Skip to content

Fix layer_config bug and CI unstable issue#1949

Merged
xin3he merged 9 commits into
mainfrom
xinhe/6-24
Jul 1, 2026
Merged

Fix layer_config bug and CI unstable issue#1949
xin3he merged 9 commits into
mainfrom
xinhe/6-24

Conversation

@xin3he

@xin3he xin3he commented Jun 24, 2026

Copy link
Copy Markdown
Contributor

Description

This pull request introduces improvements to model layer handling, offloading mechanism, and test helper utilities. The main focus is on enhancing the reliability of layer selection, switching to the safer safetensors format for offloading model weights, and refining test prompt defaults.

Offloading mechanism improvements:

  • Switched offloaded model weight saving/loading from PyTorch .pt files to the safer and faster safetensors format in auto_round/utils/offload.py. This includes updating both saving (_save_to_disk) and loading (_load_from_disk) methods to use safetensors functions and file extensions. [1] [2]

Model layer handling:

  • Fixed handling of predefined ignore layers in auto_round/compressors/base.py to avoid misinterpreting compressed layer names as regex character classes, improving accuracy of layer filtering. Now, the raw (uncompressed) layer names are joined for exact substring matching.

Test and utility enhancements:

  • Updated the default prompt and expected target in eval_generated_prompt within test/helpers.py to use a more realistic question/answer pair ("What is the capital of France?" / "Paris").
  • Simplified logging in update_parameter_data in auto_round/experimental/utils.py by removing the use of layer_idx in the warning message, making logs clearer.

Type of Change

Bug fix

Related Issues

Fixes or relates to #1915

Checklist Before Submitting

  • My code has been tested locally.
  • Documentation has been updated as needed.
  • New or updated tests are included where applicable.
  • The CUDA CI has passed. You can trigger it by commenting /azp run Unit-Test-CUDA-AutoRound.

Signed-off-by: Xin He <xin3.he@intel.com>
@xin3he xin3he requested a review from lvliang-intel June 24, 2026 12:08
@chensuyue

Copy link
Copy Markdown
Contributor

/azp run Unit-Test-CUDA-AutoRound

@azure-pipelines

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

Signed-off-by: Xin He <xin3.he@intel.com>
@chensuyue chensuyue added this to the 0.14.0 milestone Jun 26, 2026
xin3he added 3 commits June 29, 2026 05:02
Signed-off-by: Xin He <xin3.he@intel.com>
Signed-off-by: Xin He <xin3.he@intel.com>

@lkk12014402 lkk12014402 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.

LGTM

Signed-off-by: Xin He <xin3.he@intel.com>
Signed-off-by: Xin He <xin3.he@intel.com>
@chensuyue

Copy link
Copy Markdown
Contributor

/azp run Unit-Test-CUDA-AutoRound

@azure-pipelines

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

Signed-off-by: Xin He <xin3.he@intel.com>
@chensuyue

Copy link
Copy Markdown
Contributor

/azp run Unit-Test-CUDA-AutoRound

@azure-pipelines

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

Signed-off-by: Xin He <xin3.he@intel.com>
@chensuyue

Copy link
Copy Markdown
Contributor

/azp run Unit-Test-CUDA-AutoRound

@azure-pipelines

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

@xin3he xin3he merged commit 05e6795 into main Jul 1, 2026
47 checks passed
@xin3he xin3he deleted the xinhe/6-24 branch July 1, 2026 02:19
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.

3 participants