Call models with a prompt and output type directly

[rlouf]

In the current re-design of the API we added Generator objects that accept a model and an output type as parameters:

from outlines import Generator, models
from pydantic import BaseModel

class Foo(BaseModel):
    bar: str

model = models.openai("gpt-4o")
generator = Generator(model, Foo)
result = generator("prompt")

The rationale behind this design is that, for open source models, the index for structured generation can take some time to compile and we hold it in this object so it can be re-used. There are a few issues with this design:

1. API models do not have this problem
2. This adds an extra step in the happy pip install outlines -> text generated path
3. It does not correspond to most people's mental model. Other libraries make you call the model instance with the prompt to generate text.

I suggest that we modify the original design to make simple use cases straightforward, but still allows flexibility in more complex cases. We can add a __call__ method to model instances so the previous workflow can be re-written as:

from outlines import Generator, models
from pydantic import BaseModel

class Foo(BaseModel):
    bar: str

model = models.openai("gpt-4o")
result = model("prompt", Foo)

While keeping the Generator pattern for cases where a user may want to call the same model repeatedly with the same structure:

from outlines import Generator, models
from pydantic import BaseModel

class Foo(BaseModel):
    bar: str

model = models.openai("gpt-4o")
generator = Generator(model, Foo)

result_1 = generator("prompt_1")
result_2 = generator("prompt_2")

[RobinPicard]

Your suggestion makes a lot of sense to me. I think that removing the generation creation step by having the possibility to call the model directly makes the basic example look a lot simpler and more intuitive to new users.