ethz-spylab/curated-harmless-dataset download history

ethz-spylab/curated-harmless-dataset is a dataset on the Hugging Face Hub. In the last 30 days it was downloaded 24 times (6 in the last 7 days), and 2,343 times in total. It ranks #326,781 among datasets by monthly downloads.

This dataset contains manually crafted pairs of comparison to address reward hacking in RLHF optimization. Examples of rejected behavior are: very long responses, empty responses, referring to companies policy of usage, etc.

Open ethz-spylab/curated-harmless-dataset on Hugging Face

Sister project: Paper Pulse, the upvote history of every Hugging Face Daily Paper.