DialoGPT
DialoGPT copied to clipboard
Multiturn mode training data
Hey guys! Awesome work.
Can you please clarify is there a reason train model with data contains not only N-turn samples if I want to use model in the N-turn mode? Does extra data with extra turns (and also samples with less number of turns) helps the model to catch the context better or there is no sense including such data into the training set?