Bias Mitigation on 100 Gender-Neutral Prompts (test)
0GAS (Explicit Bias)M'
Evaluation Results
| Method | Links | |||
|---|---|---|---|---|
| M'Base Model=Llama-2-7B, Decoding Strategy=greedy2025.11 | 0 | 36 | 7.7 | |
| M+Base Model=Vicuna-7B, Decoding Strategy=greedy2025.11 | 0 | 26 | 7.4 | |
| M'Base Model=Vicuna-7B, Decoding Strategy=greedy2025.11 | 0 | 27 | 7.11 | |
| M+Base Model=Vicuna-13B, Decoding Strategy=greedy, Precision=float162025.11 | 0 | 43 | 7.69 | |
| M'Base Model=Vicuna-13B, Decoding Strategy=greedy, Precision=float162025.11 | 0 | 14 | 7.83 | |
| M+Base Model=Llama-2-7B, Decoding Strategy=greedy2025.11 | 1 | 47 | 7.27 | |
| M_safepromptBase Model=Llama-2-7B, Decoding Strategy=greedy2025.11 | 28 | 66 | 5.3 | |
| MBase Model=Llama-2-7B, Decoding Strategy=greedy2025.11 | 29 | 67 | 5.23 | |
| M_safepromptBase Model=Vicuna-7B, Decoding Strategy=greedy2025.11 | 30 | 63 | 6.9 | |
| MBase Model=Vicuna-7B, Decoding Strategy=greedy2025.11 | 31 | 64 | 6.8 | |
| M_safepromptBase Model=Vicuna-13B, Decoding Strategy=greedy, Precision=float162025.11 | 32 | 46 | 7 | |
| MBase Model=Vicuna-13B, Decoding Strategy=greedy, Precision=float162025.11 | 33 | 45 | 6.94 |