Meta’s official Llama field guide, freshly reorganized

What it does

Llama Cookbook is a repository of Jupyter notebooks and reference code organized into getting-started guides for inference, fine-tuning, and RAG, plus end-to-end use cases and third-party integration examples. It covers both text and vision models across the Llama family, including the latest Llama 4 recipes for long-context work, research analysis, and API bots. The src directory still houses the original llama-recipes library code and a fine-tuning FAQ, preserved after the project’s recent rename and reorganization.

The interesting bit

The repo straddles both open-weight models and Meta’s newer hosted Llama API, offering recipes that range from local fine-tuning to a WhatsApp-integrated bot. That breadth makes it less a single framework and more a curated field guide for wherever you happen to be deploying Llama.

Key highlights

• Covers the full lifecycle from inference and fine-tuning to RAG and deployed end-to-end use cases
• Latest Llama 4 recipes include 5-million-token context windows with Scout, research-paper analysis with Maverick, and a WhatsApp bot integration
• Includes third-party provider integrations alongside Meta’s own Llama API and open-weight model workflows
• Preserves the original llama-recipes library source and fine-tuning FAQs in src after the recent refactor

Caveats

• The repo recently underwent a major refactor; the README warns that some links may still be broken or folders missing, with an archive-main snapshot available for the old layout
• Several Llama 4 recipes rely on the Llama API, which currently requires joining a waitlist

Verdict

Worth bookmarking if you are actively building with Llama and prefer to start from a working notebook rather than a blank page. Skip it if you need a unified framework or are not interested in Meta’s model family.