PR #33469 [CUDA] Add RMSNormRotary contrib operator

[CUDA] Add RMSNormRotary contrib operator #33469

apsonawane merged 9 commits into main from asonawane/main-ops-normalization-rotary

Add exact normalization and rotary CUDA operator modes

63aa76df

Cover exact and packed RMS normalization on standalone main branch

80988834

apsonawane requested a review from

copilot-pull-request-reviewer 15 days ago

copilot-pull-request-reviewer commented on 2026-10-19

Fix normalization determinism and preserve exact rotary fusion semantics

17662fc0

apsonawane requested a review from

copilot-pull-request-reviewer 15 days ago

apsonawane marked this pull request as ready for review 15 days ago

copilot-pull-request-reviewer commented on 2026-10-19

apsonawane enabled auto-merge (squash) 14 days ago

Add fused BF16 RMS normalization and rotary embedding

5c027174

github-actions commented on 2026-10-22

github-advanced-security commented on 2026-10-22

apsonawane added model:qwen

[CUDA] Remove unused RotaryEmbedding(ver 2) from PR #33469 and update…

d68e93b5

Merge remote-tracking branch 'origin/main' into tlwu/pr33469-updates

e32ec3cf

github-actions commented on 2026-10-23

Apply lintrunner formatting

52b3e0d7

[CUDA] Revert unused BranchwiseRMSNorm(ver 2) and unfused RMSNormRota…

12caf709

tianleiwu changed the title ~~Add exact normalization and rotary CUDA operator modes~~ [CUDA] Add RMSNormRotary contrib operator 10 days ago

fix(docs): sync RMSNormRotary schema docs in ContribOperators.md

6c2f985b

jiafatom approved these changes on 2026-10-24

apsonawane merged c7873286 into main 9 days ago

apsonawane deleted the asonawane/main-ops-normalization-rotary branch 9 days ago

Reviewers

jiafatom

github-actions

github-advanced-security

copilot-pull-request-reviewer

Assignees

No one assigned

Labels

model:qwen

Milestone

No milestone

onnxruntime [CUDA] Add RMSNormRotary contrib operator #33469 Merged

[CUDA] Add RMSNormRotary contrib operator #33469

onnxruntime
[CUDA] Add RMSNormRotary contrib operator
#33469

Merged