Report non-finite AutoQuantize sensitivity scores before optimization

[meenchen]

What does this PR do?

Type of change: Bug fix

Reject non-finite AutoQuantize sensitivity scores before score normalization and solver construction. The error identifies the hyperparameter and candidate recipe instead of failing later inside PuLP. Checking before min(score, prev_score) also prevents an infinite score from being hidden by a previous finite score.

Draft: This improves diagnosis of the reported Qwen3.6 AutoQuantize failure. It does not yet fix the upstream numerical operation producing invalid scores; full-model reproduction and validation remain outstanding.

Usage

No API changes. Invalid scoring results now raise a descriptive ValueError.

Testing

• AutoQuantize unit suite: 102 passed; the two distributed tests passed separately outside the filesystem sandbox, which otherwise blocks socket binding (104 total).
• The new test runs calibration and gradient scoring with a non-finite loss. It fails on unmodified main with a downstream solver error and passes with this change.
• All applicable pre-commit hooks passed.
• Full-model GPU E2E has not run: the requested allocation remained pending; alternate clusters had authentication/connectivity failures.

Before your PR is "Ready for review"

• Is this change backward compatible?: Yes; finite-score behavior is unchanged.
• If you copied code from any other sources or added a new PIP dependency, did you follow guidance in CONTRIBUTING.md: N/A.
• Did you write any new necessary tests?: Yes.
• Did you update Changelog?: N/A; diagnostic improvement only at this stage.
• Did you get Claude approval on this PR?: No; draft pending review.

Additional Information

Do not treat this draft as evidence that Qwen3.6 quantization succeeds. The remaining task is to locate the first invalid model output or gradient and validate the numerical fix end to end.

[copy-pr-bot]

Auto-sync is disabled for draft pull requests in this repository. Workflows must be run manually.

Contributors can view more details about this message here.

[coderabbitai]

Important

Draft PR not reviewed

Draft PRs are not automatically reviewed by default.

• Trigger a manual review

To automatically review draft PRs, update your CodeRabbit configuration:

reviews:
  auto_review:
    drafts: true

---

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

[github-actions]

PR Preview Action v1.8.1
🚀 View preview at https://NVIDIA.github.io/Model-Optimizer/pr-preview/pr-2432/
Built to branch gh-pages at 2026-09-14 17:23 UTC. Preview will be ready when the GitHub Pages deployment is complete.

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 71.42%. Comparing base (700e188) to head (09d7e8e).
⚠️ Report is 2 commits behind head on main.

Additional details and impacted files

@@            Coverage Diff             @@
##             main    #2432      +/-   ##
==========================================
+ Coverage   71.41%   71.42%   +0.01%     
==========================================
  Files         590      590              
  Lines       64692    64694       +2     
==========================================
+ Hits        46197    46205       +8     
+ Misses      18495    18489       -6     

Flag	Coverage Δ
unit	57.81% <100.00%> (+0.01%)	⬆️

Flags with carried forward coverage won't be shown. Click here to find out more.

☔ View full report in Codecov by Harness.
📢 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.