The Llama Cookbook is Meta's official guide for building applications with the Llama model family. Written primarily in Jupyter Notebook format, the repository serves as a comprehensive resource for developers working with Llama models across inference, fine-tuning, retrieval-augmented generation (RAG), and end-to-end problem solving. The repository was recently renamed from llama-recipes to llama-cookbook to better reflect its expanded scope as a practical guide rather than just a collection of recipes.
The repository is organized into several key sections. The Getting Started section provides reference implementations for inferencing, fine-tuning, and RAG examples. The End to End Use Cases section spans various domains and applications, including recent additions like WhatsApp integration with Llama 4, research paper analysis with Llama 4 Maverick, and character mind mapping from books. The 3P Integrations section offers recipes and use cases from various Llama service providers. The src directory contains the original llama-recipes library code along with frequently asked questions addressing fine-tuning concerns.
Recent additions highlight support for Llama 4 models, including the Llama 4 Scout variant which supports 5 million token context windows, and Llama 4 Maverick for more advanced use cases. The repository also maintains documentation and links to model cards for multiple Llama versions including Llama 3.3, 3.2, 3.1, 3, and 2, each with associated license and acceptable use policy documentation.
GitGenius activity data reveals the repository maintains active community engagement with 340 tracked issues and pull requests. The median response latency for issues and PRs is 154.2 hours, though the mean extends to 2562.6 hours, indicating some longer-running discussions. The most frequently applied label is triaged with 87 instances, followed by question with 18 and enhancement with 12, showing a well-organized triage process. The most active contributors tracked are wukaixingxp with 323 events, init27 with 216 events, and mreso with 105 events, demonstrating consistent maintenance and community support.
The repository's classification spans multiple machine learning and AI development categories including fine-tuning models, instruction tuning, quantization techniques, transformer models, and model deployment. Integration points include Hugging Face, LangChain, and PyTorch, reflecting the broader ecosystem the Llama models operate within. The repository also links to related Meta projects including the synthetic-data-kit and llama-prompt-ops tools, positioning itself as part of a larger Llama development ecosystem.
The repository underwent a significant refactor, with an archive-main branch preserving the previous structure for reference. This reflects the project's evolution from a recipes collection to a more comprehensive cookbook format designed to serve developers at various skill levels building production applications with Llama models.
