This is correct from what I've seen, but it's not well documented. Also, that fine-tuning is "better for training " style than knowledge."