[Bug] Type Validation not enforced in Python 3.14

[timo-obrecht]

What happened?

When using dspy.Signature classes to enforce structured data, the type are not correctly parsed and converted in Python 3.14. It works fine in 3.13.

Steps to reproduce

import os
import dspy


lm = dspy.LM(
    model="openrouter/openai/gpt-4o-mini",
    api_base="https://openrouter.ai/api/v1",
    api_key=os.getenv("OPENROUTER_API_KEY"),
    temperature=0.0,
)


class ObjectInfo(dspy.Signature):
    text: str = dspy.InputField(desc="A given object")
    is_blue: bool = dspy.OutputField(desc="Is this object blue ?")
    is_big: bool = dspy.OutputField(desc="Is this object big ?")


def main(padding: int) -> bool:
    module = dspy.Predict(
        ObjectInfo.with_instructions(
            "The user gives an object and you must return if the object is blue and if the object is big"
        )
    )
    with dspy.context(lm=lm, num_retries=0):
        object_info = module(text="The sky" + " " * padding)  # add padding to avoid prompt caching
    return not all((isinstance(object_info.is_blue, bool), isinstance(object_info.is_big, bool)))


if __name__ == "__main__":
    print(f"Got {sum(main(i) for i in range(10))} errors")

With this setup, I always get 100% errors in python 3.14. The object returned is

Prediction(
    is_blue='Yes',
    is_big='Yes'
)

I get 0 errors executing the same code in Python 3.13. The object returned is :

Prediction(
    is_blue=True,
    is_big=True
)

DSPy version

3.0.4