nvidia/OpenMathReasoning
Dataset · https://huggingface.co/datasets/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.
First seen by Oriel on .
AI summary
NVIDIA’s OpenMathReasoning is a large-scale dataset of synthetic mathematical reasoning traces, intended for training and evaluating language models and used in an AIMO-2 winning submission.
- The release contains 306K AoPS problems with 3.2M chain-of-thought solutions, 1.7M tool-integrated reasoning solutions, and 566K GenSelect samples, plus 193K problems without solutions.
- Qwen2.5-32B-Instruct preprocesses problems; DeepSeek-R1 and QwQ-32B generate solutions.
- NVIDIA corrected an initial 540K problem count: filtering reduced the released solution set to 306K problems, and a pipeline bug lost 137K proof-based questions whose recovery regressed SFT performance.
- The dataset includes problem type, expected answer, generation model, inference mode, and other metadata; it is released under cc-by-4.0 for community training and evaluation.
- OpenMath-Nemotron models trained on the data report benchmark results across AIME24, AIME25, HMMT-24-25, and HLE-Math, with GenSelect improving several scores.
#math reasoning #synthetic data #llm training #tool use
Generated by gpt-5.6-sol. Check the original for details.
Metric trends
Last 30 days · Up to 120 recorded samples per source and metric, with reposts kept separate · First and latest retained
Likes
Total likes reported by the source.
- Hugging Face Datasets: 500, latest observation ; +2 over observed period since
Downloads (30d)
Rolling 30-day downloads. A change in this value is not new downloads during the observed period.
- Hugging Face Datasets: 68,644, latest observation ; -3,443 over observed period since
Aliases
| Source | External ID | URL |
|---|---|---|
| Hugging Face Datasets | nvidia/OpenMathReasoning | https://huggingface.co/datasets/nvidia/OpenMathReasoning Read discussion |