GRACE-MoE: Grouping and Replication with Locality-Aware Routing for Efficient Distributed MoE Inference
For practitioners deploying large SMoE models, GRACE-MoE addresses the critical trade-off between communication overhead and load imbalance in distributed inference.
GRACE-MoE reduces end-to-end inference latency for distributed SMoE models by up to 4.66x over existing systems through a lossless co-optimization of expert grouping, dynamic replication, and locality-aware routing.
Sparse Mixture of Experts (SMoE) enables scalable parameter growth in large language models (LLMs) by selectively activating a subset of experts, and its large parameter count necessitates distributed deployment for inference. However, distributed inference faces a critical dilemma: although communication overhead constitutes the primary bottleneck, reducing it often exacerbates computational load imbalance, leading to resource waste. In this paper, we present GRACE-MoE, which stands for Grouping and Replication with Locality-Aware Routing for SMoE inference. GRACE-MoE is a lossless co-optimization framework that integrates expert grouping to reduce communication and dynamic replication to correct load skew, together with locality-aware routing to resolve replica selection. To underpin this coordinated optimization in multi-node settings, GRACE-MoE adopts a hierarchical sparse communication design that reduces cross-node traffic while implicitly aligning execution across nodes, thereby mitigating synchronization overhead. Experiments on diverse models and multi-node, multi-GPU environments demonstrate that GRACE-MoE efficiently reduces end-to-end inference latency, achieving up to 4.66x speedup over existing systems, and the code will be released upon acceptance.