Remove ReCoRD from the common sense datasets and the custom
dataset preparation. Unlike the other common sense datasets, each
instance from ReCoRD has multiple correct answers, this means that
ReCoRD is evaluated differently and requires separate logic for
things like early stopping and evaluation scripts. So, we're
cutting it from the scope.
Remove ReCoRD from the common sense datasets and the custom dataset preparation. Unlike the other common sense datasets, each instance from ReCoRD has multiple correct answers, this means that ReCoRD is evaluated differently and requires separate logic for things like early stopping and evaluation scripts. So, we're cutting it from the scope.