Training PEGASUS on unlabeled data

Hi,

I was wondering if there is any way to load a PEGASUS model and further train it on unlabeled data with its training objective rather than supervised finetuning with labeled data ?