HierMoE: Accelerating MoE Training with Hierarchical Token Deduplication and Expert Swap

Wenxiang Lin, Xinglin Pan, Lin Zhang 0059, Shaohuai Shi, Xuan Wang 0002, Xiaowen Chu 0001. HierMoE: Accelerating MoE Training with Hierarchical Token Deduplication and Expert Swap. In IEEE INFOCOM 2026 - IEEE Conference on Computer Communications, Tokyo, Japan, May 18-21, 2026. pages 1-10, IEEE, 2026. [doi]

Authors

Wenxiang Lin

This author has not been identified. Look up 'Wenxiang Lin' in Google

Xinglin Pan

This author has not been identified. Look up 'Xinglin Pan' in Google

Lin Zhang 0059

This author has not been identified. Look up 'Lin Zhang 0059' in Google

Shaohuai Shi

This author has not been identified. Look up 'Shaohuai Shi' in Google

Xuan Wang 0002

This author has not been identified. Look up 'Xuan Wang 0002' in Google

Xiaowen Chu 0001

This author has not been identified. Look up 'Xiaowen Chu 0001' in Google