AI Paper Atlas

Research map / Vision-language and multimodal / Multimodal LLMs

DAVID: Dual-stage Adaptive Vision-text Integrated Decoupling for Multimodal KV Cache Eviction

Yifeng Gu, Jianxiu Jin, Kailing Guo, Xiangmin Xu

AAAI 2026 · Vision-language and multimodal › Multimodal LLMs · approach: Training-free token reduction for MLLMs · 0 citations (OpenAlex)

Read the paperCheck your idea against 137,217 papers

AI-extracted summary. Written by AI from the paper's full text. Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.

Proposes DAVID, a layer-adaptive KV cache eviction strategy for multimodal LLMs that adapts to the degree of vision-text fusion across layers.

Problem

KV cache memory overhead from long multimodal inputs makes deploying multimodal LLMs on low-resource devices difficult.

Approach

Eviction adapts to the degree of modality fusion at each layer, based on analysis of features across layers, rather than applying a uniform policy.

Measured effects reported in the paper

What was changedWhereMeasuredEffect
add NCAR-based layer switching to decoupled eviction (ablation)DAVID (LLaVA-1.5-7B) · OCRBenchfinal score: 298 -> 305Improved
add decoupled vision/text eviction in all layers (ablation)SnapKV-style eviction (LLaVA-1.5-7B) · OCRBenchfinal score: 295 -> 298Improved
apply DAVID on top of H2O scoring (ablation)H2O (LLaVA-1.5-7B) · OCRBenchfinal score: 271 -> 285Improved
DAVID vs SnapKV on Qwen-VLDAVID · OCRBenchscore: 469 -> 478Improved
DAVID vs SnapKV on Qwen-VLDAVID · MM-Vetscore: 38.43 -> 40.31Improved
DAVID vs SnapKV at 20% sparsityDAVID (LLaVA-1.5-13B) · OCRBenchfinal score: 322 -> 324No clear effect
DAVID vs full cacheDAVID (LLaVA-1.5-7B) · OCRBenchfinal score: 308 -> 305Hurt
modality-fixed LOOK-M-style merging (LOOK-M, MEDA) vs unimodal SnapKVLOOK-M / MEDA · OCRBenchfinal score: SnapKV 295 -> LOOK-M 158, MEDA 156Hurt
reduce KV budget to 20% with DAVIDDAVID · MathVista subsetdecoding latency: 26.23 -> 19.17 ms/tokenImproved
reduce KV budget to 20% with DAVIDDAVID · MathVista subsetGPU memory: 1.26GB -> 0.34GBImproved

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