AI Paper Atlas

Research map / Architectures and efficiency / Attention and transformers

Making Every Head Count: Sparse Attention Without the Speed-Performance Trade-off

Mingkuan Zhao, Wentao Hu, Jiayin Wang, Xin Lai, Tianchen Huang, Yuheng Min, Rui Yan, Xiaoyan Zhu

AAAI 2026 · Architectures and efficiency › Attention and transformers · approach: Sparse attention for long-context LLMs · 8 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
replace dense attention with SPAttentionOLMoE 0.25B-1.75B · Average of six metricsaccuracy: 0.3610 -> 0.3684Improved
replace dense attention with SPAttentionOLMoE 1B-7B · Average of six metricsaccuracy: 0.3402 -> 0.3532Improved
replace dense attention with SPAttentionOLMoE 1B-7B · Winograndeaccuracy: 0.5040 -> 0.4956Hurt
replace dense attention with SPAttentionOLMoE 0.25B-1.75B · MMLU SocSci and Humanities (5-shot)accuracy: 0.2640 -> 0.2552; 0.2520 -> 0.2468Hurt
replace dense attention with SPAttention8-layer Transformer · training throughputtokens/second: 3,214 -> 6,167Improved
use SPAttention instead of Longformer8-layer Transformer · Averageaccuracy: 0.3967 -> 0.4048Improved
use SPAttention instead of BigBird8-layer Transformer · HellaSwagaccuracy: 0.2772 -> 0.2856Improved
use balanced full-coverage bands instead of sliding window (ablation)SPAttention · Average (0.25B-1.75B)accuracy: 0.3376 -> 0.3434Improved
use complete bands instead of gapped bands (first half only) (ablation)SPAttention · Average (0.25B-1.75B)accuracy: 0.3315 -> 0.3434Improved
use balanced partitioning instead of exclusive unbalanced bands (ablation)SPAttention · Average (0.25B-1.75B)accuracy: 0.3289 -> 0.3434Improved
use distance-band partitioning instead of dense attentionsimulated 4-layer 8-head Transformer · head diversitystd of head behaviour: 0.0000-0.0005 -> 0.1845-0.1847Improved
use distance-band partitioning instead of dense attentionsimulated 4-layer 8-head Transformer · attention entropyentropy: 4.5461 -> 3.6276Improved

Earlier methods this paper compares with

Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 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