Accelerating Prefilling for Long-Context LLMs via Sparse Pattern Sharing

Fuente: arXiv
Enregistré dans:
Détails bibliographiques
Auteurs principaux: Peng, Dan, Fu, Zhihui, Ye, Zewen, Song, Zhuoran, Wang, Jun
Format: Preprint
Publié: 2025
Sujets:
Accès en ligne:
Tags: Ajouter un tag
Pas de tags, Soyez le premier à ajouter un tag!
_version_ 1866908380038168576
author Peng, Dan
Fu, Zhihui
Ye, Zewen
Song, Zhuoran
Wang, Jun
author_facet Peng, Dan
Fu, Zhihui
Ye, Zewen
Song, Zhuoran
Wang, Jun
contents Sparse attention methods exploit the inherent sparsity in attention to speed up the prefilling phase of long-context inference, mitigating the quadratic complexity of full attention computation. While existing sparse attention methods rely on predefined patterns or inaccurate estimations to approximate attention behavior, they often fail to fully capture the true dynamics of attention, resulting in reduced efficiency and compromised accuracy. Instead, we propose a highly accurate sparse attention mechanism that shares similar yet precise attention patterns across heads, enabling a more realistic capture of the dynamic behavior of attention. Our approach is grounded in two key observations: (1) attention patterns demonstrate strong inter-head similarity, and (2) this similarity remains remarkably consistent across diverse inputs. By strategically sharing computed accurate patterns across attention heads, our method effectively captures actual patterns while requiring full attention computation for only a small subset of heads. Comprehensive evaluations demonstrate that our approach achieves superior or comparable speedup relative to state-of-the-art methods while delivering the best overall accuracy.
format Preprint
id arxiv_https___arxiv_org_abs_2505_19578
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle Accelerating Prefilling for Long-Context LLMs via Sparse Pattern Sharing
Peng, Dan
Fu, Zhihui
Ye, Zewen
Song, Zhuoran
Wang, Jun
Machine Learning
Artificial Intelligence
Computation and Language
Sparse attention methods exploit the inherent sparsity in attention to speed up the prefilling phase of long-context inference, mitigating the quadratic complexity of full attention computation. While existing sparse attention methods rely on predefined patterns or inaccurate estimations to approximate attention behavior, they often fail to fully capture the true dynamics of attention, resulting in reduced efficiency and compromised accuracy. Instead, we propose a highly accurate sparse attention mechanism that shares similar yet precise attention patterns across heads, enabling a more realistic capture of the dynamic behavior of attention. Our approach is grounded in two key observations: (1) attention patterns demonstrate strong inter-head similarity, and (2) this similarity remains remarkably consistent across diverse inputs. By strategically sharing computed accurate patterns across attention heads, our method effectively captures actual patterns while requiring full attention computation for only a small subset of heads. Comprehensive evaluations demonstrate that our approach achieves superior or comparable speedup relative to state-of-the-art methods while delivering the best overall accuracy.
title Accelerating Prefilling for Long-Context LLMs via Sparse Pattern Sharing
topic Machine Learning
Artificial Intelligence
Computation and Language
url https://arxiv.org/abs/2505.19578