# Sparse Activation Parameters

MoE-LLaVA is a large vision-language model based on the Mixture of Experts architecture, achieving efficient multimodal learning through sparse activation parameters