List your company

Reward Scoring on HHA benchmark

66.97Harmlessness Score (Base)SFT+DPO
31.672440.83625059.1638Feb 16, 2025

Evaluation results

MethodLinks
SFT+DPOModel=Qwen2.5-14B2025.0266.9767.8368.1668.9468.3167.97
SFT+VARModel=Qwen2.5-14B2025.0266.3767.7471.771.0870.1769.22
SFT+VARModel=Qwen2.5-3B2025.0265.1567.8671.2470.7869.9668.75
SFT+VARModel=Qwen2.5-7B2025.0264.9666.4669.8169.3968.5567.65
VARModel=Qwen2.5-7B2025.0264.8666.4170.8769.7869.0267.98
VARModel=Qwen2.5-14B2025.0264.2465.2369.7968.567.8466.94
VARModel=Qwen2.5-3B2025.0263.366.2970.0769.3268.5667.24
SFT+VARModel=Qwen2.5-1.5B2025.0262.7666.0769.1368.7867.9966.69
DPOModel=Qwen2.5-7B2025.0261.5166.0966.567.2866.6265.35
VARModel=Qwen2.5-1.5B2025.0261.3364.7268.8768.0167.265.73
SFT+DPOModel=Qwen2.5-7B2025.0260.7862.0261.6262.6762.161.77
DPOModel=Qwen2.5-3B2025.0260.5863.3763.8464.964.0363.17
SFT+VARModel=Qwen2.5-0.5B2025.0258.2261.7263.5863.5662.9561.77
SFT+DPOModel=Qwen2.5-0.5B2025.0256.4258.0260.3859.9159.4458.68
DPOModel=Qwen2.5-14B2025.0255.8355.6359.0457.6157.4357.03
VARModel=Qwen2.5-0.5B2025.0255.2258.0962.3260.3860.2659
DPOModel=Qwen2.5-0.5B2025.0255.2155.556.2456.7556.1755.93
SFT+DPOModel=Qwen2.5-3B2025.0254.6158.4558.6960.1959.1157.98
SFT+DPOModel=Qwen2.5-1.5B2025.0254.5161.1361.5462.2661.6459.86
DPOModel=Qwen2.5-1.5B2025.0253.45757.3857.8557.4156.41
SFT+VARModel=Qwen2.5-32B-Int42025.0253.1849.0755.6651.952.2152.45
VARModel=Qwen2.5-32B-Int42025.0250.0345.3651.7747.7748.348.73
BaseModel=Qwen2.5-3B2025.0247.0734.2741.8636.6137.5839.96
BaseModel=Qwen2.5-7B2025.0242.4942.0247.9644.2144.7344.17
BaseModel=Qwen2.5-14B2025.0239.2932.7339.0335.1635.6436.55
BaseModel=Qwen2.5-32B-Int42025.0238.834.3639.1336.7836.7737.27
SFT+DPOModel=Qwen2.5-32B-Int42025.0237.9527.728.6328.1628.1630.61
DPOModel=Qwen2.5-32B-Int42025.0237.0931.3834.5833.043334.02
BaseModel=Qwen2.5-1.5B2025.0235.0126.1832.1128.1328.8130.36
BaseModel=Qwen2.5-0.5B2025.0233.0325.4430.8626.9427.7529.06