1,200+ notebooks that actually show you how wandb works

What it does

This is the official examples repo for Weights & Biases. It collects Jupyter notebooks and Python scripts demonstrating how to instrument training runs across PyTorch, TensorFlow/Keras, Hugging Face Transformers, and plain Python. Think of it as the “show, don’t tell” companion to wandb’s documentation.

The interesting bit

The repo doesn’t just dump snippets. Each example is a complete, runnable Colab or script you can execute without reverse-engineering the docs. The Keras integration, for instance, shows the exact callback wiring (WandbMetricsLogger, WandbModelCheckpoint) that uploads checkpoints and logs metrics every 5 steps—useful if you’ve ever wondered “where does the magic actually happen?”

Key highlights

• Framework-specific folders with working integrations for PyTorch, Keras, and Hugging Face
• Colab links for nearly every example; zero local setup required to explore
• Covers the full wandb surface: wandb.init, wandb.config, run.watch(), run.log(), and artifact tracking
• Includes hyperparameter sweep examples and model versioning workflows
• Free tier supported; examples work with personal/academic accounts

Caveats

• README is mostly a landing page; you’ll need to dig into subdirectories for actual code
• Some framework sections are stubbed or link externally rather than hosting code in-repo
• No explicit versioning or compatibility matrix for framework/wandb version pairs

Verdict

Grab this if you’re evaluating wandb or onboarding a team and need “copy-paste, then modify” starter code. Skip it if you already know the API and just need reference docs—these are tutorials, not a cookbook for advanced patterns.