MiniCPM-V 4.5: Cooking Efficient MLLMs via Architecture, Data, and Training Recipe
Fuente:
arXiv
Saved in:
| Main Authors: | , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , |
|---|---|
| Format: | Preprint |
| Published: |
2025
|
| Subjects: | |
| Online Access: | |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| _version_ | 1866914051307601920 |
|---|---|
| author | Yu, Tianyu Wang, Zefan Wang, Chongyi Huang, Fuwei Ma, Wenshuo He, Zhihui Cai, Tianchi Chen, Weize Huang, Yuxiang Zhao, Yuanqian Xu, Bokai Cui, Junbo Xu, Yingjing Ruan, Liqing Zhang, Luoyuan Liu, Hanyu Tang, Jingkun Liu, Hongyuan Guo, Qining Hu, Wenhao He, Bingxiang Zhou, Jie Cai, Jie Qi, Ji Guo, Zonghao Chen, Chi Zeng, Guoyang Li, Yuxuan Cui, Ganqu Ding, Ning Han, Xu Yao, Yuan Liu, Zhiyuan Sun, Maosong |
| author_facet | Yu, Tianyu Wang, Zefan Wang, Chongyi Huang, Fuwei Ma, Wenshuo He, Zhihui Cai, Tianchi Chen, Weize Huang, Yuxiang Zhao, Yuanqian Xu, Bokai Cui, Junbo Xu, Yingjing Ruan, Liqing Zhang, Luoyuan Liu, Hanyu Tang, Jingkun Liu, Hongyuan Guo, Qining Hu, Wenhao He, Bingxiang Zhou, Jie Cai, Jie Qi, Ji Guo, Zonghao Chen, Chi Zeng, Guoyang Li, Yuxuan Cui, Ganqu Ding, Ning Han, Xu Yao, Yuan Liu, Zhiyuan Sun, Maosong |
| contents | Multimodal Large Language Models (MLLMs) are undergoing rapid progress and represent the frontier of AI development. However, their training and inference efficiency have emerged as a core bottleneck in making MLLMs more accessible and scalable. To address the challenges, we present MiniCPM-V 4.5, an 8B parameter model designed for high efficiency and strong performance. We introduce three core improvements in model architecture, data strategy and training method: a unified 3D-Resampler model architecture for highly compact encoding over images and videos, a unified learning paradigm for document knowledge and text recognition without heavy data engineering, and a hybrid reinforcement learning strategy for proficiency in both short and long reasoning modes. Comprehensive experimental results in OpenCompass evaluation show that MiniCPM-V 4.5 surpasses widely used proprietary models such as GPT-4o-latest, and significantly larger open-source models such as Qwen2.5-VL 72B. Notably, the strong performance is achieved with remarkable efficiency. For example, on the widely adopted VideoMME benchmark, MiniCPM-V 4.5 achieves state-of-the-art performance among models under 30B size, using just 46.7\% GPU memory cost and 8.7\% inference time of Qwen2.5-VL 7B. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2509_18154 |
| institution | arXiv |
| publishDate | 2025 |
| record_format | arxiv |
| spellingShingle | MiniCPM-V 4.5: Cooking Efficient MLLMs via Architecture, Data, and Training Recipe Yu, Tianyu Wang, Zefan Wang, Chongyi Huang, Fuwei Ma, Wenshuo He, Zhihui Cai, Tianchi Chen, Weize Huang, Yuxiang Zhao, Yuanqian Xu, Bokai Cui, Junbo Xu, Yingjing Ruan, Liqing Zhang, Luoyuan Liu, Hanyu Tang, Jingkun Liu, Hongyuan Guo, Qining Hu, Wenhao He, Bingxiang Zhou, Jie Cai, Jie Qi, Ji Guo, Zonghao Chen, Chi Zeng, Guoyang Li, Yuxuan Cui, Ganqu Ding, Ning Han, Xu Yao, Yuan Liu, Zhiyuan Sun, Maosong Machine Learning Computer Vision and Pattern Recognition Multimodal Large Language Models (MLLMs) are undergoing rapid progress and represent the frontier of AI development. However, their training and inference efficiency have emerged as a core bottleneck in making MLLMs more accessible and scalable. To address the challenges, we present MiniCPM-V 4.5, an 8B parameter model designed for high efficiency and strong performance. We introduce three core improvements in model architecture, data strategy and training method: a unified 3D-Resampler model architecture for highly compact encoding over images and videos, a unified learning paradigm for document knowledge and text recognition without heavy data engineering, and a hybrid reinforcement learning strategy for proficiency in both short and long reasoning modes. Comprehensive experimental results in OpenCompass evaluation show that MiniCPM-V 4.5 surpasses widely used proprietary models such as GPT-4o-latest, and significantly larger open-source models such as Qwen2.5-VL 72B. Notably, the strong performance is achieved with remarkable efficiency. For example, on the widely adopted VideoMME benchmark, MiniCPM-V 4.5 achieves state-of-the-art performance among models under 30B size, using just 46.7\% GPU memory cost and 8.7\% inference time of Qwen2.5-VL 7B. |
| title | MiniCPM-V 4.5: Cooking Efficient MLLMs via Architecture, Data, and Training Recipe |
| topic | Machine Learning Computer Vision and Pattern Recognition |
| url | https://arxiv.org/abs/2509.18154 |