Anyone have idea how we can finetune a model using Trainer API?

I was trying to finetune a T5 small model using WMT-14 dataset but I am running into disk space issues due to size of the dataset hence I was wondering is there any way we can use dataset streaming for training?