lmms-lab/MMSearch-R1-7B

MMSearch-R1-7B

Introduction

MMSearch-R1-7B is a search-augmented LMM trained with end-to-end reinforcement learning, equipped with the ability to invoke multimodal search tools on demand. The model can dynamically decide whether to perform image or text search based on the question and integrate the retrieved external information into its reasoning process, enabling more accurate answers for knowledge-intensive VQA tasks. For more details on the training process and model evaluation, please refer to the blog or the paper.

Model Details

• Model name: MMSearch-R1-7B
• Architecture: Qwen2.5-VL-7B base model, fine-tuned with Reinforcement Learning (GRPO)
• Model type: Multimodal Large Language Model with Search-Augmentation
• Languages: English(primary), multilingual(partially)
• License: Apache license 2.0
• Paper: MMSearch-R1: Incentivizing LMMs to Search
• Code: EvolvingLMMs-Lab/multimodal-search-r1

Training Details

• Dataset: FVQA-train
• RL Framework: veRL
• GPUs: 32 * H100

Citation

@article{wu2025mmsearch,
  title={MMSearch-R1: Incentivizing LMMs to Search},
  author={Wu, Jinming and Deng, Zihao and Li, Wei and Liu, Yiding and You, Bo and Li, Bo and Ma, Zejun and Liu, Ziwei},
  journal={arXiv preprint arXiv:2506.20670},
  year={2025}
}