Fix MXFP4 quantizer to support variable num_local_experts and hidden_size #41795
+6
−2
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
The quantizer hardcoded 32 experts and 2880 hidden_size in the reshape operations. This caused failures when quantizing models with different numbers of experts (e.g., averaged single-expert models). Changes: - Read num_local_experts and hidden_size from model.config - Use dynamic values in reshape operations instead of hardcoded constants - Defaults to 32 and 2880 for backward compatibility This enables quantizing averaged/merged MoE models with fewer experts. 🤖 Generated with [Claude Code](https://claude.com/claude-code) Co-Authored-By: Claude <noreply@anthropic.com>
|
cc @MekkCyber for quantization |
|
run-slow: mxfp4 |
|
This comment contains run-slow, running the specified jobs: models: [] |
|
[For maintainers] Suggested jobs to run (before merge) run-slow: mxfp4 |
|
run-slow: mxfp4 |
|
This comment contains run-slow, running the specified jobs: models: [] |
Regarding the tests: |
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Outdated suggestions cannot be applied.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.
What does this PR do?
This PR replaces hardcoded values
num_local_expertsandhidden_sizeinMXFP4ConfigforGPT-OSStype models.I discovered this when experimenting with non-standard configs of GPT-OSS architecture but i'm pretty sure it'll break for openai/gpt-oss-120b as well since it's number of experts is different from the hardcoded value.
The quantizer hardcoded 32 experts and 2880 hidden_size in the reshape operations. This caused failures when quantizing models with different numbers of experts.
Changes:
This enables quantizing averaged/merged MoE models with fewer experts.
Passed all tests that I was able to run locally on 24gb of vram.
Before submitting
Pull Request section? - yes
to it if that's the case. - I looked and didn't find an issue
documentation guidelines, and
here are tips on formatting docstrings. - likely not necessary
Who can review?
Anyone in the community is free to review the PR once the tests have passed. Feel free to tag
members/contributors who may be interested in your PR.