VecAttention: Vector-wise Sparse Attention for Accelerating Long Context Inference

Fuente: arXiv
Guardado en:
Detalles Bibliográficos
Autores principales: Liu, Anmin, Yang, Ruixuan, Jiang, Huiqiang, Lin, Bin, Sun, Minmin, Li, Yong, Zhang, Chen, Xie, Tao
Formato: Preprint
Publicado: 2026
Materias:
Acceso en línea:
Etiquetas: Agregar Etiqueta
Sin Etiquetas, Sea el primero en etiquetar este registro!
_version_ 1866915901376299008
author Liu, Anmin
Yang, Ruixuan
Jiang, Huiqiang
Lin, Bin
Sun, Minmin
Li, Yong
Zhang, Chen
Xie, Tao
author_facet Liu, Anmin
Yang, Ruixuan
Jiang, Huiqiang
Lin, Bin
Sun, Minmin
Li, Yong
Zhang, Chen
Xie, Tao
contents Long-context video understanding and generation pose a significant computational challenge for Transformer-based video models due to the quadratic complexity of self-attention. While existing sparse attention methods employ coarse-grained patterns to improve efficiency, they typically incur redundant computation and suboptimal performance. To address this issue, in this paper, we propose \textbf{VecAttention}, a novel framework of vector-wise sparse attention that achieves superior accuracy-efficiency trade-offs for video models. We observe that video attention maps exhibit a strong vertical-vector sparse pattern, and further demonstrate that this vertical-vector pattern offers consistently better accuracy-sparsity trade-offs compared with existing coarse-grained sparse patterns. Based on this observation, VecAttention dynamically selects and processes only informative vertical vectors through a lightweight important-vector selection that minimizes memory access overhead and an optimized kernel of vector sparse attention. Comprehensive evaluations on video understanding (VideoMME, LongVideoBench, and VCRBench) and generation (VBench) tasks show that VecAttention delivers a 2.65$\times$ speedup over full attention and a 1.83$\times$ speedup over state-of-the-art sparse attention methods, with comparable accuracy to full attention. Our code is available at https://github.com/anminliu/VecAttention.
format Preprint
id arxiv_https___arxiv_org_abs_2603_29494
institution arXiv
publishDate 2026
record_format arxiv
spellingShingle VecAttention: Vector-wise Sparse Attention for Accelerating Long Context Inference
Liu, Anmin
Yang, Ruixuan
Jiang, Huiqiang
Lin, Bin
Sun, Minmin
Li, Yong
Zhang, Chen
Xie, Tao
Computer Vision and Pattern Recognition
Long-context video understanding and generation pose a significant computational challenge for Transformer-based video models due to the quadratic complexity of self-attention. While existing sparse attention methods employ coarse-grained patterns to improve efficiency, they typically incur redundant computation and suboptimal performance. To address this issue, in this paper, we propose \textbf{VecAttention}, a novel framework of vector-wise sparse attention that achieves superior accuracy-efficiency trade-offs for video models. We observe that video attention maps exhibit a strong vertical-vector sparse pattern, and further demonstrate that this vertical-vector pattern offers consistently better accuracy-sparsity trade-offs compared with existing coarse-grained sparse patterns. Based on this observation, VecAttention dynamically selects and processes only informative vertical vectors through a lightweight important-vector selection that minimizes memory access overhead and an optimized kernel of vector sparse attention. Comprehensive evaluations on video understanding (VideoMME, LongVideoBench, and VCRBench) and generation (VBench) tasks show that VecAttention delivers a 2.65$\times$ speedup over full attention and a 1.83$\times$ speedup over state-of-the-art sparse attention methods, with comparable accuracy to full attention. Our code is available at https://github.com/anminliu/VecAttention.
title VecAttention: Vector-wise Sparse Attention for Accelerating Long Context Inference
topic Computer Vision and Pattern Recognition
url https://arxiv.org/abs/2603.29494