The T5-base model was trained on cnn dailymail dataset and then fine tuned on dataset containing 11k articles. The data fields were: