Tag
This paper introduces PAJAMA, a system that distills LLM-as-a-judge decision logic into programmatic judges, eliminating per-sample API costs while matching the performance of a 13B LLM judge, and providing transparency and efficiency improvements for scalable evaluation.