Dataset Card for Summarize from Feedback

Dataset Description

In the Learning to Summarize from Human Feedback paper , a reward model was trained from human feedback. The reward model was then used to train a summarization model to align with human preferences. This is the dataset of human feedback that was released for reward modelling. There are two parts of this dataset: comparisons and axis . In the comparisons part, human annotators were asked to choose the best out of two summaries. In the axis part, human annotators gave scores on a likert scale for the quality of a summary. The comparisons part only has a train and validation split, and the axis part only has a test and validation split.

The summaries used for training the reward model in the paper come from the TL;DR dataset. Additional validation and test data come from the TL;DR dataset, CNN articles, and Daily Mail articles.

For more information, see the repo here .

Citation Information

https://arxiv.org/abs/2009.01325

@inproceedings{stienon2020learning,
  author = {Nisan Stiennon and Long Ouyang and Jeff Wu and Daniel M. Ziegler and Ryan Lowe and Chelsea Voss and Alec Radford and Dario Amodei and Paul Christiano},
  title = {Learning to summarize from human feedback},
  booktitle = {NeurIPS},
  year = 2020,
}

Dataset added to the Hugging Face Hub with help from @Tristan