AnyExperts: On-Demand Expert Allocation for Multimodal Language Models with Mixture of Expert

TL;DR

AnyExperts introduces a dynamic, importance-aware expert routing framework for multimodal MoE models, optimizing resource allocation and maintaining high performance across vision, audio, and NLP tasks.

Contribution

It proposes a novel on-demand, budget-aware routing strategy that adaptively allocates real and virtual experts based on semantic importance, improving efficiency.

Findings

Achieves 40% fewer real expert activations on image/video tasks.

Maintains performance while reducing real expert usage by 10% on text-dense tasks.

Enhances efficiency and effectiveness of multimodal MoE models.

Abstract

Multimodal Mixture-of-Experts (MoE) models offer a promising path toward scalable and efficient large vision-language systems. However, existing approaches rely on rigid routing strategies (typically activating a fixed number of experts per token) ignoring the inherent heterogeneity in semantic importance across modalities. This leads to suboptimal compute allocation, where redundant tokens consume as many resources as critical ones. To address this, we propose AnyExperts, a novel on-demand, budget-aware dynamic routing framework that allocates a variable total number of expert slots per token based on its semantic importance. Crucially, to prevent uncontrolled compute growth, the total slots per token are constrained within a fixed range, and each slot is filled by either a real expert or a virtual expert, with the virtual share capped at a small maximum (e.g., 20%). The model then adaptively balances the real-to-virtual ratio per token, assigning more real experts to semantically rich regions and relying more on virtual experts for redundant content. Evaluated across diverse tasks in visual understanding, audio understanding, and NLP understanding, AnyExperts improves performance under the same compute budget. Notably, on general image/video tasks, it achieves comparable accuracy with 40% fewer real expert activations; on text-dense tasks (OCR and NLP), it maintains performance while reducing real expert usage by 10%. These results demonstrate that fine-grained, importance-driven expert allocation significantly enhances both the efficiency and effectiveness of multimodal MoE models.