Skip to content

[Attention] Refactor CUDA attention backend selection logic #24794

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Jump to bottom
Merged
mgoin merged 121 commits into from
Nov 11, 2025
Merged

[Attention] Refactor CUDA attention backend selection logic #24794

mgoin merged 121 commits into from
Nov 11, 2025

Conversation

@MatthewBonanni
Copy link
Contributor

@MatthewBonanni MatthewBonanni commented Sep 13, 2025 •
edited by github-actions bot
Loading

Purpose

CudaPlatformBase.get_attention_backend_cls has gotten complex and messy over time. This PR cleans up the logic (without changing the behavior) and standardizes the interface.

Test Plan

Test Result

Essential Elements of an Effective PR Description Checklist
  • The purpose of the PR, such as "Fix some issue (link existing issues this PR will resolve)".
  • The test plan, such as providing test command.
  • The test results, such as pasting the results comparison before and after, or e2e results
  • (Optional) The necessary documentation update, such as updating supported_models.md and examples for a new model.
  • (Optional) Release notes update. If your change is user facing, please update the release notes draft in the Google Doc.
@mergify mergify bot added rocm Related to AMD ROCm speculative-decoding v1 tpu Related to Google TPUs labels Sep 13, 2025
@mergify
Copy link

mergify bot commented Sep 13, 2025

This pull request has merge conflicts that must be resolved before it can be
merged. Please rebase the PR, @MatthewBonanni.

https://docs.github.com/en/pull-requests/collaborating-with-pull-requests/working-with-forks/syncing-a-fork
@mergify mergify bot added the needs-rebase label Sep 13, 2025
njhill
njhill reviewed Sep 13, 2025
View reviewed changes
@MatthewBonanni MatthewBonanni changed the title [Attention] Refactor CUDA attention backend selection logic [WIP][Attention] Refactor CUDA attention backend selection logic Sep 13, 2025
@MatthewBonanni MatthewBonanni changed the title [WIP][Attention] Refactor CUDA attention backend selection logic [Attention] Refactor CUDA attention backend selection logic Sep 16, 2025
@MatthewBonanni MatthewBonanni marked this pull request as ready for review September 16, 2025 13:22
@mergify mergify bot removed the needs-rebase label Sep 16, 2025
@LucasWilkinson LucasWilkinson mentioned this pull request Nov 10, 2025
Merged
@LucasWilkinson
Copy link
Collaborator

LucasWilkinson commented Nov 10, 2025

@MatthewBonanni how hard would it be to keep backwards compatibility between _Backend and AttentionBackendEnum for a version with a warning?
@LucasWilkinson
Copy link
Collaborator

LucasWilkinson commented Nov 10, 2025

With #26487 potential in the pipe what do we think about having a get_mla_attn_backend_cls instead of is_mla? @Yikun ?
@MatthewBonanni
add _Backend backward compatibility
d0f4698 
Signed-off-by: Matthew Bonanni <mbonanni@redhat.com>
@MatthewBonanni
Copy link
Contributor Author

MatthewBonanni commented Nov 10, 2025

@MatthewBonanni how hard would it be to keep backwards compatibility between _Backend and AttentionBackendEnum for a version with a warning?

@LucasWilkinson done in d0f4698
@NickLucche
Copy link
Collaborator

NickLucche commented Nov 10, 2025

Discussed offline thanks for the work @MatthewBonanni !
wangxiyuan
wangxiyuan approved these changes Nov 11, 2025
View reviewed changes
vllm/attention/backends/registry.py
return AttentionBackendEnum[name]


class _Backend(metaclass=_BackendMeta):
Copy link
Contributor

@wangxiyuan wangxiyuan Nov 11, 2025

Choose a reason for hiding this comment

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

Nice change
@mergify mergify bot added the nvidia label Nov 11, 2025
mgoin
mgoin approved these changes Nov 11, 2025
View reviewed changes
Copy link
Member

