Skip to content

add gptqmodel 6.0 compatible change and MTP quantization skip warning. - #1663

Merged
xin3he merged 8 commits into
mainfrom
xinhe/4-3
Apr 7, 2026
Merged

add gptqmodel 6.0 compatible change and MTP quantization skip warning.#1663
xin3he merged 8 commits into
mainfrom
xinhe/4-3

Conversation

@xin3he

@xin3he xin3he commented Apr 7, 2026

Copy link
Copy Markdown
Contributor

Description

Please briefly describe your main changes, the motivation.

Type of Change

  • Bug fix
  • New feature
  • Documentation update
  • Performance improvement
  • Code refactoring
  • Other (please specify):

Related Issues

Fixes or relates to #

Checklist Before Submitting

  • My code has been tested locally.
  • Documentation has been updated as needed.
  • New or updated tests are included where applicable.

Copilot AI review requested due to automatic review settings April 7, 2026 01:34
@xin3he xin3he added this to the 0.12.1 milestone Apr 7, 2026

Copilot AI left a comment

Copy link
Copy Markdown
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Pull request overview

This PR updates AutoRound’s GPTQ-Marlin integration to be compatible with gptqmodel 6.0 API changes and improves WOQ[RTN] behavior by explicitly skipping unsupported bit-width quantization with a one-time warning.

Changes:

  • Skip WOQ[RTN] quantization for unsupported bits values (only 4/8 supported) with a warning_once.
  • Add a gptqmodel >= 6.0.0 compatibility path for MarlinQuantLinear construction using validate_kwargs and manual attribute initialization.

Reviewed changes

Copilot reviewed 2 out of 2 changed files in this pull request and generated 2 comments.

File Description
auto_round/utils/missing_tensors.py Adds an early skip + warning for unsupported WOQ[RTN] bit-widths when packing missing tensors.
auto_round_extension/cuda/gptqmodel_marlin.py Introduces a gptqmodel 6.0-specific initialization path for Marlin quant linear layers.

Comment thread auto_round/utils/missing_tensors.py Outdated
Comment thread auto_round_extension/cuda/gptqmodel_marlin.py
Comment thread auto_round/utils/missing_tensors.py Outdated
@xin3he

xin3he commented Apr 7, 2026

Copy link
Copy Markdown
Contributor Author

/azp run Unit-Test-CUDA-AutoRound

@azure-pipelines

Copy link
Copy Markdown
Azure Pipelines successfully started running 1 pipeline(s).

xin3he added 2 commits April 7, 2026 11:25
Signed-off-by: Xin He <xin3.he@intel.com>
@xin3he
xin3he requested a review from wenhuach21 April 7, 2026 05:51
@xin3he

xin3he commented Apr 7, 2026

Copy link
Copy Markdown
Contributor Author

/azp run Unit-Test-CUDA-AutoRound

@azure-pipelines

Copy link
Copy Markdown
Azure Pipelines successfully started running 1 pipeline(s).

Comment thread auto_round/inference/backend.py
Comment thread auto_round/utils/missing_tensors.py Outdated
@wenhuach21
wenhuach21 self-requested a review April 7, 2026 06:01
xin3he and others added 3 commits April 7, 2026 15:00
Signed-off-by: Xin He <xin3.he@intel.com>
Signed-off-by: Xin He <xin3.he@intel.com>
@xin3he

xin3he commented Apr 7, 2026

Copy link
Copy Markdown
Contributor Author

/azp run Unit-Test-CUDA-AutoRound

@azure-pipelines

Copy link
Copy Markdown
Azure Pipelines successfully started running 1 pipeline(s).

@xin3he
xin3he merged commit ac7aa28 into main Apr 7, 2026
42 checks passed
@xin3he
xin3he deleted the xinhe/4-3 branch April 7, 2026 13:07
XuehaoSun pushed a commit that referenced this pull request Apr 8, 2026
#1663)

Signed-off-by: Xin He <xin3.he@intel.com>
Co-authored-by: pre-commit-ci[bot] <66853113+pre-commit-ci[bot]@users.noreply.github.com>
@Qubitium

Qubitium commented Apr 9, 2026

Copy link
Copy Markdown

@xin3he Please we have fixed this in GPT-QModel main. v6.1 will have more lifeycle updates which affect other ops but the mtp issues should be fixed.

ModelCloud/GPTQModel#2677

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

4 participants