AI RESEARCH
MACS: Modality-Aware Capacity Scaling for Efficient Multimodal MoE Inference
arXiv CS.LG
•
ArXi:2605.05225v1 Announce Type: new Mixture-of-Experts Multimodal Large Language Models (MoE MLLMs) suffer from a significant efficiency bottleneck during Expert Parallelism (EP) inference due to the straggler effect. This issue is worsened in the multimodal context, as existing token-count-based load balancing methods fail to address two unique challenges: (1) Information Heterogeneity, where numerous redundant visual tokens are treated equally to semantically critical ones, and (2) Modality Dynamics, where varying visual to text ratios across tasks lead to resource misallocation.