Fact Verification on SciFact (test)
84AccuracyDesign B
Evaluation results
| Method | Links | |||||
|---|---|---|---|---|---|---|
| Design BModel=Qwen3-4B2026.06 | 84 | 50.5 | 2.3 | 48.4 | 1.3 | |
| Design A rescueModel=Qwen3-14B2026.06 | 84 | 66.7 | 2.1 | 63.8 | 0.7 | |
| Design A 2-sidedModel=Qwen3-14B2026.06 | 84 | 66.7 | 2.1 | 63.8 | 0.7 | |
| Design BModel=Qwen3-14B2026.06 | 84 | 66.7 | 2.1 | 63.8 | 0.7 | |
| Design A rescueModel=Qwen3-4B2026.06 | 83.2 | 50 | 1.8 | 48.9 | 1.8 | |
| Design A 2-sidedModel=Qwen3-4B2026.06 | 83.2 | 50 | 1.8 | 48.9 | 1.8 | |
| SIFTModel=Qwen3-14B2026.06 | 80.9 | 64.6 | — | 63.1 | — | |
| SIFTModel=Qwen3-4B2026.06 | 79.4 | 48.2 | — | 47.1 | — | |
| Design A rescueModel=Gemma2-9B2026.06 | 79.4 | 46.6 | 0.5 | 47.6 | 0.5 | |
| Design A 2-sidedModel=Gemma2-9B2026.06 | 79.4 | 46.6 | 0.5 | 47.6 | 0.5 | |
| Design BModel=Gemma2-9B2026.06 | 79.4 | 46.6 | 0.5 | 47.6 | 0.5 | |
| DirectModel=Qwen3-14B2026.06 | 79.4 | 41.5 | — | — | — | |
| SIFTModel=Gemma2-9B2026.06 | 78.6 | 46.1 | — | 47.1 | — | |
| DirectModel=Gemma2-9B2026.06 | 77.9 | 38.7 | — | 32 | — | |
| DirectModel=Qwen3-4B2026.06 | 71.8 | 45.6 | — | 50.9 | — | |
| Design BModel=Llama3.1-8B2026.06 | 70.2 | 63.1 | 5.1 | 56.9 | -1.1 | |
| NLI-as-judge (DeBERTa)2026.06 | 69.7 | — | — | — | — | |
| DirectModel=Llama3.1-8B2026.06 | 67.9 | 30.9 | — | 27.9 | — | |
| Design A rescueModel=Llama3.1-8B2026.06 | 64.1 | 59.6 | 1.6 | 58 | 0 | |
| Design A 2-sidedModel=Llama3.1-8B2026.06 | 64.1 | 59.6 | 1.6 | 58 | 0 | |
| SIFTModel=Llama3.1-8B2026.06 | 58.8 | 58 | — | 58 | — |