Multi label emotion classification - BERT?

Question

natural language processing

My dataset looks like this:

sentence emotions
bla bla bla {'kind':105, 'sad':32, 'angry':5}
..... {'surprised': 65, 'kind': 10}

That is, each post has several emotions, for example: kind:158, smile:32, angry:5 ...
The numbers next to the emotions are the number of people who were evoked these emotions.

Essentially, my goal is to predict the type of emotion a post might evoke. and I'm going to use a BERT transformer with Deep Pavlov

Question: How to be in this case? Is it okay to give BERT data in this format? Or is it better to remove the numbers? or leave only the emotion that scored the maximum number of votes? Anyone with similar experience please share.