introvoyz041/OpenMathReasoning

OpenMathReasoning是一个大规模数学推理数据集，专为训练大型语言模型（LLMs）而设计。该数据集包含来自AoPS论坛的30.6万个独特数学问题，其中包括320万条长链式思考（CoT）解决方案、170万条工具集成推理（TIR）解决方案以及56.6万个从多个候选方案中选择最有希望解决方案的样本（GenSelect）。此外，还包含来自AoPS论坛的19.3万个问题（仅问题，无解决方案）。该数据集是我们在AIMO-2 Kaggle竞赛中获胜的基础。数据集还包含多个字段，如问题陈述、生成的解决方案、问题类型、预期答案等。

OpenMathReasoning is a large-scale mathematical reasoning dataset specifically designed for training large language models (LLMs). It contains 306,000 unique mathematical problems sourced from the AoPS Forum, including 3.2 million long-chain thinking (CoT) solution samples, 1.7 million tool-integrated reasoning (TIR) solution samples, and 566,000 GenSelect samples that select the most promising solution from multiple candidate solutions. Additionally, it includes 193,000 problems (only questions without solutions) from the AoPS Forum. This dataset served as the foundation for our winning submission to the AIMO-2 Kaggle Competition. The dataset also contains multiple fields such as problem statement, generated solutions, problem type, expected answer, and so on.