AI Paper Atlas

Research map / Architectures and efficiency / Attention and transformers

RAT+: Train Dense, Infer Sparse - Recurrence Augmented Attention for Dilated Inference

Xiuying Wei, Caglar Gulcehre

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 recurrence to dense attention (ablation)attention 1B from scratch · Val PPL T=4096perplexity: 7.44 -> 7.38Improved
add active recurrence learning (joint sparse+dense training) (ablation)RAT+ L=T · Val PPL at D=64perplexity: 9.96 -> 7.66Improved
use fixed overlapped chunk L=64 instead of L=D at inference (ablation)RAT+ · Val PPL at D=1perplexity: 8.39 -> 7.4Improved
per-batch joint training instead of 0.5/0.5 loss mixing (ablation)RAT+ · Val PPL at D=64perplexity: 7.82 -> 7.66Improved
add 1B-token resolution adaptation (ablation)RAT+ L=T L*=64 · Val PPL at D=16perplexity: 8.03 -> 7.52Improved
add recurrence on attention keys (ablation)RAT+ 200M · PG19 PPL at D=1perplexity: 11.68 -> 10.71Improved
apply recurrence on qkv inputs instead of k,v (ablation)RAT+ 200M · PG19 PPL at D=128perplexity: 11.39 -> 12.09Hurt
replace RoPE with NoPE (ablation)RAT+ · Val PPL at D=64perplexity: 7.66 -> 7.63No clear effect
add recurrence to top-k block attention (RAT+ vs attention) (ablation)Quest D=64 K=16 · NIAH-MK-2accuracy: 63.2 -> 93.8Improved
combine dilation with top-k block (ablation)Quest D=64 K=8 on RAT+ · NIAH-MK-2accuracy: 57.4 -> 97.4Improved
combine dilation with top-k block (ablation)Quest D=64 K=8 on RAT+ · NIAH-MK-3accuracy: 11.0 -> 79.6Improved
increase dilation from 1 to 64RAT+ 1.5B · commonsense avgaccuracy: 59.21 -> 57.46Hurt

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