Similarity-Aware Token Pruning: Your VLM but Faster

Fuente: arXiv
Guardado en:
Detalles Bibliográficos
Autores principales: Jeddi, Ahmadreza, Baghbanzadeh, Negin, Dolatabadi, Elham, Taati, Babak
Formato: Preprint
Publicado: 2025
Materias:
Acceso en línea:
Etiquetas: Agregar Etiqueta
Sin Etiquetas, Sea el primero en etiquetar este registro!
_version_ 1866915198457085952
author Jeddi, Ahmadreza
Baghbanzadeh, Negin
Dolatabadi, Elham
Taati, Babak
author_facet Jeddi, Ahmadreza
Baghbanzadeh, Negin
Dolatabadi, Elham
Taati, Babak
contents The computational demands of Vision Transformers (ViTs) and Vision-Language Models (VLMs) remain a significant challenge due to the quadratic complexity of self-attention. While token pruning offers a promising solution, existing methods often introduce training overhead or fail to adapt dynamically across layers. We present SAINT, a training-free token pruning framework that leverages token similarity and a graph-based formulation to dynamically optimize pruning rates and redundancy thresholds. Through systematic analysis, we identify a universal three-stage token evolution process (aligner-explorer-aggregator) in transformers, enabling aggressive pruning in early stages without sacrificing critical information. For ViTs, SAINT doubles the throughput of ViT-H/14 at 224px with only 0.6% accuracy loss on ImageNet-1K, surpassing the closest competitor by 0.8%. For VLMs, we apply SAINT in three modes: ViT-only, LLM-only, and hybrid. SAINT reduces LLaVA-13B's tokens by 75%, achieving latency comparable to LLaVA-7B with less than 1% performance loss across benchmarks. Our work establishes a unified, practical framework for efficient inference in ViTs and VLMs.
format Preprint
id arxiv_https___arxiv_org_abs_2503_11549
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle Similarity-Aware Token Pruning: Your VLM but Faster
Jeddi, Ahmadreza
Baghbanzadeh, Negin
Dolatabadi, Elham
Taati, Babak
Computer Vision and Pattern Recognition
The computational demands of Vision Transformers (ViTs) and Vision-Language Models (VLMs) remain a significant challenge due to the quadratic complexity of self-attention. While token pruning offers a promising solution, existing methods often introduce training overhead or fail to adapt dynamically across layers. We present SAINT, a training-free token pruning framework that leverages token similarity and a graph-based formulation to dynamically optimize pruning rates and redundancy thresholds. Through systematic analysis, we identify a universal three-stage token evolution process (aligner-explorer-aggregator) in transformers, enabling aggressive pruning in early stages without sacrificing critical information. For ViTs, SAINT doubles the throughput of ViT-H/14 at 224px with only 0.6% accuracy loss on ImageNet-1K, surpassing the closest competitor by 0.8%. For VLMs, we apply SAINT in three modes: ViT-only, LLM-only, and hybrid. SAINT reduces LLaVA-13B's tokens by 75%, achieving latency comparable to LLaVA-7B with less than 1% performance loss across benchmarks. Our work establishes a unified, practical framework for efficient inference in ViTs and VLMs.
title Similarity-Aware Token Pruning: Your VLM but Faster
topic Computer Vision and Pattern Recognition
url https://arxiv.org/abs/2503.11549