[ananyaa-ms]

Describe your changes

• Register the onnxruntime-ep-openvino plugin for ONNX Runtime GenAI vision evaluation when the NPU target selects OpenVINOExecutionProvider, matching the existing QNN registration path.
• Report missing-image and inference errors per sample instead of silently scoring blank predictions.
• Add unit coverage for OpenVINO registration (single and split vision configurations), CPU behavior, and sample errors.

User-facing change: OpenVINO NPU GenAI vision models can be evaluated through Olive; individual sample failures now stop evaluation with a descriptive error.

Validation:

• C:\Users\aamancherla\repos\olive-recipes\google-gemma-4-E2B-it\multi_comp\.venv\Scripts\python.exe -m pytest -q test\evaluator\test_olive_evaluator.py -k genai_vision --tb=short — passed (8 tests).
• C:\Users\aamancherla\repos\olive-recipes\google-gemma-4-E2B-it\multi_comp\.venv\Scripts\python.exe -m ruff check --ignore CPY001 olive\evaluator\olive_evaluator.py test\evaluator\test_olive_evaluator.py — passed.
• C:\Users\aamancherla\repos\olive-recipes\google-gemma-4-E2B-it\multi_comp\.venv\Scripts\python.exe -m ruff format --check olive\evaluator\olive_evaluator.py test\evaluator\test_olive_evaluator.py — passed (2 files).
• Full test suite and lintrunner -a — not run locally.

Checklist before requesting a review

• Add unit tests for this change.
• Make sure all tests can pass.
• Update documents if necessary.
• Lint and apply fixes to your code by running lintrunner -a
• Is this a user-facing change? If yes, give a description of this change to be included in the release notes.

(Optional) Issue link

[Copilot]

🟢 Approval recommended

The focused changes have corresponding unit coverage, and no blocking issues were identified.

No open issues 🎉

What changed in this PR

Adds OpenVINO plugin registration to Olive’s GenAI vision evaluation for NPU targets and makes sample failures explicit.

Changes:

• Registers OpenVINO once when selected for NPU evaluation.
• Raises descriptive errors for missing images and inference failures.
• Adds tests for single/split vision configurations, CPU behavior, and sample errors.

File	Description
test/​evaluator/​test_olive_evaluator.py	Tests OpenVINO registration and sample failure handling.
olive/​evaluator/​olive_evaluator.py	Adds OpenVINO registration and replaces blank predictions on failure with errors.

_{Review effort: Balanced}