Harvard has made its textbook on ML systems publicly available. It's extremely practical: not just about how to train models, but how to build production systems around them - what really matters.

The topics there are really top-notch:

> Building autograd, optimizers, attention, and mini-PyTorch from scratch to understand how the framework is structured internally. (This is really awesome)
> Basic things about DL: batches, computational accuracy, model architectures, and training
> Optimizing ML performance, hardware acceleration, benchmarking, and efficiency

So this isn't just an introductory course on ML, but a complete cycle from start to practical application. You can already read the book and view the code for free. For 2025, this is one of the strongest textbooks to have been released, so it's best not to miss out.

The repository is here, with a link to the book inside 👏