HierMoE: Accelerating MoE Training with Hierarchical Token Deduplication and Expert Swap.
Wenxiang Lin, Xinglin Pan, Lin Zhang, Shaohuai Shi, Xuan Wang, Xiaowen Chu
Browse the full INFOCOM paper archive.
Wenxiang Lin, Xinglin Pan, Lin Zhang, Shaohuai Shi, Xuan Wang, Xiaowen Chu
Browse the full INFOCOM paper archive.