MiniGuard Training Dataset
Training data for MiniGuard-v0.1, a compact content safety classifier.
Subset
Samples
Description
Standard
40,000
English subset of nvidia/Nemotron-Safety-Guard-Dataset-v3
Thinking-Augmented
34,658
Reasoning traces from openai/gpt-oss-safeguard-120b
MiniGuard Targeted
1,199
Synthetic hard examples for edge cases generated using Hermes-4.3-36B
Each example contains a conversations list… See the full description on the dataset page:
https://huggingface.co/datasets/prem-research/MiniGuard-Safety-Dataset.