How to finetune it for a multivariate feature space?

[MOvi996]

Hi,
I am trying to use this for a project that generates youtube comments based on parameters like videoCategory, channelName, videoTitle.
For this i want to fine-tune this gpt2-simple on a dataset with following columns:
videoCategory | channelName | videoTitle | commentText .

For text generation the model will take in the first three values and generate a relevant comment.

I want to know how to finetune the model for such a multicolumn dataset? Should I add special tokens in places of column separations? If so, how to do that?
Would really appreciate your help.