vocab/gpt_neox_japanese_2.7b/__init__.py

from transformers import AutoTokenizer

tokenizer = AutoTokenizer.from_pretrained("abeja/gpt-neox-japanese-2.7b")