MMLU-Pro
收藏资源简介:
MMLU-Pro是由滑铁卢大学等机构创建的综合性语言理解与推理基准数据集,包含12,032个问题,覆盖14个不同的学科领域。该数据集通过增加选项数量和引入更多需要深思熟虑推理的大学水平考试问题,显著提高了挑战性和鲁棒性。创建过程中,数据集经历了两轮专家评审,以减少数据噪声并确保问题质量。MMLU-Pro旨在通过提供更复杂、更具挑战性的问题,推动大型语言模型在多学科语言理解和推理能力上的进步。
MMLU-Pro is a comprehensive language understanding and reasoning benchmark dataset created by institutions including the University of Waterloo, containing 12,032 questions spanning 14 distinct academic disciplines. This dataset significantly enhances challenge and robustness by increasing the number of answer options and introducing more college-level exam questions that require deliberate, well-considered reasoning. During its development, the dataset underwent two rounds of expert reviews to reduce data noise and ensure high-quality questions. MMLU-Pro aims to advance the multi-disciplinary language understanding and reasoning capabilities of large language models (LLMs) by providing more complex and challenging questions.




