List your company

Safety Alignment on PKU-SafeRLHF (IID)

89.26WRShaPO-T
21.233638.894356.55574.2157Feb 7, 2026

Evaluation results

MethodLinks
ShaPO-TBackbone=Qwen2.5-7B2026.0289.260.750.13
ShaPO-RBackbone=Qwen2.5-7B2026.0289.231.130
Dr.DPOBackbone=Qwen2.5-7B2026.0288.932.010.13
ShaPO-RBackbone=LLaMA-3-8B2026.0287.451.130.13
ShaPO-TBackbone=LLaMA-3.2-3B2026.0285.881.250.13
ShaPO-TBackbone=LLaMA-3-8B2026.0285.452.760.13
ShaPO-RBackbone=LLaMA-3.2-3B2026.0285.291.750.5
Dr.DPOBackbone=LLaMA-3.2-3B2026.0283.772.380.5
rDPOBackbone=Qwen2.5-7B2026.0282.744.512.01
Dr.DPOBackbone=LLaMA-3-8B2026.0282.54.390.5
rDPOBackbone=LLaMA-3-8B2026.0282.136.643.26
rDPOBackbone=LLaMA-3.2-3B2026.0281.34.141.75
DPOBackbone=Qwen2.5-7B2026.0275.8120.557.27
IPOBackbone=Qwen2.5-7B2026.0273.9721.38.02
DPOBackbone=LLaMA-3.2-3B2026.0273.711.286.14
cDPOBackbone=LLaMA-3.2-3B2026.0272.9125.4412.78
IPOBackbone=LLaMA-3.2-3B2026.0272.711.786.14
IPOBackbone=LLaMA-3-8B2026.0270.2226.8213.53
DPOBackbone=LLaMA-3-8B2026.0269.1926.5714.41
VanillaBackbone=Qwen2.5-7B2026.0269.0932.2113.16
cDPOBackbone=Qwen2.5-7B2026.0267.6531.8313.66
cDPOBackbone=LLaMA-3-8B2026.0259.4837.9423.43
ShaPO-RBackbone=Pythia-2.8B2026.0257.1411.42.76
ShaPO-TBackbone=Pythia-2.8B2026.0256.1415.161.76
Dr.DPOBackbone=Pythia-2.8B2026.0252.6629.9518.17
rDPOBackbone=Pythia-2.8B2026.0246.5457.7723.43
IPOBackbone=Pythia-2.8B2026.0245.2761.6522.68
cDPOBackbone=Pythia-2.8B2026.0245.2760.2816.92
VanillaBackbone=LLaMA-3.2-3B2026.0244.9366.7918.42
DPOBackbone=Pythia-2.8B2026.0244.2357.1419.17
VanillaBackbone=LLaMA-3-8B2026.0242.8267.5424.69
SFTBackbone=Pythia-2.8B2026.0241.0571.326.32
VanillaBackbone=Pythia-2.8B2026.0240.1873.3126.94
SFTBackbone=LLaMA-3.2-3B2026.0235.9368.6745.36
SFTBackbone=Qwen2.5-7B2026.0225.0970.5552.88
SFTBackbone=LLaMA-3-8B2026.0223.8569.9252.51