List your company

Safety Alignment Evaluation on PKU-SafeRLHF 30K (test)

90.23Win Rate (WR)wDPO
21.194839.117457.0474.9626Mar 7, 2026

Evaluation results

MethodLinks
wDPOBackbone=Qwen2.5-7B2026.0390.231.630
Dr.DPOBackbone=Qwen2.5-7B2026.0388.932.010.13
wDPOBackbone=Llama-3.2-3B2026.0385.451.630.13
wDPOBackbone=Llama-3-8B2026.0384.014.510.38
Dr.DPOBackbone=Llama-3.2-3B2026.0383.772.380.5
rDPOBackbone=Qwen2.5-7B2026.0382.744.512.01
Dr.DPOBackbone=Llama-3-8B2026.0382.54.390.5
rDPOBackbone=Llama-3-8B2026.0382.136.643.26
rDPOBackbone=Llama-3.2-3B2026.0381.34.141.75
DPOBackbone=Qwen2.5-7B2026.0375.8120.557.27
IPOBackbone=Qwen2.5-7B2026.0373.9721.38.02
DPOBackbone=Llama-3.2-3B2026.0373.711.286.14
cDPOBackbone=Llama-3.2-3B2026.0372.9125.4412.78
IPOBackbone=Llama-3.2-3B2026.0372.711.786.14
IPOBackbone=Llama-3-8B2026.0370.2226.8213.53
DPOBackbone=Llama-3-8B2026.0369.1926.5714.41
VanillaBackbone=Qwen2.5-7B2026.0369.0932.2113.16
cDPOBackbone=Qwen2.5-7B2026.0367.6531.8313.66
wDPOBackbone=Pythia-2.8B2026.0359.9213.033.01
cDPOBackbone=Llama-3-8B2026.0359.4837.9423.43
Dr.DPOBackbone=Pythia-2.8B2026.0357.4117.044.14
rDPOBackbone=Pythia-2.8B2026.0346.5457.7723.43
IPOBackbone=Pythia-2.8B2026.0345.2761.6522.68
cDPOBackbone=Pythia-2.8B2026.0345.2760.2816.92
VanillaBackbone=Llama-3.2-3B2026.0344.9366.7918.42
DPOBackbone=Pythia-2.8B2026.0344.2357.1419.17
VanillaBackbone=Llama-3-8B2026.0342.8267.5424.69
SFTBackbone=Pythia-2.8B2026.0341.0571.326.32
VanillaBackbone=Pythia-2.8B2026.0340.1873.3126.94
SFTBackbone=Llama-3.2-3B2026.0335.9368.6745.36
SFTBackbone=Qwen2.5-7B2026.0325.0970.5552.88
SFTBackbone=Llama-3-8B2026.0323.8569.9252.51