Meta's Llama is a family of open-source generative AI models, with the latest version being Llama 4. Unlike many other AI models that are accessible only via APIs, Llama allows developers to download and use it with certain limitations. Meta partners with vendors like AWS, Google Cloud, and Microsoft Azure to make cloud-hosted versions of Llama available.
Key Features of Llama 4:
- Scout: 17 billion active parameters, 109 billion total parameters, and a 10 million token context window.
- Maverick: 17 billion active parameters, 400 billion total parameters, and a 1 million token context window.
- Behemoth: (Not yet released) 288 billion active parameters and 2 trillion total parameters.
Llama 4 models are trained on large amounts of unlabeled text, image, and video data, supporting over 200 languages. These models can perform tasks such as coding, answering math questions, and summarizing documents. Llama 4 Scout is designed for longer workflows, Maverick balances reasoning and speed, and Behemoth is intended for advanced research.
Llama is used in Meta AI chatbot experiences on Facebook Messenger, WhatsApp, Instagram, Oculus, and Meta.ai. It is also available on Llama.com and platforms like Hugging Face. Meta provides tools like Llama Guard, Prompt Guard, CyberSecEval, Llama Firewall, and Code Shield to enhance model safety and security.
Limitations: While powerful, Llama has limitations, including primary support for English in multimodal features. It was trained using datasets that include pirated e-books and articles. Additionally, Llama may produce buggy or insecure code and generate plausible but false information.