Well, if you enable the model fine-tuning, it should be fine. I do not commit a lot of mistakes, and I usually correct manually, though.