List your company

Adversarial Safety Evaluation on AdvBench (test)

73.86GCG Success RateSFT
-2.954416.987836.9356.8722Mar 3, 2026

Evaluation results

MethodLinks
SFTBase Model=LLaMA-3.1-8B-Instruct, Safety Alignment=SFT2026.0373.8672.8880.4896.4428.5786.651.2449.62
NemoGuardBase Model=LLaMA-3.1-8B-Instruct, Safety Alignment=NemoGuard2026.0359.576581.5394.6257.1464.5226.4231.35
RLHFBase Model=LLaMA-2-7B-Chat, Safety Alignment=RLHF2026.0347.8451.0231.3580.1624.178.4227.10
SFTBase Model=Qwen2.5-7B-Instruct, Safety Alignment=SFT2026.0343.482747.2193.2736.7388.6525.628.72
Egida-DPOBase Model=Qwen2.5-7B-Instruct, Safety Alignment=Egida-DPO2026.0339.3629.550.7889.8128.5783.6514.686.92
PSR (N=8)Base Model=LLaMA-3.1-8B-Instruct, Safety Alignment=PSR (N=8)2026.0325.02010.0535.5822.4532.690.260
SFTBase Model=Qwen2.5-14B-Instruct, Safety Alignment=SFT2026.0323.8518.524.7174.6229.2572.611.351.73
PSA DetectorBase Model=Qwen2.5-14B-Instruct, Safety Alignment=PSA Detector2026.0323.65013.04026.5364.042.320
ShieldGemmaBase Model=Qwen2.5-14B-Instruct, Safety Alignment=ShieldGemma2026.0322.121424.5174.8128.5770.774.511.73
TSC-GRPOBase Model=LLaMA-2-7B-Chat, Safety Alignment=TSC-GRPO2026.0318.6215.5618.1725.5112.466.4700
TSC-GRPOBase Model=LLaMA-3.1-8B-Instruct, Safety Alignment=TSC-GRPO2026.0317.9404.6324.4817.2824.4900
PSR (N=8)Base Model=Qwen2.5-7B-Instruct, Safety Alignment=PSR (N=8)2026.033.42110.0535.5822.450.711.860
TSC-GRPOBase Model=Qwen2.5-7B-Instruct, Safety Alignment=TSC-GRPO2026.031.3502.5826.3216.45000
PSR (N=8)Base Model=Qwen2.5-14B-Instruct, Safety Alignment=PSR (N=8)2026.031.35014.5937.4419.396.5400
TSC-GRPOBase Model=Qwen2.5-14B-Instruct, Safety Alignment=TSC-GRPO2026.03004.4725.3714.480.7100