Building BERT classification task without segmentation and with sequence information

Question

I\'m looking at creating my own NLP task based on BERT or similar model. My model is unique in that it is not language terms, but other identifiers (like a hash). For example: