killdevil111/DataShield-Sample-Risk

DataShield This dataset releases sample-level risk scores for DataShield: Uncovering Risky Fine-Tuning Data Across LLMs Through Consensus Subspace Alignment, accepted to the EMNLP Main Conference. For the method, code, and complete documentation, see the DataShield GitHub repository. Dataset configurations Configuration Source dataset Rows dolly15k databricks/databricks-dolly-15k 15,011 alpaca52k tatsu-lab/alpaca 51,974 from datasets import… See the full description on the dataset page: https://huggingface.co/datasets/killdevil111/DataShield-Sample-Risk.