AI Paper Atlas

Research map / Architectures and efficiency / Attention and transformers

EntroKV: Entropy-Guided Dynamic Budget Allocation for KV-Cache Compression

Wenhao Gao, Haoran Cao, Yueyan Li, YongGao Xiao, Caixia Yuan, Xiaojie Wang

ICML 2026 · Architectures and efficiency › Attention and transformers · approach: Long-context KV and memory attention

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 entropy-guided head-wise budget allocation to SnapKVSnapKV · RULER-4Kscore: 86.91 -> 89.33Improved
Add entropy-guided budget allocation to CaMCaM · RULER-16Kscore: 67.18 -> 91.57Improved
Add entropy-guided budget allocation to D2OD2O · LongBenchscore: 41.23 -> 45.89Improved
Add entropy-guided allocation to SnapKV on Qwen3-8BSnapKV · LongBenchscore: 48.25 -> 49.05Improved
Use entropy head-level allocation instead of PyramidKV or AdaKV layer/head allocationSnapKV · LongBench (Llama-3.1-8B)score: 47.79 -> 48.15Improved
Use LAVa or Task-KV allocation instead of uniform SnapKVSnapKV · RULER-16K (Llama-3.1-8B)score: 90.88 -> 87.12Hurt
Vary adaptivity alpha from 0 to 1 (mix of global baseline and instance entropy) (ablation)EntroKV · LongBench (Llama-3.1-8B, Qwen2.5-7B)accuracyMixed
Increase total budget ratio beta (ablation)EntroKV · LongBenchaccuracyImproved
Use Entro-SnapKV at long context (128K) instead of SnapKVSnapKV · RULER-128K (Llama-3.1-8B)score: 72.68 -> 74.00Improved
Add entropy-guided allocation under few-shot MATH-500 and SQuADSnapKV, CaM, D2O · MATH-500, SQuAD (0-5 shot)accuracyImproved
Use entropy allocation instead of full cache at budget 0.4-0.5EntroKV on Llama · LongBenchaccuracyImproved

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