca85	Ready to go	2023-03-14 15:10:33 -07:00
Eric Wang	648af26073	update hyperparams	2023-03-14 08:51:30 -07:00
Eric Wang	5cd474bcc0	lr=2e-5	2023-03-14 08:47:49 -07:00
Jan Malte Lichtenberg	a3b80fdbd5	Fix bug in generate promp using 'instruction' instead of 'input'	2023-03-14 15:14:37 +01:00
Eric Wang	41e0ff6c78	tokenizer changes	2023-03-13 21:53:19 -07:00
Eric Wang	df2a5dc4be	cleanup notebooks	2023-03-13 17:33:27 -07:00
Eric Wang	357ec81a17	decapoda	2023-03-13 17:23:29 -07:00
Eric Wang	63121244c8	Licenses and whatnot	2023-03-13 15:00:05 -07:00
Eric Wang	26f64780ad	initial commit	2023-03-13 14:34:26 -07:00