TA-MoE: Topology-Aware Large Scale Mixture-of-Expert Training
This work addresses a bottleneck in scaling MoE models for AI researchers and practitioners by optimizing training efficiency in heterogeneous network environments, representing an incremental advancement.
The paper tackles the problem of inefficient dispatch patterns in large-scale Mixture-of-Expert (MoE) training by proposing TA-MoE, a topology-aware routing strategy that dynamically adjusts dispatch based on network topology, resulting in performance improvements of 1.01x-4.77x over existing methods.
Sparsely gated Mixture-of-Expert (MoE) has demonstrated its effectiveness in scaling up deep neural networks to an extreme scale. Despite that numerous efforts have been made to improve the performance of MoE from the model design or system optimization perspective, existing MoE dispatch patterns are still not able to fully exploit the underlying heterogeneous network environments. In this paper, we propose TA-MoE, a topology-aware routing strategy for large-scale MoE trainging, from a model-system co-design perspective, which can dynamically adjust the MoE dispatch pattern according to the network topology. Based on communication modeling, we abstract the dispatch problem into an optimization objective and obtain the approximate dispatch pattern under different topologies. On top of that, we design a topology-aware auxiliary loss, which can adaptively route the data to fit in the underlying topology without sacrificing the model accuracy. Experiments show that TA-MoE can substantially outperform its counterparts on various hardware and model configurations, with roughly 1.01x-1.61x, 1.01x-4.77x, 1.25x-1.54x improvements over the popular DeepSpeed-MoE, FastMoE and FasterMoE.