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. T o 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.

Mixture of experts (MoE) is a popular technique to improve capacity of large models with conditionally-activated parallel neural network modules (experts). Due to its remarkable scaling performance with sparse computation, it is widely used in modern Large Language Models (LLMs) and Large Vision Models (LVMs). However, serving such large models on edge devices is challenging due to memory constraints. Typical solutions like memory swapping or weight pruning may lead to significantly higher latency or severe accuracy loss. In this paper, we introduce SwapMoE, a framework for efficient continuous MoE-based large models serving with tunable memory budgets. The main idea of SwapMoE is to keep a small dynamic set of important experts, namely Virtual Experts, in the main memory for inference, while seamlessly maintaining how the Virtual Experts map to the actual experts. We use a profiling-guided planner to allocate the resources for SwapMoE that can fully utilize the memory budgets and bandwidth, and an importance-aware scheduler to efficiently identify, update, and use the Virtual Experts for accurate inference. To evaluate SwapMoE, we conduct experiments on multiple edge devices with state-of-the-art MoE-based Large Language Models and Large Vision Models. The results demonstrate remarkable performance of SwapMoE under various memory constraints. Specifically, SwapMoE can enable running large MoE models under tight memory budgets with similar latency to pruned compact models, while with significantly higher accuracy.

By now, you must have heard the term "virtual assistants." The natural evolution of virtual assistants is the virtual experts. Going from former to the latter is a substantial technical challenge not many companies are willing to meet yet, because the simpler "assistant" version has untapped commercial potential and a quick ROI. Nevertheless, virtual experts is the real game changer – a paradigm shift – that will have social and economic impact beyond our wildest imagination. The rise of virtual experts is just hiding behind the puzzle of the most effective machine learning approach.