AIME25-CoT-CN

# Sci-Bench-AIME25'

This repo is a branch of Sci Bench made by IPF team. Mainly include the AIME 25' solution with multi-modal CoT and diverse solving path.

## 📚 Cite

If you use the **Sci-Bench-AIME25 (IPF/AIME25-CoT-CN)** dataset in your research, please cite:

```bibtex
@dataset{zhang2025scibench_aime25,
title = {{Sci-Bench-AIME25}: A Multi-Modal Chain-of-Thought Dataset for Advanced Tool-Intergrated Mathematical Reasoning},
author = {Zhang, Haoxiang and Wang, Siyuan and Fang, Xueji and Zou, Xinkai and Lyu, Tiange and Cao, Siyuan and Huang, Jingyuan and Xing, Jie},
year = {2025},
publisher = {Hugging Face},
doi = {10.5281/zenodo.17112481},
url = {https://huggingface.co/datasets/IPF/AIME25-CoT-CN},
note = {Also available on Zenodo}
}
```
-- -- --

# Brief intro

## 💻 Overview

A brief template and final report will be posted in [Isaac's Blog](https://isaacghx.github.io/2025/08/08/solutions/Hand-made-Solution-and-CoT-for-AIME25/)

And the markdown template can be found in data/I_2

## ❓ Why we do this?
- The multi-lingual datasets are scarce, while the CoT of Math is even less, no matter whether the CoT or the solution contains pictures, code blocks, etc.
- It can be used as a base benchmark for reasoning LLM post-training: SFT & RL.
- It's tiny but entropy-rich.
- It can solve the issue of overfitting on certain single, text-based solution CoT.

-- -- --

# 🗡 Template

This repo's structure is formed as:

```
/Sci-Bench-AIME25
├── data
│ ├── I_1
│ │ ├── problem.md
│ │ ├── ground_truth.md
│ │ ├── solution1
│ │ ├── description.md (if it only bears (a) handwritten picture(s), please fill this md with "ONLY HANDWRITTEN PIC" at the very beginning. )
│ │ ├── code1.py
│ │ ├── code2.py
│ ├── I_2
│ ├── ...
│ ├── II_1
│ ├── ...
│
├── images(image should be higher than 1080P resolution and ensure the words are clear. Tips: it can be unease to recognize but should be formed relative structured)
│ ├── gen_images
│ │ ├── I_2
│ │ │ ├── code_1.png
│ │ ├── └── code_2.png
│ │ ├── I_6
│ │ ├── └── code_1.png
│ │ └── ...
│ └── hand_written
│ ├── II_12
│ │ ├── description_1.png
│ │ └── description_2.png
│ ├── I_11
│ │ └── description_1.png
│ └── ...
└── README.md
```

- Detail format of `problem.md`

```markdown
求所有整数底数 $b>9$ 的和，使得 $17_b$ 是 $97_b$ 的因数。
```

- Detail format of `ground_truth.md`

```markdown
70
```

- Detail format of `description.md`: please use `## step_n. xxx` to make the step-by-step subtitle.
More to mention: when there exists python code in the markdown `description.md`, you should also format into a step-wise title `## step_n. xxx` and elaborate the function of it.
Furthermore, Python code should be encapsulated by md tag: ```python ./code_n.py ```

```markdown
## 1. 线段比例分析与平行关系

首先，分析边 $\overline{AB}$ 和 $\overline{AC}$ 上的线段长度和比例。

* **边 $\overline{AB}$**:
...

* **边 $\overline{AC}$**:
...

根据以上比例，我们发现：
$\frac{AD}{AB} = \frac{AF}{AC} = \frac{1}{7} \quad \text{以及} \quad \frac{AE}{AB} = \frac{AG}{AC} = \frac{5}{7}$

根据泰勒斯定理（逆定理），这些比例关系意味着：
...

## 2. 计算 $\triangle ABC$ 的面积

...

## 5. 答案/结论

七边形 $AFNBCEM$ 的面积等于 $\triangle ABC$ 的面积。
$S_{AFNBCEM} = S_{ABC} = 588$

