huglabs/math_250

---
license: mit
datasets:
- openai/prm800k
language:
- en
---

# MATH - Processed Subset (250 Samples)

This dataset is a curated selection of 250 samples from the **MATH** dataset (specifically the splits found in the PRM800K repository). It is organized into three nested tiers for progressive benchmarking.

## Dataset Description

This version provides a controlled set of 250 mathematical problems divided into three tiers: **Small**, **Medium**, and **Large**.

**Nested Structure:**
The tiers are designed to be cumulative. Each subsequent tier includes all samples from the previous ones:
* **Small Tier:** The base selection of foundational samples.
* **Medium Tier:** Includes all samples from "Small" plus additional problems.
* **Large Tier:** Includes all samples from "Medium" (and "Small"), representing the full 250-sample subset.

## Dataset Sources

This subset is derived from the MATH dataset splits provided by OpenAI:
* **Repository:** [OpenAI PRM800K - MATH Splits](https://github.com/openai/prm800k/tree/main?tab=readme-ov-file#math-splits)
* **Paper:** [Let's Verify Step by Step (Lightman et al., 2023)](https://arxiv.org/abs/2305.20050)

## Uses

This dataset is intended for:
* Benchmarking LLM mathematical reasoning.
* Experiments involving scalable oversight and process-based supervision.
* Quick evaluation runs where the full MATH dataset is too large for rapid iteration.

## Modifications and Attribution

This repository contains a **modified subset** of the original MATH data. The primary modifications include the selection of 250 specific rows and their categorization into the Small, Medium, and Large hierarchical tiers.

## Citation

If you use this dataset, please cite the original work:

```bibtex
@article{lightman2023lets,
title={Let's Verify Step by Step},
author={Lightman, Hunter and Kosaraju, Vineet and Burda, Yura and Edwards, Harri and Baker, Bowen and Lee, Teddy and Leike, Jan and Schulman, John and Sutskever, Ilya and Cobbe, Karl},
journal={arXiv preprint arXiv:2305.20050},
year={2023}
}