Problem with `add` method and `Rouge`

[lvwerra]

As reported here the following code yields an error:

 rouge = evaluate.load('rouge')
 rouge.add(predictions="sentence 1", references="sentence 2")

ValueError: Evaluation module inputs don't match the expected format.
Expected format: {'predictions': Value(dtype='string', id='sequence'), 'references': Value(dtype='string', id='sequence')},
Input predictions: sentence 1,
Input references: sentence 2

[Sumanth077]

Hi @lvwerra, Happy to work on this Issue. Would be really great to get more information about the issue and also guidance from you. Thankyou

[lvwerra]

Hi @Sumanth077, that would be great. Under the hood this add method is called. It could be either _infer_feature_from_example or _enforce_nested_string_type that is not working as expected. The former infers which feature type is used if a metric defines several (as in ROUGE) and the latter makes sure that we don't accidentally cast a type that's not a string to string (since everything can be cast as a string). I would check if you can find anything unusual there.

[Sumanth077]

Thankyou @lvwerra. Could you also kindly attach the exact file link of Rouge here. It would be great.

[lvwerra]

Sure, it's here.

[hazrulakmal]

Hi @Sumanth077, just a friendly pin to ask if you have started working on this. would like to help and collab here.