[Kernel] Support decode context parallelism on Blackwell with CUTLASS MLA #24385
+63
−32
Conversation
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
Signed-off-by: Ming Yang <minos.future@gmail.com>
minosfuture
requested review from
WoosukKwon,
alexm-redhat,
comaniac,
njhill,
robertgshaw2-redhat and
ywang96
as code owners
![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 adds support for returning log-sum-exp (LSE) values from the CUTLASS MLA decode kernel, which is a key requirement for enabling decode context parallelism on GB200. The changes are well-contained and correctly plumb the new lse tensor through the C++ kernel, PyTorch bindings, and the Python attention backend. I have identified one critical issue related to tensor shape consistency that needs to be addressed.
youkaichao
reviewed
Sep 7, 2025
youkaichao
changed the title
Co-authored-by: youkaichao <youkaichao@gmail.com> Signed-off-by: Ming Yang <minos.future@gmail.com>
youkaichao
reviewed
Sep 7, 2025
Co-authored-by: youkaichao <youkaichao@gmail.com> Signed-off-by: Ming Yang <minos.future@gmail.com>
youkaichao
approved these changes
Sep 7, 2025
github-actions
bot
added
the
ready
Signed-off-by: youkaichao <youkaichao@gmail.com>
Signed-off-by: youkaichao <youkaichao@gmail.com>
Signed-off-by: youkaichao <youkaichao@gmail.com>
LucasWilkinson
approved these changes
Sep 7, 2025
eicherseiji
pushed a commit
to eicherseiji/vllm
that referenced
this pull request
Sep 9, 2025
… MLA (vllm-project#24385) Signed-off-by: Ming Yang <minos.future@gmail.com> Signed-off-by: youkaichao <youkaichao@gmail.com> Co-authored-by: youkaichao <youkaichao@gmail.com>
skyloevil
pushed a commit
to skyloevil/vllm
that referenced
this pull request
Sep 13, 2025
… MLA (vllm-project#24385) Signed-off-by: Ming Yang <minos.future@gmail.com> Signed-off-by: youkaichao <youkaichao@gmail.com> Co-authored-by: youkaichao <youkaichao@gmail.com>
FeiDaLI
pushed a commit
to FeiDaLI/vllm
that referenced
this pull request
Sep 25, 2025
… MLA (vllm-project#24385) Signed-off-by: Ming Yang <minos.future@gmail.com> Signed-off-by: youkaichao <youkaichao@gmail.com> Co-authored-by: youkaichao <youkaichao@gmail.com>
xuebwang-amd
pushed a commit
to xuebwang-amd/vllm
that referenced
this pull request
Oct 10, 2025
… MLA (vllm-project#24385) Signed-off-by: Ming Yang <minos.future@gmail.com> Signed-off-by: youkaichao <youkaichao@gmail.com> Co-authored-by: youkaichao <youkaichao@gmail.com> Signed-off-by: xuebwang-amd <xuebwang@amd.com>
xuebwang-amd
pushed a commit
to xuebwang-amd/vllm
that referenced
this pull request
Oct 24, 2025
… MLA (vllm-project#24385) Signed-off-by: Ming Yang <minos.future@gmail.com> Signed-off-by: youkaichao <youkaichao@gmail.com> Co-authored-by: youkaichao <youkaichao@gmail.com> Signed-off-by: xuebwang-amd <xuebwang@amd.com>
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.
Purpose
This PR supports decode context parallelism with CUTLASS MLA kernels on GB200
credits to #22789 from @LucasWilkinson, and @youkaichao
Test Plan
pytest -v -s tests/distributed/test_context_parallel.py
note: on GB200, needs to modify this UT to use only two GPUs. This can be added later or in a follow-up PR.
Test Result
both
-tp 2 -dcp 2and-tp 2workEssential Elements of an Effective PR Description Checklist
supported_models.mdandexamplesfor a new model.