AI Paper Atlas

Research map / Architectures and efficiency / Attention and transformers

Smooth Reading: Bridging the Gap of Recurrent LLM to Self-Attention LLM on Long-Context Understanding

Kai Liu, Zhan Su, Peijie Dong, Fengran Mo, Jianfei Gao, Shaoting Zhang, Kai Chen

ICLR 2026 · Architectures and efficiency › Attention and transformers · approach: Sparse attention for long-context LLMs

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 One-Round with End-to-End Multi-Round inferenceSWA-3B-4k · LongBenchaverage score: 41.70 -> 50.99Improved
replace One-Round with End-to-End Multi-Round inferenceRWKV-7-3B · LongBenchaverage score: 41.43 -> 48.03Improved
replace One-Round with End-to-End Multi-Round inferenceSWA-3B-4k · NIAH 64k-256kaverage accuracy: 3.40 -> 99.80Improved
preserve hidden states across rounds (EMR instead of NMR) (ablation)SWA-3B-4k vs Qwen-2.5-3B · LongBenchaverage score: 48.37 -> 50.99Improved
use Non-End-to-End Multi-Round inference instead of One-RoundQwen-2.5-3B · LongBenchaverage score: 47.38 -> 48.37Improved
use Non-End-to-End Multi-Round inference instead of One-RoundQwen-2.5-7B · LongBenchaverage score: 54.60 -> 51.33Hurt
add early stopping (ablation)SWA-3B-4k-EMR · NIAHaverage accuracy: 99.87 -> 99.70No clear effect
chunk size 4096 instead of 512 with window 512 (ablation)SWA EMR · NIAHaccuracy: 97.0 -> 0.0Hurt
scale to 7B parametersSWA-7B-4k-EMR · LongBenchaverage score: 46.17 -> 53.86Improved
increase window size from 512 to 4096 at chunk 2048 (ablation)SWA EMR · NIAHaccuracy: 0% -> 100%Improved
increase chunk size from 512 to 4096 at window 4096 (ablation)SWA EMR · NIAHinference time: 646s -> 366sImproved
use EMR instead of CompACT compressionSWA-3B-4k · HotpotQAF1: 36.57 -> 54.25Improved

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