onnxruntime
[CPU] GQA supports head_sink input for smooth softmax
#25269

Merged

[CPU] GQA supports head_sink input for smooth softmax #25269

tianleiwu merged 11 commits into main from tlwu/gqa_head_sink

update spec

1d84450c

tianleiwu marked this pull request as draft 173 days ago

Implement CPU

a1b51b74

github-actions commented on 2025-07-05

merge main

4a9f45c4

add cpu test

e8b1490c

Merge branch 'main' into tlwu/gqa_head_sink

fdb8619f

reduce tests

fc6d1b3f

tianleiwu changed the title ~~[WIP] GQA supports per head smooth softmax~~ [CPU] GQA supports per head smooth softmax 167 days ago

lintrunner

489dba20

tianleiwu marked this pull request as ready for review 167 days ago

tianleiwu requested a review from

fs-eire 167 days ago

tianleiwu requested a review from

kunal-vaishnavi 167 days ago

revert some files

1a1421d0

tianleiwu changed the title ~~[CPU] GQA supports per head smooth softmax~~ [CPU] GQA supports head_sink input for smooth softmax 167 days ago

revert cuda/rocm

d09d8049

update doc

19e3f55f

update cuda script

60de045c

fs-eire commented on 2025-07-09

nenad1002 approved these changes on 2025-07-09

tianleiwu merged cd5f91fe into main 167 days ago

tianleiwu deleted the tlwu/gqa_head_sink branch 167 days ago

Reviewers

nenad1002

fs-eire

github-actions

kunal-vaishnavi

Assignees

No one assigned

Labels

None yet

Milestone

No milestone

onnxruntime [CPU] GQA supports head_sink input for smooth softmax #25269 Merged

[CPU] GQA supports head_sink input for smooth softmax #25269

onnxruntime
[CPU] GQA supports head_sink input for smooth softmax
#25269

Merged