LLMC+: Benchmarking Vision-Language Model Compression with a Plug-and-play Toolkit

TL;DR

LLMC+ is a comprehensive benchmark and toolkit for evaluating and combining various vision-language model compression techniques, addressing current limitations and promoting fair, realistic assessments of model efficiency.

Contribution

Introduces LLMC+, a versatile benchmark with a plug-and-play toolkit supporting over 20 algorithms for systematic VLM compression evaluation.

Findings

Spatial and temporal redundancies require different strategies.

Token reduction impacts multi-turn and detail-sensitive tasks.

Combining token and model compression yields high efficiency with minimal performance loss.

Abstract

Large Vision-Language Models (VLMs) exhibit impressive multi-modal capabilities but suffer from prohibitive computational and memory demands, due to their long visual token sequences and massive parameter sizes. To address these issues, recent works have proposed training-free compression methods. However, existing efforts often suffer from three major limitations: (1) Current approaches do not decompose techniques into comparable modules, hindering fair evaluation across spatial and temporal redundancy. (2) Evaluation confined to simple single-turn tasks, failing to reflect performance in realistic scenarios. (3) Isolated use of individual compression techniques, without exploring their joint potential. To overcome these gaps, we introduce LLMC+, a comprehensive VLM compression benchmark with a versatile, plug-and-play toolkit. LLMC+ supports over 20 algorithms across five representative VLM families and enables systematic study of token-level and model-level compression. Our benchmark reveals that: (1) Spatial and temporal redundancies demand distinct technical strategies. (2) Token reduction methods degrade significantly in multi-turn dialogue and detail-sensitive tasks. (3) Combining token and model compression achieves extreme compression with minimal performance loss. We believe LLMC+ will facilitate fair evaluation and inspire future research in efficient VLM. Our code is available at https://github.com/ModelTC/LightCompress.