Model Steering on Steering Evaluation Concepts Llama 3.3-70B (test)
1.2331GranularityDarkness Centric
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| Darkness CentricVector Type=Unit Mean, Coef.=4.5, Layer=292026.05 | 1.2331 | 81.7 | |
| NumericalVector Type=PV, Coef.=2.0, Layer=362026.05 | 1.2059 | 78.3 | |
| Golden Gate CentricVector Type=PV, Coef.=4.0, Layer=342026.05 | 1.198 | 66.1 | |
| Media CentricVector Type=PV, Coef.=3.5, Layer=402026.05 | 1.1935 | 69.8 | |
| EvilVector Type=PV, Coef.=2.0, Layer=302026.05 | 1.1857 | 59.9 | |
| GraphicalVector Type=PV, Coef.=3.0, Layer=452026.05 | 1.1298 | 81.7 | |
| HumorousVector Type=PV, Coef.=2.5, Layer=292026.05 | 1.1206 | 77.5 | |
| AnthropomorphicVector Type=PV, Coef.=4.0, Layer=242026.05 | 1.1158 | 85.7 | |
| ApatheticVector Type=PV, Coef.=3.0, Layer=252026.05 | 1.1048 | 69.9 | |
| Google CentricVector Type=Cluster, Coef.=3.0, Layer=382026.05 | 1.0942 | 78.6 | |
| EconomicVector Type=Unit Mean, Coef.=2.5, Layer=382026.05 | 1.0739 | 81.4 | |
| SycophanticVector Type=Unit Mean, Coef.=2.0, Layer=322026.05 | 1.0596 | 67.6 | |
| HedgingVector Type=PV, Coef.=2.5, Layer=442026.05 | 1.0469 | 90.3 | |
| MaritimeVector Type=PV, Coef.=4.0, Layer=292026.05 | 1.046 | 83.5 | |
| Self DeprecatingVector Type=PV, Coef.=3.0, Layer=272026.05 | 1.0436 | 62.6 | |
| MoralizingVector Type=Unit Mean, Coef.=3.0, Layer=282026.05 | 1.0122 | 86 | |
| CryptographicVector Type=PV, Coef.=2.5, Layer=432026.05 | 1.0072 | 83.5 | |
| HallucinatingVector Type=PV, Coef.=2.5, Layer=292026.05 | 1.0012 | 84.6 | |
| Culinary CentricVector Type=PV, Coef.=4.5, Layer=342026.05 | 0.9617 | 78.1 | |
| ProfessionalVector Type=Unit Mean, Coef.=1.0, Layer=52026.05 | 0.9552 | 93.5 |