xcfeng

Thank you for this excellent job, I still have some questions about rl_loss, `rl_loss = neg_reward * sample_out.loss`, the `neg_reward` is obtained by `greedy_rouge - sample_rouge`, and the `sample_out.loss` means...

Recommendation

1

[NLP Course | For You](https://lena-voita.github.io/nlp_course.html)

Hi, thanks for this wonderful dataset. How can I get the train/valid/test split as mentioned in the paper?

Hi, is there any documentation about how to use this web?