SentencePieceTokenizer should also be able to directly take a vocab. Now it looks only take a vocab file as here