AI Paper Atlas

Research map / Architectures and efficiency / Attention and transformers

Scout Before You Attend: Sketch-and-Walk Sparse Attention for Efficient LLM Inference

Hoang Anh Duy Le, Sahil Joshi, Zeyu Yang, Zhaozhuo Xu, Anshumali Shrivastava

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 Sketch&Walk prefill sparsity instead of dense attentionLlama-3.1-8B-Instruct · RULER prefill averageaccuracy: 91.44 -> 91.35No clear effect
use Sketch&Walk instead of FlexPrefillLlama-3.2-1B-Instruct · RULER prefill averageaccuracy: 63.25 -> 67.59Improved
use Sketch&Walk instead of MInferenceLlama-3.1-8B-Instruct · RULER prefill averageaccuracy: 90.03 -> 91.35Improved
use Sketch&Walk instead of Quest for decodingLlama-3.1-8B-Instruct · RULER decode averageaccuracy: 89.06 -> 91.08Improved
use Sketch&Walk instead of Adamas for decodingLlama-3.1-8B-Instruct · RULER decode averageaccuracy: 90.74 -> 91.08Improved
apply end-to-end Sketch&Walk at 80% sparsityQwen2-7B-Instruct · LongBench AVGaccuracy: 40.85 -> 43.22Improved
apply end-to-end Sketch&Walk at 80% sparsityLlama-3.1-8B-Instruct · LongBench AVGaccuracy: 47.95 -> 47.63No clear effect
apply end-to-end Sketch&Walk at 80% sparsityLlama-3.2-1B-Instruct · LongBench AVGaccuracy: 29.75 -> 30.10No clear effect
reduce Hadamard sketch dimension to 16-32 (ablation)Sketch&Walk · LongBench/RULER (3 datasets)accuracyNo clear effect
increase walk degree (ablation)Sketch&Walk · 3 datasetsaccuracyImproved
increase sparsity ratio (ablation)Sketch&Walk · 3 datasetsaccuracyNo clear effect

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