Gendered-term mention ratio control on GCGender
-0.92Delta M Mention RatiosNeuron-TST
Evaluation results
| Method | Links | |||
|---|---|---|---|---|
| sNeuron-TSTModel=Llama 3.1, Target=Masculine2026.05 | -0.92 | -0.27 | -0.34 | |
| sNeuron-TSTModel=Qwen 2.5, Target=Masculine2026.05 | -0.83 | -0.54 | 3.39 | |
| LAPEModel=Llama 3.1, Target=Masculine2026.05 | -0.62 | -0.03 | 0.07 | |
| LAPEModel=Qwen 2.5, Target=Masculine2026.05 | -0.57 | -0.38 | -0.16 | |
| LAPEModel=Qwen 2.5, Target=Feminine2026.05 | -0.06 | -1.83 | 0.13 | |
| neuron-level intervention approachModel=Qwen 2.5, Target=Neutral2026.05 | -0.02 | -0.04 | -0.01 | |
| LAPEModel=Llama 3.1, Target=Feminine2026.05 | -0.02 | -0.33 | 0.15 | |
| neuron-level intervention approachModel=Llama 3.1, Target=Feminine2026.05 | -0.02 | 0.1 | -0.01 | |
| neuron-level intervention approachModel=Qwen 2.5, Target=Feminine2026.05 | -0.01 | 0.07 | -0.05 | |
| LAPEModel=Qwen 2.5, Target=Neutral2026.05 | 0 | 0.02 | -1.05 | |
| neuron-level intervention approachModel=Llama 3.1, Target=Neutral2026.05 | 0 | 0 | 0 | |
| sNeuron-TSTModel=Qwen 2.5, Target=Feminine2026.05 | 0.02 | -1.53 | -0.34 | |
| LAPEModel=Llama 3.1, Target=Neutral2026.05 | 0.02 | -0.03 | -0.46 | |
| sNeuron-TSTModel=Llama 3.1, Target=Neutral2026.05 | 0.04 | -0.04 | -1.16 | |
| neuron-level intervention approachModel=Qwen 2.5, Target=Masculine2026.05 | 0.06 | -0.11 | -0.02 | |
| sNeuron-TSTModel=Qwen 2.5, Target=Neutral2026.05 | 0.24 | -0.04 | 1.61 | |
| neuron-level intervention approachModel=Llama 3.1, Target=Masculine2026.05 | 0.76 | -0.38 | -0.02 | |
| sNeuron-TSTModel=Llama 3.1, Target=Feminine2026.05 | 2.45 | -4.14 | 0.45 |