@mgoin mgoin left a comment

Choose a reason for hiding this comment

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

Release has been cut, let's go for it on main
@github-project-automation github-project-automation bot moved this to Ready in NVIDIA Nov 11, 2025
@mgoin mgoin merged commit b30dfa0 into vllm-project:main Nov 11, 2025
68 checks passed
@github-project-automation github-project-automation bot moved this from Ready to Done in NVIDIA Nov 11, 2025
@hmellor
Copy link
Member

hmellor commented Nov 11, 2025

The merge commit of this PR failed pre-commit because the base of the branch was out of date
@MatthewBonanni MatthewBonanni deleted the backend_selection_refactor branch November 11, 2025 14:29
@MatthewBonanni MatthewBonanni mentioned this pull request Nov 11, 2025
Merged
5 tasks
@hl475 hl475 mentioned this pull request Nov 12, 2025
Merged
5 tasks
@mgoin mgoin mentioned this pull request Nov 12, 2025
Merged
5 tasks
@njhill njhill mentioned this pull request Nov 12, 2025
Merged
xuebwang-amd pushed a commit to xuebwang-amd/vllm that referenced this pull request Nov 13, 2025
@MatthewBonanni @ProExpertProg @xuebwang-amd
[Attention] Refactor CUDA attention backend selection logic (vllm-pro…
9ac7938 
…ject#24794) Signed-off-by: Matthew Bonanni <mbonanni@redhat.com> Signed-off-by: Matthew Bonanni <mbonanni001@gmail.com> Co-authored-by: Luka Govedič <ProExpertProg@users.noreply.github.com> Signed-off-by: xuebwang-amd <xuebwang@amd.com>
@MatthewBonanni MatthewBonanni mentioned this pull request Nov 13, 2025
Merged
5 tasks
leo-pony added a commit to 22dimensions/vllm-ascend that referenced this pull request Nov 19, 2025
@leo-pony
vllm break of PR:vllm-project/vllm#24794
111f445 
Signed-off-by: leo-pony <nengjunma@outlook.com>
@NickLucche NickLucche mentioned this pull request Nov 20, 2025
Merged
wangxiyuan pushed a commit to wangxiyuan/vllm-ascend that referenced this pull request Nov 24, 2025
@leo-pony @wangxiyuan
vllm break of PR:vllm-project/vllm#24794
1fe271a 
Signed-off-by: leo-pony <nengjunma@outlook.com>
wangxiyuan pushed a commit to wangxiyuan/vllm-ascend that referenced this pull request Nov 24, 2025
@leo-pony @wangxiyuan
vllm break of PR:vllm-project/vllm#24794
58303c8 
Signed-off-by: leo-pony <nengjunma@outlook.com>
devpatelio pushed a commit to SumanthRH/vllm that referenced this pull request Nov 29, 2025
@MatthewBonanni @ProExpertProg @devpatelio
[Attention] Refactor CUDA attention backend selection logic (vllm-pro…
eea77ba 
…ject#24794) Signed-off-by: Matthew Bonanni <mbonanni@redhat.com> Signed-off-by: Matthew Bonanni <mbonanni001@gmail.com> Co-authored-by: Luka Govedič <ProExpertProg@users.noreply.github.com>
charlotte12l pushed a commit to charlotte12l/vllm that referenced this pull request Dec 5, 2025
@MatthewBonanni @ProExpertProg @charlotte12l
[Attention] Refactor CUDA attention backend selection logic (vllm-pro…
e35f6dd 
…ject#24794) Signed-off-by: Matthew Bonanni <mbonanni@redhat.com> Signed-off-by: Matthew Bonanni <mbonanni001@gmail.com> Co-authored-by: Luka Govedič <ProExpertProg@users.noreply.github.com> Signed-off-by: Xingyu Liu <charlotteliu12x@gmail.com>
@MatthewBonanni MatthewBonanni mentioned this pull request Dec 15, 2025
Open
1 task
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

ci/build deepseek Related to DeepSeek models documentation Improvements or additions to documentation frontend gpt-oss Related to GPT-OSS models kv-connector multi-modality Related to multi-modality (#4194) new-model Requests to new models nvidia performance Performance-related issues qwen Related to Qwen models ready ONLY add when PR is ready to merge/full CI is needed rocm Related to AMD ROCm speculative-decoding structured-output tpu Related to Google TPUs v1

9 participants

@MatthewBonanni @ILikeIneine @mgoin @LucasWilkinson @NickLucche @hmellor @wangxiyuan @ProExpertProg @njhill