Skip to content

Source paper

MiniCPM-V 4.5: Cooking Efficient MLLMs via Architecture, Data, and Training Recipe

Tianyu Yu, Zefan Wang, Chongyi Wang and 31 more

Tianyu Yu, Zefan Wang, Chongyi Wang, Fuwei Huang, Wenshuo Ma, Zhihui He, Tianchi Cai, Weize Chen, Yuxiang Huang, Yuanqian Zhao, Bokai Xu, Junbo Cui, Yingjing Xu, Liqing Ruan, Luoyuan Zhang, Hanyu Liu, Jingkun Tang, Hongyuan Liu, Qining Guo, Wenhao Hu, Bingxiang He, Jie Zhou, Jie Cai, Ji Qi, Zonghao Guo, Chi Chen, Guoyang Zeng, Yuxuan Li, Ganqu Cui, Ning Ding, Xu Han, Yuan Yao, Zhiyuan Liu, Maosong Sun

arXiv 2509.18154 · 202554 upvotes on Hugging Face

Research ideas from this paper (3)

Generated by AI. The paper’s authors did not write these.

AI summary

MiniCPM-V 4.5 is an 8B parameter Multimodal Large Language Model (MLLM) designed to maximize performance and efficiency through architectural and training innovations. The paper introduces a unified 3D-Resampler for compact spatial-temporal visual encoding, a document learning paradigm that uses dynamic noise to unify OCR and context reasoning, and a hybrid Reinforcement Learning strategy that alternates between short and long chain-of-thought reasoning modes.

How it works (AI summary)

The model integrates a lightweight visual encoder with a Unified 3D-Resampler that compresses images and videos into a small, fixed number of tokens (e.g., 128 tokens for video) to reduce computational overhead. The training pipeline utilizes a 'corruption-based' document learning objective to fuse OCR and semantic understanding, followed by Reinforcement Learning with Preference Optimization (GRPO) to align both fast (short) and slow (long chain-of-thought) reasoning capabilities.