在ICLR 2025的学术舞台上,一项名为LLaVA-MoD(Lightweight Language-Vision Architecture with Mixture-of-Experts Distillation)的研究引发了广泛关注。该模型通过结合混合专家(Mixture-of-Experts, MoE)架构……