summary |
shortlog | log |
commit |
commitdiff |
tree
first ⋅ prev ⋅ next
Francois Fleuret [Sat, 27 Aug 2022 09:28:03 +0000 (11:28 +0200)]
The "mask" array actually specifies what attention to discard.
Francois Fleuret [Sat, 20 Aug 2022 05:47:14 +0000 (07:47 +0200)]
Replaced --synthesis_sampling with --deterministic_synthesis.
Francois Fleuret [Mon, 8 Aug 2022 15:59:08 +0000 (17:59 +0200)]
Added args.learning_rate_end for an exponential decay.
Francois Fleuret [Mon, 8 Aug 2022 05:13:54 +0000 (07:13 +0200)]
Added the small weight embedding + id layer norm inits.
Francois Fleuret [Sun, 7 Aug 2022 19:50:36 +0000 (21:50 +0200)]
Added the rng state in the checkpoint.
Francois Fleuret [Sun, 7 Aug 2022 19:50:15 +0000 (21:50 +0200)]
Added the small-weight embedding initialization.
Francois Fleuret [Sat, 30 Jul 2022 08:35:46 +0000 (10:35 +0200)]
Update.
Francois Fleuret [Sat, 30 Jul 2022 08:32:20 +0000 (10:32 +0200)]
Update.
Francois Fleuret [Sat, 30 Jul 2022 07:37:08 +0000 (09:37 +0200)]
Update.
Francois Fleuret [Sat, 30 Jul 2022 06:06:11 +0000 (08:06 +0200)]
Fixed a bug when there are no squares.
Francois Fleuret [Fri, 29 Jul 2022 08:07:59 +0000 (10:07 +0200)]
OCDC
Francois Fleuret [Fri, 29 Jul 2022 04:13:14 +0000 (06:13 +0200)]
Update.
Francois Fleuret [Fri, 29 Jul 2022 04:05:01 +0000 (06:05 +0200)]
Update.
Francois Fleuret [Fri, 29 Jul 2022 04:04:57 +0000 (06:04 +0200)]
OCDC
Francois Fleuret [Fri, 29 Jul 2022 03:45:46 +0000 (05:45 +0200)]
Update.
Francois Fleuret [Thu, 28 Jul 2022 19:53:21 +0000 (21:53 +0200)]
OCDC
Francois Fleuret [Thu, 28 Jul 2022 06:50:21 +0000 (08:50 +0200)]
Update.
Francois Fleuret [Wed, 27 Jul 2022 16:52:54 +0000 (18:52 +0200)]
Fixed stuff.
Francois Fleuret [Wed, 27 Jul 2022 14:42:47 +0000 (16:42 +0200)]
OCDC
Francois Fleuret [Wed, 27 Jul 2022 14:42:28 +0000 (16:42 +0200)]
OCDC
Francois Fleuret [Wed, 27 Jul 2022 14:22:26 +0000 (16:22 +0200)]
OCDC
Francois Fleuret [Wed, 27 Jul 2022 14:15:39 +0000 (16:15 +0200)]
Cleaning up more.
Francois Fleuret [Wed, 27 Jul 2022 14:07:36 +0000 (16:07 +0200)]
OCD cosmectics
Francois Fleuret [Wed, 27 Jul 2022 13:58:34 +0000 (15:58 +0200)]
Cosmetics.
Francois Fleuret [Wed, 27 Jul 2022 09:18:06 +0000 (11:18 +0200)]
Update.
Francois Fleuret [Wed, 27 Jul 2022 04:57:23 +0000 (06:57 +0200)]
OCD cosmetics.
Francois Fleuret [Wed, 27 Jul 2022 04:56:40 +0000 (06:56 +0200)]
Compute both the average number of requested and obtained properties.
Francois Fleuret [Tue, 26 Jul 2022 21:05:46 +0000 (23:05 +0200)]
Update.
Francois Fleuret [Tue, 26 Jul 2022 19:26:38 +0000 (21:26 +0200)]
Update.
Francois Fleuret [Tue, 26 Jul 2022 15:21:55 +0000 (17:21 +0200)]
Update.
Francois Fleuret [Tue, 26 Jul 2022 15:16:19 +0000 (17:16 +0200)]
Update.
Francois Fleuret [Tue, 26 Jul 2022 15:06:43 +0000 (17:06 +0200)]
Update.
Francois Fleuret [Tue, 26 Jul 2022 15:06:13 +0000 (17:06 +0200)]
Fixed the size of w_o.
Francois Fleuret [Tue, 26 Jul 2022 13:49:54 +0000 (15:49 +0200)]
Update.
Francois Fleuret [Tue, 26 Jul 2022 11:27:58 +0000 (13:27 +0200)]
Added a null token, which is the one to predict.
Francois Fleuret [Tue, 26 Jul 2022 10:47:26 +0000 (12:47 +0200)]
Removed the Linear transformation since there is now w_o.
Francois Fleuret [Tue, 26 Jul 2022 10:37:41 +0000 (12:37 +0200)]
Update.
Francois Fleuret [Tue, 26 Jul 2022 10:35:19 +0000 (12:35 +0200)]
Moved the input/output shift in the forward of the model.
