WMDP
Benchmarks
Task NameDataset NameSOTA ResultTrendResults
WMDP bio
71.2Accuracy
93
WMDP
24.8Bio Accuracy
74
WMDP evil
65.37WMDP-evil Score
60
WMDP cyber
47.21Accuracy
47
WMDP Biology
64.5Default Score
38
WMDP Cyber QA
44.3Default Accuracy
38
WMDP retain
48.3Retain
36
WMDP-Bio 100-sample subset
0.93ASR
36
WMDP Cyber (test)
61.15MMLU
29
WMDP-cyber 1.0 (test)
53.7BF16 Score
28
WMDP-chem 1.0 (test)
0.56BF16
28
WMDP-bio 1.0 (test)
80.3BF16 Accuracy
28
WMDP
75.9Performance (Bio)
26
WMDP
68.98Accuracy
26
WMDP retain
55.6Retain
22
WMDP (forget split)
55.4BF16 Precision
22
WMDP
3.46Mean Fluency
22
WMDP
74.16Acc (Bio)
21
WMDP
43S-unlearning Score
16
WMDP cyber (retain)
54.1Rt
16
WMDP-cyber forget-set
53.7BF16 Performance
16
WMDP-chem (retain)
56Rt (Knowledge Retention)
16
WMDP chem forget-set
56BF16 Score
16
WMDP bio (retain)
80.3Rt
16
WMDP-cyber
0CAD
16