Fix GPTQ Hessian coordinates after AWQ smoothing. - #2401
Draft
sarahpardo wants to merge 1 commit into
Draft
Conversation
Signed-off-by: Sarah Pardo <spardo@nvidia.com>
Contributor
|
Important Draft PR not reviewedDraft PRs are not automatically reviewed by default.
To automatically review draft PRs, update your CodeRabbit configuration: reviews:
auto_review:
drafts: trueComment |
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
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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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?
Type of change: Bug fix
GPTQ previously bypassed disabled input quantizers while collecting
Hessians. Weight-only AWQ disables input quantization but retains a
pre-quant scale that transforms inputs before the linear operation.
Consequently, an AWQ→GPTQ chain collected the Hessian in the original
input coordinates while GPTQ operated on AWQ-smoothed weights.
This change always passes Hessian inputs through the input quantizer when
one exists. A disabled quantizer without transforms remains an identity,
while AWQ pre-quant scaling is now correctly represented.
Usage
No API changes.
Testing
Added an end-to-end AWQ→GPTQ regression test that verifies the Hessian
matches the transformed input coordinates and differs from the raw-input
Hessian.
Tested with:
pytest tests/gpu/torch/quantization/test_gptq.pyBefore your PR is "Ready for review"
Make sure you read and follow Contributor guidelines and your commits are signed (
git commit -s -S).Make sure you read and follow the Security Best Practices (e.g. avoiding hardcoded
trust_remote_code=True,torch.load(..., weights_only=False),pickle, etc.).CONTRIBUTING.md: N/A