[OMNIML-2917] export layer config using actual prefix instead of hard…

[shengliangxu]

…coded model.layers

This is change set 1 from working on OMNIML-2917.

When we export quantized model to hf unified format, we hard code modules with a "model.layers" prefix. This is something completely odd and unnecessary. The major problem is that we may output some quant config that have completely wrong prefix, such as in exclude_modules. For example, for the Qwen3-VL models, there are 2 transformer blocks: language_model and vision. Before this change, for language_model, we will output:

model.layers.language_model.layers.0.xxx
model.layers.language_model.layers.1.xxx

The prefixes are completely wrong therefore when inference systems such as vllm try to read the quant config, it will fail.

Fix it by simply use the prefixes from parsing the model itself.

What does this PR do?

Type of change: ?

Overview: ?

Usage

# Add a code snippet demonstrating how to use this

Testing

Before your PR is "Ready for review"

• Make sure you read and follow Contributor guidelines and your commits are signed.
• Is this change backward compatible?: Yes/No
• Did you write any new necessary tests?: Yes/No
• Did you add or update any necessary documentation?: Yes/No
• Did you update Changelog?: Yes/No

Additional Information

[copy-pr-bot]

This pull request requires additional validation before any workflows can run on NVIDIA's runners.

Pull request vetters can view their responsibilities here.

Contributors can view more details about this message here.

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 73.38%. Comparing base (14fa1e5) to head (a254129).
⚠️ Report is 2 commits behind head on main.

Additional details and impacted files

@@           Coverage Diff           @@
##             main     #470   +/-   ##
=======================================
  Coverage   73.38%   73.38%           
=======================================
  Files         180      180           
  Lines       18111    18111           
=======================================
  Hits        13290    13290           
  Misses       4821     4821           

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.

[shengliangxu]

abandon