Tiny Aya Korean safety steering: datasets Collection English vs. Korean refusal and over-refusal of Tiny Aya Global, before and after steering, and the source prompt sets. • 4 items • Updated 4 days ago
Tiny Aya Korean safety steering: models Collection Inference-time safety steering for Tiny Aya Global (SafeTune CAST), and the models it was built and judged with. • 3 items • Updated 4 days ago