AI Paper Atlas

Research map / Architectures and efficiency / Attention and transformers

A Unified Sparse Attention via Multi-Granularity Compression

Siran Liu, Zane Cao, Yongchao He

ICML 2026 · Architectures and efficiency › Attention and transformers · approach: Sparse attention for long-context LLMs · 0 citations (OpenAlex)

Read the paperCheck your idea against 137,217 papers

AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.

Measured effects reported in the paper

What was changedWhereMeasuredEffect
use average pooling instead of max pooling for composite tokens (ablation)UniSparse (0.95) · RULER, Qwen2.5-7B-Instructoverall accuracy: 81.01 -> 84.26Improved
use average pooling instead of stochastic pooling (ablation)UniSparse (0.95) · HELMET, Qwen2.5-7B-Instructoverall accuracy: 44.05 -> 47.03Improved
use average pooling instead of max pooling (ablation)UniSparse (0.95) · HELMET, Llama-3.1-8B-Instructoverall accuracy: 54.83 -> 56.21Improved
use average pooling instead of stochastic pooling (ablation)UniSparse (0.95) · RULER, Llama-3.1-8B-Instructoverall accuracy: 92.21 -> 92.56Improved
balanced Q-K compression (cq=ck=8) instead of query-heavy cq=32, ck=2 (ablation)UniSparse (0.95) · RULER, Llama-3.1-8B-Instructoverall accuracy: 91.37 -> 92.56Improved
balanced Q-K compression instead of cq=32, ck=2 (ablation)UniSparse (0.95) · RULER, Qwen2.5-7B-Instructoverall accuracy: 81.93 -> 84.26Improved
key-heavy compression cq=4, ck=16 instead of balanced (ablation)UniSparse (0.95) · HELMET, Qwen2.5-7B-Instructoverall accuracy: 47.03 -> 47.27No clear effect
compression factor c=16 instead of c=8 (ablation)UniSparse (0.95) · RULER, Qwen2.5-7B-Instructoverall accuracy: 84.26 -> 83.43Hurt
compression factor c=4 instead of c=8 (ablation)UniSparse (0.95) · RULER, Qwen2.5-7B-Instructoverall accuracy: 84.26 -> 84.51No clear effect
add head compression ch=2 at Top-P 0.95 (ablation)UniSparse (0.95) · RULER, Llama-3.1-8B-Instructsparsity: 46.33% -> 45.13%Hurt
add head compression ch=4 at Top-P 0.9 (ablation)UniSparse (0.95) · RULER, Qwen2.5-7B-Instructsparsity: 60.12% -> 46.82%Hurt
add head compression ch=2 (ablation)UniSparse (0.95) · block selection overhead, 128K tokensspeedup over XAttention-0.95: 1.62x -> 2.64xImproved

Earlier methods this paper compares with

Named baselines from earlier accepted papers in this paper's results tables: better than 2 of 3 on most shared results.

Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.

Closest papers by meaning

Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy