Dataset · nvidia
OpenMathReasoning Mathematical Reasoning
nvidia/OpenMathReasoning
306k tasks and millions of CoT/TIR solutions for training LLM mathematical reasoning.
What's inside
A large English-language dataset for training LLMs in mathematical reasoning. It contains 306k unique tasks from AoPS forums, 3.2M long CoT solutions, 1.7M tool-integrated reasoning (TIR) solutions, and 566k GenSelect best-solution selection examples. Additionally, there are 193k tasks without solutions.
What you can do with it
- SFT for LLM mathematical reasoning
- Training tool-integrated reasoning
- Evaluating GenSelect solution choices
- Tasks
- Question answering, Text generation
- Languages
- English
- Size
- 1M–10M rows
- Downloads this month
- 91.6K
- Likes
- 505
Models trained on it
- cybertruck32489/LFM3-1.7B-RDText generation 3.9K
- mradermacher/Qwen3-0.9B-A0.6B-GGUF 3.1K
- nvidia/OpenMath-Nemotron-14BText generation 2.5K
- nvidia/OpenMath-Nemotron-14B-KaggleText generation 2.3K
- mradermacher/OpenMath-Nemotron-14B-Kaggle-i1-GGUF 1.9K
- cybertruck32489/LFM3-1.7B-RD-v2Text generation 1.8K
- bartowski/nvidia_OpenMath-Nemotron-14B-Kaggle-GGUFText generation 1.8K
- nvidia/OpenMath-Nemotron-1.5BText generation 1.6K
How to load it
from datasets import load_dataset
ds = load_dataset("nvidia/OpenMathReasoning")