Question Answering on COPA (Accuracy)
96AccuracyPaLM-2 L
Evaluation Results
| Method | Links | |
|---|---|---|
| PaLM-2 Lprotocol=1-shot2023.11 | 96 | |
| Falcon-180Bprotocol=1-shot2023.11 | 96 | |
| Spherical SteeringZero-shot=true, Backbone=LLaMA-3.1-8B-Instruct2026.02 | 95 | |
| PaLMprotocol=1-shot2023.11 | 91 | |
| PaLM-2 Mprotocol=1-shot2023.11 | 90 | |
| PaLM-2 Sprotocol=1-shot2023.11 | 89 | |
| BaseBackbone=Qwen-3-4B2026.01 | 86.21 | |
| LFTFBackbone=Qwen-3-4B2026.01 | 85.95 | |
| FairSteerBackbone=Qwen-3-4B2026.01 | 85 | |
| KnowBiasBackbone=Qwen-3-4B2026.01 | 84.7 | |
| CAAZero-shot=true, Backbone=LLaMA-3.1-8B-Instruct2026.02 | 84 | |
| SADI-HEADZero-shot=true, Backbone=LLaMA-3.1-8B-Instruct2026.02 | 84 | |
| CRISPRBackbone=Qwen-3-4B, Debias target=race2026.01 | 83.3 | |
| CRISPRBackbone=Qwen-3-4B, Debias target=gender2026.01 | 83.05 | |
| BaselineZero-shot=true, Backbone=LLaMA-3.1-8B-Instruct2026.02 | 83 | |
| ITIZero-shot=true, Backbone=LLaMA-3.1-8B-Instruct2026.02 | 83 | |
| CRISPRBackbone=Qwen-3-4B, Debias target=religion2026.01 | 82.7 | |
| PEFTBackbone=Qwen-3-4B, Debias target=race2026.01 | 81 | |
| ReGiFTBackbone=Qwen-3-4B2026.01 | 80.1 | |
| SDBackbone=Qwen-3-4B2026.01 | 73.4 | |
| CRISPRBackbone=Llama-3.1-8B, Debias target=gender2026.01 | 72.2 | |
| BaseBackbone=Llama-3.1-8B2026.01 | 71.32 | |
| BiasEditBackbone=Llama-3.1-8B, Debias target=religion2026.01 | 71.18 | |
| LFTFBackbone=Llama-3.1-8B2026.01 | 71.05 | |
| PEFTBackbone=Qwen-3-4B, Debias target=religion2026.01 | 70.8 | |
| PEFTBackbone=Llama-3.1-8B, Debias target=religion2026.01 | 70.24 | |
| CRISPRBackbone=Llama-3.1-8B, Debias target=religion2026.01 | 68.5 | |
| FairSteerBackbone=Llama-3.1-8B2026.01 | 68.12 | |
| KnowBiasBackbone=Llama-3.1-8B2026.01 | 67.8 | |
| BiasEditBackbone=Llama-3.1-8B, Debias target=race2026.01 | 67.14 | |
| CRISPRBackbone=Llama-3.1-8B, Debias target=race2026.01 | 66.5 | |
| BiasEditBackbone=Qwen-3-4B, Debias target=religion2026.01 | 64.9 | |
| PEFTBackbone=Qwen-3-4B, Debias target=gender2026.01 | 63.8 | |
| BiasEditBackbone=Llama-3.1-8B, Debias target=gender2026.01 | 62.8 | |
| PEFTBackbone=Llama-3.1-8B, Debias target=gender2026.01 | 61.44 | |
| FairSteerBackbone=Llama-3.2-3B2026.01 | 60.95 | |
| LFTFBackbone=Llama-3.2-3B2026.01 | 60.75 | |
| BaseBackbone=Llama-3.2-3B2026.01 | 60.73 | |
| ReGiFTBackbone=Llama-3.1-8B2026.01 | 59.62 | |
| CRISPRBackbone=Llama-3.2-3B, Debias target=race2026.01 | 59 | |
| PEFTBackbone=Llama-3.1-8B, Debias target=race2026.01 | 58.29 | |
| KnowBiasBackbone=Llama-3.2-3B2026.01 | 57.7 | |
| CRISPRBackbone=Llama-3.2-3B, Debias target=gender2026.01 | 57.62 | |
| SDBackbone=Llama-3.1-8B2026.01 | 55.5 | |
| CRISPRBackbone=Llama-3.2-3B, Debias target=religion2026.01 | 53.17 | |
| BiasEditBackbone=Qwen-3-4B, Debias target=race2026.01 | 51 | |
| BiasEditBackbone=Qwen-3-4B, Debias target=gender2026.01 | 49.6 | |
| PEFTBackbone=Llama-3.2-3B, Debias target=race2026.01 | 49.5 | |
| SDBackbone=Llama-3.2-3B2026.01 | 47.1 | |
| PEFTBackbone=Llama-3.2-3B, Debias target=gender2026.01 | 45.73 | |
| BiasEditBackbone=Llama-3.2-3B, Debias target=religion2026.01 | 44.8 | |
| ReGiFTBackbone=Llama-3.2-3B2026.01 | 44.37 | |
| PEFTBackbone=Llama-3.2-3B, Debias target=religion2026.01 | 41.56 | |
| BiasEditBackbone=Llama-3.2-3B, Debias target=race2026.01 | 33 | |
| BiasEditBackbone=Llama-3.2-3B, Debias target=gender2026.01 | 31.21 | |
| MUPPET2022.12 | 0.65 | |
| ColD-Fusion2022.12 | 0.644 | |
| Multitask2022.12 | 0.6 | |
| Finetune2022.12 | 0.47 |