OpenMathReasoning is a large-scale math reasoning dataset for training large language models (LLMs).
This dataset contains
306K unique mathematical problems sourced from AoPS forums with:
3.2M long chain-of-thought (CoT) solutions
1.7M long tool-integrated reasoning (TIR) solutions
566K samples that select the most promising solution out of many candidates (GenSelect)
Additional 193K problems sourced from AoPS forums (problems only, no solutions)
