1 paper
Bowen Dong, Yilong Fan, Yutao Sun +4
Routing networks in sparsely activated mixture-of-experts (MoE) dynamically allocate input tokens to top-k experts through differentiable sparse transformations, enabling scalable…