NOTE: The tokenizer generated by zipf tokenizer will need to have the word "vocab" and the first and last brackets removed to make it compatible with the trainer, not worth hunting down the issue when ...