MindOmni: Unleashing Reasoning Generation in Vision Language Models with RGPO

Yuxin Chen (University of Pennsylvania) · Yicheng Xiao (Southern University of Science and Technology) · Xiu Li (Bytedance) · Yukang Chen (NVIDIA Research) · Wei Huang (RIKEN AIP) · Ying Shan (Tencent) · Xiaojuan Qi (The University of Hong Kong) · Lin Song (Tencent) · Yingmin Luo (Tencent ARC Lab) · Yukang Gan (Tencent)
chain-of-thoughtcomplex reasoning tasksdecoder-only diffusion modulefine-grained reasoning generationgeneration benchmarkslarge language modelmultimodal feedbackmultimodal inputspolicy updatesreasoning generationreasoning generation policy optimizationreinforcement learningsupervised fine-tuningunderstanding benchmarksvision language model

Recent text-to-image systems face limitations in handling multimodal inputs and complex reasoning tasks. We introduce MindOmni, a unified multimodal large language model that addresses these challenges by incorporating reasoning generation through reinforcement learning. MindOmni leverages a three-phase training strategy: i) design of a unified vision language model with a decoder-only diffusion module, ii) supervised fine-tuning with Chain-of-Thought (CoT) instruction data, and iii) our proposed Reasoning Generation Policy Optimization (RGPO) algorithm, utilizing multimodal feedback to effectively guide policy updates. Experimental results demonstrate that MindOmni outperforms existing models, achieving impressive performance on both understanding and generation benchmarks, meanwhile showcasing advanced fine-grained reasoning generation capabilities, especially with mathematical reasoning instruction. All codes will be made public.