Runtime execution vs. model export

It’s very important to understand the distinction between running your model directly within a development environment (typically on your own PC) and running an exported version on an embedded target.

Runtime execution (development mode)

During development and testing, you’ll mostly run your AI models directly within the Aidge framework.

This allows you to:

• Iterate quickly: test changes, debug, and validate model behavior without the need for a full export process.

• Access full debugging tools: utilize the rich debugging capabilities of the development environment.

This “runtime execution” is ideal for the iterative development cycle, but it doesn’t represent how the model will perform on a constrained embedded device.

Export (deployment mode)

Once your model is refined and ready for deployment, Aidge’s export strategy comes into play. The goal of exporting is to generate a source code tailored for specific hardware, allowing inference of your AI model on embedded device.

Then, you could also use this exported version in a runtime execution mode. To do so, create a backend plugin by converting the export of one operator into a compiled kernel library.

Summary