A (nicer) tokenizer you want to use for model `inference` and `training`: with all known peventable `gotchas` normalized or auto-fixed.
pip install tokenicer