FSMoE co-schedules intra-node and inter-node communication with expert computation and adaptively partitions gradients, achieving 1.18x-1.22x speedups over Tutel and up to 3.01x over DeepSpeed-MoE in MoE training.
Centauri: Enabling efficient sched- uling for communication-computation overlap in large model train- ing via communication partitioning
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
citation-role summary
background 1
citation-polarity summary
fields
cs.LG 1years
2025 1verdicts
CONDITIONAL 1roles
background 1polarities
background 1representative citing papers
citing papers explorer
-
FSMoE: A Flexible and Scalable Training System for Sparse Mixture-of-Experts Models
FSMoE co-schedules intra-node and inter-node communication with expert computation and adaptively partitions gradients, achieving 1.18x-1.22x speedups over Tutel and up to 3.01x over DeepSpeed-MoE in MoE training.