Specify attention-23 kernel and relax assertion in prepare qkv#27217
Merged
titaiwangms merged 1 commit intomainfrom Feb 3, 2026
Merged
Specify attention-23 kernel and relax assertion in prepare qkv#27217titaiwangms merged 1 commit intomainfrom
titaiwangms merged 1 commit intomainfrom
Conversation
Contributor
There was a problem hiding this comment.
Pull request overview
This PR improves the attention kernel selection logic for the CUDA Attention operator by explicitly setting kernel types and removing outdated code. The changes enhance code clarity and maintainability while enabling proper support for causal (unidirectional) attention with the unfused kernel path.
Changes:
- Explicitly set
kernel_typetoAttentionKernel_Unfusedin the Attention operator to clarify kernel selection - Remove outdated TODO comments about parameter handling and kernel selection that are now properly implemented
- Relax the assertion in
PrepareQkv_MHA_NoPastthat incorrectly prevented causal attention, replacing it with a clarifying comment about which kernels support unidirectional attention
Reviewed changes
Copilot reviewed 2 out of 2 changed files in this pull request and generated no comments.
| File | Description |
|---|---|
| onnxruntime/core/providers/cuda/llm/attention.cc | Removes obsolete TODOs and explicitly sets kernel_type to AttentionKernel_Unfused, making kernel selection more explicit and maintainable |
| onnxruntime/contrib_ops/cuda/bert/attention_prepare_qkv.cu | Removes overly restrictive assertion blocking causal attention and adds explanatory comment about kernel support for is_unidirectional flag |
💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.
tianleiwu
approved these changes
Jan 31, 2026
xadupre
approved these changes
Feb 2, 2026
Contributor
Author
|
Windows GPU CI error will be fixed in #27206 |
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
3 participants
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
This pull request updates the attention kernel selection logic and clarifies support for unidirectional (causal) attention in the CUDA attention implementation. The main changes focus on improving documentation, removing outdated comments, and explicitly setting the kernel type for better maintainability and clarity.
Kernel selection and configuration improvements:
kernel_typefield toAttentionKernel_Unfusedin theAttentionDatastructure to clarify which kernel is being used and improve future extensibility.Documentation and code clarity: