polym-eval-run-evaluation

What problem does it solve?

It helps you run a complete benchmark loop that generates outputs for two models, compares them pairwise with an evaluator, and persists the scored results for later analysis.

Core Features & Use Cases

• End-to-end evaluation orchestration: Generate image/video candidates for Model A and Model B, then score them pairwise.
• Pairwise Gemini scoring: Produce a winner (A/B/tie) with a per-dimension breakdown and rationale.
• Evaluation persistence to eval-db: Save each comparison into polym-eval-db (SQLite) for querying and aggregated rankings.

Quick Start

Use the skill to compare two models on a single prompt by telling it to generate results for both models, score them pairwise, and save the comparison report into polym-eval-db.