How to do token type prediction task?

[oathaha]

In the paper, there is a task that lets a model to predict whether a code token is identifier or not. Can you explain more detail about how to do it? I would like to know the data format and how to use the data to train a model.