Fix assertion bug with new_signature

[Shangyint]

Co-authored-by: chiragshah285
@chiragshah285

[Shangyint]

@arnavsinghvi11 Would there be an issue if we change target_module to be the actual module in Assertions854a259? It seems assert_transform will re-assign the wrapped Retry module to all modules defined in __init__ thus we shouldn't have the problem?

[arnavsinghvi11]

Thanks @chiragshah285 @Shangyint. This change makes sense, but will require some refactoring to existing assertion notebooks examples as the assertion statements currently rely on the target_module mapped to corresponding module signature. I believe this logic was kept originally as with 2 assertion statements referring to the same Module type (e.g. ChainOfThought), Retry was not identifying the correct module to backtrack on and hence relied on the signature identifier.

will require some further testing to verify if the proposed logic correctly maintains this. Feel free to check out an example notebook using assertions (LongFormQA - particularly where it uses dspy.Suggest on target_module=GenerateCitedParagraph while the module includes 2 chain of thought layers, dspy.ChainOfThought(GenerateSearchQuery) and dspy.ChainOfThought(GenerateCitedParagraph)

[okhat]

What's the urgency and status here?