Enable gpt-oss mxfp4 on older hardware (sm75+)#39940
Merged
Conversation
|
The docs for this PR live here. All of your documentation changes will be reflected on that endpoint. The docs are available until 30 days after the last update. |
MekkCyber
approved these changes
Aug 6, 2025
Contributor
MekkCyber
left a comment
MekkCyber
left a comment
There was a problem hiding this comment.
Sounds Good ! Thanks for lowering the hardware requirements 🤗
SunMarc
approved these changes
Aug 6, 2025
Member
SunMarc
left a comment
SunMarc
left a comment
There was a problem hiding this comment.
Nice ! Thanks for testing this
Contributor
|
[For maintainers] Suggested jobs to run (before merge) run-slow: mxfp4 |
8 tasks
SunMarc
added
the
for patch
1 task
Closed
4 tasks
ArthurZucker
pushed a commit
that referenced
this pull request
Aug 13, 2025
Co-authored-by: Marc Sun <57196510+SunMarc@users.noreply.github.com>
1 task
8 tasks
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
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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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.
What does this PR do?
Currently the MXFP4 quantized version of gpt-oss models is restricted to newer GPUs (Hopper and Blackwell). This PR enables the MXFP4 version on Turing, Ampere, and Ada GPUs.
Tested with the gpt-oss-20b on RTX 4090 and T4.
If the user has the kernels installed but the hardware is too old, we'll fall back to dequantizing instead of raising an error.
There's additionally a fix included here for a device mismatch issue that occurred when running with
device_map="auto"on 2x4090.Before submitting
Pull Request section?
to it if that's the case.
documentation guidelines, and
here are tips on formatting docstrings.
Who can review?
@SunMarc @MekkCyber