How to use this resource

There are two ways to work through this course. The first is the standard route. The second uses an AI coding agent to help inspect the repository and provide additional explanation.

Standard route

Complete Modules 0–4 in order. Follow the commands, run the scripts, read the explanations, and consider the deployment decisions and trade-offs presented in each task.

Make sure to spend time reading and understanding the code. Open the main preparation, export, quantization, lowering, and runtime scripts and identify the ExecuTorch APIs used at each stage of the ahead-of-time flow. For the example applications, inspect the Python, Java, and C++ code to see how each application loads a .pte program, creates its inputs, invokes the runtime, and turns model output into application behavior.

AI-assisted route

The course repository is structured so that an AI coding agent can trace commands from the module text into the scripts and application code. Use an agent to probe the implementation, ask follow-up questions, compare deployment choices, and test your understanding. The agent should supplement the practical work, not replace running the steps and reading the code yourself.

Two portable agent skills are provided under .agents/skills. Both use .agents/course-context.md, which maps the important scripts, identifies the ExecuTorch APIs, and explains the repeated course structure:

• $explain-mode turns the agent into an interactive ExecuTorch tutor. It reads the relevant module and code, traces the real APIs, explains command output and artifacts, and asks short questions to check your understanding.
• $build-mode helps adapt the course patterns into a new model pipeline or application. It checks current official documentation before carrying course APIs, dependencies, or workarounds into a new application.

If your agent supports named skill invocation, try:

    

        
        
Use $explain-mode to walk me through Module 2's INT8 flow. Show me the relevant
code, explain which parts are the ExecuTorch API, and ask me
questions along the way.

    

If named invocation is not supported, point the agent directly at the skill file:

    

        
        
Read .agents/skills/explain-mode/SKILL.md and follow it while teaching me
Module 2.

    

Build mode can produce a working application quickly, but code generation by itself does not provide the same understanding as tracing the implementation. Spend time with explain mode before or after a build so that you understand the model contract, ahead-of-time pipeline, backend choice, runtime calls, and application integration.

Recommendation

Combine both approaches. Work through the course normally, interpret the scripts yourself, and then use explain mode as a one-to-one ExecuTorch tutor. When you understand the workflow, use build mode to begin prototyping your own application.