Efficient Inference for Large Vision-Language Models: Bottlenecks, Techniques, and Prospects

TL;DR

This paper systematically analyzes the efficiency bottlenecks in large vision-language models during inference, proposing a structured taxonomy of techniques and outlining future research directions.

Contribution

It provides the first end-to-end analysis of inference bottlenecks in LVLMs, integrating various optimization strategies and proposing a comprehensive taxonomy.

Findings

Identifies visual token dominance as a key efficiency barrier.

Analyzes how upstream decisions affect downstream bottlenecks.

Outlines future research directions with empirical insights.

Abstract

Large Vision-Language Models (LVLMs) enable sophisticated reasoning over images and videos, yet their inference is hindered by a systemic efficiency barrier known as visual token dominance. This overhead is driven by a multi-regime interplay between high-resolution feature extraction, quadratic attention scaling, and memory bandwidth constraints. We present a systematic taxonomy of efficiency techniques structured around the inference lifecycle, consisting of encoding, prefilling, and decoding. Unlike prior reviews focused on isolated optimizations, we analyze the end-to-end pipeline to reveal how upstream decisions dictate downstream bottlenecks, covering compute-bound visual encoding, the intensive prefilling of massive contexts, and the ''visual memory wall'' in bandwidth-bound decoding. By decoupling the efficiency landscape into the axes of shaping information density, managing long-context attention, and overcoming memory limits, this work provides a structured analysis of how isolated optimizations compose to navigate the trade-off between visual fidelity and system efficiency. The survey concludes by outlining four future frontiers supported by pilot empirical insights, including hybrid compression based on functional unit sensitivity, modality-aware decoding with relaxed verification, progressive state management for streaming continuity, and stage-disaggregated serving through hardware-algorithm co-design. Our literature repository is at https://github.com/SuDIS-ZJU/Efficient-LVLMs-Inference.