Towards Efficient Large Vision-Language Models: A Comprehensive Survey on Inference Strategies

TL;DR

This survey reviews recent techniques for improving the inference efficiency of large vision-language models, focusing on visual token compression, memory management, architecture, and decoding strategies.

Contribution

It provides a systematic taxonomy of current optimization methods and critically analyzes their limitations and open challenges.

Findings

Categorizes inference optimization techniques into four main groups.

Highlights limitations and open problems in current methodologies.

Provides insights to guide future research in efficient multimodal systems.

Abstract

Although Large Vision Language Models (LVLMs) have demonstrated impressive multimodal reasoning capabilities, their scalability and deployment are constrained by massive computational requirements. In particular, the massive amount of visual tokens from high-resolution input data aggravates the situation due to the quadratic complexity of attention mechanisms. To address these issues, the research community has developed several optimization frameworks. This paper presents a comprehensive survey of the current state-of-the-art techniques for accelerating LVLM inference. We introduce a systematic taxonomy that categorizes existing optimization frameworks into four primary dimensions: visual token compression, memory management and serving, efficient architectural design, and advanced decoding strategies. Furthermore, we critically examine the limitations of these current methodologies and identify critical open problems to inspire future research directions in efficient multimodal systems.