musumecmtcd/MathNet

MathNet是一个高质量、大规模、多模态、多语言的奥林匹克数学问题数据集，包含30,676个专家编写的问题及其解决方案，涵盖17种语言和47个国家的竞赛，时间跨度为20年。数据集支持三个基准任务：问题解决、数学感知检索和检索增强的问题解决。数据来源于官方问题手册，确保专家级质量和风格一致性。数据处理流程包括文档摄取、问题-解决方案提取和验证阶段。预览版包含100个多样化的问题样本，并嵌入了图像。数据集模式详细描述了问题陈述、解决方案、主题和元数据的结构。该数据集适用于奥林匹克级推理评估、多语言数学评估和检索基准测试，但需注意数据污染和LLM辅助元数据的局限性。

MathNet is a high-quality, large-scale, multimodal, and multilingual dataset of Olympiad-level math problems, comprising 30,676 expert-authored problems with solutions across 17 languages and 47 countries, spanning two decades of competitions. It supports three benchmark tasks: problem solving, math-aware retrieval, and retrieval-augmented problem solving. The dataset is sourced from official problem booklets, ensuring expert-level quality and stylistic consistency. The data pipeline involves document ingestion, problem-solution extraction, and verification stages. The preview includes a diverse 100-problem slice with embedded images. The schema details the structure of the dataset, including problem statements, solutions, topics, and metadata. The dataset is intended for Olympiad-level reasoning evaluation, multilingual math evaluation, and retrieval benchmarks, with noted caveats regarding contamination and LLM-assisted metadata.