Selective Calibration on Model-specific (val)
11.7ECEPCHI
Evaluation Results
| Method | Links | ||||
|---|---|---|---|---|---|
| PCHIModel=Qwen3-4B2026.06 | 11.7 | 90.3 | 82.2 | 10.8 | |
| PCHIModel=Gemma3-4B2026.06 | 16.9 | 81.7 | 52.5 | 11.6 | |
| No interventionModel=Qwen3-4B2026.06 | 21.9 | 66.5 | — | — | |
| Activation steeringModel=Qwen3-4B2026.06 | 21.9 | 66.7 | 0.1 | 0 | |
| Activation steeringModel=Gemma3-4B2026.06 | 26.2 | 76.8 | 0.5 | 0 | |
| No interventionModel=Gemma3-4B2026.06 | 26.3 | 76.8 | — | — |