6 datasets found

The PARLAI SINGLE ADVERSARIALbenchmark dataset consists of single-turn conversations annotated based on offensiveness.

The PARLAI SINGLE STANDARDBenchmark dataset consists of single-turn conversations annotated based on offensiveness.

The PARLAI dataset consists of single-turn conversations annotated based on offensiveness.

The ToxicSpans dataset is a subset of the Civil Comments dataset, containing toxic spans.

The Jigsaw dataset is a collection of text, where each text is labeled as toxic or non-toxic.

RealToxicityPrompts constitutes a collection of 100k naturally occurring sentences, amassed from various internet sources and designed to function as LM prompts.