LLM Steering on BASE→SYS.NEG (out-of-distribution)
2.944SteerabilityL2S
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| L2SModel=Llama-2-7B-Chat, Layer Selection Strategy=W2S2026.04 | 2.944 | 96.4 | |
| L2SModel=Llama-2-7B-Chat, Layer Selection Strategy=Fixed Layer2026.04 | 2.62 | 94.5 | |
| CAAModel=Llama-2-7B-Chat, Layer Selection Strategy=W2S2026.04 | 1.902 | 78 | |
| L2SModel=Qwen1.5-14B-Chat, Layer Selection Strategy=W2S2026.04 | 1.897 | 91.4 | |
| L2SModel=Qwen1.5-14B-Chat, Layer Selection Strategy=Fixed Layer2026.04 | 1.683 | 87.9 | |
| CAAModel=Llama-2-7B-Chat, Layer Selection Strategy=Fixed Layer2026.04 | 1.503 | 73.6 | |
| CAAModel=Qwen1.5-14B-Chat, Layer Selection Strategy=W2S2026.04 | 1.472 | 72.3 | |
| CAAModel=Qwen1.5-14B-Chat, Layer Selection Strategy=Fixed Layer2026.04 | 1.143 | 69.3 |