Astrea: A MOE-based Visual Understanding Model with Progressive Alignment

Xiaoda Yang, JunYu Lu, Hongshun Qiu +12

Vision-Language Models (VLMs) based on Mixture-of-Experts (MoE) architectures have emerged as a pivotal paradigm in multimodal understanding, offering a powerful framework for inte…