```

# Sci-Bench-AIME25'
本仓库为IPF团队打造的Sci Bench分支项目，主要收录AIME 25'的多模态思维链（Chain-of-Thought, CoT）解题方案与多样化求解路径。

## 📚 引用规范
若您在研究中使用**Sci-Bench-AIME25 (IPF/AIME25-CoT-CN)** 数据集，请引用如下文献：
bibtex
@dataset{zhang2025scibench_aime25,
title = {{Sci-Bench-AIME25}: A Multi-Modal Chain-of-Thought Dataset for Advanced Tool-Intergrated Mathematical Reasoning},
author = {Zhang, Haoxiang and Wang, Siyuan and Fang, Xueji and Zou, Xinkai and Lyu, Tiange and Cao, Siyuan and Huang, Jingyuan and Xing, Jie},
year = {2025},
publisher = {Hugging Face},
doi = {10.5281/zenodo.17112481},
url = {https://huggingface.co/datasets/IPF/AIME25-CoT-CN},
note = {Also available on Zenodo}
}

-- -- --

# 简介
## 💻 项目概览
简短模板与最终报告将发布于[Isaac个人博客](https://isaacghx.github.io/2025/08/08/solutions/Hand-made-Solution-and-CoT-for-AIME25/)，Markdown模板可在data/I_2目录中获取。

## ❓ 项目初衷
- 当前多语言数学数据集较为稀缺，而兼具图文、代码块等多模态形式的数学思维链（CoT）资源更是寥寥无几。
- 本数据集可作为推理型大语言模型（Large Language Model, LLM）后训练（包括监督微调SFT与强化学习RL）的基础基准测试集。
- 数据集体量小巧但信息熵丰富，极具研究价值。
- 可有效规避单一文本式思维链解题方案导致的过拟合问题。

-- -- --

# 🗂 项目目录结构与文件格式规范
本仓库的目录结构如下：

/Sci-Bench-AIME25
├── data
│ ├── I_1
│ │ ├── problem.md
│ │ ├── ground_truth.md
│ │ ├── solution1
│ │ ├── description.md (若该目录仅包含手写图片，请在文件开头填写"ONLY HANDWRITTEN PIC"字样)
│ │ ├── code1.py
│ │ ├── code2.py
│ ├── I_2
│ ├── ...
│ ├── II_1
│ ├── ...
│
├── images(图片分辨率需不低于1080P，且文字清晰可辨。提示：图片内容无需完全易读，但需具备相对结构化的呈现形式)
│ ├── gen_images
│ │ ├── I_2
│ │ │ ├── code_1.png
│ │ ├── └── code_2.png
│ │ ├── I_6
│ │ ├── └── code_1.png
│ │ └── ...
│ └── hand_written
│ ├── II_12
│ │ ├── description_1.png
│ │ └── description_2.png
│ ├── I_11
│ │ └── description_1.png
│ └── ...
└── README.md

### `problem.md` 文件详细格式示例
markdown
求所有整数底数 $b>9$ 的和，使得 $17_b$ 是 $97_b$ 的因数。

### `ground_truth.md` 文件详细格式示例
markdown
70

### `description.md` 文件详细格式说明
需使用 `## step_n. 主题` 的格式编写分步小标题。特别注意：若Markdown文件中包含Python代码，也需将其归入对应步骤标题下，并说明代码功能。此外，Python代码需使用Markdown代码块格式包裹：` python ./code_n.py `

markdown
## 1. 线段比例分析与平行关系
首先，分析边 $\overline{AB}$ 和 $\overline{AC}$ 上的线段长度和比例。
* **边 $\overline{AB}$**:
...

* **边 $\overline{AC}$**:
...
根据以上比例，我们发现：
$\frac{AD}{AB} = \frac{AF}{AC} = \frac{1}{7} \quad ext{以及} \quad \frac{AE}{AB} = \frac{AG}{AC} = \frac{5}{7}$
根据泰勒斯定理（逆定理），这些比例关系意味着：
...

## 2. 计算 $\triangle ABC$ 的面积
...

## 5. 答案/结论
七边形 $AFNBCEM$ 的面积等于 $\triangle ABC$ 的面积。
$S_{AFNBCEM} = S_{ABC} = 588$