The LLaMA tokenizer is a BPE model based on sentencepiece.