AI Paper Atlas

Research map / Architectures and efficiency / Pruning and efficient tuning

Getting ViT in Shape: Scaling Laws for Compute-Optimal Model Design

Ibrahim Alabdulmohsin, Xiaohua Zhai, Alexander Kolesnikov, Lucas Beyer

NeurIPS 2023 · Architectures and efficiency › Pruning and efficient tuning · approach: Efficient ViT adaptation · 4 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
use compute-optimal shape SoViT-400m/14 instead of ViT-L/16 at 384pxfine-tuning · ImageNet v2accuracy: 80.4 -> 83.2Improved
use SoViT-400m/14 instead of ViT-g/14 at 518px (2.4x smaller)fine-tuning · ImageNet Valaccuracy: 90.2 -> 90.3No clear effect
use SoViT-400m/14 instead of ViT-L/16 in linear probelinear probe · ImageNet-Aaccuracy: 67.8 -> 76.4Improved
use SoViT-400m/14 instead of ViT-g/14 in linear probelinear probe · ImageNet-Raccuracy: 90.3 -> 89.0Hurt
use SoViT-400m/14 instead of ViT-g/1410-shot linear transfer · ImageNetaccuracy: 84.0 -> 84.1No clear effect
use SoViT-400m/14 instead of ViT-L/1610-shot linear transfer · CIFAR100accuracy: 82.2 -> 86.7Improved
use SoViT-400m/14 instead of ViT-g/1410-shot linear transfer · Birdsaccuracy: 88.5 -> 88.8No clear effect
use SoViT-400m/14 instead of ViT-L/16 in LiTzero-shot ImageNet · accuracyaccuracy: 79.9 -> 82.2Improved
use SoViT-400M instead of ViT-g/14 in multitask decodingOCR-VQA · accuracy [%]accuracy: 52.5 -> 52.9Improved
use SoViT-400M instead of ViT-g/14 in multitask decodingGQA · accuracy [%]accuracy: 58.0 -> 56.0Hurt
use SoViT-400M instead of ViT-g/14 in multitask decodingCOCO Captioning · CIDErCIDEr: 126 -> 125No clear effect
use SoViT-400m/14 instead of ViT-g/14 as UViM encoderCOCO panoptic · PQPQ: 44.8 -> 43.7Hurt

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