VisMMOE: Exploiting Visual-Expert Affinity for Efficient Visual-Language MoE Offloading

Fuente: arXiv
Guardado en:
Detalles Bibliográficos
Autores principales: Xu, Cheng, Hou, Xiaofeng, Liu, Jiacheng, Li, Chao
Formato: Preprint
Publicado: 2026
Materias:
Acceso en línea:
Etiquetas: Agregar Etiqueta
Sin Etiquetas, Sea el primero en etiquetar este registro!
_version_ 1866914537752494080
author Xu, Cheng
Hou, Xiaofeng
Liu, Jiacheng
Li, Chao
author_facet Xu, Cheng
Hou, Xiaofeng
Liu, Jiacheng
Li, Chao
contents Large-scale vision-language mixture-of-experts (VL-MoE) models provide strong multimodal capability, but efficient deployment on memory-constrained platforms remains difficult. Existing MoE offloading systems are largely designed for text-centric workloads and become much less effective for visual-heavy inputs, where large numbers of visual tokens induce broader and less predictable expert accesses. We present VisMMoE, a VL-MoE offloading system built on a single systems insight: pruning redundant visual tokens can improve offloading not only by reducing computation, but also by reshaping expert demand. We refer to this effect as \textit{visual-expert affinity}: token pruning makes expert accesses more concentrated within layers and more stable across layers, producing a smaller and more predictable expert working set. Guided by this insight, VisMMoE combines affinity-aware token compression, lookahead expert prediction, and cache/pipeline orchestration to improve expert locality and prefetch effectiveness under tight memory budgets. We implement VisMMoE on multiple frameworks and evaluate it on representative VL-MoE models and benchmarks. VisMMoE improves end-to-end inference performance by up to 2.68x and 1.61x, respectively, over strong baselines for today's VL-MoE deployments while maintaining competitive accuracy.
format Preprint
id arxiv_https___arxiv_org_abs_2605_05899
institution arXiv
publishDate 2026
record_format arxiv
spellingShingle VisMMOE: Exploiting Visual-Expert Affinity for Efficient Visual-Language MoE Offloading
Xu, Cheng
Hou, Xiaofeng
Liu, Jiacheng
Li, Chao
Machine Learning
Large-scale vision-language mixture-of-experts (VL-MoE) models provide strong multimodal capability, but efficient deployment on memory-constrained platforms remains difficult. Existing MoE offloading systems are largely designed for text-centric workloads and become much less effective for visual-heavy inputs, where large numbers of visual tokens induce broader and less predictable expert accesses. We present VisMMoE, a VL-MoE offloading system built on a single systems insight: pruning redundant visual tokens can improve offloading not only by reducing computation, but also by reshaping expert demand. We refer to this effect as \textit{visual-expert affinity}: token pruning makes expert accesses more concentrated within layers and more stable across layers, producing a smaller and more predictable expert working set. Guided by this insight, VisMMoE combines affinity-aware token compression, lookahead expert prediction, and cache/pipeline orchestration to improve expert locality and prefetch effectiveness under tight memory budgets. We implement VisMMoE on multiple frameworks and evaluate it on representative VL-MoE models and benchmarks. VisMMoE improves end-to-end inference performance by up to 2.68x and 1.61x, respectively, over strong baselines for today's VL-MoE deployments while maintaining competitive accuracy.
title VisMMOE: Exploiting Visual-Expert Affinity for Efficient Visual-Language MoE Offloading
topic Machine Learning
url https://arxiv.org/abs/2605.05899