Skip to content

Conversation

@shaopeng-666
Copy link
Contributor

@shaopeng-666 shaopeng-666 commented Nov 27, 2025

What this PR does / why we need it?

Qwen2.5-VL mrope precision problem would been solved once this pr is merged

Does this PR introduce any user-facing change?

No

How was this patch tested?

Test on G8600 with textVQA dataset

Signed-off-by: 李少鹏 <[email protected]>
@github-actions
Copy link

👋 Hi! Thank you for contributing to the vLLM Ascend project. The following points will speed up your PR merge:‌‌

  • A PR should do only one thing, smaller PRs enable faster reviews.
  • Every PR should include unit tests and end-to-end tests ‌to ensure it works and is not broken by other future PRs.
  • Write the commit message by fulfilling the PR description to help reviewer and future developers understand.

If CI fails, you can run linting and testing checks locally according Contributing and Testing.

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 addresses a precision issue with the mrope fusion operator on Ascend NPUs. The changes involve removing a workaround that disabled the operator on x86 platforms, indicating the underlying precision problem has been resolved for the specific qwen3vl model configuration. Additionally, an unused import is cleaned up, and a call to .contiguous() is added for the positions tensor before passing it to the npu_mrope operator. This is a good defensive measure to ensure correctness by providing a contiguous tensor as expected by the kernel. The changes are logical and well-contained.

Signed-off-by: 李少鹏 <[email protected]>
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant