For PFunc APIs, please refer to the pfunc part of the user documentation. In system design, we mainly dicuss the implementation of the pfunc.
We can use the @P.function annotation to decorate a Python function and
make it a SemanticFunction. This parsing process is done in parrot/frontend/pfunc/function.
We parse a SemanticFunction into several FuncBodyPiece, which can be a Constant or a ParameterLoc. We can bind a Parameter to a SemanticVarible by passing the SV as the argument of the function.
Sometimes we need to formalize / transform our prompts by some certain formats.
We abstract them as the transformations on SemanticFunctions. For example, the chat
template of some chat models (e.g. Vicuna) are implemented as a ConversationTemplate transform.