AI Paper Atlas

Research map / Vision-language and multimodal / Multimodal LLMs

Not All Attention is Needed: Parameter and Computation Efficient Tuning for Multi-modal Large Language Models via Effective Attention Skipping

Qiong Wu, Yiyi Zhou, Weihao Ye, Xiaoshuai Sun, Rongrong Ji

IJCV 2026 · Vision-language and multimodal › Multimodal LLMs · approach: Unified any-modality language models · 6 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
skip MHAs instead of FFNs (12 skipped) (ablation)EAS on LaVIN · ScienceQA (LaVIN-7B tuning)average accuracy: 82.1 -> 90.0Improved
restrict skipping candidates to MHA instead of MHA or FFN (12 skipped) (ablation)EAS on LaVIN · ScienceQA (LaVIN-7B tuning)average accuracy: 88.6 -> 90.0Improved
add information exchange (two-path average pooling) to PIA (ablation)EAS-12 on LaVIN · ScienceQA (LaVIN-7B tuning)average accuracy: 87.1 -> 90.1Improved
freeze the average-pooling bias in PIA (ablation)EAS-12 on LaVIN · ScienceQA (LaVIN-7B tuning)average accuracy: 90.1 -> 90.0No clear effect
re-parameterize PIA into FFN at inference (ablation)EAS-12 on LaVIN · ScienceQA (LaVIN-7B tuning)inference time (s): 2.97s -> 1.69sImproved
skip 16 MHAs instead of 12 (ablation)EAS on LaVIN · ScienceQA (LaVIN-7B tuning)average accuracy: 90.0 -> 88.4Hurt
skip 8 MHAs instead of 4 (ablation)EAS on LaVIN · ScienceQA (LaVIN-7B tuning)average accuracy: 89.4 -> 89.8Improved
increase PIA hidden dimension from 8 to 64 (ablation)EAS-8 on LaVIN · ScienceQA (LaVIN-7B tuning)average accuracy: 87.6 -> 90.1Improved
replace LaVIN adapters with EAS-12LaVIN-7B · ScienceQA (LaVIN-7B tuning)inference time (s): 3.70s -> 1.69sImproved
use EAS-12 instead of DAS-4 on LaVINLaVIN-7B · ScienceQA (LaVIN-7B tuning)average accuracy: 89.2 -> 90.0Improved
replace MAM with EAS-8 on LLaVALLaVA-7B · ScienceQA (LLaVA-7B tuning)inference time (s): 2.87s -> 1.98sImproved
skip 12 modules instead of 8EAS on LLaVA · ScienceQA (LLaVA-7B tuning)average accuracy: 88.7 -> 88.1Hurt

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