OREAL-RL-Prompts

# OREAL-RL-Prompts

## Links

- [Arxiv](https://arxiv.org/abs/2502.06781)
- [Github](https://github.com/InternLM/OREAL)
- [OREAL-7B Model](https://huggingface.co/internlm/OREAL-7B)
- [OREAL-32B Model](https://huggingface.co/internlm/OREAL-32B)
- [Data](https://huggingface.co/datasets/InternLM/OREAL-RL-Prompts)

## Introduction

This repository contains the prompts used in the RL training phase of the OREAL project. The prompts are collected from MATH, Numina, and historical AMC/AIME (2024 is excluded). The pass rate of the prompts are calculated with 16 times of inference with OREAL-7B-SFT.

# OREAL-RL-Prompts

## 链接
- [arXiv（国际学术预印本平台）](https://arxiv.org/abs/2502.06781)
- [GitHub（全球代码托管平台）](https://github.com/InternLM/OREAL)
- [OREAL-7B 模型](https://huggingface.co/internlm/OREAL-7B)
- [OREAL-32B 模型](https://huggingface.co/internlm/OREAL-32B)
- [数据集](https://huggingface.co/datasets/InternLM/OREAL-RL-Prompts)

## 简介
本仓库收录了OREAL项目强化学习（Reinforcement Learning，RL）训练阶段使用的提示词（Prompt）。该批提示词采集自MATH数据集、Numina数据集以及历史美国数学竞赛（American Mathematics Competition，AMC）与美国邀请赛数学考试（American Invitational Mathematics Examination，AIME，合称AMC/AIME）试题（2024年试题除外）。提示词的通过率通过使用OREAL-7B-SFT模型进行16轮模型推理计算得出。