Question Answering on BBQ (test)
98.86Accuracy (amb)MAD
Evaluation Results
| Method | Links | ||||||
|---|---|---|---|---|---|---|---|
| MADModel=Llama-3.1-70B-Instruct2026.01 | 98.86 | 81.78 | 90.32 | 0.0055 | 0.0127 | 0.0091 | |
| MPTModel=Llama-3.1-70B-Instruct2026.01 | 98.46 | 87.02 | 92.74 | 0.003 | 0.0077 | 0.0053 | |
| Re-PromptingVariant=Debias, Model=Llama-3.1-70B-Instruct2026.01 | 98.33 | 80.87 | 89.6 | 0.0063 | 0.0341 | 0.0202 | |
| Re-PromptingVariant=Persona, Model=Llama-3.1-70B-Instruct2026.01 | 96.86 | 71.79 | 84.32 | 0.0104 | 0.0373 | 0.0238 | |
| Self-ConsistencyVariant=Debias, Model=Llama-3.1-70B-Instruct2026.01 | 96.49 | 86.23 | 91.36 | 0.0223 | 0.0112 | 0.0168 | |
| Self-ConsistencyVariant=Standard, Model=Llama-3.1-70B-Instruct2026.01 | 94.29 | 92.19 | 93.24 | 0.0424 | 0.0136 | 0.028 | |
| Direct-PromptingVariant=Debias, Model=Llama-3.1-70B-Instruct2026.01 | 92.82 | 79.72 | 86.27 | 0.0211 | 0.0109 | 0.016 | |
| Direct-PromptingVariant=Standard, Model=Llama-3.1-70B-Instruct2026.01 | 92.73 | 88.56 | 90.64 | 0.0406 | 0.0115 | 0.026 | |
| Self-ConsistencyVariant=Persona, Model=Llama-3.1-70B-Instruct2026.01 | 91.26 | 75.43 | 83.35 | 0.0156 | 0.0063 | 0.0109 | |
| MPTModel=Llama-3.1-8B-Instruct2026.01 | 90.46 | 87.55 | 89.01 | 0.0275 | 0.0849 | 0.0562 | |
| Direct-PromptingVariant=Persona, Model=Llama-3.1-70B-Instruct2026.01 | 83.96 | 71 | 77.48 | 0.0166 | 0.0108 | 0.0137 | |
| Re-PromptingVariant=Debias, Model=Llama-3.1-8B-Instruct2026.01 | 83.36 | 80.77 | 82.07 | 0.0387 | 0.0985 | 0.0686 | |
| MADModel=Llama-3.1-8B-Instruct2026.01 | 83.29 | 83.11 | 83.2 | 0.0357 | 0.1158 | 0.0758 | |
| Re-PromptingVariant=Persona, Model=Llama-3.1-8B-Instruct2026.01 | 73.3 | 70.55 | 71.92 | 0.0359 | 0.1164 | 0.0761 | |
| Self-ConsistencyVariant=Debias, Model=Llama-3.1-8B-Instruct2026.01 | 64.61 | 90.5 | 77.55 | 0.0811 | 0.0793 | 0.0802 | |
| Self-ConsistencyVariant=Standard, Model=Llama-3.1-8B-Instruct2026.01 | 64.48 | 91.6 | 78.04 | 0.0841 | 0.0677 | 0.0759 | |
| Direct-PromptingVariant=Standard, Model=Llama-3.1-8B-Instruct2026.01 | 61.75 | 89.76 | 75.75 | 0.0839 | 0.0721 | 0.078 | |
| Direct-PromptingVariant=Debias, Model=Llama-3.1-8B-Instruct2026.01 | 61.03 | 88.01 | 74.52 | 0.0676 | 0.087 | 0.0773 | |
| Self-ConsistencyVariant=Persona, Model=Llama-3.1-8B-Instruct2026.01 | 55.74 | 90.11 | 72.92 | 0.1213 | 0.0704 | 0.0958 | |
| Direct-PromptingVariant=Persona, Model=Llama-3.1-8B-Instruct2026.01 | 53.36 | 86.57 | 69.96 | 0.088 | 0.0754 | 0.0817 |