HEX-PHI
Benchmarks
Task NameDataset NameSOTA ResultTrendResults
HEx-PHI
97.2HEx-PHI Score
162
HEx-PHI
1.8Attack Success Rate (ASR)
107
HEx-PHI
0Attack Success Rate
63
HEX-PHI (test)
2Harmfulness Score (Llama-Guard-3B)
56
HEx-PHI (out-of-domain)
0ASR (OM)
26
HEx-PHI
95.1Attack Success Rate (ASR)
24
HEx-PHI Alpaca risk-ranked subsets
13.1S1 ASR (%)
21
HEx-PHI Dolly risk-ranked
8.97S1 ASR
21
HEx-PHI (150 questions)
18.94ASR (No Trigger)
20
HEx-PHI Identity Shifting Attack (300 questions)
22.83ASR
20
HEx-PHI
0.7Harmful Response Rate
18
HEx-PHI
98.8HEx-PHI Score
18
HEx-PHI
2.06Harmfulness Score
16
HEX-PHI
54.9ASR
16
HEX-PHI
1.74Harmful Score
16
HEx-PHI
1.26Harm Score
15
HEx-PHI (test)
76.3MIP
14
HEx-PHI (test)
86.67ASR Category 1
12
HEx-PHI
69.87Safety Score (HEx-PHI)
10
HEx-PHI
99.06Accuracy
9
HEx-PHI (first 100 harmful prompts)
45ASR
8
HEx-PHI Direct
99.67Safety Score (1-ASR)
8
HEx-PHI Harmful
95.76Attack Success Rate
6
HEx-PHI
1.26Harmfulness Score
6
HEx-PHI out
67.29Attack Success Rate (OM)
6