AI Paper Atlas

Research map / Architectures and efficiency / Attention and transformers

Tactic: Adaptive Sparse Attention with Clustering and Distribution Fitting for Long-Context LLMs

Kan Zhu, Tian Tang, Qinyu Xu, Zhan Jin, Yile Gu, Zhichen Zeng, Rohan Kadekodi, Liangyu Zhao, Ang Li, Arvind Krishnamurthy, Baris Kasikci

ICLR 2026 · Architectures and efficiency › Attention and transformers · approach: Sparse attention for long-context LLMs · 27 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 union of selected tokens across grouped query heads (ablation)Tactic on GQA model · attention latency across context lengths and Pattention speedup: up to 1.65xImproved
use cumulative-attention-score thresholds instead of fixed token budgets (Tactic vs Quest at 75%)Llama-3.1-8B-Instruct · RULER averageaccuracy: 69.2 -> 84.7Improved
use Tactic instead of PyramidKV at 90% budgetLlama-3.1-8B-Instruct · RULER averageaccuracy: 73.0 -> 84.6Improved
use Tactic instead of Quest at 75% budgetMegaBeam-Mistral-7B-512k · RULER averageaccuracy: 75.4 -> 85.2Improved
raise threshold from 75% to 90%Tactic Llama-3.1-8B-Instruct · RULER averageaccuracy: 84.7 -> 84.6No clear effect
raise threshold from 75% to 90%Tactic MegaBeam-Mistral-7B-512k · RULER averageaccuracy: 85.2 -> 85.4No clear effect
raise threshold from 70% to 90%Tactic Llama-3.1-8B-Instruct · LongBench Qasperscore: 43.30 -> 44.20Improved
raise threshold from 70% to 90%Tactic Llama-3.1-8B-Instruct · LongBench Musiquescore: 28.57 -> 30.71Improved
use Tactic instead of Quest at 90% budgetMegaBeam-Mistral-7B-512k · LongBench HotpotQAscore: 51.69 -> 49.59Hurt
use Tactic instead of PyramidKV at 70% budgetLlama-3.1-8B-Instruct · LongBench Qasperscore: 27.50 -> 43.30Improved
use Tactic instead of PyramidKV at 70% budgetMegaBeam-Mistral-7B-512k · LongBench MultiFieldQAscore: 36.74 -> 50.50Improved
use Tactic at 90% instead of Ada-SnapKVMegaBeam-Mistral-7B-512k · RULER 96Kaccuracy: 87.1 -> 82.6Hurt

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