Decorator for writing simpler calls as functions instead of classes

[willbakst]

Description

Originally requested by @jbbakst in the slack community

Original request:

@mirascope.openai_call(model="gpt-4o", max_tokens=4096, temperature=0.3)
def recommend_book(genre: str) -> str:
    """
    SYSTEM:
    You are a literary expert, with both a deep and broad knowledge and love of books, particularly in the {genre} genre.

    USER:
    Recommend a {genre} book. Respond with ONLY the title, no other prose.
    """

book = recommend_book(genre="fantasy")

Initial thoughts:

• For return type str we can use a standard call, but for anything else we'll need to use an extractor.
• Ideally we can make this extremely configurable for easy swapping across providers

I'm thinking maybe we want something like this?

from mirascope.openai import openai_fn


@openai_fn(model="gpt-4o", max_tokens=4096, temperature=0.3)
def recommend_book(genre: str) -> str:
    """
    SYSTEM:
    You are a literary expert, with both a deep and broad knowledge and love of books, particularly in the {genre} genre.

    USER:
    Recommend a {genre} book. Respond with ONLY the title, no other prose.
    """

book = recommend_book(genre="fantasy")

We should also make the following work:

from mirascope.openai import openai_fn
from pydantic import BaseModel


class Book(BaseModel):
    title: str
    author: str


@openai_fn(model="gpt-4o", max_tokens=4096, temperature=0.3)
def recommend_book(genre: str) -> Book:
    """Recommend a {genre} book."""

book = recommend_book(genre="fantasy")
assert isinstance(book, Book)

Ideally we could also make the following work (using some other features in progress):

from mirascope.openai import OpenAIStream, openai_fn


@openai_fn(model="gpt-4o", max_tokens=4096, temperature=0.3)
def recommend_book(genre: str) -> OpenAIStream[str]:
    """Recommend a {genre} book. Response with ONLY the title. No other text."""

for chunk_content in recommend_book(genre="fantasy"):
    print(chunk_content, end="", flush=True)

Lastly, a general decorator could be cool:

from mirascope.base import llm_fn
from mirascope.openai import OpenAICallParams


@llm_fn(provider="openai", call_params=OpenAICallParams(...))
def recommend_book(genre: str) -> str:
    """Recommend a {genre} book."""

We could also look into doing something like this lol:

from mirascope.base import llm_fn
from mirascope.openai import OpenAICallParams


@llm_fn(provider="openai", call_params=OpenAICallParams(...))
def recommend_book(genre: str) -> str:
    """Recommend a {formatted_genre} book."""

    def formatted_genre(genre: str) -> str:
        return f"Formatted: {genre}"

[willbakst]

Moving work on this to #322