AI Paper Atlas

Research map / Vision-language and multimodal / Multimodal LLMs

LLaVA-PruMerge: Adaptive Token Reduction for Efficient Large Multimodal Models

Yuzhang Shang, Mu Cai, Bingxin Xu, Yong Jae Lee, Yan Yan

ICCV 2025 · Vision-language and multimodal › Multimodal LLMs · approach: Training-free token reduction for MLLMs · 374 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 token supplement (key-similarity merging) to adaptive token selection (ablation)LLaVA-1.5 Vicuna-7B + AITS · MMEscore: 1221.6 -> 1350.3Improved
add token supplement to adaptive token selection (ablation)LLaVA-1.5 Vicuna-7B + AITS · ScienceQA-IMGaccuracy: 66.5 -> 68.5Improved
add token supplement to adaptive token selection (ablation)LLaVA-1.5 Vicuna-7B + AITS · TextVQAaccuracy: 54.8 -> 56.0Improved
add token supplement to adaptive token selection (ablation)LLaVA-1.5 Vicuna-7B + AITS · POPEaccuracy: 75.7 -> 76.3Improved
use adaptive outlier-based selection instead of sequential sampling (ablation)LLaVA-PruMerge (40 tokens) · TextVQAaccuracy: 42.72 -> 54.00Improved
use adaptive selection instead of spatial sampling (ablation)LLaVA-PruMerge (40 tokens) · MMEscore: 1180.23 -> 1250.07Improved
use adaptive selection instead of sequential sampling (ablation)LLaVA-PruMerge (35 tokens) · POPEaccuracy: 11.7 -> 76.2Improved
use adaptive selection instead of spatial sampling (ablation)LLaVA-PruMerge (16 tokens) · ScienceQA-IMGaccuracy: 66.29 -> 68.07Improved
add LoRA fine-tuning with reduced tokens (ablation)LLaVA-PruMerge · MMEscore: 1250.1 -> 1350.3Improved
add LoRA fine-tuning with reduced tokens (ablation)LLaVA-PruMerge · TextVQAaccuracy: 54.0 -> 56.0Improved
add LoRA fine-tuning with reduced tokens (ablation)LLaVA-PruMerge · POPEaccuracy: 76.2 -> 76.3No clear effect
add spatially uniform extra tokens (PruMerge+ over PruMerge)LLaVA-1.5 Vicuna-7B · POPEaccuracy: 76.3 -> 84.0Improved

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