Data/output Structure

Hey, I found out the format of Target and source file from previous issues.
but what would be the format of output after preprocess? is it something like tokenizing inputs?
And if I want to create dataset for training should it be in source/target file format or M2 format? which one is helpful?