hi @PalGitts - yes, if you change the stories (training data) or domain, you will have to re-train the dialogue model. Though if you haven’t changed anything to do with NLU, that will be skipped.
Yes, I was training with 15 intents and 16 entity along with DIETClassifier. It is taking almost 13-15 sec per iteration. May be as I am using pretty old Laptop and training is done on a non English language using FastText vectors.
On the Positive side the accuray of intent classification and NER is pretty much satisfactory.