// the find
mrdbourke/pytorch-deep-learning
Materials for the Learn PyTorch for Deep Learning: Zero to Mastery course.
A complete course repo for learning PyTorch from scratch, built around a food image classification project called FoodVision. It covers fundamentals through model deployment across 10 sections, with matching slides, exercises, solutions, and 300+ hours of video. Aimed squarely at beginners with some Python experience.
The progression from notebook code to modular Python scripts in section 05 is genuinely useful — most courses skip this and leave students unable to write non-notebook code. Section 08 walks through replicating the Vision Transformer paper equation by equation, which is a concrete way to learn to read ML papers. Every section has exercises with worked solutions, not just lecture content. The companion site at learnpytorch.io makes the material searchable and readable without needing to run notebooks.
The material is frozen at 2023 — no coverage of modern training patterns like mixed precision by default, gradient checkpointing, or anything that emerged post-PyTorch 2.0 beyond that one intro notebook. The course uses Google Colab throughout, which is fine for learning but means students never deal with local environment setup, CUDA driver issues, or multi-GPU anything. The 'going modular' section produces a workable but fairly naive file structure that doesn't scale past toy projects — no config management, no logging framework, just raw Python files. At 18k stars it's popular, but you'd outgrow it fast and need to look elsewhere for anything production-adjacent.