1 paper · 1 filter
Weilin Cai, Juyong Jiang, Le Qin +3
Expert parallelism has emerged as a key strategy for distributing the computational workload of sparsely-gated mixture-of-experts (MoE) models across multiple devices, enabling the…