Guardado en:
Detalles Bibliográficos
Autores principales: Tan, Haoyue, Wang, Shengnan, Qiao, Yulin, Zhang, Juncheng, Bai, Youhui, Gong, Ping, Jin, Zewen, Li, Cheng
Formato: Preprint
Publicado: 2026
Materias:
Acceso en línea:https://arxiv.org/abs/2604.18348
Etiquetas: Agregar Etiqueta
Sin Etiquetas, Sea el primero en etiquetar este registro!
_version_ 1866908980395114496
author Tan, Haoyue
Wang, Shengnan
Qiao, Yulin
Zhang, Juncheng
Bai, Youhui
Gong, Ping
Jin, Zewen
Li, Cheng
author_facet Tan, Haoyue
Wang, Shengnan
Qiao, Yulin
Zhang, Juncheng
Bai, Youhui
Gong, Ping
Jin, Zewen
Li, Cheng
contents Video diffusion transformers (DiTs) suffer from prohibitive inference latency due to quadratic attention complexity. Existing sparse attention methods either overlook semantic similarity or fail to adapt to heterogeneous token distributions across layers, leading to model performance degradation. We propose AdaCluster, a training-free adaptive clustering framework that accelerates the generation of DiTs while preserving accuracy. AdaCluster applies an angle-similarity-preserving clustering method to query vectors for higher compression, and designs a euclidean-similarity-preserving clustering method for keys, covering cluster number assignment, threshold-wise adaptive clustering, and efficient critical cluster selection. Experiments on CogVideoX-2B, HunyuanVideo, and Wan-2.1 on one A40 GPU demonstrate up to 1.67-4.31x speedup with negligible quality degradation.
format Preprint
id arxiv_https___arxiv_org_abs_2604_18348
institution arXiv
publishDate 2026
record_format arxiv
spellingShingle AdaCluster: Adaptive Query-Key Clustering for Sparse Attention in Video Generation
Tan, Haoyue
Wang, Shengnan
Qiao, Yulin
Zhang, Juncheng
Bai, Youhui
Gong, Ping
Jin, Zewen
Li, Cheng
Computer Vision and Pattern Recognition
Artificial Intelligence
Video diffusion transformers (DiTs) suffer from prohibitive inference latency due to quadratic attention complexity. Existing sparse attention methods either overlook semantic similarity or fail to adapt to heterogeneous token distributions across layers, leading to model performance degradation. We propose AdaCluster, a training-free adaptive clustering framework that accelerates the generation of DiTs while preserving accuracy. AdaCluster applies an angle-similarity-preserving clustering method to query vectors for higher compression, and designs a euclidean-similarity-preserving clustering method for keys, covering cluster number assignment, threshold-wise adaptive clustering, and efficient critical cluster selection. Experiments on CogVideoX-2B, HunyuanVideo, and Wan-2.1 on one A40 GPU demonstrate up to 1.67-4.31x speedup with negligible quality degradation.
title AdaCluster: Adaptive Query-Key Clustering for Sparse Attention in Video Generation
topic Computer Vision and Pattern Recognition
Artificial Intelligence
url https://arxiv.org/abs/2604.18348