List your company

Safety Detection on Alpaca + XSTest full (val-selected)

0.991AUROCwopt
0.49180.62140.7510.8806Apr 20, 2026

Evaluation results

MethodLinks
woptModel=Llama-3.2 (abliterated), Layer=52026.040.9910.894
θ two-classModel=Llama-3.2 (instruct), Layer=52026.040.9890.809
wLDAModel=Llama-3.2 (abliterated), Layer=52026.040.9890.869
θ two-classModel=Llama-3.2 (abliterated), Layer=52026.040.9890.816
wLDAModel=Llama-3.2 (instruct), Layer=52026.040.9880.857
woptModel=Llama-3.2 (instruct), Layer=52026.040.9880.857
woptModel=Gemma-3 (instruct), Layer=192026.040.9810.766
woptModel=Llama-3.2 (base), Layer=62026.040.980.819
woptModel=Gemma-3 (abliterated), Layer=192026.040.980.742
wLDAModel=Gemma-3 (instruct), Layer=192026.040.9790.76
θ two-classModel=Llama-3.2 (base), Layer=62026.040.9770.718
wLDAModel=Gemma-3 (abliterated), Layer=192026.040.9770.718
wLDAModel=Llama-3.2 (base), Layer=62026.040.9740.676
θ two-classModel=Gemma-3 (instruct), Layer=192026.040.9730.61
θ two-classModel=Gemma-3 (abliterated), Layer=192026.040.9710.672
woptModel=Gemma-3 (base), Layer=192026.040.970.713
wLDAModel=Gemma-3 (base), Layer=192026.040.9690.708
θ two-classModel=Gemma-3 (base), Layer=192026.040.9630.538
wPC1Model=Llama-3.2 (abliterated), Layer=52026.040.7560.021
wPC1Model=Llama-3.2 (base), Layer=62026.040.7450.037
RandomModel=Gemma-3 (instruct), Layer=192026.040.7410.078
wPC1Model=Gemma-3 (instruct), Layer=192026.040.7210.015
wPC1Model=Gemma-3 (base), Layer=192026.040.7110.024
RandomModel=Llama-3.2 (base), Layer=62026.040.7020.07
wPC1Model=Gemma-3 (abliterated), Layer=192026.040.6850.01
wPC1Model=Llama-3.2 (instruct), Layer=52026.040.6770
RandomModel=Llama-3.2 (abliterated), Layer=52026.040.6670.031
RandomModel=Llama-3.2 (instruct), Layer=52026.040.6420.019
PerplexityModel=Llama-3.2 (abliterated), Layer=—2026.040.6030.004
PerplexityModel=Gemma-3 (abliterated), Layer=—2026.040.590.003
RandomModel=Gemma-3 (abliterated), Layer=192026.040.5750.034
θ-normModel=Gemma-3 (abliterated), Layer=192026.040.570.01
RandomModel=Gemma-3 (base), Layer=192026.040.5690.013
PerplexityModel=Gemma-3 (base), Layer=—2026.040.5680.003
θ-normModel=Llama-3.2 (instruct), Layer=52026.040.5570.012
PerplexityModel=Llama-3.2 (base), Layer=—2026.040.5530.004
PerplexityModel=Gemma-3 (instruct), Layer=—2026.040.540.001
θ-normModel=Llama-3.2 (base), Layer=62026.040.5340.003
θ-normModel=Llama-3.2 (abliterated), Layer=52026.040.5170.004
θ-normModel=Gemma-3 (instruct), Layer=192026.040.5170
PerplexityModel=Llama-3.2 (instruct), Layer=—2026.040.5150
θ-normModel=Gemma-3 (base), Layer=192026.040.5110.001