MACS: Modality-Aware Capacity Scaling for Efficient Multimodal MoE Inference
For practitioners deploying MoE MLLMs, MACS provides a training-free solution to improve inference efficiency by handling visual token redundancy and varying modality ratios.
MACS addresses the straggler effect in Mixture-of-Experts Multimodal Large Language Models during expert parallelism inference by introducing an entropy-weighted load mechanism and dynamic modality-adaptive capacity, achieving significant efficiency gains across multimodal benchmarks.
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. To address these challenges, we propose MACS (Modality-Aware Capacity Scaling), a training-free inference framework. Specifically, MACS introduces an Entropy-Weighted Load mechanism to quantify the semantic value of visual tokens, addressing information heterogeneity. Additionally, the Dynamic Modality-Adaptive Capacity mechanism allocates expert resources based on the real-time modal composition of the input. Extensive experiments demonstrate that MACS significantly outperforms existing methods on various multimodal benchmarks, providing a novel and robust solution for the efficient deployment of MoE MLLMs in EP inference.