We-Math2.0-Pro

# Dataset Card for We-Math 2.0

[GitHub](https://github.com/We-Math/We-Math2.0) | [Paper](https://arxiv.org/abs/2508.10433) | [Website](https://we-math2.github.io/)

We-Math 2.0 is a unified system designed to comprehensively enhance the mathematical reasoning capabilities of Multimodal Large Language Models (MLLMs).
It integrates a structured mathematical knowledge system, model-centric data space modeling, and a reinforcement learning (RL)-based training paradigm to achieve both broad conceptual coverage and robust reasoning performance across varying difficulty levels.

The key contributions of We-Math 2.0 are fourfold:

- MathBook Knowledge System — A five-level hierarchical structure encompassing 491 knowledge points and 1,819 fundamental principles.
- MathBook-Standard & MathBook-Pro — MathBook-Standard ensures wide conceptual coverage and flexibility via dual expansion, while MathBook-Pro defines a three-dimensional difficulty space and generates 7 progressive variants per problem for robust training.
- MathBook-RL — A two-stage RL framework comprising Cold-Start Fine-tuning for knowledge-oriented chain-of-thought alignment, and Progressive Alignment RL with average-reward learning and dynamic data scheduling for gradual alignment across difficulty levels.
- MathBookEval — A comprehensive benchmark covering all 491 knowledge points with diverse reasoning step distributions.

## Citation
If you find the content of this project helpful, please cite our paper as follows:

```
@misc{qiao2025wemath20versatilemathbook,
title={We-Math 2.0: A Versatile MathBook System for Incentivizing Visual Mathematical Reasoning},
author={Runqi Qiao and Qiuna Tan and Peiqing Yang and Yanzi Wang and Xiaowan Wang and Enhui Wan and Sitong Zhou and Guanting Dong and Yuchen Zeng and Yida Xu and Jie Wang and Chong Sun and Chen Li and Honggang Zhang},
year={2025},
eprint={2508.10433},
archivePrefix={arXiv},
primaryClass={cs.AI},
url={https://arxiv.org/abs/2508.10433},
}
```

# We-Math 2.0 数据集卡片

[GitHub仓库](https://github.com/We-Math/We-Math2.0) | [研究论文](https://arxiv.org/abs/2508.10433) | [项目官网](https://we-math2.github.io/)

We-Math 2.0 是一款统一化系统，旨在全面提升多模态大语言模型（Multimodal Large Language Models, MLLMs）的数学推理能力。该系统整合了结构化数学知识体系、以模型为中心的数据集空间建模方法，以及基于强化学习（Reinforcement Learning, RL）的训练范式，可在覆盖广泛数学概念的同时，在不同难度层级下均实现稳健的推理性能。

We-Math 2.0 的核心贡献共四项：
- **MathBook 知识体系**：采用五级层级结构，涵盖491个数学知识点与1819条基础原理。
- **MathBook-Standard 与 MathBook-Pro**：MathBook-Standard 通过双扩展机制实现广泛的概念覆盖与灵活性；MathBook-Pro 则定义了三维难度空间，并为每个问题生成7种渐进式变体以支撑高质量训练。
- **MathBook-RL**：采用两阶段强化学习框架，其一为面向知识导向思维链（Chain-of-Thought, CoT）对齐的冷启动微调，其二为结合平均奖励学习与动态数据调度的渐进式对齐强化学习，用于在不同难度层级上逐步实现对齐。
- **MathBookEval**：覆盖全部491个知识点的综合基准测试集，包含多样化的推理步骤分布。

## 引用
若您认为本项目内容对您的研究有所帮助，请按如下格式引用我们的论文：

@misc{qiao2025wemath20versatilemathbook,
title={We-Math 2.0: A Versatile MathBook System for Incentivizing Visual Mathematical Reasoning},
author={Runqi Qiao and Qiuna Tan and Peiqing Yang and Yanzi Wang and Xiaowan Wang and Enhui Wan and Sitong Zhou and Guanting Dong and Yuchen Zeng and Yida Xu and Jie Wang and Chong Sun and Chen Li and Honggang Zhang},
year={2025},
eprint={2508.10433},
archivePrefix={arXiv},
primaryClass={cs.AI},
url={https://arxiv.org/abs/2508.10433},
}