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

- Headlines
- ArticleBody