Share your thoughts, 1 month free Claude Pro on usSee more
WorkDL logo mark

Jailbreak Attack on JBB (Multi-Model ASR/Loss Evaluation)

91.7Llama2-7B ASR

Faster-I-GCG

21.60439.8025876.198Oct 20, 2024
Updated 14d ago

Evaluation Results

MethodLinks
2024.10
91.70.1061.20148.41.35276.3-----
2024.10
89.60.1081.45343.11.31577.11.16956.10.34958.264.8
2024.10
71.40.170.65377.60.34884.20.46668.60.14988.778.1
2024.10
68.90.2780.9550.10.4483.60.67256.60.29674.566.7
2024.10
660.2680.86354.20.44581.40.51467.10.25178.769.5
2024.10
58.20.350.98147.70.48580.60.63964.50.31277.965.8
2024.10
49.40.5311.77629.40.66171.81.16841.5---
2024.10
42.9--65.3-96.8-85-4767.4
2024.10
35--37.8-88-57.5-49.253.5
2024.10
27.71.8254.34123.15.7471.81.85436.1---
2024.10
26.9--49.7-90.5-89.7-42.859.9
2024.10
24.3--28.6-93.1-84.9-7.547.7