Share your thoughts, 1 month free Claude Pro on us
See more
Home
/
Benchmarks
Hazard Knowledge Evaluation on WMDP
Loading...
68.98
Accuracy
CoA
22.5648
34.6149
46.665
58.7151
Mar 24, 2026
Accuracy
Updated 25d ago
Evaluation Results
Method
Method
Links
Accuracy
CoA
Backbone=Mistral 7B
2026.03
68.98
SFT
Backbone=Llama3.1 8B
2026.03
68.16
CoA
Backbone=Llama3.1 8B
2026.03
67.76
Extra
Backbone=Llama3.1 8B
2026.03
67.48
SFT
Backbone=Mistral 7B
2026.03
67.48
PermLM
Backbone=Llama3.1 8B
2026.03
65.17
Extra
Backbone=Mistral 7B
2026.03
65.03
SFT
Backbone=Qwen3 1.7B
2026.03
61.77
Extra
Backbone=Qwen3 1.7B
2026.03
61.22
PermLM
Backbone=Mistral 7B
2026.03
60.41
CoA
Backbone=Qwen3 1.7B
2026.03
59.59
PermLM
Backbone=Qwen3 1.7B
2026.03
51.84
Base
Backbone=Llama3.1 8B
2026.03
48
Base
Backbone=Mistral 7B
2026.03
42
sudoLM
Backbone=Mistral 7B
2026.03
35.24
sudoLM
Backbone=Qwen3 1.7B
2026.03
29.39
Base
Backbone=Qwen3 1.7B
2026.03
28
sudoLM
Backbone=Llama3.1 8B
2026.03
24.35
Feedback
Search any
task
Search any
task