10 results

abliterix

inall models datasets apps agents people projects

Datasets

All datasets matching “abliterix”

BIGJUTT /abliterix-datasets Abliterix Bilingual Refusal Datasets ⚠️ Safety notice. The harmful_* splits contain prompts designed to elicit genuinely harmful, hateful, or dangerous content (CBRNE, self-harm, cyber-attacks, etc.). They exist for alignment / abliteration / red-team research only — specifically to compute refusal directions and to evaluate whether a model still refuses. Do not use them for capability training or for any non-research purpose. Bilingual (English / Chinese / mixed) harmful +… See the full description on the dataset page: https://huggingface.co/datasets/BIGJUTT/abliterix-datasets.texttext-generation1K<n<10K1 likes27 downloads4mo agoHugging Face