List your company

Function Deletion on HarmfulBench

-99Delta Refusal Rate (%)CNT
-102.0532-81.4441-60.835-40.2259Mar 19, 2026

Evaluation results

MethodLinks
CNTModel=Llama3-8B, Transfer Rate (TR)=0.00232026.03-9991.3495
EDModel=Llama3-8B2026.03-98.678678.34
EDModel=Llama3.2-3B2026.03-97.338373.66
CNTModel=Llama3.2-3B, Transfer Rate (TR)=0.00112026.03-97.337667.66
RDModel=Llama3-8B2026.03-96.6782.3487.34
CNTModel=DeepSeek-V2-Lite, Transfer Rate (TR)=0.00082026.03-8787.3481.67
EDModel=DeepSeek-V2-Lite2026.03-84.677970
CNTModel=Average, Transfer Rate (TR)=0.00142026.03-81.676.877.13
RDModel=Llama3.2-3B2026.03-81.3356.3449.66
EDModel=Average2026.03-81.271.1362.73
RDModel=Average2026.03-77.1358.860.34
ActSVD-OPModel=Mistral-Nemo-Instruct2026.03-73.3454.6766
RDModel=Mistral-Nemo-Instruct2026.03-73.3455.6766
CNTModel=Mistral-Nemo-Instruct, Transfer Rate (TR)=0.00242026.03-72.6763.6775.67
RDModel=DeepSeek-V2-Lite2026.03-7041.6735.34
EDModel=Yi-1.5-6B2026.03-66.3362.6761.34
RDModel=Yi-1.5-6B2026.03-64.335863.34
EDModel=Mistral-Nemo-Instruct2026.03-594530.33
ActSVD-OPModel=Yi-1.5-6B2026.03-57.6663.6765.67
ActSVD-OPModel=Llama3.2-3B2026.03-545349.66
CNTModel=Yi-1.5-6B, Transfer Rate (TR)=0.00022026.03-5265.6765.67
ActSVD-OPModel=Average2026.03-51.9251.4254.5
ActSVD-OPModel=Llama3-8B2026.03-22.6734.3436.67