pytorch-book

Why we included this project

This repository works as a full PyTorch course, not just a pile of examples. It starts with tensor basics and autograd in Jupyter notebooks that a complete beginner can run and tinker with, then moves into distributed training and CUDA extensions. The later chapters are where it gets fun: a GAN that generates anime faces, a Transformer that composes classical-style poems, and a CenterNet object detector, each with complete code and pretrained models ready to run. Since about 60 percent of the book's explanatory text lives in the repo, you can learn entirely from the notebooks without buying the print edition. It is a solid ramp-up for someone new to deep learning, while people who already know PyTorch can skip ahead to a specific chapter for a complete, tested implementation.