4d6f2b65c8
chore(settings): remove the bmm attention methods
...
Batch matrix-matrix and Dynamic Attention BMM applied a legacy Attention
processor to pipe.unet, which a diffusion transformer does not have, so
they served unet models alone and said nothing elsewhere. The choices, the
processor and its slice helper are removed, an unrecognized method now
warns rather than selecting nothing, and a stored value is rewritten to
Scaled-Dot-Product on load.
2026-08-24 19:59:41 +01:00
76941efc7e
strict typechecking
...
Signed-off-by: Vladimir Mandic <mandic00@live.com >
2026-08-16 12:45:50 +02:00
6aa5b45b80
remove directml
...
Signed-off-by: Vladimir Mandic <mandic00@live.com >
2026-08-10 13:43:45 +02:00
bfe014f5da
modernize typing
2026-02-19 09:15:37 +01:00
bc7c89c070
add typing to sdpa hijacks
2025-09-10 04:43:57 +03:00
ad716b118b
fix enable_gqa with dyn atten
2025-07-26 01:49:17 +03:00
90f887ac4a
Add dim checks to ck flash atten and fix dim check on dyn atten
2025-03-25 03:50:21 +03:00
675956010f
Multiply the slice rate for dyn bmm by 4
2025-01-30 18:44:26 +03:00
6960b05129
lint and update changelog
...
Signed-off-by: Vladimir Mandic <mandic00@live.com >
2025-01-28 09:02:44 -05:00
b0ecdf2f1c
Better dyn atten sdpa usage estimation logic
2025-01-26 16:57:26 +03:00
bb07dd7a8f
Add trigger rate control to dyn atten
2025-01-26 03:48:01 +03:00
3f2c9c4e22
Fix dyn atten not registering slice rate changes
2025-01-25 22:44:29 +03:00
d266143f20
Cleanup
2025-01-25 21:46:23 +03:00
02ecc6710d
Fix missing unsqueeze
2025-01-25 21:43:38 +03:00
a770b1c888
More correct Dynamic Atten SDPA implementation and deprecate IPEX Diffusers attention
2025-01-25 21:33:42 +03:00
85c3a76c1a
Dynamic Atten use expand instead of repeat
2024-10-25 21:06:46 +03:00
7d56d5b72c
Dynamic atten don't repeat the last shape of atten masks
2024-10-25 13:45:20 +03:00
3e268e7cea
Cleanup
2024-10-24 21:53:33 +03:00
3195e8ad1f
Dynamic Atten fix OmniGen
2024-10-24 21:22:06 +03:00
2e2cb43406
Make SDPA hijacks chainable and add Sage Attention
2024-10-12 21:19:38 +03:00
f2c5cbbb36
lint updates and diffusers installer
2024-09-06 14:10:53 -04:00
a3f26c9df0
Convert Dynamic Attention SDP to a global SDP option
2024-08-18 01:44:27 +03:00
bbaec29720
Cleanup
2024-06-21 21:07:15 +03:00
c9e01b4a32
OpenVINO fix --device-id with single GPU
2024-02-12 01:57:27 +03:00
3f8da51eac
Set default atten for ROCm to SDP
2024-02-11 13:44:39 +03:00
0ac2dfbcaa
Dyn Atten don't synchronize if not slicing
2024-02-10 20:02:29 +03:00
43c5be76ca
Unite attention optimization settings
2024-02-10 10:32:53 -05:00
CalamitousFelicitousness
Vladimir Mandic
Disty0
Disty0