作者位次 #4/8ACL 2026Findings· Findings of ACL
From Pseudo-Balancing to True Specialization: Memory-Aware Routing for Mixture-of-Experts
Peixuan Hou · Yunbo Hou · Bin Chen · LI He · Yundu Huang (Taobao & · Weiping Li · Zerui Huang (Taobao & · Guojie Song
摘要
Mixture-of-Experts (MoE) efficiently trains large models by using sparse activation to lower costs, selecting a few experts based on data characteristics. For MoE, an unbalanced expert load will lead to inefficient expert utilization and routing collapse. Existing methods commonly achieve an expe...