AI Paper Atlas

Research map / Architectures and efficiency / Attention and transformers

KeepKV: Achieving Periodic Lossless KV Cache Compression for Efficient LLM Inference

Yuxuan Tian, Zihan Wang, Yebo Peng, Aomufei Yuan, Zhiming Wang, Bairen Yi, Xin Liu, Yong Cui, Tong Yang

AAAI 2026 · Architectures and efficiency › Attention and transformers · approach: KV cache eviction and compression · 14 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
KeepKV merging versus H2O evictionLlama-2-7B, 20% KV budget · LongBench NrtvQAscore: 16.54 -> 17.32Improved
KeepKV versus D2O mergingLlama-2-7B · LongBench TRECscore: 63.5 -> 60.5Hurt
KeepKV versus D2O mergingMistral-7B · LongBench HotpotQAscore: 26.41 -> 26.56No clear effect
KeepKV versus D2O mergingLlama-3-8B · LongBench Qasperscore: 8.86 -> 10.63Improved
KeepKV versus D2O mergingLlama-3-8B · LongBench RB-Pscore: 64.84 -> 65.68Improved
KeepKV compression versus full cacheLlama-2-7B · throughput at 4k contexttokens/s: 116.54 -> 255.99Improved
KeepKV versus H2O eviction on throughputLlama-2-7B · throughputtokens/s: 317.33 -> 255.99Hurt
KeepKV versus D2O on throughputLlama-2-7B · throughputtokens/s: 214.8 -> 255.99Improved
combine KeepKV with H2O or PyramidInfer eviction (ablation)H2O, PyramidInfer · XSUM Llama-7BROUGEImproved

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