Hi ! For your interesting work. I wonder how much resources are needed for a complete training tokenizer?