Concept Alignment on RepE (test)
100HARMSelf Modulation
Evaluation Results
| Method | Links | ||||
|---|---|---|---|---|---|
| Self ModulationTarget LLM (mt)=Llama3.1, Source LLM (ms)=Llama3.12025.01 | 100 | 36 | 8.74 | 9.28 | |
| Self ModulationTarget LLM (mt)=Llama2, Source LLM (ms)=Llama22025.01 | 96 | 56 | 8.52 | 7.26 | |
| L-Cross ModulationTarget LLM (mt)=Llama2, Source LLM (ms)=Qwen22025.01 | 96 | 64 | 9.16 | 8.84 | |
| L-Cross ModulationTarget LLM (mt)=Llama2, Source LLM (ms)=Llama3.12025.01 | 96 | 54 | 8.92 | 7.96 | |
| Self ModulationTarget LLM (mt)=Qwen2, Source LLM (ms)=Qwen22025.01 | 88 | 50 | 7.04 | 6.26 | |
| No ModulationTarget LLM (mt)=Llama2, Source LLM (ms)=None2025.01 | 0 | 98 | 5.56 | 5.74 |