List your company

Safety Evaluation on Harmfulness Evaluation Sequences

0.79Harmfulness Scorellama2-13b-chat
-0.03160.18170.3950.6083May 23, 2024

Evaluation results

MethodLinks
llama2-13b-chatLearning Rate=3e-5, Attack Steps=1k2024.050.79
Qwen-4B-chatLearning Rate=8e-5, Attack Steps=1k2024.050.77
llama2-13b-chatLearning Rate=3e-5, Attack Steps=10k2024.050.76
Qwen-1.8B-chatLearning Rate=6e-5, Attack Steps=1k2024.050.76
Qwen-1.8B-chatLearning Rate=6e-5, Attack Steps=10k2024.050.76
Qwen-4B-chatLearning Rate=6e-5, Attack Steps=1k2024.050.76
Qwen-7B-chatLearning Rate=6e-5, Attack Steps=10k2024.050.76
Qwen-7B-chatLearning Rate=8e-5, Attack Steps=1k2024.050.76
Qwen-7B-chatLearning Rate=8e-5, Attack Steps=10k2024.050.76
llama2-13b-chatLearning Rate=8e-5, Attack Steps=1k2024.050.75
llama2-13b-chatLearning Rate=8e-5, Attack Steps=10k2024.050.75
Qwen-1.8B-chatLearning Rate=8e-5, Attack Steps=1k2024.050.75
Qwen-1.8B-chatLearning Rate=8e-5, Attack Steps=10k2024.050.75
Qwen-4B-chatLearning Rate=6e-5, Attack Steps=10k2024.050.75
Qwen-4B-chatLearning Rate=8e-5, Attack Steps=10k2024.050.75
llama2-7b-chatLearning Rate=3e-5, Attack Steps=10k2024.050.74
llama2-7b-chatLearning Rate=8e-5, Attack Steps=1k2024.050.74
Qwen-1.8B-chatLearning Rate=3e-5, Attack Steps=10k2024.050.74
Qwen-4B-chatLearning Rate=3e-5, Attack Steps=10k2024.050.74
Qwen-7B-chatLearning Rate=6e-5, Attack Steps=1k2024.050.74
llama2-7b-chatLearning Rate=6e-5, Attack Steps=1k2024.050.73
llama2-7b-chatLearning Rate=8e-5, Attack Steps=10k2024.050.73
llama2-13b-chatLearning Rate=6e-5, Attack Steps=1k2024.050.73
llama2-13b-chatLearning Rate=6e-5, Attack Steps=10k2024.050.73
Qwen-0.5B-chatLearning Rate=8e-5, Attack Steps=10k2024.050.73
llama2-7b-chatLearning Rate=6e-5, Attack Steps=10k2024.050.72
Qwen-1.8B-chatLearning Rate=3e-5, Attack Steps=1k2024.050.72
Qwen-4B-chatLearning Rate=3e-5, Attack Steps=1k2024.050.72
Qwen-7B-chatLearning Rate=3e-5, Attack Steps=10k2024.050.72
Qwen-0.5B-chatLearning Rate=6e-5, Attack Steps=10k2024.050.71
Qwen-0.5B-chatLearning Rate=8e-5, Attack Steps=1k2024.050.7
Qwen-7B-chatLearning Rate=3e-5, Attack Steps=1k2024.050.7
Qwen-0.5B-chatLearning Rate=6e-5, Attack Steps=1k2024.050.68
Qwen-0.5B-chatLearning Rate=3e-5, Attack Steps=10k2024.050.66
Qwen-0.5B-chatLearning Rate=3e-5, Attack Steps=1k2024.050.5
llama2-7b-chatLearning Rate=3e-5, Attack Steps=1k2024.050.47
RepNoise (llama2-7b-chat)Learning Rate=6e-5, Attack Steps=10k2024.050.13
RepNoise (llama2-7b-chat)Learning Rate=3e-5, Attack Steps=10k2024.050.12
RepNoise (llama2-7b-chat)Learning Rate=8e-5, Attack Steps=10k2024.050.12
RepNoise (Qwen-0.5B-chat)Learning Rate=8e-5, Attack Steps=1k2024.050.12
RepNoise (llama2-7b-chat)Learning Rate=8e-5, Attack Steps=1k2024.050.11
