Maeve is a language model that is similar to BART in structure but trained specially using a CAT (Conditionally Adversarial Transformer).

This allows the model to learn to create long-form text from short entries with high degrees of control and coherence that are impossible to achieve with traditional transformers.

This specific model has been trained on the XSUM dataset, and can invert summaries into full-length news articles. Feel free to try examples on the right!