DeepScaleR-Preview-Dataset

## Data

Our training dataset consists of approximately 40,000 unique mathematics problem-answer pairs compiled from:

- AIME (American Invitational Mathematics Examination) problems (1984-2023)
- AMC (American Mathematics Competition) problems (prior to 2023)
- Omni-MATH dataset
- Still dataset

## Format

Each row in the JSON dataset contains:

- **problem**: The mathematical question text, formatted with LaTeX notation.
- **solution**: Offical solution to the problem, including LaTeX formatting and boxed final answers. If there is no solution, the `solution` field is an empty string.
- **answer**: The answer to the problem, usually extracted from the solution.

## Example

```json
{
"problem": "Let $a_n=6^{n}+8^{n}$. Determine the remainder upon dividing $a_ {83}$ by $49$.",
"solution": "$6^{83} + 8^{83} = (6+8)(6^{82}-6^{81}8+\\ldots-8^{81}6+8^{82})$\n Becuase $7|(6+8)$, we only consider $6^{82}-6^{81}8+\\ldots-8^{81}6+8^{82} \\pmod{7}$\n$6^{82}-6^{81}8+\\ldots-8^{81}6+8^{82} \\equiv (-1)^{82} - (-1)^{81}+ \\ldots - (-1)^1 + 1 = 83 \\equiv 6 \\pmod{7}$\n$6^{83} + 8^{83} \\equiv 14 \\cdot 6 \\equiv \\boxed{035} \\pmod{49}$",
"answer": "35",
}
```

## Citation

```bibtex
@misc{deepscaler2025,
title={DeepScaleR: Surpassing O1-Preview with a 1.5B Model by Scaling RL},
author={Michael Luo and Sijun Tan and Justin Wong and Xiaoxiang Shi and William Tang and Manan Roongta and Colin Cai and Jeffrey Luo and Tianjun Zhang and Erran Li and Raluca Ada Popa and Ion Stoica},
year={2025},
howpublished={\url{https://pretty-radio-b75.notion.site/DeepScaleR-Surpassing-O1-Preview-with-a-1-5B-Model-by-Scaling-RL-19681902c1468005bed8ca303013a4e2}},
note={Notion Blog}
year={2025}
}
```

## 数据
我们的训练数据集包含约40000组独特的数学题-答案配对数据，其来源如下：
- AIME（美国数学邀请赛，American Invitational Mathematics Examination）：1984年至2023年的试题
- AMC（美国数学竞赛，American Mathematics Competition）：2023年之前的试题
- Omni-MATH数据集
- Still数据集

## 格式
该JSON数据集的每一行均包含以下字段：
- **problem（试题）**：数学问题文本，采用LaTeX排版格式。
- **solution（解答）**：该试题的官方解答，包含LaTeX格式与带框的最终答案。若无官方解答，则`solution`字段为空字符串。
- **answer（答案）**：该试题的答案，通常从对应解答中提取得到。

## 示例
json
{
"problem": "设 $a_n=6^{n}+8^{n}$。求 $a_{83}$ 除以49的余数。",
"solution": "$6^{83} + 8^{83} = (6+8)(6^{82}-6^{81}8+ldots-8^{81}6+8^{82})$
由于 $7|(6+8)$，我们仅需考虑 $6^{82}-6^{81}8+ldots-8^{81}6+8^{82} pmod{7}$
$6^{82}-6^{81}8+ldots-8^{81}6+8^{82} equiv (-1)^{82} - (-1)^{81}+ ldots - (-1)^1 + 1 = 83 equiv 6 pmod{7}$
$6^{83} + 8^{83} equiv 14 cdot 6 equiv oxed{035} pmod{49}$",
"answer": "35",
}

## 引用
bibtex
@misc{deepscaler2025,
title={DeepScaleR: 通过强化学习缩放实现超越O1-Preview的1.5B参数模型},
author={Michael Luo、Sijun Tan、Justin Wong、Xiaoxiang Shi、William Tang、Manan Roongta、Colin Cai、Jeffrey Luo、Tianjun Zhang、Erran Li、Raluca Ada Popa、Ion Stoica},
year={2025},
howpublished={url{https://pretty-radio-b75.notion.site/DeepScaleR-Surpassing-O1-Preview-with-a-1-5B-Model-by-Scaling-RL-19681902c1468005bed8ca303013a4e2}},
note={Notion博客}
year={2025}
}

搜集汇总

背景与挑战

背景概述

该数据集由约4万条数学问题-答案对组成，数据来源于AIME、AMC、Omni-MATH和Still等公开数学竞赛与资源。每条数据以JSON格式存储，包含问题文本、官方解答和最终答案，适用于数学问题求解模型的训练。