datasets
Training and evaluation data, with the modality, task and licence stated up front. Listed live from the Hugging Face Hub.
cleaned_auditor
Cleaned Auditor Dataset (29,953 samples)
This dataset is a cleaned and filtered version of the OpenSafetyLab/t2i_safety_dataset.
It contains 29,953 samples curated for binary and multi-class safety auditing.
Dataset Structure
The dataset contains the following columns:
image: The generated image (PIL Image).
prompt: The text prompt used to generate the image.
disturbing: 1 if the image category is 'disturbing', else 0.
hate: 1 if the image category is 'hate', else 0.… See the full description on the dataset page: https://huggingface.co/datasets/kricko/cleaned_auditor.cleaned_auditor
Balanced Auditor Training Dataset (Curated 200)
This is a highly curated version of the Auditor training data, balanced to contain exactly 50 samples for each of the 4 primary safety labels.
Dataset Summary
The dataset contains 200 rows in total, focusing on the most critical safety categories for auditing.
Labels
Safe: Images that do not violate safety policies. (50 samples)
Violence: Images containing violent content. (50 samples)
Sexual: Images containing… See the full description on the dataset page: https://huggingface.co/datasets/AIISC1/cleaned_auditor.simo2-contrastive-audit
SimO2 Contrastive Audit Artifacts
This dataset repo backs up experiment artifacts for the SimO2 / AFCL contrastive-learning paper audit.
Archive:
simo_contrastive_cifar_stl_tiny_outputs_20260610.tar
Contents include:
outputs/ run directories with model_final.pt, metrics.json, embeddings, and logs where present.
CIFAR-100, STL-10, and Tiny-ImageNet-200 AFCL/Yat-SupCon/SupCon/CE audit runs.
Retrieval and gradient-mass summary CSV/JSON files under outputs/_summary/.
Queue… See the full description on the dataset page: https://huggingface.co/datasets/mlnomad/simo2-contrastive-audit.