Proposition with a distinct ModelFormatter class

outlines/models/base.py

@@ -1,9 +1,23 @@
-from abc import ABC
+from abc import ABC, abstractmethod
+
+
+class ModelFormatter(ABC):
+    """Base class for all model formatters."""
+
+    @abstractmethod
+    def format_input(self, model_input):
+        ...
+
+    @abstractmethod
+    def format_output_type(self, output_type):
+        ...
 
 
 class Model(ABC):
     """Base class for all models."""
 
+    formatter: ModelFormatter
+
     def __call__(self, model_input, output_type=None, **inference_kwargs):
         from outlines.generate import Generator
 

outlines/models/openai.py

@@ -1,18 +1,18 @@
 """Integration with OpenAI's API."""
 from functools import singledispatchmethod
-from types import NoneType
 from typing import Optional, Union
 
 from pydantic import BaseModel
 
 from outlines.models import Model
+from outlines.models.base import ModelFormatter
 from outlines.prompts import Vision
 from outlines.types import Json
 
 __all__ = ["OpenAI"]
 
 
-class OpenAIBase:
+class OpenAIFormatter(ModelFormatter):
     """Base class for the OpenAI clients.
 
     `OpenAI` base is responsible for preparing the arguments to OpenAI's
@@ -83,7 +83,7 @@ def format_output_type(self, output_type):
             f"The type {output_type} is not available with OpenAI. The only output type available is `Json`."
         )
 
-    @format_output_type.register(NoneType)
+    @format_output_type.register(type(None))
     def format_none_output_type(self, _: None):
         """Generate the `response_format` argument to the client when no
         output type is specified by the user.
@@ -115,7 +115,7 @@ def format_json_output_type(self, output_type: Json):
         }
 
 
-class OpenAI(Model, OpenAIBase):
+class OpenAI(Model):
     """Thin wrapper around the `openai.OpenAI` client.
 
     This wrapper is used to convert the input and output types specified by the
@@ -128,15 +128,16 @@ def __init__(self, model_name: str, *args, **kwargs):
 
         self.client = OpenAI(*args, **kwargs)
         self.model_name = model_name
+        self.formatter = OpenAIFormatter()
 
     def generate(
         self,
         model_input: Union[str, Vision],
         output_type: Optional[Union[type[BaseModel], str]] = None,
         **inference_kwargs,
     ):
-        messages = self.format_input(model_input)
-        response_format = self.format_output_type(output_type)
+        messages = self.formatter.format_input(model_input)
+        response_format = self.formatter.format_output_type(output_type)
         result = self.client.chat.completions.create(
             model=self.model_name, **messages, **response_format, **inference_kwargs
         )