from transformers import AutoTokenizer

tokenizer = AutoTokenizer.from_pretrained("google-t5/t5-base")
print(tokenizer._tokenizer.model)

# Unigram(...)