HiPrune: Training-Free Visual Token Pruning via Hierarchical Attention in Vision-Language Models
This addresses efficiency issues for users of VLMs by enabling faster inference without retraining, though it is incremental as it builds on prior token pruning methods.
The paper tackles the problem of excessive computational overhead in Vision-Language Models (VLMs) due to lengthy visual token sequences by proposing HiPrune, a training-free token pruning framework that preserves up to 99.3% task accuracy with only 33.3% tokens and reduces inference FLOPs and latency by up to 9×.
Vision-Language Models (VLMs) encode images into lengthy sequences of visual tokens, leading to excessive computational overhead and limited inference efficiency. While prior efforts prune or merge tokens to address this issue, they often rely on special tokens (e.g., CLS) or require task-specific training, hindering scalability across architectures. In this paper, we propose HiPrune, a training-free and model-agnostic token Pruning framework that exploits the Hierarchical attention structure within vision encoders. We identify that middle layers attend to object-centric regions, while deep layers capture global contextual features. Based on this observation, HiPrune selects three types of informative tokens: (1) Anchor tokens with high attention in object-centric layers, (2) Buffer tokens adjacent to anchors for spatial continuity, and (3) Register tokens with strong attention in deep layers for global summarization. Our method requires no retraining and integrates seamlessly with any ViT-based VLM. Extensive experiments on LLaVA-1.5, LLaVA-NeXT, and Qwen2.5-VL demonstrate that HiPrune achieves state-of-the-art pruning performance, preserving up to 99.3% task accuracy with only 33.3% tokens, and maintaining 99.5% accuracy with just 11.1% tokens. Meanwhile, it reduces inference FLOPs and latency by up to 9$\times$, showcasing strong generalization across models and tasks. Code is available at https://github.com/Danielement321/HiPrune.