Skip to content

fix auto_scheme gguf format wrong bits if use tiny model#1953

Merged
XuehaoSun merged 8 commits into
mainfrom
hengguo/gguf_fix_625
Jun 29, 2026
Merged

fix auto_scheme gguf format wrong bits if use tiny model#1953
XuehaoSun merged 8 commits into
mainfrom
hengguo/gguf_fix_625

Conversation

@n1ck-guo

Copy link
Copy Markdown
Contributor

Description

Please briefly describe your main changes, the motivation.

Type of Change

Bug fix

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.
  • The CUDA CI has passed. You can trigger it by commenting /azp run Unit-Test-CUDA-AutoRound.

Signed-off-by: n1ck-guo <heng.guo@intel.com>

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 fixes AutoScheme’s bit-budget accounting for tiny models when using GGUF-related options by ensuring embedding layers remain eligible for scheme assignment (even after they’re removed from the DP scoring layer list) and by applying the lm_head “head trick” after embedding-layer budget has been deducted.

Changes:

  • Fix embedding eligibility computation so embedding layers are assigned an AutoScheme option even though they’re removed from quant_layer_names for DP scoring.
  • Move the lm_head option restriction (_apply_head_trick) to run after embedding-layer scheme selection so feasibility/budget checks use the true remaining budget.
  • Refactor embedding eligibility into a helper and update lm_head rule comments.

Comment thread auto_round/auto_scheme/delta_loss.py Outdated
Comment thread auto_round/auto_scheme/delta_loss.py Outdated
@n1ck-guo n1ck-guo requested a review from wenhuach21 June 25, 2026 06:25
Signed-off-by: n1ck-guo <heng.guo@intel.com>
Comment thread auto_round/auto_scheme/delta_loss.py Outdated
@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).

@n1ck-guo n1ck-guo requested a review from wenhuach21 June 25, 2026 08:30
Signed-off-by: n1ck-guo <heng.guo@intel.com>
Comment thread auto_round/auto_scheme/delta_loss.py Outdated
@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: n1ck-guo <heng.guo@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).

@n1ck-guo n1ck-guo requested a review from wenhuach21 June 26, 2026 08:02
@chensuyue chensuyue added this to the 0.14.0 milestone Jun 26, 2026
n1ck-guo added 2 commits June 26, 2026 16:48
Signed-off-by: n1ck-guo <heng.guo@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: n1ck-guo <heng.guo@intel.com>
@n1ck-guo

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).

@XuehaoSun XuehaoSun merged commit b1a6baf into main Jun 29, 2026
45 of 47 checks passed
@XuehaoSun XuehaoSun deleted the hengguo/gguf_fix_625 branch June 29, 2026 06:01
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.

6 participants