arXiv 2511.163342025-12-05cs.AIcs.CL

OpenMMReasoner: 推动多模态推理的前沿研究：一个开放且通用的配方

OpenMMReasoner: Pushing the Frontiers for Multimodal Reasoning with an Open and General Recipe

MiroMind AI；Nanyang Technological University（南洋理工大学）；Tsinghua University（清华大学）；LMMs-Lab Team（多模态实验室团队）

Kaichen Zhang, Keming Wu, Zuhao Yang, Bo Li, Kairui Hu, Bin Wang, Ziwei Liu, Xingxuan Li, Lidong Bing

阅读 PDF

arXiv 页面 ↗HTML 原文 ↗

AI总结：

OpenMMReasoner提出了一种开放且通用的多模态推理训练配方，通过两阶段方法提升推理性能，实现在多个基准测试中超越现有基线。

摘要：

近年来，大推理模型的进步推动了将其能力扩展到多模态领域的需求增长。然而，尽管在视觉推理方面取得了显著进展，缺乏透明且可重复的数据整理和训练策略仍然是可扩展研究的主要障碍。在本工作中，我们介绍了OpenMMReasoner，一个完全透明的两阶段配方，涵盖监督微调（SFT）和强化学习（RL）。在SFT阶段，我们构建了一个包含874,000个样本的冷启动数据集，通过严格的分步验证，为推理能力提供了坚实的基础。随后的RL阶段利用了一个包含74,000个样本的跨多领域数据集，进一步提高并稳定这些能力，从而实现更稳健和高效的训练过程。广泛评估表明，我们的训练配方不仅超越了强大的基线，还突显了数据质量和训练设计在塑造多模态推理性能中的关键作用。值得注意的是，我们的方法在九个多模态推理基准测试中，相对于Qwen2.5-VL-7B-Instruct基线实现了11.6%的提升，为未来大规模多模态推理研究奠定了坚实的实证基础。我们已开源所有代码、流程和数据，地址为https://github.com/EvolvingLMMs-Lab/OpenMMReasoner。

英文摘要：

Recent advancements in large reasoning models have fueled growing interest in extending such capabilities to multimodal domains. However, despite notable progress in visual reasoning, the lack of transparent and reproducible data curation and training strategies remains a major barrier to scalable research. In this work, we introduce OpenMMReasoner, a fully transparent two-stage recipe for multimodal reasoning spanning supervised fine-tuning (SFT) and reinforcement learning (RL). In the SFT stage, we construct an 874K-sample cold-start dataset with rigorous step-by-step validation, providing a strong foundation for reasoning capabilities. The subsequent RL stage leverages a 74K-sample dataset across diverse domains to further sharpen and stabilize these abilities, resulting in a more robust and efficient learning process. Extensive evaluations demonstrate that our training recipe not only surpasses strong baselines but also highlights the critical role of data quality and training design in shaping multimodal reasoning performance. Notably, our method achieves a 11.6% improvement over the Qwen2.5-VL-7B-Instruct baseline across nine multimodal reasoning benchmarks, establishing a solid empirical foundation for future large-scale multimodal reasoning research. We open-sourced all our codes, pipeline, and data at https://github.com/EvolvingLMMs-Lab/OpenMMReasoner.

机构、总结、中文摘要均由AI分析而来，可能存在错误，请以论文正文为准。