Francois Fleuret [Mon, 25 Jul 2022 19:44:26 +0000 (21:44 +0200)]
Update.
Francois Fleuret [Mon, 25 Jul 2022 19:04:30 +0000 (21:04 +0200)]
OCD update
Francois Fleuret [Mon, 25 Jul 2022 19:03:46 +0000 (21:03 +0200)]
Added --no_checkpoint
Francois Fleuret [Mon, 25 Jul 2022 16:15:00 +0000 (18:15 +0200)]
Update.
Francois Fleuret [Mon, 25 Jul 2022 16:14:57 +0000 (18:14 +0200)]
Initialize properly w_o.
Francois Fleuret [Mon, 25 Jul 2022 13:44:50 +0000 (15:44 +0200)]
Update.
Francois Fleuret [Mon, 25 Jul 2022 13:31:58 +0000 (15:31 +0200)]
Update.
Francois Fleuret [Mon, 25 Jul 2022 13:31:18 +0000 (15:31 +0200)]
Added the (missing) W_o
Francois Fleuret [Mon, 25 Jul 2022 13:29:51 +0000 (15:29 +0200)]
Removed the default image size.
Francois Fleuret [Sat, 16 Jul 2022 14:34:28 +0000 (16:34 +0200)]
Update.
Francois Fleuret [Sat, 16 Jul 2022 10:12:06 +0000 (12:12 +0200)]
Update.
Francois Fleuret [Sat, 16 Jul 2022 09:49:57 +0000 (11:49 +0200)]
Update.
Francois Fleuret [Sat, 16 Jul 2022 08:51:13 +0000 (10:51 +0200)]
Update.
Francois Fleuret [Sat, 16 Jul 2022 08:47:27 +0000 (10:47 +0200)]
Update.
Francois Fleuret [Sat, 16 Jul 2022 08:14:21 +0000 (10:14 +0200)]
Update.
Francois Fleuret [Fri, 15 Jul 2022 16:10:14 +0000 (18:10 +0200)]
Update.
Francois Fleuret [Fri, 15 Jul 2022 15:52:40 +0000 (17:52 +0200)]
Update.
Francois Fleuret [Fri, 15 Jul 2022 15:07:47 +0000 (17:07 +0200)]
Update.
Francois Fleuret [Sat, 2 Jul 2022 19:07:54 +0000 (21:07 +0200)]
Update.
Francois Fleuret [Fri, 1 Jul 2022 08:01:27 +0000 (10:01 +0200)]
Update.
Francois Fleuret [Fri, 1 Jul 2022 08:01:12 +0000 (10:01 +0200)]
Add cross-attention to QKVAttention.
Francois Fleuret [Mon, 20 Jun 2022 06:14:46 +0000 (08:14 +0200)]
Finalized PicoCLVR with "many colors".
Francois Fleuret [Fri, 17 Jun 2022 11:54:07 +0000 (13:54 +0200)]
Update.
Francois Fleuret [Fri, 17 Jun 2022 11:53:10 +0000 (13:53 +0200)]
Update.
Francois Fleuret [Fri, 17 Jun 2022 11:31:08 +0000 (13:31 +0200)]
Update.
Francois Fleuret [Mon, 13 Jun 2022 13:33:56 +0000 (15:33 +0200)]
Update.
Francois Fleuret [Fri, 10 Jun 2022 09:18:26 +0000 (11:18 +0200)]
Update.
Francois Fleuret [Thu, 12 May 2022 06:28:24 +0000 (08:28 +0200)]
Update.
Francois Fleuret [Wed, 11 May 2022 20:21:00 +0000 (22:21 +0200)]
Update.
Francois Fleuret [Fri, 29 Apr 2022 12:08:20 +0000 (14:08 +0200)]
Update.
Francois Fleuret [Fri, 29 Apr 2022 11:59:10 +0000 (13:59 +0200)]
Update.
Francois Fleuret [Fri, 29 Apr 2022 11:58:55 +0000 (13:58 +0200)]
Update.
Francois Fleuret [Tue, 26 Apr 2022 14:44:05 +0000 (16:44 +0200)]
Update.
Francois Fleuret [Tue, 26 Apr 2022 08:05:59 +0000 (10:05 +0200)]
Update.
Francois Fleuret [Tue, 26 Apr 2022 07:22:25 +0000 (09:22 +0200)]
Update.
Francois Fleuret [Mon, 25 Apr 2022 18:30:25 +0000 (20:30 +0200)]
Fixed an asymmetry between top / bottom and between left / right.
Francois Fleuret [Mon, 25 Apr 2022 18:29:28 +0000 (20:29 +0200)]
Update.
Francois Fleuret [Sun, 24 Apr 2022 12:04:04 +0000 (14:04 +0200)]
Update.
Francois Fleuret [Sun, 24 Apr 2022 12:03:08 +0000 (14:03 +0200)]
Update.
Francois Fleuret [Sun, 24 Apr 2022 08:31:41 +0000 (10:31 +0200)]
Update.
Francois Fleuret [Sun, 24 Apr 2022 08:18:51 +0000 (10:18 +0200)]
Initial commit
projects / mygpt.git / log