Question Answering on TruthfulQA (Truth & Info and Diversity Metrics)
45.17Truth & Info ScoreTypical Sampling + DFD
Evaluation Results
| Method | Links | ||||
|---|---|---|---|---|---|
| Typical Sampling + DFDDecoding Strategy=Typical, Dynamic Focus Decoding=true2025.03 | 45.17 | 74.33 | 84.54 | 14.43 | |
| Top-k Sampling + DFDDecoding Strategy=Top-k, Dynamic Focus Decoding=true2025.03 | 44.55 | 75.71 | 86.69 | 11.29 | |
| Nucleus Sampling + DFDDecoding Strategy=Nucleus, Dynamic Focus Decoding=true2025.03 | 44.19 | 77.57 | 88.03 | 10.67 | |
| Temperature Sampling + DFDDecoding Strategy=Temperature, Dynamic Focus Decoding=true2025.03 | 41.62 | 77.55 | 88.78 | 9.77 | |
| Top-k SamplingDecoding Strategy=Top-k2025.03 | 41.04 | 71.63 | 82.49 | 16.56 | |
| Typical SamplingDecoding Strategy=Typical2025.03 | 40.72 | 73.65 | 83.08 | 15.98 | |
| Nucleus SamplingDecoding Strategy=Nucleus2025.03 | 40.31 | 72.23 | 82.35 | 16.67 | |
| Temperature SamplingDecoding Strategy=Temperature2025.03 | 39.66 | 75.18 | 87.24 | 11.38 |