Using mixup on RoBERTa

[replydotai]

Thanks for sharing @sgugger.
Data augmentation for text classification really is a tough one. Is there anything you consider promising?

@franborjavalero you might want to checkout this thread