SKU SFT-2207 · Sold by External
MathInstruct
Product specifications
| SKU | SFT-2207 |
|---|---|
| Data type | SFT dataset |
| Volume | 262k instruction-response examples |
| Size on disk | 212 MB |
| Format | Parquet (Hugging Face) |
| Access model | PUBLIC LICENSE |
| Pricing | Free · open-source license |
| Quality score | — |
| License | MIT |
MathInstruct (TIGER-Lab) is the 262k-example math SFT dataset behind the MAmmoTH models, compiled from 13 math rationale datasets. It is deliberately hybrid: chain-of-thought (CoT) examples give step-by-step natural-language reasoning, while program-of-thought (PoT) examples solve problems with executable Python code blocks. PoT here is code-interpreter / program-of-thought reasoning, not API function-calling. The overall dataset is MIT-licensed, though constituent sources carry mixed licenses.