PPE-MATH-Best-of-K

# Overview

This contains the MATH correctness preference evaluation set for Preference Proxy Evaluations.

The prompts are sampled from [MATH](https://huggingface.co/datasets/hendrycks/competition_math).

This dataset is meant for benchmarking and evaluation, not for training.

[Paper](https://arxiv.org/abs/2410.14872)

[Code](https://github.com/lmarena/PPE)

# License

User prompts are licensed under MIT, and model outputs are governed by the terms of use set by the respective model providers.

# Citation

```
@misc{frick2024evaluaterewardmodelsrlhf,
title={How to Evaluate Reward Models for RLHF},
author={Evan Frick and Tianle Li and Connor Chen and Wei-Lin Chiang and Anastasios N. Angelopoulos and Jiantao Jiao and Banghua Zhu and Joseph E. Gonzalez and Ion Stoica},
year={2024},
eprint={2410.14872},
archivePrefix={arXiv},
primaryClass={cs.LG},
url={https://arxiv.org/abs/2410.14872},
}
```

# 数据集概览
本数据集收录了面向偏好代理评估（Preference Proxy Evaluations）的MATH竞赛数学正确性偏好评估集。
其提示语均采样自MATH竞赛数学数据集（https://huggingface.co/datasets/hendrycks/competition_math）。
本数据集仅用于基准测试与模型评估，不得用于模型训练。
论文链接：https://arxiv.org/abs/2410.14872
代码链接：https://github.com/lmarena/PPE

# 授权协议
用户提示语采用MIT协议授权，模型输出需遵循对应模型服务商的使用条款。

# 引用文献

@misc{frick2024evaluaterewardmodelsrlhf,
title={如何评估面向强化学习从人类反馈（Reinforcement Learning from Human Feedback, RLHF）的奖励模型},
author={埃文·弗里克、李天乐、康纳·陈、魏林·江、阿纳斯塔西奥斯·N·安杰洛普洛斯、焦建涛、朱邦华、约瑟夫·E·冈萨雷斯、伊恩·斯托伊卡},
year={2024},
eprint={2410.14872},
archivePrefix={arXiv},
primaryClass={cs.LG},
url={https://arxiv.org/abs/2410.14872},
}

搜集汇总

背景与挑战

背景概述

该数据集是MATH正确性偏好评估集，用于偏好代理评估的基准测试和评估。其提示词源自MATH数据集，且仅适用于评估目的，不应用于训练。