pytorch/examples ? reverse-engineered prompt

I want a new PyTorch example in this repo that shows how to train a simple image classifier from scratch on a small dataset, in the same style as the other examples here.

Make it easy to run, with a clear README, one main training script, and basic evaluation so someone can see it work quickly. Keep the code short and beginner friendly, with sensible defaults and comments that explain what each part is doing. If it helps, use the same kind of setup and conventions as the other vision examples in this repository.

Please also include a tiny way to save and reload the model, and make sure the example runs cleanly from the command line without extra setup. If you need to check current PyTorch docs or best practices, go ahead and look them up online.