InternVL3: Exploring Advanced Training and Test-Time Recipes for Open-Source Multimodal Models

作者：Jinguo Zhu, Weiyun Wang, Zhe Chen, Zhaoyang Liu, Sheng‐Long Ye, Lixin Gu, Hao Tian, Yuchen Duan, Weijie Su, Shao, Jie, Zhangwei Gao, Erfei Cui, Xuehui Wang, Yue Cao, Yangzhou Liu, Xingguang Wei, Hongjie Zhang, Haomin Wang, Weiye Xu, Hao Li, Jiahao Wang, Deng, Nianchen, Songze Li, Yinan He, Tan Jiang, Jiapeng Luo, Zhichao Wang, Conghui He, Botian Shi, Zhang, Xingcheng, Wei Shao, Junjun He, Yingtong Xiong, Wenwen Qu, Peng Sun, Penglong Jiao, Han Lv, WU Li-jun, Kaipeng Zhang, Huipeng Deng, Jiaye Ge, Kai Chen, Limin Wang, Min Dou, Lewei Lu, Zhu, Xizhou, Tong Lü, Dahua Lin, Yu Qiao, Jifeng Dai, Wenhai Wang · 发表于：arXiv (Cornell University) · 年份：2025 · DOI：10.48550/arxiv.2504.10479 · 被引用次数：8 · 研究领域：Multimodal Machine Learning Applications、Topic Modeling、Domain Adaptation and Few-Shot Learning

We introduce InternVL3, a significant advancement in the InternVL series featuring a native multimodal pre-training paradigm. Rather than adapting a text-only large language model (LLM) into a multimodal large language model (MLLM) that supports visual inputs, InternVL3 jointly acquires multimodal and linguistic capabilities from both diverse multimodal data and pure-text corpora during a single pre-training stage. This unified training paradigm effectively addresses the complexities and alignment challenges commonly encountered in conventional post-hoc training pipelines for MLLMs. To further improve performance and scalability, InternVL3 incorporates variable visual position encoding (V2PE) to support extended multimodal contexts, employs advanced post-training techniques such as supervised fine-tuning (SFT) and mixed preference optimization (MPO), and adopts test-time scaling strategies alongside an optimized training infrastructure. Extensive empirical evaluations demonstrate that InternVL3 delivers superior performance across a wide range of multi-modal tasks. In particular, InternVL3-78B achieves a score of 72.2 on the MMMU benchmark, setting a new state-of-the-art among open-source MLLMs. Its capabilities remain highly competitive with leading proprietary models, including ChatGPT-4o, Claude 3.5 Sonnet, and Gemini 2.5 Pro, while also maintaining strong pure-language proficiency. In pursuit of open-science principles, we will publicly release both the training data and ...