feat(text-metrics): split text_score pair

[davidberenstein1957]

Summary

Splits the text_score pair into a dedicated stacked PR, adds OCR-based text metrics and shared utilities, and wires Long Text Bench + OneIG Text Rendering.

Stack Position

• Base: PR feat(text-metrics): split oneig_alignment #646 (feat/vlm-pr-3b-oneig-alignment)
• Next: PR feat(text-metrics): split oneig_reasoning #648 (feat/vlm-pr-3d-oneig-reasoning)
• Final integration: PR feat(e2e-tests): stacked e2e after split metrics #641 (feat/vlm-pr-5-e2e-tests)
• Canonical umbrella reference: PR feat(evaluation): add VLMMetrics #545 (feat/metrics-vlm-support)

Files

• src/pruna/evaluation/metrics/metric_text_score.py
• src/pruna/evaluation/metrics/metric_text_score_utils.py
• src/pruna/evaluation/benchmarks.py

Test Plan

uv run pytest tests/evaluation/test_text_metrics.py -k text_score

Review Focus

• OCR scoring behavior
• Long Text Bench and OneIG Text Rendering wiring

Review Flow (Order)

Review the stack in this exact order:

1. feat(vendor): add LLM2Vec embedding model #637 vendor
2. feat(infrastructure): add VLM base classes and utilities #638 infrastructure
3. feat(text-metrics): split qa_accuracy #645 qa_accuracy
4. feat(text-metrics): split oneig_alignment #646 oneig_alignment
5. feat(text-metrics): split text_score pair #647 text_score pair
6. feat(text-metrics): split oneig_reasoning #648 oneig_reasoning
7. feat(vision-metrics): split vqa #649 vqa
8. feat(vision-metrics): split vie_score #650 vie_score
9. feat(vision-metrics): split img_edit_score #651 img_edit_score
10. feat(e2e-tests): stacked e2e after split metrics #641 e2e tests

This PR in the flow (5/10)

• Review after PR feat(text-metrics): split oneig_alignment #646.
• Next PR to review: feat(text-metrics): split oneig_reasoning #648.
• Confirm this PR's tests and scope before continuing.

[cursor]

Cursor Bugbot has reviewed your changes and found 1 potential issue.

^{❌ Bugbot Autofix is OFF. To automatically fix reported issues with cloud agents, enable autofix in the Cursor dashboard.}

Comment @cursor review or bugbot run to trigger another review on this PR

^{Reviewed by Cursor Bugbot for commit 946b068. Configure here.}

[cursor]

Hallucination keyword replacement order creates dead code

Low Severity

In clean_oneig_ocr_hallucinations, the chained .replace() calls have the wrong order. The first call out.replace(keyword, "") removes all occurrences of keyword, so the subsequent .replace(f"\n{keyword}", "") and .replace(f"{keyword}\n", "") can never find a match — they are dead code. The intent was to cleanly remove adjacent newlines together with the keyword, but the current order leaves orphan newlines behind instead. The practical impact is mitigated by downstream whitespace normalization in preprocess_string_oneig, but the logic is inverted from its clear intent.

 

^{Reviewed by Cursor Bugbot for commit 946b068. Configure here.}