transformers
🚨 EP: fix EP router contract for many models + honor FP8 scale format
#46818
Merged
Merged
🚨 EP: fix EP router contract for many models + honor FP8 scale format #46818
vasqu
merged
32 commits into
main
from
fix-glm-dsa
honor the quant config's scale format and refuse
777c2726
IlyasMoutawwakil
changed the title 
fix fp4 specific
122adb59
strict
98034c9d
deeper ep fix
f1e2235c
test
83625bce
style
b34872ee
IlyasMoutawwakil
changed the title
add assertion
0e0550a2
3outeille
commented
on 2026-06-23
3outeille
commented
on 2026-06-23
3outeille
commented
on 2026-06-23
more ep plans
4a585fe6
fold tp+ep checks and ep assert into one helper
0288a11a
style
7d5976d4
rasie propper error upon ep request with no ep plan
8ea20bf5
Merge branch 'main' into fix-glm-dsa
d288a9e6
IlyasMoutawwakil
changed the title
address anton's comments and make more modular
d7a2fea9
fix repo
99f28f14
vasqu
commented
on 2026-06-23
more modular
b8f8eedd
more modular dsv2 topK router
4b04debf
IlyasMoutawwakil
commented
on 2026-06-24
IlyasMoutawwakil
commented
on 2026-06-24
modular phimoe router
c015bd2d
fix
6cf58567
IlyasMoutawwakil
commented
on 2026-06-24
reverting phimoe changes
c8379360
last modular attempt
9cd6feb3
correct fix ?
31f21395
add BC variation just in case
5c2b2aff
clearer message
918d6c47
post init workaround?
c54aee90
remove the warning
6de8bffd
Merge branch 'main' into fix-glm-dsa
d1dcde04
fix CI
54ab3ee2
ci
8d9846bf
fixup modular so we don't need to mess up more attribute maps
dd65d891
Merge branch 'main' into fix-glm-dsa
15af0007
fix glm4v moe
a5be1ba8
fix exaone
d71be264
Reviewers
Assignees
No one assigned
Labels
for patch
Milestone
No milestone
Login to write a write a comment.
Login via GitHub