Aya Red-teaming
Multilingual red teaming prompts across harm categories, from Cohere Labs.
Made by CohereDatasets and benchmarks
About Aya Red-teaming
The dataset is published by Cohere Labs on Hugging Face under the CohereLabs organization.
The repository is marked as containing sensitive content.
Drafted from huggingface.co · not yet confirmed by Cohere
How it works
Aya Red-teaming from Cohere Labs is a set of harmful prompts written by native speakers in eight languages (English, Hindi, French, Spanish, Arabic, Russian, Serbian, Filipino), labeled by harm category and whether the harm is global or culturally local. It's built to test whether safety holds outside English.
If you serve non English users, this is one of the few sets that doesn't rely on machine translation.
From the project
Reviews
Be the first to review it
AnonymousKnow a team weighing Aya Red-teaming? Send them this page.