mmlu_pro_mcq
mmlu_pro_mcq
¶
MMLU-Pro MCQ scorer — LLM-based letter extraction + exact match.
Adapted from IPW's mcq.py evaluation handler.
Classes¶
MMLUProScorer
¶
MMLUProScorer(judge_backend: InferenceBackend, judge_model: str)
Bases: LLMJudgeScorer
Score MMLU-Pro responses by extracting answer letter via LLM.