The SentencePiece library also implements Byte-pair-encoding. That's what the LLaMA models use and the original Mistral models were essentially a copy of LLaMA2.