nvidia/OpenMathReasoning

OpenMathReasoning 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) We used… See the full description on the dataset page: https://huggingface.co/datasets/nvidia/OpenMathReasoning.