Bump flashinfer to 0.4.0rc2 to support determinism #26306
Conversation
![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 bumps the version of flashinfer-python to 0.4.0rc2. My review highlights the risk of using a release candidate version for a dependency, which can affect stability. I also point out that the Dockerfile's flashinfer version has not been updated in sync with this change, as suggested by a code comment, which could lead to inconsistent behavior across different installation methods.
![chatgpt-codex-connector[bot]](https://avatars.githubusercontent.com/in/1144995?s=60&v=4){kind=small}
There was a problem hiding this comment.
💡 Codex Review
Here are some automated review suggestions for this pull request.
ℹ️ About Codex in GitHub
Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you
- Open a pull request for review
- Mark a draft as ready
- Comment "@codex review".
If Codex has suggestions, it will comment; otherwise it will react with 👍.
| ], # Required for audio processing | ||
| "video": [], # Kept for backwards compatibility | ||
| # FlashInfer should be updated together with the Dockerfile | ||
| "flashinfer": ["flashinfer-python==0.3.1"], | ||
| "flashinfer": ["flashinfer-python==0.4.0rc2"], | ||
| # Optional deps for AMD FP4 quantization support |
There was a problem hiding this comment.
The change bumps the flashinfer extra to 0.4.0rc2, but the repo’s Dockerfile still installs flashinfer_python-0.3.1. The comment above this stanza explicitly says FlashInfer should be updated together with the Dockerfile, so Docker builds will continue to use the old wheel and miss the determinism fix this commit is targeting. Consider updating the pinned version in docker/Dockerfile to avoid version skew between local installs and the official image.
Useful? React with 👍 / 👎.
|
Don't worry about DCO, we can force it to pass |
mgoin
changed the title
Signed-off-by: Bram Wasti <[email protected]>
bwasti
force-pushed
the
flashinfer_bump
branch
from
6783809 to
7fa856c
Compare
|
@bwasti you forced pushed my dockerfile change out... so the dockerfile is still running with 0.3.1 |
|
sorry about the confusion here, I thought this was confirmed dead. (didn't realize I had overwritten your changes) will close due to this effort: #26326 |
Previous attempt to land #25769 caused CI failures because the version for flashinfer was not bumped.
Purpose
This PR tests and bumps a release candidate version of flashinfer.
Test Plan
Run CI
Test Result
[unknown]
Essential Elements of an Effective PR Description Checklist
supported_models.mdandexamplesfor a new model.