onnxruntime
Use fp32 accumulation in SkipLayerNorm/EmbedLayerNorm CUDA kernels
#28682

Merged

Use fp32 accumulation in SkipLayerNorm/EmbedLayerNorm CUDA kernels #28682

tianleiwu merged 4 commits into main from tlwu/sln_fp32_compute_type

use float as compute type in layer norm

8ef5ab44

add profiling script

fb33e823

github-advanced-security commented on 2026-05-26

tianleiwu requested a review from

copilot-pull-request-reviewer 61 days ago

copilot-pull-request-reviewer commented on 2026-05-27

address review feedback for profiling scripts

01336e2a

fallback if no nvtx

5549a98c

tianleiwu requested a review from

copilot-pull-request-reviewer 61 days ago

copilot-pull-request-reviewer commented on 2026-05-27

tianleiwu requested a review from

nenad1002 61 days ago

tianleiwu requested a review from

kunal-vaishnavi 61 days ago

kunal-vaishnavi approved these changes on 2026-05-27

tianleiwu merged 26871ab4 into main 60 days ago

tianleiwu deleted the tlwu/sln_fp32_compute_type branch 60 days ago

Reviewers

kunal-vaishnavi

nenad1002

Assignees

No one assigned

Labels

None yet

Milestone

No milestone