---
title: "MM-Eureka: Exploring the Frontiers of Multimodal Reasoning with Rule-based Reinforcement Learning"
canonical_url: "https://www.modelscope.cn/papers/125149"
md_url: "https://www.modelscope.cn/papers/125149.md"
arxiv_id: 2503.07365
published: 2025-04-15
last_updated: 2025-04-15
authors:
  - "Fanqing Meng"
  - "Lingxiao Du"
  - "Zongkai Liu"
  - "Zhixiang Zhou"
  - "Quanfeng Lu"
  - "Daocheng Fu"
  - "Tiancheng Han"
  - "Botian Shi"
  - "Wenhai Wang"
  - "Junjun He"
  - "Kaipeng Zhang"
  - "Ping Luo"
  - "Yu Qiao"
  - "Qiaosheng Zhang"
  - "Wenqi Shao"
model_name: MM-Eureka
model_developer: "上海人工智能实验室、上海创新研究院、上海交通大学、香港大学"
domain:
  - "计算机视觉"
  - "多模态推理"
  - "强化学习"
  - "数学推理"
  - "视觉语言模型"
type:
  - "计算机视觉"
  - "多模态推理"
  - "强化学习"
  - "数学推理"
  - "视觉语言模型"
  - "Computer Vision and Pattern Recognition"
arxiv_url: "https://arxiv.org/abs/2503.07365"
pdf_url: "https://arxiv.org/pdf/2503.07365"
code_link: "https://github.com/ModalMinds/MM-EUREKA"
---

# MM-Eureka: Exploring the Frontiers of Multimodal Reasoning with Rule-based Reinforcement Learning

> DeepSeek R1, and o1 have demonstrated powerful reasoning capabilities in the text domain through stable large-scale reinforcement learning. To enable broader applications, some works have attempted to transfer these capabilities to multimodal reasoning.…

「MM-Eureka: Exploring the Frontiers of Multimodal Reasoning with Rule-based Reinforcement Learning」是 ModelScope 魔搭社区收录的论文，arXiv 2503.07365，作者为 Fanqing Meng, Lingxiao Du, Zongkai Liu et al.，发表于 2025-04-15，属于 计算机视觉、多模态推理、强化学习 领域。

- **ArXiv**: 2503.07365
- **Published**: 2025-04-15
- **Authors**: Fanqing Meng, Lingxiao Du, Zongkai Liu, Zhixiang Zhou, Quanfeng Lu, Daocheng Fu, Tiancheng Han, Botian Shi, Wenhai Wang, Junjun He, Kaipeng Zhang, Ping Luo, Yu Qiao, Qiaosheng Zhang, Wenqi Shao
- **Model**: MM-Eureka
- **Developer**: 上海人工智能实验室、上海创新研究院、上海交通大学、香港大学
- **Domain**: 计算机视觉, 多模态推理, 强化学习, 数学推理, 视觉语言模型
- **ArXiv URL**: https://arxiv.org/abs/2503.07365
- **PDF**: https://arxiv.org/pdf/2503.07365
- **Code**: https://github.com/ModalMinds/MM-EUREKA

Source: https://www.modelscope.cn/papers/125149

---

> MM-Eureka：基于规则的强化学习探索多模态推理前沿

## 摘要

本文提出了MMK12数据集和MM-Eureka多模态推理模型（包含7B和32B两个版本）。MMK12是一个高质量的K12级别多模态数学推理数据集，涵盖函数、几何、方程等领域。MM-Eureka模型基于Qwen2.5-VL构建，采用GRPO算法进行基于规则的强化学习训练，并引入在线过滤策略和两阶段训练方法以提升训练稳定性。实验表明，MM-Eureka-32B在MathVista、WeMath及MMK12等多学科评测中超越了绝大多数开源和闭源模型，展现出强大的跨学科泛化能力。

## Abstract

DeepSeek R1, and o1 have demonstrated powerful reasoning capabilities in the text domain through stable large-scale reinforcement learning. To enable broader applications, some works have attempted to transfer these capabilities to multimodal reasoning. However, these efforts have been limited by the limited difficulty of selected tasks and relatively small training scales, making it challenging to demonstrate strong multimodal reasoning abilities. To address this gap, we introduce the MMK12 dataset and MM-EUREKA with 7B and 32B parameters. The former is a high-quality multimodal mathematics reasoning dataset featuring diverse knowledge domains with human-verified answers and solution processes. The latter is a multimodal model employing rule-based reinforcement learning on MMK12, utilizing online filtering and two-stage training strategy to enhance training stability. MM-EUREKA demonstrates remarkable performance gains in multimodal mathematical reasoning, outperforming previous powerful models like InternVL2.5-78B or InternVL2.5-38B-MPO. In particular, MM-EUREKA achieves competitive or superior performance compared to both open-source and closed-source models, and trails slightly behind o1 in multidisciplinary reasoning tasks. We open-source our complete pipeline to foster further research in this area. We release all our codes, models, data, etc. at https://github.com/ModalMinds/MM-EUREKA