How do you get single embedding vector for each word (token) from RoBERTa?

Question

As you may know, RoBERTa (BERT, etc.) has its own tokenizer and sometimes you get pieces of given word as tokens, e.g. embeddings » embed, #dings

Since the nature of the