Mathematical Topics Tree (MaTT) benchmark

MaTT数据集由卡内基梅隆大学创建，包含1958个问题，覆盖纯数学和应用数学的12个主要领域。数据集通过提取Wikipedia的数学主题列表和相关参考书籍构建，每个问题都配有详细的层次化主题链。MaTT旨在全面评估大型语言模型在数学推理上的能力，特别是解决复杂和多步骤数学问题的能力。

The MaTT dataset was developed by Carnegie Mellon University. It contains 1958 questions spanning 12 primary domains of pure and applied mathematics. The dataset is constructed by extracting mathematical topic lists and related reference books from Wikipedia, with each question paired with a detailed hierarchical topic chain. The MaTT dataset is designed to comprehensively evaluate the mathematical reasoning capabilities of large language models, particularly their ability to solve complex, multi-step mathematical problems.