Fei Chao

Large Language Models continue to scale in size and capability, driving substantial computational and memory demands.Mixture-of-Experts (MoE) architectures alleviate this cost by activating only a sparse subset of experts per token, enabling efficient scaling without proportional increases in inf...