I trained a distilgpt2 model (trained it on CPU since that was the easiest for me to train it on) on a bunch of 5 minute improv sketches in an attempt to generalize the data. Example Results (Temperature 0.8):
Scene: Sunshine Person1: Ahh beautiful sunshine Person2: I love the way it bounces off your - beautiful face Person1: You were going to say something else there Person2: I resisted the bald joke because it's
Scene: Body Cancel: It's okay, pal. Person1: Hey it's okay Person2:I just um... I just need this time. Person1: You want us to leave? Person2: Yeah Person: It's okay, pal. Person1: Hey it's okay Person2:I just um... I just need this time. Person1: You want us to leave? Person2: Yeah
how are you feeling bad for doing you's's? Person2: I know you might not be as bad as I thought. Person1: I just um... I just need this time. Person2: You want us to leave for doing you's's? Person2: I know you might not be as bad as I thought. Person1: I just um... I just need this time. Person2: You want us to leave?
Memes: I ain't saying we're gonna not have a fight we lost, is it? Person1: We ain't saying we're gonna have a fight we lost, is it? End Scene: Digging I ain't saying we're gonna not have a fight we lost, is it? Person1: We ain't saying we're gonna have a fight we lost, is it? End Scene: Digging Person