How can I fine tune with my own dataset?

I see the the tutorials and there are all dataset using api, so I wanna ask if I want to load a dataset in my computer, and I wanna use Trainer API, should I rename my dataset’s column name?
By the way, I’am confused about the column name of “train_dataset” in Trainer API parameter, how it know which column is my input data and which column is my label?(seems different dataset have different column name because of different tasks(some are classification, some are next sentence prediction), but they are all used as “train_dataset” in Trainer API? Did I miss something important?)
And how can we know the batch_size of Trainer.train()? I only know the Dataloader in pytorch.
Thanks so much if someone could answer my question!(something like blog or tutorial would be better!)