fix bug of quantizing Z-image

[xin3he]

Description

1. save config in FrozenDict type (ZImageTransformer2DModel.config)
2. stop evaluation if no args.tasks
3. skip .state_dict calling if module is an parameter object. (x_pad_token, cap_pad_token)

Type of Change

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

Related Issues

Fixes or relates to #1507

Checklist Before Submitting

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

[xin3he]

Raw model output:

W4A16 output (auto-round Tongyi-MAI/Z-Image-Turbo --iters 0 --disable_opt_rtn --format fake):

[xin3he]

revert save_config change, and use #1519 insteadly.

[Copilot]

Pull request overview

This PR targets export/evaluation correctness for quantized models (including diffusion/Z-Image related flows), adjusting how export artifacts are written and when evaluation is executed.

Changes:

• Introduces new export utilities (model saving, AutoGPTQ QuantLinear selection, quantization-config filtering, layer release helper).
• Adjusts evaluation flow to skip evaluation when no args.tasks are provided.
• Hardens shard-writer offload logic to avoid calling .state_dict() on non-Module objects.

Reviewed changes

Copilot reviewed 2 out of 3 changed files in this pull request and generated 1 comment.

File	Description
auto_round/export/utils.py	Adds shared export helpers for saving models and cleaning quantization configs.
auto_round/eval/evaluation.py	Reorders early-return logic around evaluation dispatch.
auto_round/compressors/shard_writer.py	Adds a type-check to avoid .state_dict() on non-modules during meta offload.

---

You can also share your feedback on Copilot code review. Take the survey.