Streamline Without Sacrifice -- Squeeze out Computation Redundancy in LMM

TL;DR

This paper introduces ProxyV, a method to reduce computation in large multimodal models by processing vision tokens more lightly, achieving efficiency gains without performance loss.

Contribution

The paper identifies computation-level redundancy in vision tokens and proposes ProxyV, a novel approach that uses proxy vision tokens to reduce computational load in LMMs.

Findings

ProxyV reduces visual token computation without performance loss

Combining ProxyV with token reduction methods further improves efficiency

ProxyV can lead to performance gains in certain scenarios

Abstract

Large multimodal models excel in multimodal tasks but face significant computational challenges due to excessive computation on visual tokens. Unlike token reduction methods that focus on token-level redundancy, we identify and study the computation-level redundancy on vision tokens to ensure no information loss. Our key insight is that vision tokens from the pretrained vision encoder do not necessarily require all the heavy operations (e.g., self-attention, FFNs) in decoder-only LMMs and could be processed more lightly with proper designs. We designed a series of experiments to discover and progressively squeeze out the vision-related computation redundancy. Based on our findings, we propose ProxyV, a novel approach that utilizes proxy vision tokens to alleviate the computational burden on original vision tokens. ProxyV enhances efficiency without compromising performance and can even yield notable performance gains in scenarios with more moderate efficiency improvements. Furthermore, the flexibility of ProxyV is demonstrated through its combination with token reduction methods to boost efficiency further. The code will be made public at this https://github.com/penghao-wu/ProxyV URL.