Search papers, labs, and topics across Lattice.
The paper introduces LAER-MoE, a framework for efficient Mixture-of-Experts (MoE) training that addresses load imbalance among experts during expert parallel training. LAER-MoE employs Fully Sharded Expert Parallel (FSEP), partitioning expert parameters across devices and restoring partial experts via All-to-All communication, enabling dynamic re-layout of experts to improve load balancing. Experiments on an A100 cluster demonstrate up to 1.69x speedup compared to existing state-of-the-art MoE training systems.
Achieve 1.69x faster Mixture-of-Experts training by dynamically re-arranging expert parameters to balance load across devices.
Expert parallelism is vital for effectively training Mixture-of-Experts (MoE) models, enabling different devices to host distinct experts, with each device processing different input data. However, during expert parallel training, dynamic routing results in significant load imbalance among experts: a handful of overloaded experts hinder overall iteration, emerging as a training bottleneck. In this paper, we introduce LAER-MoE, an efficient MoE training framework. The core of LAER-MoE is a novel parallel paradigm, Fully Sharded Expert Parallel (FSEP), which fully partitions each expert parameter by the number of devices and restores partial experts at expert granularity through All-to-All communication during training. This allows for flexible re-layout of expert parameters during training to enhance load balancing. In particular, we perform fine-grained scheduling of communication operations to minimize communication overhead. Additionally, we develop a load balancing planner to formulate re-layout strategies of experts and routing schemes for tokens during training. We perform experiments on an A100 cluster, and the results indicate that our system achieves up to 1.69x acceleration compared to the current state-of-the-art training systems. Source code available at https://github.com/PKU-DAIR/Hetu-Galvatron/tree/laer-moe.
