Skip to content

Comments

[ROCm] Check that AITER MHA is not selected with sinks#34998

Open
gshtras wants to merge 2 commits intovllm-project:mainfrom
ROCm:check_sinks_aiter_mha
Open

[ROCm] Check that AITER MHA is not selected with sinks#34998
gshtras wants to merge 2 commits intovllm-project:mainfrom
ROCm:check_sinks_aiter_mha

Conversation

@gshtras
Copy link
Collaborator

@gshtras gshtras commented Feb 20, 2026

AITER MHA does not support sinks. Gating its selector
This should unbreak VLLM_ROCM_USE_AITER=1 vllm serve openai/gpt-oss-120b

Signed-off-by: Gregory Shtrasberg <Gregory.Shtrasberg@amd.com>
Signed-off-by: Gregory Shtrasberg <Gregory.Shtrasberg@amd.com>
@gshtras gshtras requested a review from tjtanaa as a code owner February 20, 2026 23:05
@dosubot
Copy link

dosubot bot commented Feb 20, 2026

Related Documentation

Checked 0 published document(s) in 1 knowledge base(s). No updates required.

How did I do? Any feedback?  Join Discord

@mergify mergify bot added the rocm Related to AMD ROCm label Feb 20, 2026
@github-project-automation github-project-automation bot moved this to Todo in AMD Feb 20, 2026
Copy link
Contributor

@gemini-code-assist gemini-code-assist bot left a comment

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Code Review

This pull request correctly adds checks to prevent the selection of the AITER MHA attention backend when sinks are enabled, as this configuration is unsupported. The changes cover both explicit backend selection and automatic selection logic. My review includes one comment pointing out a significant code duplication where a block of logic is redundant and unreachable. I've suggested its removal to improve code clarity and maintainability.

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

rocm Related to AMD ROCm

Projects

Status: Todo

Development

Successfully merging this pull request may close these issues.

1 participant