axiom-foundation-models-ref

To generate structured outputs from on-device foundation models, initialize a LanguageModelSession and use the respond(to:generating:) method with the @Generable macro to produce typed results directly within your iOS app.

The Tool protocol in the Foundation Models API defines structured interfaces that allow a LanguageModelSession to interact with external functions, enabling on-device models to execute specific actions and fetch dynamic data during generation.

Yes, the Foundation Models API supports constrained decoding and dynamic schemas, allowing developers to enforce specific output structures and adapt generation logic dynamically during a LanguageModelSession on iOS and macOS.

You can stream responses from a LanguageModelSession by utilizing the built-in streaming capabilities provided by the Foundation Models API, which delivers generated tokens incrementally for real-time UI updates in iOS applications.

Performance considerations for Generable types include profiling schema complexity and managing constrained decoding overhead, ensuring the on-device LanguageModelSession maintains responsive generation times across iOS, macOS, and iPadOS deployments.

Yes, SystemLanguageModel provides the interface to verify on-device foundation model availability and capabilities, ensuring your iOS app safely initializes a LanguageModelSession only when the underlying generable hardware and software support are present.