Mathlib Conjectures

该数据集包含了从40个Mathlib种子文件生成的总共12,289个猜想，其中有3,776个被认定为在语法上有效且非平凡的猜想。此外，该数据集还包括了与拓扑学中半开集、α-开集和预开集属性相关的猜想，这些猜想是通过结合基于规则的上下文提取与基于大型语言模型的定理陈述生成管道产生的。规模上，数据集涵盖了来自40个文件的12,289个猜想。任务方面，该数据集旨在用于定理证明和猜想生成。

This dataset contains a total of 12,289 conjectures generated from 40 Mathlib seed files, among which 3,776 are identified as syntactically valid and non-trivial. In addition, it also includes conjectures related to the properties of semi-open sets, α-open sets, and pre-open sets in topology, which were generated by combining rule-based context extraction with a large language model-based theorem statement generation pipeline. In terms of scale, the dataset covers 12,289 conjectures sourced from 40 files. This dataset is designed for theorem proving and conjecture generation.