OctoMed/OrzMath

---
license: apache-2.0
dataset_info:
features:
- name: question
dtype: string
- name: options
sequence: string
- name: answer
dtype: string
- name: responses
sequence: string
splits:
- name: train
num_examples: 16000
configs:
- config_name: default
data_files:
- split: train
path: data/train-*
---

# OrzMath - Open Reasoning Zero Math

## Description

This dataset contains math problem-solving data with explicit chain-of-thought reasoning traces.
Problems span arithmetic, algebra, geometry, and competition-level mathematics.
5 reasoning traces were collected on a subset of the problems by sampling from DeepSeek-R1-0528,
available in the `responses` column.

We greatly appreciate and build upon the original data source:
[orz-math-57k](https://huggingface.co/datasets/Open-Reasoner-Zero/orz_math_72k_collection_extended),
curated by the [Open-Reasoner-Zero](https://huggingface.co/Open-Reasoner-Zero) team.

## Data Fields
- `question`: Math problem text
- `answer`: The ground truth answer
- `responses`: Up to 5 DeepSeek-R1-0528 reasoning traces per problem (in train split)

## Splits
- `train`: Training data with reasoning traces

## Usage
```python
from datasets import load_dataset

dataset = load_dataset("OctoMed/OrzMath")
```

## Citation

If you find our work helpful, feel free to give us a cite!

```
@article{ossowski2025octomed,
title={OctoMed: Data Recipes for State-of-the-Art Multimodal Medical Reasoning},
author={Ossowski, Timothy and Zhang, Sheng and Liu, Qianchu and Qin, Guanghui and Tan, Reuben and Naumann, Tristan and Hu, Junjie and Poon, Hoifung},
journal={arXiv preprint arXiv:2511.23269},
year={2025}
}
```