Gender Neuron Masking on InclusiveGender Feminine
0.8Count of Gendered Terms per Response (M)sNeuron-TST
Evaluation Results
| Method | Links | |||||||||
|---|---|---|---|---|---|---|---|---|---|---|
| sNeuron-TSTBase Model=Llama 3.12026.05 | 0.8 | 0.66 | 0.32 | 6.14 | 3.18 | 2.21 | 3.54 | -4.37 | -0.02 | |
| BaselineBase Model=Llama 3.12026.05 | 0.38 | 1.1 | 0.33 | 2.6 | 7.55 | 2.23 | — | — | — | |
| OursBase Model=Llama 3.12026.05 | 0.33 | 1.22 | 0.31 | 2.27 | 8.44 | 2.19 | -0.33 | 0.89 | -0.04 | |
| LAPEBase Model=Llama 3.12026.05 | 0.28 | 1.03 | 0.35 | 2.24 | 7.36 | 2.27 | -0.36 | -0.19 | 0.04 |