AI Paper Atlas

Research map / Vision-language and multimodal / Multimodal LLMs

ShortV: Efficient Multimodal Large Language Models by Freezing Visual Tokens in Ineffective Layers

Qianhao Yuan, Qingyu Zhang, Yanjiang Liu, Jiawei Chen, Yaojie Lu, Hongyu Lin, Jia Zheng, Xianpei Han, Le Sun

ICCV 2025 · Vision-language and multimodal › Multimodal LLMs · approach: Training-free token reduction for MLLMs · 9 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
select replaced layers by Layer Contribution instead of random choice (ablation)ShortV (N=19) · MMBench ENaccuracy: 58.4 -> 64.8Improved
select replaced layers by Layer Contribution instead of cosine similarity (ablation)ShortV (N=19) · MMBench ENaccuracy: 60.8 -> 64.8Improved
select replaced layers by Layer Contribution instead of cosine similarity (ablation)ShortV (N=19) · GQAaccuracy: 59.5 -> 60.9Improved
freeze visual tokens rather than text tokens (ablation)ShortV (N=19) · MMBench ENaccuracy: 2.1 -> 64.8Improved
freeze visual tokens rather than all tokens (ablation)ShortV (N=19) · GQAaccuracy: 0.0 -> 60.9Improved
freeze visual tokens rather than the same number of random tokens (ablation)ShortV (N=19) · SEED-Benchaccuracy: 5.5 -> 66.2Improved
add FastV on top of ShortVShortV on LLaVA-1.5-7B · FLOPs ratioFLOPs ratio: 55% -> 29%Improved
add FastV on top of ShortVShortV on LLaVA-1.5-7B · GQAaccuracy: 60.9 -> 59.3Hurt
ShortV instead of FastV at similar FLOPsLLaVA-1.5-7B · Flickr30KCIDEr: 67.5 -> 71.3Improved
ShortV instead of VTW at similar FLOPsLLaVA-1.5-7B · GQAaccuracy: 55.1 -> 60.9Improved
increase number of ShortV layers from 19 to 24 (ablation)ShortV on LLaVA-1.5-7B · MMEscore: 1503.1 -> 1341.7Hurt
increase number of ShortV layers from 0 to 24 on LLaVA-1.5-13B (ablation)ShortV on LLaVA-1.5-13B · 7-benchmark retentionrelative performance (Per.): 100.0 -> 99.2No clear effect

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