Gendered-term mention ratio control on InclusiveGender
-2.37Δ M Mention RatiosNeuron-TST
Evaluation results
| Method | Links | |||
|---|---|---|---|---|
| sNeuron-TSTModel=Llama 3.1, Target=Masculine2026.05 | -2.37 | 1.87 | 0.08 | |
| LAPEModel=Qwen 2.5, Target=Masculine2026.05 | -2.03 | -0.11 | 0.26 | |
| LAPEModel=Llama 3.1, Target=Masculine2026.05 | -1.65 | 0.59 | 0.42 | |
| LAPEModel=Llama 3.1, Target=Feminine2026.05 | -0.36 | -0.19 | 0.04 | |
| neuron-level intervention approachModel=Llama 3.1, Target=Feminine2026.05 | -0.33 | 0.89 | -0.04 | |
| LAPEModel=Qwen 2.5, Target=Neutral2026.05 | -0.16 | 0.06 | 5.81 | |
| sNeuron-TSTModel=Qwen 2.5, Target=Feminine2026.05 | -0.09 | -3.8 | 2.14 | |
| neuron-level intervention approachModel=Qwen 2.5, Target=Feminine2026.05 | -0.04 | 0.17 | -0.09 | |
| neuron-level intervention approachModel=Qwen 2.5, Target=Neutral2026.05 | -0.03 | -0.01 | 0.22 | |
| neuron-level intervention approachModel=Llama 3.1, Target=Neutral2026.05 | -0.02 | 0 | -0.31 | |
| neuron-level intervention approachModel=Qwen 2.5, Target=Masculine2026.05 | 0.02 | -0.03 | -0.03 | |
| LAPEModel=Llama 3.1, Target=Neutral2026.05 | 0.02 | 0.01 | -0.63 | |
| LAPEModel=Qwen 2.5, Target=Feminine2026.05 | 0.07 | -4 | 0.23 | |
| sNeuron-TSTModel=Qwen 2.5, Target=Masculine2026.05 | 0.17 | 0.07 | 2.23 | |
| sNeuron-TSTModel=Llama 3.1, Target=Neutral2026.05 | 0.17 | 0.25 | -10.42 | |
| neuron-level intervention approachModel=Llama 3.1, Target=Masculine2026.05 | 0.33 | -0.65 | -0.1 | |
| sNeuron-TSTModel=Qwen 2.5, Target=Neutral2026.05 | 1.11 | 0.3 | 3.4 | |
| sNeuron-TSTModel=Llama 3.1, Target=Feminine2026.05 | 3.54 | -4.37 | -0.02 |