SKU SFT-2205 · Sold by External
OpenMathReasoning
Product specifications
| SKU | SFT-2205 |
|---|---|
| Data type | SFT dataset |
| Volume | ~5.68M solutions |
| Size on disk | Not published by source |
| Format | Parquet (Hugging Face) |
| Access model | PUBLIC LICENSE |
| Pricing | Free · open-source license |
| Quality score | — |
| License | CC-BY-4.0 |
OpenMathReasoning is NVIDIA's math reasoning SFT corpus built from ~306k unique Art of Problem Solving (AoPS) forum problems, with solutions generated by DeepSeek-R1 and QwQ-32B. It totals ~5.68M rows across four kinds of solutions: CoT (~3.2M, plain text chain-of-thought), TIR (~1.72M, tool-integrated reasoning that interleaves executable Python run through a code interpreter), GenSelect (~566k, selecting the best of several candidate solutions), plus ~193k additional problems. TIR is a code-interpreter / tool-integrated format, not API function-calling.