List your company

Fine-grained Safety Control on CoSApien Allowed instructions

88Accuracy (GD)PALETTE
44.3255.666778.34May 22, 2026

Evaluation results

MethodLinks
PALETTEBase Model=LLaMA2-7B-Chat2026.058810088.20.353
CASTBase Model=LLaMA2-7B-Chat2026.058282.284.30.326
AutoDANBase Model=LLaMA2-7B-Chat2026.054810068.60.168
BaseBase Model=LLaMA2-7B-Chat2026.054610066.70.365