MMLU_PRO_IDK

NAME = MMLU_PRO_IDK
DATASET_PATH = TIGER-Lab/MMLU-Pro
SAMPLE_SPLIT = test
FEWSHOT_SPLIT = test
RESPONSE_TYPE = LOGLIKELIHOODS
METRICS = [AccuracyLoglikelihood, AccuracyNormLoglikelihood, ConfidenceWeightedAccuracy, DistributionalCorrectnessScore, TernaryScore]
SUBJECTS = ['engineering', 'physics', 'psychology', 'chemistry', 'biology', 'law', 'philosophy', 'computer science', 'other', 'economics', 'business', 'history', 'math', 'health']
LANGUAGE = <Language.ENG: 'English'>

• Module: eval_framework.tasks.benchmarks.mmlu_pro

• File: src/eval_framework/tasks/benchmarks/mmlu_pro.py | View on GitHub

• Link to dataset: https://huggingface.co/datasets/TIGER-Lab/MMLU-Pro

More detailed documentation, with prompt examples and ground truth completions, can be generated with uv run -m eval_framework.utils.generate_task_docs --add-prompt-examples --only-tasks "MMLU_PRO_IDK".