Commit graph

60 commits

Author SHA1 Message Date
C43H66N12O12S2
623251ce2b allow pascal onwards 2022-10-10 19:54:07 +03:00
hentailord85ez
d5c14365fd Add back in output hidden states parameter 2022-10-10 18:54:48 +03:00
hentailord85ez
460bbae587 Pad beginning of textual inversion embedding 2022-10-10 18:54:48 +03:00
hentailord85ez
b340439586 Unlimited Token Works
Unlimited tokens actually work now. Works with textual inversion too. Replaces the previous not-so-much-working implementation.
2022-10-10 18:54:48 +03:00
Fampai
1824e9ee3a Removed unnecessary tmp variable 2022-10-09 22:31:23 +03:00
Fampai
ad3ae44108 Updated code for legibility 2022-10-09 22:31:23 +03:00
Fampai
e59c66c008 Optimized code for Ignoring last CLIP layers 2022-10-09 22:31:23 +03:00
Fampai
1371d7608b Added ability to ignore last n layers in FrozenCLIPEmbedder 2022-10-08 22:10:37 +03:00
AUTOMATIC
3061cdb7b6 add --force-enable-xformers option and also add messages to console regarding cross attention optimizations 2022-10-08 19:22:15 +03:00
C43H66N12O12S2
cc0258aea7 check for ampere without destroying the optimizations. again. 2022-10-08 17:54:16 +03:00
C43H66N12O12S2
017b6b8744 check for ampere 2022-10-08 17:54:16 +03:00
AUTOMATIC
cfc33f99d4 why did you do this 2022-10-08 17:29:06 +03:00
AUTOMATIC
27032c47df restore old opt_split_attention/disable_opt_split_attention logic 2022-10-08 17:10:05 +03:00
AUTOMATIC
dc1117233e simplify xfrmers options: --xformers to enable and that's it 2022-10-08 17:02:18 +03:00
AUTOMATIC1111
48feae37ff
Merge pull request #1851 from C43H66N12O12S2/flash
xformers attention
2022-10-08 16:29:59 +03:00
C43H66N12O12S2
970de9ee68
Update sd_hijack.py 2022-10-08 16:29:43 +03:00
C43H66N12O12S2
26b459a379
default to split attention if cuda is available and xformers is not 2022-10-08 16:20:04 +03:00
MrCheeze
5f85a74b00 fix bug where when using prompt composition, hijack_comments generated before the final AND will be dropped 2022-10-08 15:48:04 +03:00
AUTOMATIC
77f4237d1c fix bugs related to variable prompt lengths 2022-10-08 15:25:59 +03:00
AUTOMATIC
4999eb2ef9 do not let user choose his own prompt token count limit 2022-10-08 14:25:47 +03:00
AUTOMATIC
706d5944a0 let user choose his own prompt token count limit 2022-10-08 13:38:57 +03:00
C43H66N12O12S2
91d66f5520
use new attnblock for xformers path 2022-10-08 11:56:01 +03:00
C43H66N12O12S2
b70eaeb200
delete broken and unnecessary aliases 2022-10-08 04:10:35 +03:00
AUTOMATIC
f7c787eb7c make it possible to use hypernetworks without opt split attention 2022-10-07 16:39:51 +03:00
C43H66N12O12S2
5e3ff846c5
Update sd_hijack.py 2022-10-07 06:38:01 +03:00
C43H66N12O12S2
5303df2428
Update sd_hijack.py 2022-10-07 06:01:14 +03:00
C43H66N12O12S2
35d6b23162
Update sd_hijack.py 2022-10-07 05:31:53 +03:00
C43H66N12O12S2
2eb911b056
Update sd_hijack.py 2022-10-07 05:22:28 +03:00
Jairo Correa
ad0cc85d1f Merge branch 'master' into stable 2022-10-02 18:31:19 -03:00
AUTOMATIC
88ec0cf557 fix for incorrect embedding token length calculation (will break seeds that use embeddings, you're welcome!)
add option to input initialization text for embeddings
2022-10-02 19:40:51 +03:00
AUTOMATIC
820f1dc96b initial support for training textual inversion 2022-10-02 15:03:39 +03:00
Jairo Correa
ad1fbbae93 Merge branch 'master' into fix-vram 2022-09-30 18:58:51 -03:00
AUTOMATIC
98cc6c6e74 add embeddings dir 2022-09-30 14:16:26 +03:00
AUTOMATIC
c715ef04d1 fix for incorrect model weight loading for #814 2022-09-29 15:40:28 +03:00
AUTOMATIC
c1c27dad3b new implementation for attention/emphasis 2022-09-29 11:31:48 +03:00
Jairo Correa
c2d5b29040 Move silu to sd_hijack 2022-09-29 01:16:25 -03:00
Liam
e5707b66d6 switched the token counter to use hidden buttons instead of api call 2022-09-27 19:29:53 -04:00
Liam
5034f7d759 added token counter next to txt2img and img2img prompts 2022-09-27 15:56:18 -04:00
AUTOMATIC
073f6eac22 potential fix for embeddings no loading on AMD cards 2022-09-25 15:04:39 +03:00
guaneec
615b2fc9ce Fix token max length 2022-09-25 09:30:02 +03:00
AUTOMATIC
254da5d127 --opt-split-attention now on by default for torch.cuda, off for others (cpu and MPS; because the option does not work there according to reports) 2022-09-21 09:49:02 +03:00
AUTOMATIC
1578859305 fix for too large embeddings causing an error 2022-09-21 00:20:11 +03:00
AUTOMATIC
90401d96a6 fix a off by one error with embedding at the start of the sentence 2022-09-20 12:12:31 +03:00
AUTOMATIC
ab38392119 add the part that was missing for word textual inversion checksums 2022-09-20 09:53:29 +03:00
AUTOMATIC
cae5c5fa8d Making opt split attention the default. Are you upset about this? Sorry. 2022-09-18 20:55:46 +03:00
C43H66N12O12S2
18d6fe4346
..... 2022-09-18 01:21:50 +03:00
C43H66N12O12S2
d63dbb3acc
Move scale multiplication to the front 2022-09-18 01:05:31 +03:00
C43H66N12O12S2
72d7f8c761 fix typo 2022-09-15 14:14:27 +03:00
C43H66N12O12S2
7ec6282ec2 pass dtype to torch.zeros as well 2022-09-15 14:14:27 +03:00
C43H66N12O12S2
3b1b1444d4
Complete cross attention update 2022-09-13 14:29:56 +03:00