List your company

Safety Alignment Robustness Evaluation on PKU-SafeRLHF (n=100 samples)

18.5Random RateSFT Base
16.73628.64340.5552.457May 11, 2026

Evaluation results

MethodLinks
SFT BaseCategory=Baselines & Neutral2026.0518.519.616.318.718.3
EGPO (HF)Category=Baselines & Neutral, Source=Hugging Face2026.0523.225.13431.828.5
NMD k = 1 NeutralCategory=Baselines & Neutral, k=1, Type=Neutral2026.0548.753.172.263.959.5
EG k = 1 NeutralCategory=Baselines & Neutral, k=1, Type=Neutral2026.0553.956.87465.362.5
IPO k = 1 NeutralCategory=Baselines & Neutral, k=1, Type=Neutral2026.0555.658.975.77265.6
IPO Entropic τ=5Objective=Risk-Trained (Entropic), τ=52026.0558.161.974.470.566.2
IPO CVaR α=0.125Objective=Risk-Trained (CVaR), α=0.1252026.0558.559.975.370.666.1
IPO Entropic τ=10Objective=Risk-Trained (Entropic), τ=102026.0561.964.478.474.269.7
IPO CVaR α=0.25Objective=Risk-Trained (CVaR), α=0.252026.0562.46175.671.567.6
EG Entropic τ=5Objective=Risk-Trained (Entropic), τ=52026.0562.657.975.768.366.1