Commit Graph

61 Commits

Author SHA1 Message Date
enhuiz
4c5ebd2f1e Add data 2023-01-16 16:44:10 +08:00
enhuiz
035f48d670 Fix device 2023-01-16 16:39:16 +08:00
enhuiz
b5e1ab8057 Fix a difference in NAR implementation 2023-01-16 16:34:05 +08:00
enhuiz
8188506440 Use different sampling temperature for AR and NAR 2023-01-16 12:35:02 +08:00
enhuiz
77b52e42ce Batch accumulation. Ignore prompt and text loss in NAR and prompt loss in AR. Sampling temperature. 2023-01-16 02:01:41 +08:00
enhuiz
b7d3c89d6d Add __init__.py 2023-01-15 22:02:36 +08:00
enhuiz
f6c6df00b5 Update default configs 2023-01-15 12:14:33 +08:00
enhuiz
7f99d692c6 Add a plot script 2023-01-15 11:42:00 +08:00
enhuiz
f33ce7e5a9 Update requirements 2023-01-14 17:23:35 +08:00
enhuiz
06544d4fb7 Remove old requirements 2023-01-14 17:17:29 +08:00
enhuiz
1c5d49f9c1 Forbid the prompt and response audio to be the same 2023-01-14 17:09:42 +08:00
enhuiz
be4619cddf Support fp16 2023-01-14 17:09:08 +08:00
enhuiz
4cb958f7ff Update docs 2023-01-14 17:08:18 +08:00
enhuiz
a5f05418bd Add clone instruction 2023-01-13 13:41:19 +08:00
enhuiz
8355bc62bb Remove dirty flag 2023-01-13 13:35:06 +08:00
enhuiz
71c85e8a2b Add AdaNorm to AdaLN 2023-01-13 02:16:28 +08:00
enhuiz
fbf023448c Update 2023-01-13 01:08:13 +08:00
enhuiz
02ed659d2b Update 2023-01-13 01:04:48 +08:00
enhuiz
1903d7bb4e Add citations 2023-01-13 01:03:09 +08:00
enhuiz
6651403910 Remove old comments 2023-01-13 00:45:46 +08:00
enhuiz
f962a9d744 Fix raise 2023-01-13 00:34:40 +08:00
enhuiz
36e8894f5c AdaLN and sample-wise quant level sampling 2023-01-13 00:33:16 +08:00
Zhe Niu
bead906d72
Update setup.py 2023-01-12 20:45:59 +08:00
enhuiz
3b6c88d897 Fix iters 2023-01-12 20:42:28 +08:00
enhuiz
e9ca0e7959 More lenient requirement 2023-01-12 20:38:16 +08:00
enhuiz
c3894a8881 Don't include cuda version 2023-01-12 20:37:34 +08:00
enhuiz
e7606d97dd Python version 2023-01-12 20:37:04 +08:00
enhuiz
6d05e7f78b Set python version >= 3.10.0 2023-01-12 20:36:04 +08:00
enhuiz
c0fff54518 Rename 2023-01-12 20:30:59 +08:00
enhuiz
85c6a11f26 Faster validation 2023-01-12 20:27:31 +08:00
Zhe Niu
946c86e422
Update README.md 2023-01-12 20:09:58 +08:00
enhuiz
52998447b7 Make more VRAM friendly 2023-01-12 20:07:44 +08:00
enhuiz
f91db1a64c Update 2023-01-12 20:06:43 +08:00
enhuiz
886c11de46 Typo 2023-01-12 19:55:04 +08:00
enhuiz
74b9dc5068 Update config 2023-01-12 19:52:21 +08:00
enhuiz
4c1695e2cf Remove duplicates, smaller bs by default 2023-01-12 19:50:25 +08:00
enhuiz
777f9d6a13 Fix feeder 2023-01-12 19:48:48 +08:00
enhuiz
958c2df660 Better prompt sampling 2023-01-12 19:46:20 +08:00
enhuiz
3a4d5be18b Add model presets 2023-01-12 19:46:20 +08:00
enhuiz
ae029c1d75 Hardcode prom levels 2023-01-12 19:46:20 +08:00
Zhe Niu
49250b3c17
Update README.md 2023-01-12 18:31:29 +08:00
enhuiz
024a064910 Fix 2023-01-12 18:24:32 +08:00
enhuiz
0b3d323371 Add setup 2023-01-12 18:21:51 +08:00
enhuiz
8edaa36f2f Skip exists 2023-01-12 18:11:22 +08:00
Zhe Niu
45fc310876
Create LICENSE 2023-01-12 15:01:26 +08:00
Zhe Niu
9ab5686d8e
Update README.md 2023-01-12 14:57:28 +08:00
enhuiz
54acc621d4 Update docs 2023-01-12 14:56:30 +08:00
enhuiz
c3bacebfab Add trainer 2023-01-12 14:41:44 +08:00
enhuiz
d19449f1f1 Checkpoint on attention and only sample one layer for NAR 2023-01-12 14:41:30 +08:00
enhuiz
5e4ef084b8 Simplify interface 2023-01-12 14:24:35 +08:00