Saved in:
Bibliographic Details
Main Authors: Liu, Siran, Cao, Zane, He, Yongchao
Format: Preprint
Published: 2025
Subjects:
Online Access:https://arxiv.org/abs/2512.14082
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866917148398452736
author Liu, Siran
Cao, Zane
He, Yongchao
author_facet Liu, Siran
Cao, Zane
He, Yongchao
contents Efficient long-context understanding and reasoning are increasingly vital for large language model (LLM) applications such as multi-turn dialogue and program analysis. However, the core self-attention mechanism scales quadratically with sequence length, creating a fundamental computational bottleneck. Existing sparse attention methods alleviate this issue but face trade-offs: training-based methods are costly and cannot be directly applied as acceleration plugins for other models, while inference-time methods often compromise efficiency or cross-modal generality. To address these limitations, we present UniSparse, a unified mechanism that introduces the notion of composite tokens--compact representations that aggregate multi-granularity contextual information. Building on this abstraction, UniSparse dynamically constructs sparse attention through multi-granularity compression and block-level selection, enabling efficient and hardware-friendly execution on GPU. Across multiple modalities and tasks ranging from synthetic benchmarks to real-world applications, UniSparse consistently surpasses state-of-the-art sparse attention methods (e.g., MInference, XAttention, FlexPrefill) in both accuracy and efficiency, achieving $\ge$ 99% of full-attention accuracy and up to 2.61$\times$ faster attention computation than FlashAttention.
format Preprint
id arxiv_https___arxiv_org_abs_2512_14082
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle A Unified Sparse Attention via Multi-Granularity Compression
Liu, Siran
Cao, Zane
He, Yongchao
Computation and Language
Efficient long-context understanding and reasoning are increasingly vital for large language model (LLM) applications such as multi-turn dialogue and program analysis. However, the core self-attention mechanism scales quadratically with sequence length, creating a fundamental computational bottleneck. Existing sparse attention methods alleviate this issue but face trade-offs: training-based methods are costly and cannot be directly applied as acceleration plugins for other models, while inference-time methods often compromise efficiency or cross-modal generality. To address these limitations, we present UniSparse, a unified mechanism that introduces the notion of composite tokens--compact representations that aggregate multi-granularity contextual information. Building on this abstraction, UniSparse dynamically constructs sparse attention through multi-granularity compression and block-level selection, enabling efficient and hardware-friendly execution on GPU. Across multiple modalities and tasks ranging from synthetic benchmarks to real-world applications, UniSparse consistently surpasses state-of-the-art sparse attention methods (e.g., MInference, XAttention, FlexPrefill) in both accuracy and efficiency, achieving $\ge$ 99% of full-attention accuracy and up to 2.61$\times$ faster attention computation than FlashAttention.
title A Unified Sparse Attention via Multi-Granularity Compression
topic Computation and Language
url https://arxiv.org/abs/2512.14082