MoDES: Accelerating Mixture-of-Experts Multimodal Large Language Models via Dynamic Expert Skipping

TL;DR

MoDES is a training-free framework that adaptively skips experts in multimodal large language models, significantly improving inference efficiency and maintaining high accuracy by considering modality-specific expert contributions.

Contribution

It introduces MoDES, the first adaptive expert skipping method for MoE multimodal LLMs that accounts for heterogeneous expert contributions and modality-specific behaviors.

Findings

Achieves up to 10.67% performance improvement with 88% expert skipping.

Reduces convergence time for threshold setting from days to hours.

Enhances inference speed by 2.16× in prefilling and 1.26× in decoding.

Abstract

Mixture-of-Experts (MoE) Multimodal large language models (MLLMs) excel at vision-language tasks, but they suffer from high computational inefficiency. To reduce inference overhead, expert skipping methods have been proposed to deactivate redundant experts based on the current input tokens. However, we find that applying these methods-originally designed for unimodal large language models (LLMs)-to MLLMs results in considerable performance degradation. This is primarily because such methods fail to account for the heterogeneous contributions of experts across MoE layers and modality-specific behaviors of tokens within these layers. Motivated by these findings, we propose MoDES, the first training-free framework that adaptively skips experts to enable efficient and accurate MoE MLLM inference. It incorporates a globally-modulated local gating (GMLG) mechanism that integrates global layer-wise importance into local routing probabilities to accurately estimate per-token expert importance. A dual-modality thresholding (DMT) method is then applied, which processes tokens from each modality separately, to derive the skipping schedule. To set the optimal thresholds, we introduce a frontier search algorithm that exploits monotonicity properties, cutting convergence time from several days to a few hours. Extensive experiments for 3 model series across 13 benchmarks demonstrate that MoDES far outperforms previous approaches. For instance, when skipping 88% experts for Qwen3-VL-MoE-30B-A3B-Instruct, the performance boost is up to 10.67% (97.33% vs. 86.66%). Furthermore, MoDES significantly enhances inference speed, improving the prefilling time by 2.16$\times$ and the decoding time by 1.26$\times$. Our code is available at https://github.com/ModelTC/MoDES.