Steering Success Rate on PopQA 150 questions
33Base SRDeepSeek 7B
Evaluation Results
| Method | Links | |||
|---|---|---|---|---|
| DeepSeek 7BModel=DeepSeek 7B2025.11 | 33 | 45.6 | 12.6 | |
| Llama 3 8BModel=Llama 3 8B2025.11 | 33 | 42.6 | 9.6 | |
| Qwen 2.5 7BModel=Qwen 2.5 7B2025.11 | 24.2 | 28.6 | 4.3 | |
| Qwen 2.5 32BModel=Qwen 2.5 32B2025.11 | 21.2 | 34.8 | 13.6 | |
| Gemma 2 9BModel=Gemma 2 9B2025.11 | 14.1 | 39.6 | 25.4 |