RepNoise (llama2-7b-chat)Learning Rate=6e-5, Attack Steps=1k2024.050.1
RepNoise (Qwen-0.5B-chat)Learning Rate=3e-5, Attack Steps=10k2024.050.1
RepNoise (Qwen-4B-chat)Learning Rate=8e-5, Attack Steps=10k2024.050.1
RepNoise (Qwen-0.5B-chat)Learning Rate=6e-5, Attack Steps=1k2024.050.09
RepNoise (Qwen-4B-chat)Condition=pre-attack2024.050.09
RepNoise (llama2-7b-chat)Learning Rate=3e-5, Attack Steps=1k2024.050.08
Qwen-7B-chatCondition=pre-attack2024.050.08
llama2-13b-chatCondition=pre-attack2024.050.07
RepNoise (Qwen-0.5B-chat)Learning Rate=3e-5, Attack Steps=1k2024.050.07
RepNoise (Qwen-0.5B-chat)Condition=pre-attack2024.050.06
llama2-7b-chatCondition=pre-attack2024.050.05
RepNoise (llama2-7b-chat)Condition=pre-attack2024.050.05
RepNoise (llama2-13b-chat)Learning Rate=6e-5, Attack Steps=10k2024.050.05
RepNoise (Qwen-4B-chat)Learning Rate=6e-5, Attack Steps=10k2024.050.04
RepNoise (Qwen-1.8B-chat)Condition=pre-attack2024.050.02
RepNoise (Qwen-4B-chat)Learning Rate=3e-5, Attack Steps=1k2024.050.02
RepNoise (Qwen-4B-chat)Learning Rate=6e-5, Attack Steps=1k2024.050.02
RepNoise (Qwen-0.5B-chat)Learning Rate=8e-5, Attack Steps=10k2024.050.01
RepNoise (Qwen-1.8B-chat)Learning Rate=3e-5, Attack Steps=1k2024.050.01
RepNoise (Qwen-7B-chat)Condition=pre-attack2024.050.01
RepNoise (Qwen-7B-chat)Learning Rate=3e-5, Attack Steps=10k2024.050.01
RepNoise (Qwen-7B-chat)Learning Rate=8e-5, Attack Steps=10k2024.050.01
RepNoise (llama2-13b-chat)Condition=pre-attack2024.050
RepNoise (llama2-13b-chat)Learning Rate=3e-5, Attack Steps=1k2024.050
RepNoise (llama2-13b-chat)Learning Rate=3e-5, Attack Steps=10k2024.050
RepNoise (llama2-13b-chat)Learning Rate=6e-5, Attack Steps=1k2024.050
RepNoise (llama2-13b-chat)Learning Rate=8e-5, Attack Steps=1k2024.050
RepNoise (llama2-13b-chat)Learning Rate=8e-5, Attack Steps=10k2024.050
Qwen-0.5B-chatCondition=pre-attack2024.050
RepNoise (Qwen-0.5B-chat)Learning Rate=6e-5, Attack Steps=10k2024.050
Qwen-1.8B-chatCondition=pre-attack2024.050
RepNoise (Qwen-1.8B-chat)Learning Rate=3e-5, Attack Steps=10k2024.050
RepNoise (Qwen-1.8B-chat)Learning Rate=6e-5, Attack Steps=1k2024.050
RepNoise (Qwen-1.8B-chat)Learning Rate=6e-5, Attack Steps=10k2024.050
RepNoise (Qwen-1.8B-chat)Learning Rate=8e-5, Attack Steps=1k2024.050
RepNoise (Qwen-1.8B-chat)Learning Rate=8e-5, Attack Steps=10k2024.050
Qwen-4B-chatCondition=pre-attack2024.050
RepNoise (Qwen-4B-chat)Learning Rate=3e-5, Attack Steps=10k2024.050
RepNoise (Qwen-4B-chat)Learning Rate=8e-5, Attack Steps=1k2024.050
RepNoise (Qwen-7B-chat)Learning Rate=3e-5, Attack Steps=1k2024.050
RepNoise (Qwen-7B-chat)Learning Rate=6e-5, Attack Steps=1k2024.050
RepNoise (Qwen-7B-chat)Learning Rate=6e-5, Attack Steps=10k2024.050
RepNoise (Qwen-7B-chat)Learning Rate=8e-5, Attack Steps=1k2024.050