AI Paper Atlas

Research map / Architectures and efficiency / Attention and transformers

Fast KV Compaction via Attention Matching

Adam Zweiger, Xinghong Fu, Han Guo, Yoon Kim

ICML 2026 · Architectures and efficiency › Attention and transformers · approach: KV cache eviction and compression · 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
use AM-HighestAttnKeys instead of KVzip at 1% cacheQwen3-4B · QASPERtoken-level F1: 0.127 -> 0.338Improved
use AM-HighestAttnKeys instead of H2O+ at 1% cacheLlama-3.1-8B-Instruct · QASPERtoken-level F1: 0.199 -> 0.317Improved
use AM-HighestAttnKeys instead of KVzip at 10% cacheLlama-3.1-8B-Instruct · QASPERtoken-level F1: 0.347 -> 0.452Improved
use AM-HighestAttnKeys instead of H2O+ at 1% cacheQwen3-4B-Instruct-2507 · LongBench v2 subsetaccuracy (%): 10.6 -> 61.7Improved
use AM-HighestAttnKeys instead of KVzip at 20% cacheQwen3-4B · QASPERtoken-level F1: 0.438 -> 0.435Mixed
apply AM-OMP on top of a summary (0.2x)Summarize · QuALITYaccuracy (%): 55.2 -> 55.7No clear effect
apply AM-OMP on top of a summary (0.05x)Summarize · QuALITYaccuracy (%): 55.2 -> 49.2Hurt
use OMP-fast (k=4, refit interval 2) instead of OMP (ablation)AM-OMP · LongHealth Gemma-3-12Bkey selection time (s): 565 -> 104Improved
use highest-attention key selection instead of OMP (ablation)AM · LongHealth Gemma-3-12Bkey selection time (s): 565 -> 3Improved
use context-prefill instead of repeat-prefill queries (ablation)AM · reference query generationtime (s): 8 -> 7Improved
add L2 regularization when fitting beta and C_v (ablation)AM · least squares fitperformanceHurt
add post-compaction attention biases and fitted values (leave-one-out) (ablation)AM-OMP · reference generationslog-perplexityImproved

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