[ckpt] fix: minor quantization export issue

[yueshen2016]

What does this PR do ?

Pass the moe_router_dtype from the model config to the quantization export call, fixing MoE model export when a custom router dtype is configured.

Changelog

• Added moe_router_dtype parameter to the export function call in examples/quantization/export.py, reading it from unwrapped_model.config with a fallback default of None.

GitHub Actions CI

See the CI section in the Contributing doc for how to trigger the CI. A Nvidia developer will need to approve and trigger the CI for external contributors.

Before your PR is "Ready for review"

Pre checks:

• Make sure you read and followed Contributor guidelines
• Did you write any new necessary tests?
• Did you add or update any necessary documentation?
• Does the PR affect components that are optional to install? (Ex: Numba, Pynini, Apex etc)
  • Reviewer: Does the PR have correct import guards for all optional libraries?

If you haven't finished some of the above items you can still open "Draft" PR.

Additional Information

• This is a minor one-line fix. Without this change, MoE models that specify a custom moe_router_dtype in their config would lose that setting during quantized export, potentially causing dtype mismatches or incorrect inference results.

Summary by CodeRabbit

• New Features
  • Added support for specifying MoE router data type during model export, automatically sourcing the configuration from the model when available.

[coderabbitai]

📝 Walkthrough

Walkthrough

A new moe_router_dtype parameter is added to the export_mcore_gpt_to_hf function call in the export pipeline. The value is derived from the model configuration's moe_router_dtype attribute if present, otherwise defaults to None.

Changes

Cohort / File(s)	Summary
Export Pipeline examples/quantization/export.py	Added moe_router_dtype keyword argument to export_mcore_gpt_to_hf call, sourced from model config with fallback to None.

Estimated code review effort

🎯 2 (Simple) | ⏱️ ~8 minutes

🚥 Pre-merge checks | ✅ 4

✅ Passed checks (4 passed)

Check name	Status	Explanation
Description Check	✅ Passed	Check skipped - CodeRabbit’s high-level summary is enabled.
Docstring Coverage	✅ Passed	Docstring coverage is 100.00% which is sufficient. The required threshold is 80.00%.
Test Results For Major Changes	✅ Passed	PR is a one-line fix adding moe_router_dtype parameter to export function. Minor change with +1/-0 lines qualifies as minor per check criteria.
Title check	✅ Passed	The title accurately describes the main change: passing moe_router_dtype to the quantization export function, which is a targeted fix for MoE model export issues.

_{✏️ Tip: You can configure your own custom pre-merge checks in the settings.}

✨ Finishing Touches

• 📝 Generate docstrings (stacked PR)
• 📝 Generate docstrings (commit on current branch)

🧪 Generate unit tests (beta)

• Create PR with unit tests
• Post copyable unit tests in a comment
• Commit unit tests in branch yueshen/fix-quantization-export

---

Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out.

❤️ Share

• X
• Mastodon
• Reddit
• LinkedIn

_{Comment @coderabbitai help to get the list of available commands and usage tips.}