Pydantic to OpenAI Function Calling

pydantic_to_openai_function_calling.py

from pydantic import BaseModel, Field
import json

class ReceiptDataExtractor(BaseModel):
    date: str = Field(description="The date.")
    amount: float = Field(description="The total amount.")
    supplier: str = Field(description="The supplier.")

    class Config:
        function_meta = {
            "function_name": "receipt_data_extractor",
            "function_description": "Extract meta data from receipts."
        }

def convert_to_custom_schema(model):
    # Check if the model is a subclass of BaseModel
    if not issubclass(model, BaseModel):
        raise TypeError("Model must be a subclass of pydantic.BaseModel")

    # Check if Config and function_meta are correctly defined
    if not hasattr(model, 'Config') or not hasattr(model.Config, 'function_meta'):
        raise ValueError("Model must have a 'Config' class with a 'function_meta' attribute")

    # Check for function_name and function_description in function_meta
    if 'function_name' not in model.Config.function_meta or 'function_description' not in model.Config.function_meta:
        raise ValueError("Model's 'Config.function_meta' must contain 'function_name' and 'function_description'")

    # Generate the schema from the Pydantic model
    schema = model.schema()

    # Retrieve custom metadata from Config
    function_name = model.Config.function_meta["function_name"]
    function_description = model.Config.function_meta["function_description"]

    # Create the custom schema structure
    custom_schema = {
        "type": "function",
        "function": {
            "name": function_name,
            "description": function_description,
            "parameters": {
                "type": "object",
                "properties": {},
                "required": schema.get('required', [])
            }
        }
    }

    # Map the properties from Pydantic schema to the custom schema
    for prop, details in schema.get('properties', {}).items():
        custom_schema["function"]["parameters"]["properties"][prop] = {
            "type": details["type"],
            "description": details.get("description", "")
        }

    return custom_schema

# Usage
try:
    custom_schema = convert_to_custom_schema(ReceiptDataExtractor)
    print(json.dumps(custom_schema, indent=4))
except (TypeError, ValueError) as e:
    print(f"Error: {e}")