This tokenizer is based on EleutherAI/gpt-j-6B and increases the vocabulary size to 91238 by adding Korean words.