[torch.compile] Unwrap fused_marlin_moe custom op #26739
Merged
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: Varun Sundar Rabindranath <[email protected]>
varun-sundar-rabindranath
requested review from
WoosukKwon,
mgoin,
robertgshaw2-redhat,
tlrmchlsmth and
yewentao256
as code owners
Contributor
Author
|
cc @youkaichao @ProExpertProg @bnellnm PTAL. Thanks 🙌 |
![gemini-code-assist[bot]](https://avatars.githubusercontent.com/in/956858?s=60&v=4){kind=small}
Contributor
There was a problem hiding this comment.
Code Review
This pull request correctly refactors the fused_marlin_moe function by removing its torch.ops custom operator wrapper. The changes are consistent across all modified files, replacing the custom op calls with direct function calls and removing the associated registration code. This is a clean and well-executed refactoring that achieves its goal of making the function more amenable to torch.compile by exposing its internal logic. The changes are sound and I found no issues.
bnellnm
approved these changes
Oct 13, 2025
mgoin
added
quantization
moe
ready
jeejeelee
approved these changes
Oct 14, 2025
1994
pushed a commit
to 1994/vllm
that referenced
this pull request
Oct 14, 2025
Signed-off-by: Varun Sundar Rabindranath <[email protected]> Co-authored-by: Varun Sundar Rabindranath <[email protected]> Signed-off-by: 1994 <[email protected]>
Dhruvilbhatt
pushed a commit
to Dhruvilbhatt/vllm
that referenced
this pull request
Oct 14, 2025
Signed-off-by: Varun Sundar Rabindranath <[email protected]> Co-authored-by: Varun Sundar Rabindranath <[email protected]> Signed-off-by: Dhruvil Bhatt <[email protected]>
bbartels
pushed a commit
to bbartels/vllm
that referenced
this pull request
Oct 16, 2025
Signed-off-by: Varun Sundar Rabindranath <[email protected]> Co-authored-by: Varun Sundar Rabindranath <[email protected]> Signed-off-by: bbartels <[email protected]>
lywa1998
pushed a commit
to lywa1998/vllm
that referenced
this pull request
Oct 20, 2025
Signed-off-by: Varun Sundar Rabindranath <[email protected]> Co-authored-by: Varun Sundar Rabindranath <[email protected]>
alhridoy
pushed a commit
to alhridoy/vllm
that referenced
this pull request
Oct 24, 2025
Signed-off-by: Varun Sundar Rabindranath <[email protected]> Co-authored-by: Varun Sundar Rabindranath <[email protected]>
xuebwang-amd
pushed a commit
to xuebwang-amd/vllm
that referenced
this pull request
Oct 24, 2025
Signed-off-by: Varun Sundar Rabindranath <[email protected]> Co-authored-by: Varun Sundar Rabindranath <[email protected]> Signed-off-by: xuebwang-amd <[email protected]>
xuebwang-amd
pushed a commit
to xuebwang-amd/vllm
that referenced
this pull request
Oct 24, 2025
Signed-off-by: Varun Sundar Rabindranath <[email protected]> Co-authored-by: Varun Sundar Rabindranath <[email protected]> Signed-off-by: xuebwang-amd <[email protected]>
0xrushi
pushed a commit
to 0xrushi/vllm
that referenced
this pull request
Oct 26, 2025
Signed-off-by: Varun Sundar Rabindranath <[email protected]> Co-authored-by: Varun Sundar Rabindranath <[email protected]> Signed-off-by: 0xrushi <[email protected]>
0xrushi
pushed a commit
to 0xrushi/vllm
that referenced
this pull request
Oct 26, 2025
Signed-off-by: Varun Sundar Rabindranath <[email protected]> Co-authored-by: Varun Sundar Rabindranath <[email protected]> Signed-off-by: 0xrushi <[email protected]>
rtourgeman
pushed a commit
to rtourgeman/vllm
that referenced
this pull request
Nov 10, 2025
Signed-off-by: Varun Sundar Rabindranath <[email protected]> Co-authored-by: Varun Sundar Rabindranath <[email protected]>
Zhathw
pushed a commit
to Zhathw/vllm
that referenced
this pull request
Nov 12, 2025
Signed-off-by: Varun Sundar Rabindranath <[email protected]> Co-authored-by: Varun Sundar Rabindranath <[email protected]>
devpatelio
pushed a commit
to SumanthRH/vllm
that referenced
this pull request
Nov 29, 2025
Signed-off-by: Varun Sundar Rabindranath <[email protected]> Co-authored-by: Varun Sundar Rabindranath <[email protected]>
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
4 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
Unwrap unnecessary CustomOP'ing of
fused_marlin_moe.fused_marlin_moewas first wrapped into a custom op in the PR #9632 asfused_marlin_moewas doing file reads to get optimal triton config. This function has evolved since and no longer needs file reads and could be unwrapped.Test Plan
CI
Test Result