Building with Llama 4

Introducing Building with Llama 4, a short course, created in collaboration with Meta and taught by Amit Sangani, Director of Partner Engineering for Meta’s AI team.

Meta’s new Llama 4 has added three new models and introduced a Mixture-of-Experts (MOE) architecture to its family of models, making them more efficient to serve.

In this course, you’ll work with two of the three new models introduced in Llama 4. First is “Maverick” a 400-billion parameter model, with 128 experts and 17 billion active parameters. Second is “Scout,” a 109-billion parameter model with 16 experts and 17 billion active parameters. Both Maverick and Scout support long context windows, of up to a million tokens and 10 million tokens respectively. The latter is enough to support very large GitHub repos for analysis.

In hands-on lessons, you’ll build apps using Llama 4’s long-context and its new multi-modal capabilities including reasoning across multiple images and “image grounding,” in which you can identify elements and reason within specific image regions. You’ll also learn about Llama’s newest tools: its prompt optimization tool that automatically improves system prompts, and synthetic data kit that generates high-quality datasets to fine-tune your model.

In detail, you’ll:

• Get an overview of Llama 4 models, how it evolved from Llama 2, and how it’s built on Mixture-of-Expert architecture.
• Use Meta’s official Llama API, now available for free in limited preview for US-based developers, to experiment with the new model’s capabilities, and build a translator chatbot that works across all the 12 languages Llama 4 supports.
• Work through several image reasoning and grounding examples such as detecting objects and their bounding boxes, and translating UI screenshots into executable code using Llama 4 via Meta’s Llama API as well as Llama 4 hosted on Together.ai.
• Understand Llama 4 special token and raw prompt format for both text-only and multimodal prompts.
• Learn how to work with long contexts of up to 10 million tokens, and ask questions across large text files, such as entire books, research papers, and GitHub repositories using Llama 4 on Together.ai.
• Automatically improve your system prompts in sentiment analysis and categorization use cases with Llama’s new prompt optimization tool.
• Use the synthetic data kit to ingest, create, curate, and save high-quality datasets for training and fine-tuning.

By the end of the course, you’ll confidently choose and call the right Llama 4 model and build production-ready features that span text, images, and massive context.

The open Llama 4 family of models is an important component of any GenAI Developer Toolkit. If you need an open model to extend, fine-tune, and customize, Llama is a top option and this course can help you learn what you can build with it.

Anyone who wants hands-on experience building with the Llama 4 family of models.