Commonsense Fact Verification on CREAK (test)
88.6AccuracyRACO
Evaluation Results
| Method | Links | |
|---|---|---|
| RACO#Total Params.=≥ 3B, Single-task Training=false2023.05 | 88.6 | |
| DECKER#Total Params.=~449M, Single-task Training=true2023.05 | 88.4 | |
| T5-3B#Total Params.=~3B, Single-task Training=false2023.05 | 85.1 | |
| UNICORN#Total Params.=~770M, Single-task Training=false2023.05 | 79.5 | |
| ICRBackbone=Qwen2.5-7B2025.09 | 79.2 | |
| Zero-shotBackbone=Qwen2.5-7B2025.09 | 77.8 | |
| GreaseLM#Total Params.=~359M, Single-task Training=true2023.05 | 77.5 | |
| I2CLBackbone=Qwen2.5-7B2025.09 | 74.8 | |
| M²IVBackbone=Qwen2.5-7B2025.09 | 74.6 | |
| LIVEBackbone=Qwen2.5-7B2025.09 | 72.8 | |
| Few-shot*Backbone=Qwen2.5-7B, Few-shot settings=3-shot OOD2025.09 | 59.7 | |
| ICRBackbone=Llama2-7B2025.09 | 56.4 | |
| M²IVBackbone=Llama2-7B2025.09 | 54 | |
| I2CLBackbone=Llama2-7B2025.09 | 52.6 | |
| Zero-shotBackbone=Llama2-7B2025.09 | 51.8 | |
| LIVEBackbone=Llama2-7B2025.09 | 51 | |
| Few-shot*Backbone=Llama2-7B, Few-shot settings=3-shot OOD2025.09 | 50.8 |