[Bugfix] skip cuda graph for drafter when running with eager #26821
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: Benjamin Chislett <bchislett@nvidia.com>
benchislett
requested review from
WoosukKwon and
robertgshaw2-redhat
as code owners
benchislett
requested review from
alexm-redhat,
comaniac,
njhill 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 correctly fixes a bug where speculative_config.enforce_eager was not being respected during the dummy run for the draft model. The change adds the necessary condition to disable CUDA graphs for the drafter's dummy run when eager mode is enforced. The implementation is correct and addresses the issue effectively. I have no further comments.
pavanimajety
approved these changes
Oct 14, 2025
|
disabling auto-merge, forgot I still need a stamp |
This was referenced Oct 21, 2025
Open
baonudesifeizhai
pushed a commit
to baonudesifeizhai/vllm
that referenced
this pull request
Oct 21, 2025
…oject#26821) Signed-off-by: Benjamin Chislett <bchislett@nvidia.com>
sstamenk
pushed a commit
to sstamenk/vllm
that referenced
this pull request
Oct 23, 2025
…oject#26821) Signed-off-by: Benjamin Chislett <bchislett@nvidia.com> Signed-off-by: sstamenk <strahinja.stamenkovic@amd.com>
usberkeley
pushed a commit
to usberkeley/vllm
that referenced
this pull request
Oct 23, 2025
…oject#26821) Signed-off-by: Benjamin Chislett <bchislett@nvidia.com>
albertoperdomo2
pushed a commit
to albertoperdomo2/vllm
that referenced
this pull request
Oct 23, 2025
…oject#26821) Signed-off-by: Benjamin Chislett <bchislett@nvidia.com> Signed-off-by: Alberto Perdomo <aperdomo@redhat.com>
0xrushi
pushed a commit
to 0xrushi/vllm
that referenced
this pull request
Oct 26, 2025
…oject#26821) Signed-off-by: Benjamin Chislett <bchislett@nvidia.com> Signed-off-by: 0xrushi <6279035+0xrushi@users.noreply.github.com>
0xrushi
pushed a commit
to 0xrushi/vllm
that referenced
this pull request
Oct 26, 2025
…oject#26821) Signed-off-by: Benjamin Chislett <bchislett@nvidia.com> Signed-off-by: 0xrushi <6279035+0xrushi@users.noreply.github.com>
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
2 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
Patch for
speculative_config.enforce_eagernot being respected in the draft model dummy run.This flag was introduced for DSV3.2 MTP which currently must run in eager mode, so it is not a common crash.