AI Paper Atlas

Research map / Architectures and efficiency / Attention and transformers

SALE : Low-bit Estimation for Efficient Sparse Attention in Long-context LLM Prefilling

Xiaodong Ji, Hailin Zhang, Fangcheng Fu, Bin CUI

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 SALE sparse attention instead of FlashAttention2Llama-3.1-8B-Instruct · LongBenchaverage score: 48.77 -> 48.39No clear effect
use SALE sparse attention instead of FlashAttention2Llama-3.1-8B-Instruct · InfiniteBenchaverage score: 37.75 -> 37.11No clear effect
use SALE sparse attention instead of FlashAttention2Qwen3-4B-Instruct · InfiniteBenchaverage score: 31.85 -> 32.09Improved
use SALE sparse attention instead of FlashAttention2Llama-3.1-8B-Instruct · RULERaverage score: 83.86 -> 84.67Improved
use SALE instead of FlashAttention2Llama-3.1-8B-Instruct · attention latency at 64Kspeedup: 1.00× -> 3.36×Improved
add per-head threshold calibration (ablation)SALE · InfiniteBench (128K latency)scoreImproved
increase context lengthSALE · single input speedupspeedup: 1.99× at 16K -> 3.88× at 128KImproved
use SALE on short inputs (under about 8K tokens)SALE · short-context prefilllatencyNo clear effect
use one-pass selection+computation kernel instead of two-pass (ablation)SALE · kernel designworkloadHurt
use 4-bit instead of higher-bit quantization for estimation (ablation)SALE · block selectionselection accuracy and speedImproved

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