1 paper
Zhexiang Zhang, Ye Wang, Yumiao Zhao +10
Serving large Mixture-of-Experts (MoE) models is challenging because of their large memory footprints, heterogeneous resource demands, and highly dynamic inference workloads. Most…