How can I train a bert model for representational learning task that is domain specific?

Question

I am trying to generate good sentence embeddings for some specific type od texts, using sentence transformer models while testing the the similarity and clustering using kme