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