Dataset · nvidia
LLM response safety: 33k dialogues
nvidia/Aegis-AI-Content-Safety-Dataset-2.0
33,416 annotated human-LLM interaction examples for classifying unsafe content.
What's inside
The dataset contains 33,416 annotated human-LLM interactions split into 30,007 training, 1,445 validation, and 1,964 test examples. It is an expansion of the Nemotron Content Safety Dataset V1, collected using human preference data from Hugging Face. The language is English.
What you can do with it
- Train a content safety classifier
- Evaluate LLM response moderation
- Fine-tune an LLM moderator on annotated dialogues
- Tasks
- Text classification
- Languages
- English
- Size
- 10K–100K rows
- Downloads this month
- 24.3K
- Likes
- 120
Models trained on it
- GaleneAI/llama-3.1-nemoguard-8b-content-safety-merged-NVFP4 83.1K
- jaredpalmer/kev-0.8bText classification 12.3K
- jaredpalmer/kev-4bText classification 9.7K
- ahczhg/Llama-3.2-1B-Aegis-SFT-DPOText generation 4.2K
- jaredpalmer/kev-27bText classification 3.4K
- jaredpalmer/kev-9bText classification 2.3K
- vllm-sr/Vela-2.0-0.8BZero shot classification 2K
- vllm-sr/Vela-2.0-0.3BZero shot classification 2K
How to load it
from datasets import load_dataset
ds = load_dataset("nvidia/Aegis-AI-Content-Safety-Dataset-2.0")