XShare: Collaborative in-Batch Expert Sharing for Faster MoE Inference

TL;DR

XShare introduces a dynamic expert sharing method for MoE models that reduces activation and GPU load, improving inference efficiency without retraining, especially under batching and speculative decoding scenarios.

Contribution

It models batch-aware expert selection as a modular optimization problem and develops greedy algorithms that adaptively maximize gating scores for improved efficiency.

Findings

Reduces expert activation by up to 30% under standard batching.

Cuts peak GPU load by up to 3x in expert-parallel deployments.

Achieves up to 14% throughput gains in speculative decoding.

Abstract

Mixture-of-Experts (MoE) architectures are increasingly used to efficiently scale large language models. However, in production inference, request batching and speculative decoding significantly amplify expert activation, eroding these efficiency benefits. We address this issue by modeling batch-aware expert selection as a modular optimization problem and designing efficient greedy algorithms for different deployment settings. The proposed method, namely XShare, requires no retraining and dynamically adapts to each batch by maximizing the total gating score of selected experts. It reduces expert activation by up to 30% under standard batching, cuts peak GPU load by up to 3x in expert-parallel deployments, and achieves up to 14% throughput gains in speculative decoding via hierarchical, correlation-aware expert selection even if requests in a batch drawn from heterogeneous datasets.