[VLMs] support attention backends #37576
zucchini-nlp
merged
25 commits into
huggingface:main
from
zucchini-nlp:new-attn-interface-vlms
update models
99eff853
why rename
99d3ff64
return attn weights when sdpa
15b3d61a
fixes
d2515838
fix attn implementation composite
cd6f3ab6
fix moshi
c456ce94
zucchini-nlp
commented
on 2025-04-18
zucchini-nlp
commented
on 2025-04-18
zucchini-nlp
commented
on 2025-04-18
add message
4a20bac6
add typings
ae3934e2
use explicitly all flags for each attn type
1bfc5b5c
fix some tests
c2516fef
import what is needed
848ff0c1
merge main
7e168d56
kosmos on main has ew attention already, yay
51d26910
new models in main, run fixup
98faf81e
won't fix kosmos yet
87747128
merge main
73b7a36c
fix-copies
0fe22a68
clean up after rebasing
c9dcdd47
fix tests
d1a79004
style
34a9419d
ArthurZucker
approved these changes
on 2025-05-08
dont cast attns to fp32
12d63ee5
merge main
d6619a37
did we update ruff? oke, let's just do what it asks
7f69fc61
fix pixtral after rebase
0d9eba82
Merge branch 'main' into new-attn-interface-vlms
b5757a41
uminaty
commented
on 2025-05-13
Reviewers
Assignees
No one assigned
Labels
None yet
Milestone
No milestone
Login to write a write a comment.
Login via GitHub