1 paper
Siyuan He, Bokai Yang, Jie Hu +2
Mixture-of-experts (MoE) architectures increase model capacity by combining a collection of expert predictors through input-dependent routing, while often activating only a small s…