It works best when jokes/anecdotes are at least 40 words or longer. It is based on Facebook's RoBerta-MUPPET. The training dataset was a private collection of around 2000 jokes. This model has not ...