DeepSeek-VL2: Mixture-of-Experts Vision-Language Models for Advanced Multimodal Understanding
This work addresses multimodal AI challenges for researchers and practitioners, presenting incremental improvements over its predecessor with new vision and language components.
The paper tackles the problem of advanced multimodal understanding by introducing DeepSeek-VL2, a series of Mixture-of-Experts vision-language models that achieve competitive or state-of-the-art performance across various tasks like visual question answering and document understanding, with variants ranging from 1.0B to 4.5B activated parameters.
We present DeepSeek-VL2, an advanced series of large Mixture-of-Experts (MoE) Vision-Language Models that significantly improves upon its predecessor, DeepSeek-VL, through two key major upgrades. For the vision component, we incorporate a dynamic tiling vision encoding strategy designed for processing high-resolution images with different aspect ratios. For the language component, we leverage DeepSeekMoE models with the Multi-head Latent Attention mechanism, which compresses Key-Value cache into latent vectors, to enable efficient inference and high throughput. Trained on an improved vision-language dataset, DeepSeek-VL2 demonstrates superior capabilities across various tasks, including but not limited to visual question answering, optical character recognition, document/table/chart understanding, and visual grounding. Our model series is composed of three variants: DeepSeek-VL2-Tiny, DeepSeek-VL2-Small and DeepSeek-VL2, with 1.0B, 2.8B and 4.5B activated parameters respectively. DeepSeek-VL2 achieves competitive or state-of-the-art performance with similar or fewer activated parameters compared to existing open-source dense and MoE-based models. Codes and pre-trained models are publicly accessible at https://github.com/deepseek-ai/DeepSeek-VL2.