List your company

Detoxification on AdvBench harmful behavior set

99.17Safety ScorePaCE
8.22231.833555.44579.0565Jun 6, 2024

Evaluation results

MethodLinks
PaCEBackbone=LlaMA2-13B-Chat2024.0699.17
OrthoProjBackbone=LlaMA2-13B-Chat2024.0696.72
PaCEBackbone=LlaMA2-7B-Chat2024.0696.65
VecAddBackbone=LlaMA2-13B-Chat2024.0695.33
VecAddBackbone=LlaMA2-7B-Chat2024.0694.51
PEBackbone=LlaMA2-13B-Chat2024.0693.86
OrthoProjBackbone=LlaMA2-7B-Chat2024.0692.81
PEBackbone=LlaMA2-7B-Chat2024.0691.9
VanillaBackbone=LlaMA2-13B-Chat2024.0618.04
VanillaBackbone=LlaMA2-7B-Chat2024.0611.72