Suppose I have an input with 10 tokens for which I know tokens [3,4,5] are very important and I want to let BERT know about this. How can I add more weight to / inject a token-level feature into the BERT pipeline?
Suppose I have an input with 10 tokens for which I know tokens [3,4,5] are very important and I want to let BERT know about this. How can I add more weight to / inject a token-level feature into the BERT pipeline?