Abstention in Question Answering on UMWP Underspecified Context
94Abstention F1TIAR
Evaluation Results
| Method | Links | ||||
|---|---|---|---|---|---|
| TIARModel=Qwen3-8B2026.05 | 94 | 97.6 | 90.7 | 98.5 | |
| TruthRLModel=Qwen3-8B2026.05 | 92.9 | 97.5 | 88.8 | 98.3 | |
| DPOModel=Qwen3-8B2026.05 | 92.5 | 97.6 | 87.9 | 98.5 | |
| R-TuningModel=Qwen3-8B2026.05 | 92.4 | 97.8 | 87.6 | 97.9 | |
| RFTModel=Qwen3-8B2026.05 | 91.9 | 92.9 | 90.9 | 95.9 | |
| TruthRLModel=Llama-3.1-8B-Instruct2026.05 | 78.9 | 65.5 | 99.2 | 90.3 | |
| TIARModel=Llama-3.1-8B-Instruct2026.05 | 77.6 | 63.8 | 99.1 | 91.7 | |
| RFTModel=Llama-3.1-8B-Instruct2026.05 | 76.1 | 61.7 | 99.2 | 86.7 | |
| DPOModel=Llama-3.1-8B-Instruct2026.05 | 75.5 | 61.1 | 98.7 | 89.7 | |
| R-TuningModel=Llama-3.1-8B-Instruct2026.05 | 75.3 | 60.7 | 99.1 | 86.2 |