AI Paper Atlas

Research map / Architectures and efficiency / Attention and transformers

AMS-KV: Adaptive KV Caching in Multi-Scale Visual Autoregressive Transformers

Boxun Xu, Yu Wang, Zihu Wang, Peng Li

AAAI 2026 · Architectures and efficiency › Attention and transformers · approach: KV cache eviction and compression · 3 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 AMS-KV caching to VAR-d30AMS-KV on VAR-d30 · ImageNet 256FID: 2.07 -> 2.09No clear effect
add AMS-KV caching to VAR-d30AMS-KV on VAR-d30 · ImageNet 256KV cache size: 22.41GB -> 4.77GB (78.72% down)Improved
add AMS-KV caching to VAR-d36AMS-KV on VAR-d30 · ImageNet 512KV cache size: 53.15GB -> 15.26GB (71.29% down)Improved
add AMS-KV caching to Infinity-2BAMS-KV on Infinity-2B · GenEvaloverall score: 0.711 -> 0.714No clear effect
add AMS-KV caching to VAR-d30AMS-KV on VAR-d30 · batch size 128 and 256feasibility: OOM -> 23.00 and 23.88 img/sImproved
add AMS-KV caching to VAR-d30AMS-KV on VAR-d30 · batch size 16throughput: 18.73 -> 18.86 img/sNo clear effect
keep the KV cache of intermediate scales (ablation)VAR-d30 · ImageNet 256FID: 2.12 -> 2.07No clear effect
keep the KV cache of local scales (ablation)VAR-d30 · ImageNet 256FID: 3.04 -> 2.07Improved
keep the KV cache of condensed scales (ablation)VAR-d30 · ImageNet 256FID: 2.48 -> 2.07Improved
use similarity-aware instead of uniform cache allocation under same budget (ablation)VAR-d30 · ImageNet 256FID: 6.65 -> 3.37Improved
add the first scale as condensed scale (ablation)VAR-d30 · ImageNet 256FID: 2.48 -> 2.16Improved
add the second scale as condensed scale (ablation)VAR-d30 · ImageNet 256FID: 2.16 -> 2.09Improved

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