AdvBench
Benchmarks
Task NameDataset NameSOTA ResultTrendResults
AdvBench (in-domain)
0ASR (OM)
26
AdvBench 20% evaluation
97.12ASR
25
AdvBench
0Harm Rate
25
AdvBench Ensemble configuration GPT-4o
0Attack Success Rate (ASR)
25
AdvBench
10.5Suffix Compression Rate (SCR)
24
AdvBench Text
98Total Success Rate
24
AdvBench
98ASR
21
AdvBench 80/20 (test)
82.7ASR@1
21
Advbench Vicuna-33B Guard 100 prompts Original
0ASR
21
Advbench Llama2-70B Guard 100 prompts Original
0ASR
21
AdvBench
0.19Attack Success Rate (ASR)
20
AdvBench
3.33Targeted Content Rate (TCR)
20
AdvBench
1.37ATGR
20
AdvBench
0ASR (Detoxify)
20
AdvBench GCG
0ASR
20
AdvBench AutoDAN
0ASR
20
AdvBench DIJA
6.94ASR
20
AdvBench
0.968AUROC
20
AdvBench (query-specific)
37MR
20
AdvBench 520
96.7ASR (%)
18
AdvBench
100Refusal Rate
18
AdvBench No Attack (test)
0ASR (GPT-4o)
18
AdvBench-x
7.94ASR (English)
18
AdvBench Llama3.1-405B 1.0 (test)
0ASR
17
AdvBench GPT-4o-mini 1.0 (test)
81.35ASR
17