53 Commits

Include renames

Author	SHA1	Message	Date
deep1	37429b7da2	💩	2023-09-24 16:39:23 +08:00
deep1	5b0c2d6ff4	fix 🐛 in datasets. random seed needs to vary	2023-09-23 16:06:18 +08:00
deep1	6314952d95	misc	2023-09-23 14:23:30 +08:00
deep1	8fd29ddc37	wip counterfactual inference debug	2023-09-17 18:32:05 +08:00
deep1	7b532b1b30	misc	2023-09-16 20:01:03 +08:00
deep1	3ac6c81d99	100%? I need a bigger ds	2023-09-10 15:18:26 +08:00
deep1	2d1c3f1868	96%?!	2023-09-10 13:04:21 +08:00
deep1	4aed182abe	wip	2023-09-09 08:08:11 +08:00
deep1	6ffae33c49	w int16	2023-09-08 15:06:40 +08:00
deep1	305bac87b2	wip	2023-09-07 12:29:57 +08:00
deep1	0dbcaa48f6	80% on linear probe... hmm	2023-09-04 15:49:18 +08:00
deep1	95536f83c9	wip	2023-09-03 20:07:20 +08:00
deep1	8522e268a0	wip	2023-09-03 16:44:31 +08:00
deep1	2e73d16932	init	2023-09-01 15:25:12 +08:00
deep1	b857ae6dd3	not working that great	2023-08-31 09:16:47 +08:00
deep1	43074f79cb	tidy and run probe	2023-08-27 13:46:26 +08:00
deep1	e7a87f7420	misc	2023-08-26 18:55:27 +08:00
deep1	28dec05dd9	new dataloading script alpha	2023-08-26 13:14:45 +08:00
deep1	4cc9daa76d	wip	2023-08-25 13:26:21 +08:00
deep1	4f84273777	wip	2023-08-18 16:01:41 +08:00
deep1	7a3a9dae73	fixed bug in batch	2023-08-11 15:56:14 +08:00
deep1	9533e61b6e	rought fix for 23 still need to tidy	2023-08-06 20:33:05 +08:00
deep1	38cf72f921	fixes	2023-08-06 18:47:42 +08:00
deep1	90d27388ee	tidy	2023-08-06 14:02:59 +08:00
deep1	d579a712b3	fixed generation, bug due to padding	2023-08-06 11:47:50 +08:00
deep1	d15b4fa54d	dataset refactored but there is a 🐛	2023-08-05 20:00:15 +08:00
deep1	8cb669f121	wip refactoring	2023-08-05 08:55:55 +08:00
deep1	217b7735e8	wip refactoring	2023-07-28 18:29:48 +08:00
deep1	add388479a	tidy	2023-07-28 17:24:43 +08:00
deep1	5c7631e8c2	tidy	2023-07-23 10:26:42 +08:00
deep1	a18ca5b822	debug the data gen ... - by using capitals - and recording a lie for every non lie to make sure the model can solve the problem	2023-07-10 08:11:45 +08:00
deep1	b1cc406376	fixed loaders	2023-07-08 12:47:02 +08:00
deep1	5a626008a9	include old experiments	2023-07-06 15:44:32 +08:00
deep1	b401b18041	new pipeline, and dual model works :)	2023-07-02 19:52:39 +08:00
wassname	e2ca9ddf71	add a supervised version	2023-06-18 18:32:04 +08:00
wassname	db8f3a8272	whole notebook is written, just need to try with more data	2023-06-18 13:18:04 +08:00
wassname	32f2f69e2a	dropout works on the falcan model... I just need to get the prompting right	2023-06-11 18:48:50 +08:00
wassname	62c1a0a582	the models work, lying sometimes works, and lie-dection leaks	2023-06-10 20:35:11 +08:00
wassname	38a5e8a15a	still not working	2023-06-10 14:59:57 +08:00
wassname	e113f2172b	easy way to download models	2023-06-10 13:44:15 +08:00
wassname	25ebc2a527	measure how often a model/prompt combo follows instructions	2023-06-09 13:03:28 +08:00
wassname	0b1ba11682	notes	2023-06-09 08:27:20 +08:00
wassname	f53fdc05f7	notes	2023-06-05 15:38:13 +08:00
deep1	9cac2c1152	I can't get the model to lie to me	2023-05-29 18:20:59 +08:00
deep1	1c6854437c	misc	2023-05-29 12:32:55 +08:00
deep1	7d2d9a4e8d	fix the id_n thing ... distinguish between _No and No	2023-05-28 19:15:13 +08:00
deep1	0f7b6459f4	4bit hf works better	2023-05-28 14:20:26 +08:00
deep1	e4d009e3d1	I want to use the hs from generate not forward	2023-05-27 20:37:00 +08:00
deep1	556e62101b	the generation seems to works!	2023-05-21 17:20:18 +08:00
wassname	32e9fb9440	trying to improve and debug generation	2023-05-20 10:28:37 +08:00