List your company

Refusal Detection on Alpaca + XSTest (val)

0.988AUROCwopt
0.524160.644580.7650.88542Apr 20, 2026

Evaluation results

MethodLinks
woptModel=Qwen2.5 (instruct), Layer=222026.040.98883.4
theta two-classModel=Qwen3.5 (base), Layer=102026.040.98881.5
woptModel=Qwen2.5 (base), Layer=222026.040.98783.6
wLDAModel=Qwen3.5 (base), Layer=102026.040.98783.3
woptModel=Qwen3.5 (base), Layer=102026.040.98784
woptModel=Qwen2.5 (abliterated), Layer=222026.040.98171.8
woptModel=Qwen3.5 (instruct), Layer=222026.040.97979.4
wLDAModel=Qwen2.5 (instruct), Layer=222026.040.97565.4
wLDAModel=Qwen2.5 (base), Layer=222026.040.97465.7
woptModel=Qwen3.5 (abliterated), Layer=222026.040.97475.4
theta two-classModel=Qwen2.5 (base), Layer=222026.040.96652.5
theta two-classModel=Qwen2.5 (instruct), Layer=222026.040.96654.2
wLDAModel=Qwen3.5 (instruct), Layer=222026.040.96464.2
wLDAModel=Qwen2.5 (abliterated), Layer=222026.040.96153
wLDAModel=Qwen3.5 (abliterated), Layer=222026.040.95857.5
theta two-classModel=Qwen2.5 (abliterated), Layer=222026.040.94536
theta two-classModel=Qwen3.5 (instruct), Layer=222026.040.91545.7
theta two-classModel=Qwen3.5 (abliterated), Layer=222026.040.90242.2
theta-normModel=Qwen3.5 (abliterated), Layer=222026.040.770.1
wPC1Model=Qwen3.5 (base), Layer=102026.040.762.2
theta-normModel=Qwen3.5 (instruct), Layer=222026.040.7580.3
RandomModel=Qwen3.5 (instruct), Layer=222026.040.6942.7
RandomModel=Qwen3.5 (abliterated), Layer=222026.040.6912.7
theta-normModel=Qwen2.5 (abliterated), Layer=222026.040.6790.3
RandomModel=Qwen3.5 (base), Layer=102026.040.6753.3
PerplexityModel=Qwen2.5 (abliterated), Layer=—2026.040.6670.1
theta-normModel=Qwen2.5 (instruct), Layer=222026.040.6610.3
PerplexityModel=Qwen3.5 (abliterated), Layer=—2026.040.6180.3
PerplexityModel=Qwen2.5 (base), Layer=—2026.040.6110.3
PerplexityModel=Qwen2.5 (instruct), Layer=—2026.040.6090
wPC1Model=Qwen2.5 (base), Layer=222026.040.6070.3
theta-normModel=Qwen2.5 (base), Layer=222026.040.6050
RandomModel=Qwen2.5 (instruct), Layer=222026.040.6042.2
theta-normModel=Qwen3.5 (base), Layer=102026.040.6020
wPC1Model=Qwen2.5 (abliterated), Layer=222026.040.60.3
RandomModel=Qwen2.5 (base), Layer=222026.040.5891
PerplexityModel=Qwen3.5 (base), Layer=—2026.040.5890.1
PerplexityModel=Qwen3.5 (instruct), Layer=—2026.040.5840.1
wPC1Model=Qwen2.5 (instruct), Layer=222026.040.5780.1
wPC1Model=Qwen3.5 (abliterated), Layer=222026.040.5680
wPC1Model=Qwen3.5 (instruct), Layer=222026.040.5670.1
RandomModel=Qwen2.5 (abliterated), Layer=222026.040.5422.1