Aegis Content Safety
NVIDIA's dataset for content moderation and toxicity detection.
Made by NVIDIADatasets and benchmarks
About Aegis Content Safety
Aegis AI Content Safety Dataset 2.0 is a dataset published under NVIDIA's account on Hugging Face. It is listed under the text classification task and uses the text modality.
Its tags include content moderation.
Drafted from huggingface.co · not yet confirmed by NVIDIA
How it works
Aegis is NVIDIA's content safety dataset: tens of thousands of human annotated interactions with an LLM, labeled against a taxonomy of 12 core risk categories plus a "needs caution" class. Version 2.0 adds more data, multi turn context and a refined taxonomy. NVIDIA used it to train its Aegis guard models, and it's meant to train yours.
Text only, English, and one of the larger open safety training sets.
From the project
From the project
Reviews
Be the first to review it
AnonymousKnow a team weighing Aegis Content Safety? Send them this page.