I am using Pydantic v1 in a Langchain implementation of Custom Tools. When I staticly define a Pydantic class it will populate the __fields__ attribute of my class with metadata about the fields — which is needed by Langchain.

My use case requires that I create Pydantic classes at runtime based on the user’s context. I have tried creating Pydantic classes using Python’s type function and Pydantic’s create_model function. Neither of these approaches populates __fields__

Can anyone fill me in on how to populate __fields__ when creating a Pydantic class dynamicly?

The following code block summarizes what I am expecting vs the actual result.

from enum import Enum

from langchain.pydantic_v1 import BaseModel, create_model


class SpecialtyEnumStatic (str, Enum):
    psychiatry = "SPECIALTY:Psychiatry"
    sports_medicine_internal_medicine = "SPECIALTY:Sports Medicine (Internal Medicine)"


class ProviderSearchInputStatic(BaseModel):
    specialty: SpecialtyEnumStatic


print(ProviderSearchInputStatic.__fields__)
# {'specialty': ModelField(name='specialty', type=SpecialtyEnumStatic, required=True)}

specialties = {
    'psychiatry': 'SPECIALTY:Psychiatry',
    'sports-medicine (Internal Medicine)': 'SPECIALTY:Sports Medicine (Internal Medicine)'
}

SpecialtyEnumDynamic = Enum('SpecialtyEnumDynamic', specialties)

ProviderSearchInputDynamic = create_model(
    'ProviderSearchInputDynamic',
    specialty=SpecialtyEnumDynamic,
    __base__=BaseModel
)

print(ProviderSearchInputDynamic.__fields__)
# {}