Question Answering on TriviaQA (AUC, F95, AUP)
89.48AUCSIVR (Ours)
Evaluation Results
| Method | Links | |||
|---|---|---|---|---|
| SIVR (Ours)Backbone=Llama-3.1-8B-Instruct2026.04 | 89.48 | 39.81 | 89.27 | |
| TADBackbone=Llama-3.1-8B-Instruct2026.04 | 87.02 | 50.53 | 89.48 | |
| SAPLMABackbone=Llama-3.1-8B-Instruct2026.04 | 83.7 | 62.38 | 85.63 | |
| SARBackbone=Llama-3.1-8B-Instruct2026.04 | 83.66 | 45.24 | 84.35 | |
| Lookback LensBackbone=Llama-3.1-8B-Instruct2026.04 | 82.94 | 59.09 | 87.25 | |
| SATMD + MSPBackbone=Llama-3.1-8B-Instruct2026.04 | 82.45 | 61.39 | 78.79 | |
| SEBackbone=Llama-3.1-8B-Instruct2026.04 | 81.65 | 54.76 | 83.37 | |
| EntropyBackbone=Llama-3.1-8B-Instruct2026.04 | 78.58 | 67.06 | 76.25 | |
| PerplexityBackbone=Llama-3.1-8B-Instruct2026.04 | 78.18 | 65.68 | 75.39 | |
| Max ProbBackbone=Llama-3.1-8B-Instruct2026.04 | 77.3 | 66.08 | 73.94 | |
| Temp-ScaledBackbone=Llama-3.1-8B-Instruct2026.04 | 76.65 | 66.86 | 72.71 | |
| EnergyBackbone=Llama-3.1-8B-Instruct2026.04 | 67.5 | 87.18 | 67.47 | |
| CoE-CBackbone=Llama-3.1-8B-Instruct2026.04 | 65.14 | 93.89 | 68.58 | |
| P(True)Backbone=Llama-3.1-8B-Instruct2026.04 | 65.04 | 93.29 | 65.35 | |
| CoE-RBackbone=Llama-3.1-8B-Instruct2026.04 | 37.75 | 100 | 44.87 |