I'm exploring the possibility of using Apple's SystemLanguageModel from the Foundation Models framework within a WidgetKit extension to generate a summary of today's activities from my app.
The API works as expected when invoked from the widget extension. However, I'm looking for guidance on whether this is a recommended approach in production.
Given the execution time and memory constraints of WidgetKit extensions, is it advisable to perform on-device inference directly in the widget? Or is the recommended pattern to generate the summary in the main app (or another process), store the result in an App Group/shared container, and have the widget simply read and display the precomputed output?