APTO-XSTest-JA is a Japanese translated and annotated version of the XSTest dataset for AI safety evaluation research.
XSTest is a test suite designed to identify exaggerated safety behaviours in large language models, including cases where models refuse clearly safe prompts because they contain sensitive wording or resemble unsafe requests.
This dataset includes:
Japanese translations of XSTest prompts
Japanese refusal / non-refusal reference responses
Refusal… See the full description on the dataset page:
https://huggingface.co/datasets/APTO-001/APTO-XSTest-JA.
