How to construct Chinese dataset with gpt2 fine tune

want to use my own dataset, run clm
must have [cls] in the beginning ?