Abstention in Question Answering on FreshQA Stale
81.2Abstention F1R-Tuning
Evaluation Results
| Method | Links | ||||
|---|---|---|---|---|---|
| R-TuningModel=Qwen3-8B2026.05 | 81.2 | 84.2 | 78.4 | 89 | |
| TruthRLModel=Qwen3-8B2026.05 | 81.1 | 82.5 | 79.8 | 90 | |
| TIARModel=Qwen3-8B2026.05 | 80.4 | 81.4 | 79.6 | 88 | |
| DPOModel=Qwen3-8B2026.05 | 79.6 | 81.4 | 77.8 | 84 | |
| TruthRLModel=Llama-3.1-8B-Instruct2026.05 | 74.7 | 75.7 | 73.6 | 54 | |
| TIARModel=Llama-3.1-8B-Instruct2026.05 | 74.5 | 75.1 | 73.9 | 55 | |
| RFTModel=Llama-3.1-8B-Instruct2026.05 | 72.2 | 66.7 | 78.7 | 50 | |
| DPOModel=Llama-3.1-8B-Instruct2026.05 | 72.1 | 70.1 | 74.3 | 48 | |
| RFTModel=Qwen3-8B2026.05 | 71.9 | 67.2 | 77.3 | 73 | |
| R-TuningModel=Llama-3.1-8B-Instruct2026.05 | 69.9 | 66.1 | 74.1 | 40 |