hello! this is the pretrained BART. The dataset used for pretraining is nonsense summary corpus with output as diff.
