List your company

Safety Evaluation on DANGEROUSQA Llama-2 base

15.3Chinese Safety ScoreSFT
0.52164.35838.19512.0317Feb 19, 2024

Evaluation results

MethodLinks
SFTFine-tuning Protocol=PEFT2024.0215.33.155.828.556.317.836.32
SFT+DAREFine-tuning Protocol=PEFT2024.0213.252.14.768.942.136.244.73
SFTFine-tuning Protocol=Full-FT2024.028.741.0410.815.292.594.843.33
SFT+DAREFine-tuning Protocol=Full-FT2024.027.441.048.195.292.073.962.45
SFT+RESTAFine-tuning Protocol=PEFT2024.027.0202.642.130.512.460.95
SFT+RESTAdFine-tuning Protocol=Full-FT2024.022.2102.141.5800.87-0.64
SFT+RESTAdFine-tuning Protocol=PEFT2024.022.110.532.640.211.561.41-0.1
SFT+RESTAFine-tuning Protocol=Full-FT2024.021.0902.140.5200.65-0.86