onnxruntime
[webgpu] Use components for VxAttentionScore
#23726
Merged
Overview
Commits
10
Changes
View On GitHub

[webgpu] Use components for VxAttentionScore #23726

guschmue merged 10 commits into main from attention_opt
qjia7
qjia7 Use components for VxAttentionScore
48e4a5f5
qjia7 use flash attention to merge softmax to the last stage
e39c6b0d
qjia7 use f32 instead of f16 for max/sum
ee42c55c
qjia7 make it work with GQA
1bf3739a
qjia7 recover ApplyFlashAttention
f774cb87
qjia7 qjia7 changed the title [webgpu] Use components for VxAttentionScore [webgpu] Use components for VxAttentionScore and merge softmax to it 1 year ago
qjia7 qjia7 requested a review from guschmue guschmue 1 year ago
qjia7 qjia7 requested a review from sushraja-msft sushraja-msft 1 year ago
qjia7
qjia7 Revert "recover ApplyFlashAttention"
abbfef21
qjia7 Revert "make it work with GQA"
ad528a7c
qjia7 Revert "use f32 instead of f16 for max/sum"
f7f758d3
qjia7 Revert "use flash attention to merge softmax to the last stage"
3c03d57b
qjia7 qjia7 changed the title [webgpu] Use components for VxAttentionScore and merge softmax to it [webgpu] Use components for VxAttentionScore 1 year ago
qjia7
guschmue
guschmue approved these changes on 2025-02-19
qjia7 Merge branch 'main' into attention_opt
33e428ee
qjia7
guschmue
guschmue guschmue added ep:WebGPU
guschmue guschmue merged 9b2b2ee8 into main 1 year ago
guschmue guschmue deleted the attention_opt branch 1 year ago

Login to write a write a comment.

Login via GitHub

Reviewers
Assignees
No one assigned
Labels
ep:WebGPU
Milestone
No milestone
Terms Of Use Privacy Policy Refund Policy Impressum