List your company

Safety Detection on Human-labeled evaluation set

92.86Unsafe Recall (HarmBench)12B Curriculum Judge
71.924877.359982.79588.2301Jun 8, 2026

Evaluation results

MethodLinks
12B Curriculum JudgeParameters=12B, Training Strategy=Curriculum2026.0692.8692.7995.652.86
12B (Fixed)Parameters=12B, Training Strategy=Fixed2026.0691.8488.7490.693.1
12B (Dynamic)Parameters=12B, Training Strategy=Dynamic2026.0690.3484.2587.346.09
Llama-Guard-3-8BParameters=8B2026.0672.7384.4295.1322.4