AI Paper Atlas

Research map / Architectures and efficiency / Attention and transformers

CriticalKV: Optimizing KV Cache Eviction from an Output Perturbation Perspective

Yuan Feng, Junlin Lv, Haoyu Guo, Yukun Cao, S Kevin Zhou, Xike Xie

ICML 2026 · Architectures and efficiency › Attention and transformers · approach: Long-context KV and memory attention · 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 CriticalKV selection to SnapKVSnapKV Qwen2.5-32B · RULER (40% cache)average score: 63.86 -> 81.09Improved
add CriticalKV selection to AdaKVAdaKV Qwen2.5-32B · RULER (40% cache)average score: 71.09 -> 83.87Improved
add CriticalKV selection to HeadKVHeadKV Qwen2.5-32B · RULER (40% cache)average score: 81.04 -> 90.69Improved
add CriticalKV selection to AdaKVAdaKV Mistral-7B · RULER (40% cache)average score: 34.88 -> 69.17Improved
add CriticalKV selection to AdaKVAdaKV Llama-3.1-8B · RULER (20% cache)average score: 57.48 -> 68.94Improved
add CriticalKV selection to AdaKV at 40% cacheAdaKV Llama-3.1-8B · LongBenchaverage loss vs full cache: 6.0% -> 2.4%Improved
add CriticalKV selection to AdaKV at 40% cacheAdaKV Llama-3.1-8B · LongBench Multi-Doc QAscore: 41.36 -> 45.11Improved
add CriticalKV selection to AdaKV at 40% cacheAdaKV Llama-3.1-8B · SCBench EN.QAscore: 15.71 -> 22.14Improved
add CriticalKV selection to AdaKV at 20% cacheAdaKV · LongBench code domainscoreMixed
add stage-1 safeguard (alpha=0.5 instead of alpha=0) (ablation)AdaKV w/ CriticalKV Mistral-7B · LongBench (20% cache)average score: 31.94 -> 42.85Improved
add stage-1 safeguard (alpha=0.5 instead of alpha=0) (ablation)AdaKV w/ CriticalKV Mistral-7B · LongBench (10% cache)average score: 35.94 -> 37.94Improved
add stage-1 safeguard alpha (0.5 vs 0) (ablation)AdaKV w/ CriticalKV Llama-3.1-8B · LongBench (20% cache)average score: 44.35 -> 43.88No clear effect

Earlier methods this paper compares with

Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.

Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.

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