from transformers import LlamaTokenizer

# Instantiate the architecture
tokenizer = LlamaTokenizer()

# Train on your own data to fill in vocab and merges
tokenizer.train(files=["my_corpus.txt"])