Gender Neuron Masking on GCGender Masculine
1.21Count of Gendered Terms per Response (M)Ours
Evaluation Results
| Method | Links | |||||||||
|---|---|---|---|---|---|---|---|---|---|---|
| OursBase Model=Llama 3.12026.05 | 1.21 | 0.15 | 0.3 | 6.62 | 0.8 | 1.69 | 76 | -38 | -2 | |
| BaselineBase Model=Llama 3.12026.05 | 1.08 | 0.22 | 0.31 | 5.86 | 1.18 | 1.71 | — | — | — | |
| sNeuron-TSTBase Model=Llama 3.12026.05 | 1.01 | 0.19 | 0.28 | 4.94 | 0.91 | 1.37 | -92 | -27 | -34 | |
| LAPEBase Model=Llama 3.12026.05 | 1 | 0.22 | 0.34 | 5.24 | 1.15 | 1.78 | -62 | -3 | 7 |