Deep Learning Lectures, But You Can Actually Run Them

What it does

The repository hosts Jupyter notebooks for the University of Amsterdam’s MSc AI Deep Learning course. Each tutorial pairs lecture concepts—transformers, graph neural networks, normalizing flows, and the like—with concrete implementations in PyTorch and PyTorch Lightning. The authors also provide one-to-one JAX+Flax translations, so you can see the same ideas expressed in two different frameworks.

The interesting bit

Most course materials are static PDFs or videos; these notebooks are built to run on a common laptop CPU without a GPU, with pretrained models and datasets clocking in under 1 GB. The tutorials have even been adopted as official PyTorch Lightning documentation, which suggests the code quality is closer to a reference implementation than a semester hack.

Key highlights

• Covers a wide arc: activation functions, ResNet, Vision Transformers, meta-learning, SimCLR, adversarial attacks, and more.
• Dual-framework support: every notebook has a PyTorch/Lightning version and a JAX+Flax translation.
• Runs fully offline on CPU; pretrained models auto-download and total disk usage stays under 1 GB.
• Integrated into the official PyTorch Lightning tutorial docs.
• Accompanied by YouTube lecture recordings and a ReadTheDocs site for browsing without executing.

Caveats

• The README notes this is the first year the tutorials are being presented, so small bugs and issues are expected.
• Some cluster-specific advice (Snellius) is baked in, which is useful if you happen to be a UvA student and less so otherwise.

Verdict Worth bookmarking if you are self-studying deep learning and want reputable, runnable examples beyond the usual documentation. If you already have a production framework locked down and hate notebooks, this is just polite academic overhead.