Add strict validation for RVC4 quantization overrides

[rolandocortez]

Summary

• Add opt-in rvc4.strict_quantization_overrides validation, disabled by default.
• Reject activation and parameter encoding names that are not present in the effective ONNX model before invoking SNPE.
• Validate overrides from both rvc4.encodings and raw --quantization_overrides arguments.
• Reject ambiguous multiple override sources in strict mode.
• Preserve the existing behavior when strict mode is disabled.

Qualcomm SNPE 2.41 accepts extra activation/parameter entries in quantization overrides without failing, and its --strict option does not apply to quantization override name matching. The validation is therefore implemented on the ModelConverter side.

Strict matching uses subset semantics: missing encodings are allowed; only provided names that do not exist in the model are rejected.

Testing

• 18 passed — tests/unit/test_rvc4_encoding_strictness.py
• 3 passed — tests/conversion/test_rvc4_encodings.py
• Validated with SNPE 2.41.0 / ONNX 1.18.0

Summary by CodeRabbit

• New Features

  • Added optional strict validation for RVC4 quantization overrides.
  • When enabled, checks override names against ONNX activation and parameter names.
  • Rejects unknown or ambiguous overrides before conversion.
  • Strict validation is disabled by default and skipped for FP16 conversions.

• Bug Fixes

  • Prevents invalid quantization overrides from reaching Qualcomm conversion tools.
  • Reports unsupported validation for non-ONNX inputs when strict mode is enabled.

[coderabbitai]

No actionable comments were generated in the recent review. 🎉

ℹ️ Recent review info

⚙️ Run configuration

Configuration used: defaults

Review profile: CHILL

Plan: Pro Plus

Run ID: 672e6045-8ba0-4420-9519-a4a1ffa8cd93

📥 Commits

Reviewing files that changed from the base of the PR and between 9ff0d11 and 0f254ac.

📒 Files selected for processing (1)

• tests/conversion/test_rvc4_encodings.py

Included review availability: Your plan provides up to 1 included review per hour; 0 remain after this review.

---

📝 Walkthrough

Walkthrough

The RVC4 exporter now supports optional strict quantization-override validation. It checks override names against the effective ONNX model and rejects invalid or ambiguous sources before Qualcomm tooling.

Changes

RVC4 quantization validation

Layer / File(s)	Summary
Override validation flow configs/defaults.yaml, modelconverter/utils/config.py, modelconverter/packages/rvc4/exporter.py	Adds the strict_quantization_overrides option. The exporter parses file-based and raw override sources and validates non-FP16 conversions before Qualcomm tooling.
ONNX namespace validation modelconverter/utils/encodings.py, tests/unit/test_rvc4_encoding_strictness.py	Collects activation and parameter names from the top-level ONNX graph without loading external data. Validation reports unknown names by encoding group.
Validation policy coverage tests/unit/test_rvc4_encoding_strictness.py, tests/conversion/test_rvc4_encodings.py	Tests strict and non-strict modes, FP16 behavior, non-ONNX inputs, empty names, source ambiguity, sorted errors, and validation against the final effective model.

Estimated code review effort: 3 (Moderate) | ~25 minutes

Merge Risk: ⚪ Minimal · up to 0f254

The PR adds opt-in validation for RVC4 quantization overrides while preserving existing behavior by default; no actionable merge-blocking risk remains after normal checks and review.

Suggested reviewers: klemen1999, kozlov721, n950

Sequence Diagram(s)

sequenceDiagram
  participant User
  participant RVC4Exporter
  participant EncodingUtilities
  participant EffectiveONNXModel
  participant QualcommTooling
  User->>RVC4Exporter: configure quantization overrides
  RVC4Exporter->>EncodingUtilities: parse override source
  EncodingUtilities->>EffectiveONNXModel: collect activation and parameter names
  EffectiveONNXModel-->>RVC4Exporter: return model namespaces
  RVC4Exporter->>RVC4Exporter: validate override names
  RVC4Exporter->>QualcommTooling: invoke conversion if validation passes

Loading

🚥 Pre-merge checks | ✅ 4 | ❌ 1

❌ Failed checks (1 warning)

Check name	Status	Explanation	Resolution
Docstring Coverage	⚠️ Warning	Docstring coverage is 5.00% which is insufficient. The required threshold is 80.00%. Docstring coverage is scoped to functions touched by this diff. Analyzed 40 functions across 5 files.	Write docstrings for the functions missing them to satisfy the coverage threshold.

✅ Passed checks (4 passed)

Check name	Status	Explanation
Description Check	✅ Passed	Check skipped - CodeRabbit’s high-level summary is enabled.
Title check	✅ Passed	The title clearly and concisely describes the main change: adding opt-in strict validation for RVC4 quantization overrides.
Linked Issues check	✅ Passed	Check skipped because no linked issues were found for this pull request.
Out of Scope Changes check	✅ Passed	Check skipped because no linked issues were found for this pull request.

• Fix all pre-merge checks with AI

✨ Finishing Touches 💡 1

📝 Generate docstrings 💡

• Create stacked PR
• Commit on current branch

🧪 Generate unit tests (beta)

• Create PR with unit tests
• Commit unit tests in branch feat/rvc4-encoding-strictness

---

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.}

[codecov-commenter]

Codecov Report

❌ Patch coverage is 96.85039% with 8 lines in your changes missing coverage. Please review.
✅ Project coverage is 99.04%. Comparing base (23d0e9f) to head (0f254ac).
⚠️ Report is 1 commits behind head on main.
✅ All tests successful. No failed tests found.

Files with missing lines	Patch %	Lines
modelconverter/packages/rvc4/exporter.py	89.09%	6 Missing ⚠️
tests/conversion/test_rvc4_encodings.py	93.54%	2 Missing ⚠️

Additional details and impacted files

@@             Coverage Diff             @@
##             main     #288       +/-   ##
===========================================
+ Coverage   79.09%   99.04%   +19.94%     
===========================================
  Files          84       85        +1     
  Lines        6875     7126      +251     
===========================================
+ Hits         5438     7058     +1620     
+ Misses       1437       68     -1369     

Flag	Coverage Δ
conversion	44.33% <38.18%> (?)
unit	79.56% <83.07%> (+0.46%)	⬆️

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.