AI Paper Atlas

Research map / Architectures and efficiency / Attention and transformers

Token Sparse Attention: Efficient Long-Context Inference with Interleaved Token Selection

Dongwon Jo, Beomseok Kang, Jiwon Song, Jae-Joon Kim

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
add Token Sparse Attention to FlexPrefillFlexPrefill, LLaMA-3.1-8B-Instruct · RULER 128K attention speedupspeedup: ×2.44 -> ×2.76Improved
add Token Sparse Attention to MInferenceMInference, LLaMA-3.1-8B-Instruct · RULER 128K attention speedupspeedup: ×1.12 -> ×1.38Improved
add Token Sparse Attention to MInferenceMInference, LLaMA-3.1-8B-Instruct · RULER avgaccuracy: 86.49 -> 86.05Hurt
add Token Sparse Attention to FlashAttentionMistral-Nemo-12B-Instruct · RULER avgaccuracy: 67.60 -> 67.37No clear effect
add Token Sparse Attention to FlexPrefillFlexPrefill, LLaMA-3.1-8B-Instruct · InfiniteBench avgaccuracy: 49.53 -> 49.23No clear effect
use dynamic token coverage instead of fixed sparsity ratio (matched ~1.5x speedup) (ablation)Token Sparse Attention · RULER avgaccuracy: 85.43% (s=0.5, ×1.57) -> 86.84% (tau=0.010, ×1.51)Improved
replace token eviction (FastKV) with interleaved Token Sparse Attention at matched speedupLLaMA-3.1-8B-Instruct prefill · RULER avgaccuracy: 85.64 -> 86.84Improved
apply token-level sparsity to all layers instead of selected stable layers (ablation)Token Sparse Attention · RULERaccuracyHurt
select layers with low inter-layer representation drift (ablation)Token Sparse Attention · RULER 4K, random 3-layer sparsificationaccuracyImproved
add Token Sparse Attention on top of FlexPrefill vs tuning FlexPrefill gammaFlexPrefill · RULER accuracy-speedup Paretoaccuracy at matched speedupImproved
increase token coverage tau from 0.005 to 0.010 (ablation)Token Sparse Attention on FlashAttention · RULER avg, LLaMA-3.1-8Bspeedup and accuracy: ×1.36 -> ×1.51; 87.02% -> 86.84%Mixed

Earlier methods this paper compares with

Named baselines from earlier accepted papers in this paper's results tables: better than 2 of 2 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