dair-ai/ML-Notebooks

Covers PyTorch fundamentals through advanced architectures (transformers, GNNs) with minimal, reproducible implementations spanning NLP, computer vision, and graph learning. Notebooks emphasize algorithmic clarity by implementing models from scratch—linear/logistic regression, neural networks, and attention mechanisms—alongside framework-based approaches using PyTorch and Hugging Face. Runs natively in Google Colab or GitHub Codespaces with conda-managed dependencies, enabling immediate experimentation without local setup.

3,436 stars. No commits in the last 6 months.