MultiChainComparison is not upgraded to work with dspy 2.5.7

[pambram]

Seems that the MultiChainComparison is lacking the path in the init for when we are using the new dspy.LM setting, producing the error AttributeError("'Prediction' object has no attribute 'rationale'") when used.

Minimal example:

import dspy

from dotenv import load_dotenv
load_dotenv()

class IsSmartGoal(dspy.Signature):
    goal = dspy.InputField(desc="The goal to evaluate", format=str)
    is_smart = dspy.OutputField(
        desc='Whether the goal is SMART. Make sure you deeply analyze this step by step, and have solid reasoning for your answer. Don\'t preface or postfix your answer with any reasoning, your allowed outputs are only "True" or "False".'
    )

def is_smart_goal(goal: str) -> bool:
    lm = dspy.LM("openai/gpt-4o")
    with dspy.context(lm=lm):
        pred = dspy.ChainOfThought(signature=IsSmartGoal, n=3)(goal=goal)
        compare_answers = dspy.MultiChainComparison(signature=IsSmartGoal, M=len(pred.completions))
        final_pred = compare_answers(completions=pred.completions, goal=goal).is_smart # errors here

        # clean up
        response = final_pred.strip()
        boolean_output = str(response).lower() == "true"

        return boolean_output

if __name__ == "__main__":
    goal = "I want to lose weight"
    print(is_smart_goal(goal))

Name: dspy-ai
Version: 2.5.7

[pambram]

the workaround I found is to add a rationale = dspy.OutputField() to the signature manually

[arnavsinghvi11]

Thanks @pambram ! fixed via #1630