AI Paper Atlas

Research map / Architectures and efficiency / Attention and transformers

LookaheadKV: Fast and Accurate KV Cache Eviction by Glimpsing into the Future without Generation

Jinwoo Ahn, Ingyu Seong, Akhil Kedia, Junhan Kim, Hyemi Jang, Kangwook Lee, Yongkweon Jeon

ICLR 2026 · Architectures and efficiency › Attention and transformers · approach: KV cache eviction and compression

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 lookahead LoRA on all linear layers to embedding-only lookahead tokens (ablation)LookaheadKV · LongBench budget 64, LLaMA3.2-1B, 32 lookaheadavg score: 26.40 -> 27.12Improved
restrict LoRA to Q and V instead of all linear layers (ablation)LookaheadKV · LongBench budget 64, LLaMA3.2-1Bavg score: 27.12 -> 26.85Hurt
increase lookahead size beyond 32 (ablation)LookaheadKV · LongBench budget 64, LLaMA3.2-1Bavg score: 27.12 -> 27.0 (128 tokens)No clear effect
increase lookahead size from 4 to 32 (ablation)LookaheadKV all · LongBench budget 64, LLaMA3.2-1Bavg score: 26.60 -> 27.12Improved
increase lookahead size beyond 32 (ablation)LookaheadKV · inference overheadTTFT overhead (%): 4.96 -> 11.0Hurt
add SnapKV-style suffix window to lookahead importance scores (ablation)LookaheadKV · LongBench budget 64, LLaMA3.2-1Bavg score: 29.10 -> 28.52Hurt
replace draft generation with learned lookahead tokensLAQ · TTFT overhead at 32Keviction overhead: 14.5x lowerImproved
use learned lookahead instead of draft-based surrogateLAQ · LongBench budget 128 greedy, LLaMA3.1-8Bavg score: 46.61 -> 47.72Improved
use learned lookahead instead of draft-based surrogateLAQ · MT-Bench budget 256, LLaMA3.2-3Bjudge score: 7.43 -> 7.30Hurt
evaluate at higher sampling temperature T=0.8LookaheadKV and baselines · LongBench budget 128, LLaMA3.1-8Bavg score: 47.72 -> 45.81Hurt
train with shorter context (2K, 4K, 8K) (ablation)LookaheadKV · RULER, LLaMA-3Bavg scoreHurt

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