Question Answering on Fairness Evaluation Suite BBQ, CrP, GMO, SSt, WnQ
98.26BBQ AccuracySD-E
Evaluation results
| Method | Links | ||||||
|---|---|---|---|---|---|---|---|
| SD-EBackbone=Qwen-2.5-7B2026.06 | 98.26 | 81.61 | 98.58 | 58 | 96.27 | 86.54 | |
| ADBPBackbone=Qwen-2.5-7B2026.06 | 97.99 | 83 | 98.58 | 60.5 | 93.68 | 86.75 | |
| Fair-GCGBackbone=Qwen-2.5-7B2026.06 | 97.9 | 91.31 | 99.15 | 82 | 97.91 | 93.65 | |
| Luo et al.Backbone=Qwen-2.5-7B2026.06 | 97.8 | 79.22 | 98.29 | 52 | 90 | 83.46 | |
| VanillaBackbone=Qwen-2.5-7B2026.06 | 96.25 | 76.57 | 98.29 | 49 | 87.45 | 81.51 | |
| IASCBackbone=Qwen-2.5-7B2026.06 | 95.61 | 72.42 | 94.3 | 45 | 82.45 | 77.96 | |
| SD-RBackbone=Qwen-2.5-7B2026.06 | 94.51 | 78.72 | 99.15 | 58.5 | 95.09 | 85.19 | |
| VanillaBackbone=DeepSeek-R1-Distill-Llama-70B2026.06 | 93.93 | 88.79 | 100 | 78.75 | 99.05 | 92.1 | |
| Fair-GCGBackbone=GPT-OSS-20B2026.06 | 93.87 | 73.2 | 99.15 | 56 | 96.5 | 83.74 | |
| VanillaBackbone=GPT-OSS-20B2026.06 | 93.41 | 72.42 | 99.15 | 55.25 | 95.36 | 83.12 | |
| Fair-GCGBackbone=DeepSeek-R1-Distill-Llama-70B2026.06 | 93.1 | 90.3 | 100 | 80.5 | 99.05 | 92.59 | |
| Luo et al.Backbone=Llama 3.1 8B2026.06 | 91.95 | 79.09 | 83.48 | 73.5 | 66.73 | 78.95 | |
| SD-EBackbone=Llama 3.1 8B2026.06 | 90.85 | 91.18 | 76.07 | 80.25 | 97.73 | 87.22 | |
| ADBPBackbone=Llama 3.1 8B2026.06 | 84.81 | 84.89 | 88.32 | 73.25 | 92.91 | 82.82 | |
| SD-RBackbone=Llama 3.1 8B2026.06 | 82.53 | 92.7 | 77.78 | 84.5 | 99.09 | 87.32 | |
| Fair-GCGBackbone=Llama 3.1 8B2026.06 | 81.34 | 94.21 | 78.35 | 88.5 | 97.55 | 87.99 | |
| VanillaBackbone=Llama 3.1 8B2026.06 | 74.38 | 78.34 | 66.38 | 59.25 | 90.55 | 73.78 | |
| IASCBackbone=Llama 3.1 8B2026.06 | 72.92 | 75.94 | 82.91 | 51.5 | 77.18 | 72.09 |