HISA: Efficient Hierarchical Indexing for Fine-Grained Sparse Attention

TL;DR

HISA introduces a hierarchical indexing method that significantly accelerates fine-grained sparse attention for long contexts, matching existing accuracy without additional training.

Contribution

HISA provides a plug-and-play hierarchical indexer that replaces flat token scans, enabling faster sparse attention with no loss in quality.

Findings

Achieves up to 64K context speedup.

Matches the quality of DeepSeek Sparse Attention.

Outperforms block-sparse baselines.

Abstract

Token-level sparse attention mechanisms, exemplified by DeepSeek Sparse Attention (DSA), achieve fine-grained key selection by scoring every historical key for each query through a lightweight indexer, then computing attention only on the selected subset. While the downstream sparse attention itself scales favorably, the indexer must still scan the entire prefix for every query, introducing an per-layer bottleneck that grows prohibitively with context length. We propose HISA (Hierarchical Indexed Sparse Attention), a plug-and-play replacement for the indexer that rewrites the search path from a flat token scan into a two-stage hierarchical procedure: (1) a block-level coarse filtering stage that scores pooled block representations to discard irrelevant regions, followed by (2) a token-level refinement stage that applies the original indexer exclusively within the retained candidate blocks. HISA preserves the identical token-level top-sparse pattern consumed by the downstream Sparse MLA operator and requires no additional training. On kernel-level benchmarks, HISA achieves up to speedup at 64K context. On Needle-in-a-Haystack and LongBench, we directly replace the indexer in DeepSeek-V3.2 and GLM-5 with our HISA indexer, without any finetuning. HISA closely matches the original DSA in quality, while substantially outperforming block-sparse baselines.