AI Paper Atlas

Research map / Architectures and efficiency / Attention and transformers

Scaling Attention via Feature Sparsity

Yan Xie, Tiansheng Wen, Tang Da Huang, Bo Chen, Chenyu You, Stefanie Jegelka, Yifei Wang

ICLR 2026 · Architectures and efficiency › Attention and transformers · approach: Sparse attention for long-context LLMs · 3 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
replace dense QK scoring with SFA k=8GPT2-124M (from scratch) · OpenWebTextvalidation perplexity: 17.29 -> 18.27Mixed
replace dense QK scoring with SFA k=16Qwen3-0.6B (from scratch) · Pilevalidation perplexity: 4.66 -> 4.81Mixed
use SFA instead of halving Q/K hidden size (ablation)GPT2-124M · OpenWebTextvalidation perplexity: 20.88 -> 18.27Improved
use SFA instead of halving Q/K hidden size (ablation)Qwen3-0.6B · Pileaverage zero-shot accuracy: 36.68 -> 38.94Improved
use SFA k=8 instead of dense attentionGPT2-124M trained on 32k NIAH · Synthetic NIAH 32kaccuracy: 80% -> 82%Improved
raise SFA k from 8 to 16 at 32k (ablation)GPT2-124M · Synthetic NIAH 32kdecoding speed: 1.3x -> 1.0xHurt
use SFA k=16 fine-tune instead of dense fine-tuneQwen3-0.6B · GSM8Kaccuracy: 63.42 -> 61.46Hurt
use SFA k=16 fine-tune instead of dense fine-tuneQwen3-8B · NIAH 32768accuracy: 95 -> 97Improved
use SFA k=16 fine-tune instead of dense fine-tuneQwen3-0.6B · Arxivaccuracy: 41.17 -> 39.14Hurt
use SFA with k=4 on inference KV cacheSFA · KV-cache memoryKV-cache size: up to ~40% savingsImproved

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