Library / SDK
Fine-tuning & TrainingInference & Serving
timm
1Pretrained 2PyTorch 3image 4classification 5model 6collection
the six Ws · specification
W1
Who
Originally created by Ross Wightman and now maintained together with Hugging Face.
W2
What
timm, short for PyTorch Image Models, provides hundreds of pretrained vision model architectures and training scripts.
W3
Where
Hosted on GitHub at huggingface/pytorch-image-models and distributed via PyPI.
W4
When
Started in 2019 and continuously updated with new architectures.
W5
Why
It gives researchers and engineers one consistent interface to state of the art vision backbones instead of reimplementing each paper.
W6
With
Built on PyTorch and integrates with the Hugging Face Hub.
W7
Watch
Apache 2.0 licensed, open source, maintained by Ross Wightman and Hugging Face, roughly 37k GitHub stars.
alternatives
works with