Share your thoughts, 1 month free Claude Pro on us
See more
Home
/
Benchmarks
Model Steering on 10-task steering suite (excluding JailbreakBench) (test)
Loading...
88.4
Accuracy
CLAS
51.064
60.757
70.45
80.143
Apr 27, 2026
Accuracy
Updated 1mo ago
Evaluation Results
Method
Method
Links
Accuracy
CLAS
Base Model=Llama-3.1-70B
2026.04
88.4
CLAS
Base Model=Qwen2.5-7B
2026.04
86.3
ReFT
Base Model=Llama-3.1-70B
2026.04
84.9
CLAS
Base Model=Llama-3.1-8B
2026.04
83.5
LoRA
Base Model=Llama-3.1-8B
2026.04
82.9
LoRA
Base Model=Qwen2.5-7B
2026.04
78.8
LoRA
Base Model=Llama-3.1-70B
2026.04
77.9
ReFT
Base Model=Qwen2.5-7B
2026.04
76.8
LAS
Base Model=Llama-3.1-70B
2026.04
75.1
LAS
Base Model=Llama-3.1-8B
2026.04
74.6
ReFT
Base Model=Llama-3.1-8B
2026.04
73.7
ReFT
Base Model=Llama-3.2-1B
2026.04
65.1
LAS
Base Model=Qwen2.5-7B
2026.04
64
LoRA
Base Model=Llama-3.2-1B
2026.04
62.4
CLAS
Base Model=Llama-3.2-1B
2026.04
58.3
LAS
Base Model=Llama-3.2-1B
2026.04
52.5
Feedback
Search any
task
Search any
task