Big-Math-RL-UNVERIFIED

# Big-Math: UNVERIFIED

> [!WARNING]
> WARNING: This dataset contains ONLY questions whose answers have not been verified to be correct.
> Use this dataset at your own caution.

## Dataset Creation

Big-Math-Unverified is created as an offshoot of the [Big-Math dataset (HuggingFace Dataset Link)](https://huggingface.co/datasets/SynthLabsAI/Big-Math-RL-Verified).
Big-Math-Unverified goes through the same filters as the rest of Big-Math (eg. remove non-English, remove multiple choice, etc.), except that these problems were not solved in any of the Llama-3.1-8B or 405B rollouts that we did. Therefore, we cannot guarantee the correctness of any answers.

## Citation
If you use this dataset in your work, please cite us using the below citation:
```bibtex
@misc{albalak2025bigmathlargescalehighqualitymath,
title={Big-Math: A Large-Scale, High-Quality Math Dataset for Reinforcement Learning in Language Models},
author={Alon Albalak and Duy Phung and Nathan Lile and Rafael Rafailov and Kanishk Gandhi and Louis Castricato and Anikait Singh and Chase Blagden and Violet Xiang and Dakota Mahan and Nick Haber},
year={2025},
eprint={2502.17387},
archivePrefix={arXiv},
primaryClass={cs.LG},
url={https://arxiv.org/abs/2502.17387},
}
```

# Big-Math：未验证版

> [!警告]
> 警告：本数据集仅包含答案未经正确性验证的问题。使用本数据集请自行承担风险。

## 数据集构建

Big-Math-未验证版是[Big-Math数据集（HuggingFace数据集链接）](https://huggingface.co/datasets/SynthLabsAI/Big-Math-RL-Verified)的衍生数据集。
Big-Math-未验证版与其余Big-Math数据集采用了相同的筛选流程（例如移除非英文内容、移除选择题题型等），但本数据集内的所有题目均未在我们开展的Llama-3.1-8B与Llama-3.1-405B模型推理测试中得到正确解答，因此无法保证任何答案的正确性。

## 引用

若您在研究工作中使用本数据集，请引用如下文献：
bibtex
@misc{albalak2025bigmathlargescalehighqualitymath,
title={Big-Math: A Large-Scale, High-Quality Math Dataset for Reinforcement Learning in Language Models},
author={Alon Albalak and Duy Phung and Nathan Lile and Rafael Rafailov and Kanishk Gandhi and Louis Castricato and Anikait Singh and Chase Blagden and Violet Xiang and Dakota Mahan and Nick Haber},
year={2025},
eprint={2502.17387},
archivePrefix={arXiv},
primaryClass={cs.LG},
url={https://arxiv.org/abs/2502.17387},
}

搜集汇总

背景与挑战

背景概述

该数据集是Big-Math的一个分支，专门收集未经验证答案的数学问题，适用于语言模型的强化学习。它经过与原始数据集类似的过滤处理，但答案正确性无法保证，因此使用时应谨慎。