timm DINOv3 - a timm Collection

timm 's Collections

Perception Encoder

Searching for Better ViT Baselines

MobileNetV4 pretrained weights

timm Top-20 ImageNet-1k Models

timm Top-20 Fastest Models

timm ImageNet-12k Models

timm Takes on the Classics

Fastest timm models > 75.3% IN-1k Top-1 (Original ResNet-50)

Fastest timm models > 80% Top-1 ImageNet-1k

Fastest timm models > 83% ImageNet-1k Top-1

Fastest timm models > 86% ImageNet-1k Top-1

Fastest timm models > 88% ImageNet-1k Top-1

All the ImageNets

Fine-Tune Image Classification Benchmark Datasets

timm tiny test models

timm DINOv3

updated Sep 19, 2025

Meta AI's DINOv3 weights in timm. ViTs with `qkvb` have a zero QV bias present, otherwise bias is disabled. QKV bias are all 0 in original weights.