HieraSparse: Hierarchical Semi-Structured Sparse KV Attention

Fuente: arXiv
Enregistré dans:
Détails bibliographiques
Auteurs principaux: Wang, Haoxuan, Wang, Chen
Format: Preprint
Publié: 2026
Sujets:
Accès en ligne:
Tags: Ajouter un tag
Pas de tags, Soyez le premier à ajouter un tag!
_version_ 1866913042333171712
author Wang, Haoxuan
Wang, Chen
author_facet Wang, Haoxuan
Wang, Chen
contents The deployment of long-context Large Language Models (LLMs) poses significant challenges due to the intense computational cost of self-attention and the substantial memory overhead of the Key-Value Cache (KV Cache). In this paper, we introduce HieraSparse, a hierarchical KV Cache compression framework with acceleration kernels that leverage GPU sparse tensor cores to speed up semi-structured KV Cache attention for both the prefill and decode phases. With the hierarchical design, our method allows for a flexible quality-sparsity trade-off and successfully converts sparsity into efficiency. Compared to the state-of-the-art decode method that utilizes unstructured sparsity, HieraSparse achieves $\mathbf{1.2\times}$ KV compression ratio and $\mathbf{4.57\times}$ attention speedup at the same sparsity level. Furthermore, we extended the semi-structured KV Cache pruning to the prefill stage, which demonstrated up to $\mathbf{1.85\times}$ attention speedup at the highest sparsity. Lastly, we evaluate the generation quality of HieraSparse with a simple magnitude-based pruning method, and the results show that $\mathbf{1.37\times}$ prefill speedup and $\mathbf{1.77\times}$ decode speedup can be achieved without significant quality drop. The codebase can be found at https://github.com/psl-ntu/HieraSparse.
format Preprint
id arxiv_https___arxiv_org_abs_2604_16864
institution arXiv
publishDate 2026
record_format arxiv
spellingShingle HieraSparse: Hierarchical Semi-Structured Sparse KV Attention
Wang, Haoxuan
Wang, Chen
Distributed, Parallel, and Cluster Computing
Hardware Architecture
The deployment of long-context Large Language Models (LLMs) poses significant challenges due to the intense computational cost of self-attention and the substantial memory overhead of the Key-Value Cache (KV Cache). In this paper, we introduce HieraSparse, a hierarchical KV Cache compression framework with acceleration kernels that leverage GPU sparse tensor cores to speed up semi-structured KV Cache attention for both the prefill and decode phases. With the hierarchical design, our method allows for a flexible quality-sparsity trade-off and successfully converts sparsity into efficiency. Compared to the state-of-the-art decode method that utilizes unstructured sparsity, HieraSparse achieves $\mathbf{1.2\times}$ KV compression ratio and $\mathbf{4.57\times}$ attention speedup at the same sparsity level. Furthermore, we extended the semi-structured KV Cache pruning to the prefill stage, which demonstrated up to $\mathbf{1.85\times}$ attention speedup at the highest sparsity. Lastly, we evaluate the generation quality of HieraSparse with a simple magnitude-based pruning method, and the results show that $\mathbf{1.37\times}$ prefill speedup and $\mathbf{1.77\times}$ decode speedup can be achieved without significant quality drop. The codebase can be found at https://github.com/psl-ntu/HieraSparse.
title HieraSparse: Hierarchical Semi-Structured Sparse KV Attention
topic Distributed, Parallel, and Cluster Computing
Hardware Architecture
url https://arxiv.org/abs/2604.16864