Understanding and Harnessing Sparsity in Unified Multimodal Models

TL;DR

This paper systematically analyzes the sparsity and compressibility of components in unified multimodal models, revealing that understanding modules are highly compressible while generation modules are sensitive, and proposes a Mixture-of-Experts adaptation to improve efficiency.

Contribution

It provides the first systematic analysis of component-wise sparsity in unified multimodal models and introduces a Mixture-of-Experts adaptation to enhance inference efficiency.

Findings

Understanding components are highly compressible in both tasks.

Generation components are sensitive to compression, performance drops with moderate pruning.

Mixture-of-Experts adaptation achieves comparable performance with only half the parameters activated.

Abstract

Large multimodal models have achieved remarkable progress in both understanding and generation. Recent efforts pursue unified multimodal models that integrate heterogeneous components to support both capabilities within a single framework. However, such unification introduces inference inefficiencies, e.g., specific tasks or samples may not require the full knowledge or capacity of the unified model. Yet, a systematic understanding of how these inefficiencies manifest across different components remains limited. In this work, we first conduct a systematic analysis of unified multimodal model components using training-free pruning as a probing methodology, considering both depth pruning and width reduction. Our study reveals that the understanding component exhibits notable compressibility in both understanding and generation tasks, which is more pronounced in the latter. In contrast, the generation components are highly sensitive to compression, with performance deteriorating sharply even under moderate compression ratios. To address this limitation, we propose the Mixture-of-Experts (MoE) Adaptation, inspired by the dynamic activation patterns observed across different samples. This approach partitions the generation module into multiple experts and enables sparse activation to restore generation quality. We validate the effectiveness of sparse activation through expert-frozen tuning and further demonstrate that a fully trainable adaptation delivers additional gains. As a result, the adapted BAGEL model achieves performance comparable to the full model while activating only about half of its parameters. The code is released at \href{https://github.com/Shwai-He/SparseUnifiedModel}{this link}.