Ask HN: Toy AIML finetuning Eureka moment: your recommendations?

I want to be able to quickly fine tune a pre-trained neural network model, and have the Eureka moment of observing it working on non-training data.

I have access to a Windows 11 laptop and Google CoLab's free tier. I know how to use WSL2 (Windows Subsystem for Linux), Docker Compose, and some Python tools (Anaconda, JupyterLab).

Is there a toy exercise you'd recommend?