AI Paper Atlas

Research map / Vision-language and multimodal / Multimodal LLMs

Prune Once: Retraining-Free Task-Agnostic Pruning for Vision-Language Models

Kang MinSeok, Hyunwoo Kim, Chanyoung Kim, Minwoo Kim, Jaekoo Lee, Dahuin Jung

ECCV 2026 · Vision-language and multimodal › Multimodal LLMs · approach: Attention-based visual token pruning · 0 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
replace Wanda magnitude score with PORTA variance score (uniform sparsity) (ablation)PORTA · CLIP 60% sparsity CIFAR-100accuracy: 70.28 -> 74.45Improved
replace Wanda magnitude score with PORTA variance score (uniform sparsity) (ablation)PORTA · CLIP 60% sparsity ImageNetaccuracy: 57.35 -> 59.35Improved
add PORTA ratio allocation to PORTA score (ablation)PORTA · CLIP 60% sparsity CIFAR-100accuracy: 74.45 -> 75.59Improved
add PORTA ratio allocation to PORTA score (ablation)PORTA · CLIP 60% sparsity ImageNetaccuracy: 59.35 -> 59.66Improved
add PORTA ratio allocation to Wanda score (ablation)Wanda · CLIP 60% sparsity CIFAR-100accuracy: 70.28 -> 74.29Improved
use Multiflow ratio allocation with either score (ablation)PORTA / Wanda · CLIP 60% sparsity CIFAR-100accuracy: 74.45 -> 1.40Hurt
drop zero-mean assumption (ablation)PORTA · CLIP 60% sparsity retrievalavg recall: 63.62 -> 63.64No clear effect
drop diagonal covariance approximation (ablation)PORTA · CLIP 60% sparsity retrievalavg recall: 63.62 -> 63.73No clear effect
PORTA vs Wanda on CLIP at 65% sparsityPORTA · CLIP 65% sparsity CIFAR-100accuracy: 28.86 -> 62.29Improved
PORTA vs SparseGPT on BLIP Flickr30k at 45% sparsityPORTA · BLIP 45% sparsity Flickr30kmean recall: 89.95 -> 90.26Improved
PORTA vs Wanda on Qwen2-VL ScienceQA at 50% sparsityPORTA · Qwen2-VL ScienceQA 50%average accuracy: 53.43 -> 54.30Improved
PORTA vs SparseGPT time costPORTA · CLIP pruning timeseconds: 773.35 -> 195.34Improved

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