2026/8/30 16:26:39
(2024|DeepSeek 北大,Loss-Free Balancing,每专家偏置,QB)MoE 的无辅助损失负载均衡策略
Auxiliary-Loss-Free Load Balancing Strategy for Mixture-of-Experts 论文地址:https://arxiv.org/abs/2408.15664
ICLR 2025:https://openreview.net/forum?idy1iU5czYpE
学术交流群:922230617 目录
1. 引言
2. 背景
2.1 MoE
2.2 用…