huggingface/pytorch-image-models ? reverse-engineered prompt

I want to build a comprehensive Python library for PyTorch that contains a massive collection of computer vision models. It should be the go-to place for high-performance, pre-trained image classification models.

Please include all the popular architectures like Vision Transformers (ViT), EfficientNet, ResNet, MobileNet, ConvNeXt, and Swin Transformers. The key feature is that I can easily load any of these models with pre-trained weights, ready for use.

Beyond just the model definitions, I'll need a set of scripts to make this library useful right away. Create a training script that supports distributed training, a validation script to test performance, and a simple inference script to run a model on a single image. Also, add a utility to export these models to the ONNX format for deployment. Structure it all as a proper Python package called timm.