List your company

Unsafety evaluation on CATQA

37.7Chinese Safety ScoreSFT+DARE
3.43212.328521.22530.1215Feb 19, 2024

Evaluation results

MethodLinks
SFT+DAREFine-tuning Protocol=PEFT2024.0237.78.9731.9848.3930.4731.529.32
SFTFine-tuning Protocol=PEFT2024.0236.0321.1633.2146.3931.0733.5731.39
SFT+DAREFine-tuning Protocol=Full-FT2024.0228.884.7336.9125.913.6622.0219.84
SFTFine-tuning Protocol=Full-FT2024.0228.878.0433.9826.9812.9122.1619.98
SFT+RESTAFine-tuning Protocol=PEFT2024.0214.986.213.5420.9310.4613.2211.04
SFT+RESTAdFine-tuning Protocol=PEFT2024.0212.734.5712.2720.1511.1512.179.99
SFT+RESTAFine-tuning Protocol=Full-FT2024.027.091.279.383.851.824.682.5
SFT+RESTAdFine-tuning Protocol=Full-FT2024.024.751.2710.173.671.824.342.16