Commonsense Reasoning on OpenBookQA (Inter/Intra-Layer Filtering)
75.6Accuracy (Inter-layer)CRFT
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| CRFTBackbone=LLaMA-3-8B, Identify=SSF2025.07 | 75.6 | 77 | |
| CRFTBackbone=LLaMA-3-8B, Identify=MAF2025.07 | 73.8 | 72.6 | |
| CRFTBackbone=Mistral-7B, Identify=MSF2025.07 | 72.8 | 68.6 | |
| CRFTBackbone=Mistral-7B, Identify=SAF2025.07 | 72.6 | 72.6 | |
| CRFTBackbone=LLaMA-3-8B, Identify=SAF2025.07 | 71 | 73.6 | |
| CRFTBackbone=LLaMA-3-8B, Identify=MSF2025.07 | 70 | 68.6 | |
| ReFTBackbone=LLaMA-3-8B, Identify=p7+s72025.07 | 66 | 66 | |
| CRFTBackbone=Mistral-7B, Identify=MAF2025.07 | 64.2 | 69.6 | |
| ReFTBackbone=Mistral-7B, Identify=p7+s72025.07 | 63.8 | 63.8 | |
| CRFTBackbone=Mistral-7B, Identify=SSF2025.07 | 61.6 | 66.4 | |
| CRFTBackbone=LLaMA-2-7B, Identify=MSF2025.07 | 59 | 56.4 | |
| ReFTBackbone=LLaMA-2-7B, Identify=p7+s72025.07 | 58.6 | 58.6 | |
| CRFTBackbone=LLaMA-2-7B, Identify=SSF2025.07 | 58.4 | 58.6 | |
| CRFTBackbone=LLaMA-2-7B, Identify=SAF2025.07 | 57 | 62.2 | |
| CRFTBackbone=LLaMA-2-7B, Identify=MAF2025.07 | 50.6 | 66.4 |