[Mamba] - refactor: Renamed mamba_attn to mamba2_attn #22818
Conversation
Signed-off-by: asafg <[email protected]>
Josephasafg
requested review from
WoosukKwon,
alexm-redhat,
comaniac,
njhill,
robertgshaw2-redhat,
tlrmchlsmth,
yewentao256 and
ywang96
as code owners
|
👋 Hi! Thank you for contributing to the vLLM project. 💬 Join our developer Slack at https://slack.vllm.ai to discuss your PR in #pr-reviews, coordinate on features in #feat- channels, or join special interest groups in #sig- channels. Just a reminder: PRs would not trigger full CI run by default. Instead, it would only run Once the PR is approved and ready to go, your PR reviewer(s) can run CI to test the changes comprehensively before merging. To run CI, PR reviewers can either: Add 🚀 |
|
@tdoublep @heheda12345 Can you please review? Thanks |
![gemini-code-assist[bot]](https://avatars.githubusercontent.com/in/956858?s=60&v=4){kind=small}
There was a problem hiding this comment.
Code Review
This pull request is a refactoring that renames vllm/v1/attention/backends/mamba_attn.py to vllm/v1/attention/backends/mamba2_attn.py. This change improves clarity by explicitly distinguishing the Mamba v2 attention backend from the Mamba v1 implementation, which is a welcome improvement for code maintainability. All corresponding import paths have been updated correctly across the codebase. The changes are consistent and well-executed. I find no issues of high or critical severity.
yewentao256
left a comment
yewentao256
left a comment
There was a problem hiding this comment.
Looks good to me, thanks for the work!
|
@yewentao256 I think CI is failing unrelated to my changes. Can you please help with force push maybe? |
|
The CI is in a very bad status. Therefore, the main branch is frozen and only accepts force merge related to CI fix now. |
|
@heheda12345 can this be merged now? I believe the failing test is unrelated |
|
Can you rebase with main branch again? The tree eagle failure is fixed on main. |
|
@heheda12345 Updated, thanks |
…2818) Signed-off-by: asafg <[email protected]> Co-authored-by: asafg <[email protected]>
…2818) Signed-off-by: asafg <[email protected]> Co-authored-by: asafg <[email protected]>
…2818) Signed-off-by: asafg <[email protected]> Co-authored-by: asafg <[email protected]> Signed-off-by: Duncan Moss <[email protected]>
…2818) Signed-off-by: asafg <[email protected]> Co-authored-by: asafg <[email protected]>
…2818) Signed-off-by: asafg <[email protected]> Co-authored-by: asafg <[email protected]> Signed-off-by: Xiao Yu <[email protected]>
…2818) Signed-off-by: asafg <[email protected]> Co-authored-by: asafg <[email protected]>
5 participants
Essential Elements of an Effective PR Description Checklist
supported_models.mdandexamplesfor a new model.Purpose
With the work on V1 on both mamba models (1 and 2) we need to have a distinction in the file names to better understand which is which. In the past mamba1 wasn't available in V1, now that it is,
mamba_attn.pywhich was for mamba2 should be renamed tomamba2_attn.pyfor clarity and distinction.