Document Continuation on WikiText-103 (test)
13.96MAUVETop-k Sampling + DFD
Evaluation Results
| Method | Links | ||||
|---|---|---|---|---|---|
| Top-k Sampling + DFDDecoding Strategy=Top-k, DFD=Enabled2025.03 | 13.96 | 55.48 | 49.73 | 3.23 | |
| Nucleus Sampling + DFDDecoding Strategy=Nucleus, DFD=Enabled2025.03 | 13.22 | 48.54 | 57.62 | 2.2 | |
| Top-k SamplingDecoding Strategy=Top-k, DFD=Disabled2025.03 | 12.74 | 53.54 | 49.04 | 3.56 | |
| Typical Sampling + DFDDecoding Strategy=Typical, DFD=Enabled2025.03 | 11.06 | 52.57 | 57.09 | 2.2 | |
| Nucleus SamplingDecoding Strategy=Nucleus, DFD=Disabled2025.03 | 10.03 | 47.29 | 56.05 | 2.37 | |
| Typical SamplingDecoding Strategy=Typical, DFD=Disabled2025.03 | 9.4 | 50.01 | 56.01 | 2.41 | |
| Temperature Sampling + DFDDecoding Strategy=Temperature, DFD=Enabled2025.03 | 7.8 | 45.09 | 64.8 | 1.4 | |
| Temperature SamplingDecoding Strategy=Temperature, DFD=Disabled2025.03 | 7.05 | 42.83 | 62.96 | 1.53 |