DeepSeek-VL2

DeepSeek-VL2: Mixture-of-Experts Vision-Language Models for Advanced Multimodal Understanding