onnxruntime
fix to avoid quantizing attention with varied q,k,v sizes
#9357

Merged

fix to avoid quantizing attention with varied q,k,v sizes #9357

viboga merged 2 commits into master from Vish/noattnquant

fix to avoid quantizing attention with varied q,k,v sizes

ccc99af4

viboga requested a review from

tianleiwu 4 years ago

viboga requested a review from

zhanghuanrong 4 years ago

zhanghuanrong commented on 2021-10-13

updated the changes to address the comments

0e0768a6

viboga requested a review from

zhanghuanrong 4 years ago

tianleiwu commented on 2021-10-13

viboga requested a review from

tianleiwu 4 years ago

zhanghuanrong approved these changes on 2021-10-13

viboga requested a review from

zhanghuanrong 4 years ago

viboga merged 4771256b into master 4 years ago

viboga deleted the Vish/noattnquant branch 4 years ago

Reviewers

zhanghuanrong

tianleiwu

Assignees

No one assigned

Labels

None yet

Milestone

No milestone