Stereotypical Bias Evaluation on CrowS-Pairs (CP)
64.7CP AccuracyCOFT
Evaluation results
| Method | Links | ||
|---|---|---|---|
| COFTBackbone=Mistral-7B-Instruct2026.05 | 64.7 | 1 | |
| COFTBackbone=LLaMA-2-13B2026.05 | 63.5 | 1 | |
| DT-CD⋆Backbone=Mistral-7B-Instruct2026.05 | 62.4 | 2.6 | |
| DExpertsBackbone=Mistral-7B-Instruct2026.05 | 62.1 | 3.1 | |
| DT-CD⋆Backbone=LLaMA-2-13B2026.05 | 61.3 | 2.8 | |
| SDDBackbone=Mistral-7B-Instruct2026.05 | 61.2 | 3.9 | |
| DExpertsBackbone=LLaMA-2-13B2026.05 | 61 | 3.3 | |
| SDDBackbone=LLaMA-2-13B2026.05 | 60.1 | 4 | |
| VanillaBackbone=Mistral-7B-Instruct2026.05 | 59.8 | 5.8 | |
| VanillaBackbone=LLaMA-2-13B2026.05 | 58.7 | 5.8 |