h0ssn/

wikitext-calibration-50-mia

This dataset contains a calibration set for training membership inference attack (MIA) classifiers.

Total Samples: 50 Members: 25 (sampled from training data) Non-members: 25 (sampled from unseen test split) Balance: 50.0% members / 50.0% non-members

	Data Sources




	


	Members (label=1)

Source: 10K training data (excluding eval set) Dataset:… See the full description on the dataset page: https://huggingface.co/datasets/h0ssn/wikitext-calibration-50-mia.