datasets
Training and evaluation data, with the modality, task and licence stated up front. Listed live from the Hugging Face Hub.
med-hallucination-detection
Medical hallucination detection
A dataset for training a small model to detect hallucinations in medical answers
and explain why, by checking each answer against the context it should be
grounded in. Each row is a (question, answer, context) triple with a row_type:
not_hallucinated -- the answer is grounded in its context.
hallucinated -- the answer is not (sourced separately; see below).
The not_hallucinated split (this build)
Derived from MedQuAD, a collection… See the full description on the dataset page: https://huggingface.co/datasets/Certops/med-hallucination-detection.med-hallucination-detection-unfiltered
Medical hallucination detection (unfiltered)
A dataset for training a small model to detect hallucinations in medical answers
and explain why, by checking each answer against the context it should be
grounded in. Each row is a (question, answer, context) triple labelled row_type.
This is the unfiltered union of two sources: 7,464 grounded positives and
10,000 planted-hallucination negatives. It is the raw pool before sampling and
judging -- the downstream step samples from here… See the full description on the dataset page: https://huggingface.co/datasets/Certops/med-hallucination-detection-unfiltered.