Google’s Gemma family album, served in Jupyter

What it does

The Gemma Cookbook is a curated collection of Jupyter notebooks, full-stack demos, and technical guides for Google’s open Gemma models. It covers the core model line—from the original Gemma through Gemma 4—alongside specialized variants such as CodeGemma, MedGemma, PaliGemma, and VaultGemma. The repository separates stable tutorials from research experiments, end-to-end apps, and responsible AI examples, with older material moved to an archive.

The interesting bit

The README functions less like a cookbook and more like a field guide to a small industry. Google has proliferated so many Gemma variants—vision-language models, medical imaging models, privacy-preserving models, even therapeutic development models—that a single repository feels necessary just to map the territory. The structure keeps the chaos navigable by walling off research experiments and historical notebooks from the latest tested tutorials.

Key highlights

• Covers the full Gemma model family, from lightweight text models to multimodal and domain-specific variants like MedGemma and TxGemma.
• Organized into distinct sections: tested tutorials, full-stack apps, research experiments, responsible AI notebooks, and core documentation.
• Maintains an explicit archive of older notebooks, acknowledging the rapid iteration of the model line.
• Models are available across Hugging Face Hub, Kaggle, Google Cloud Vertex AI, and NVIDIA platforms.
• Community translations exist in Traditional and Simplified Chinese.

Verdict

Developers already working with Gemma models will find this a useful index of tested patterns and domain-specific examples. If you are not using Gemma, this is essentially a very thorough product documentation repo—useful for comparison, but not a reason to switch ecosystems.