# Sparse Activated Mixture of Experts

MoE-LLaVA is a large-scale vision-language model based on a mixture of experts architecture, achieving efficient multimodal learning through sparsely activated parameters.