Dialogue Response Selection on DailyDialog reformatted multiple-choice (test)
90.35AccuracySirLLM
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| SirLLMBackbone=Yi-34b, Attention Sink Size=4, # Entropy=508, # Recent=0, ηdecay=0.72024.05 | 90.35 | 5 | |
| StreamLLMBackbone=Yi-34b, Attention Sink Size=4, # Entropy=0, # Recent=508, ηdecay=12024.05 | 85.35 | — | |
| SirLLMBackbone=Yi-6b, Attention Sink Size=4, # Entropy=508, # Recent=0, ηdecay=0.72024.05 | 83.85 | 6.95 | |
| RandomLLMBackbone=Yi-34b, Attention Sink Size=4, # Entropy=508, # Recent=0, ηdecay=12024.05 | 82.17 | -3.18 | |
| StreamLLMBackbone=Yi-6b, Attention Sink Size=4, # Entropy=0, # Recent=508, ηdecay=12024.05 | 76.9 | — | |
| SirLLMBackbone=Vicuna-13b, Attention Sink Size=4, # Entropy=508, # Recent=0, ηdecay=0.62024.05 | 71.4 | 0.3 | |
| RandomLLMBackbone=Yi-6b, Attention Sink Size=4, # Entropy=508, # Recent=0, ηdecay=12024.05 | 71.1 | -5.8 | |
| StreamLLMBackbone=Vicuna-13b, Attention Sink Size=4, # Entropy=0, # Recent=508, ηdecay=12024.05 | 71.1 | — | |
| IntervalLLMBackbone=Yi-34b, Attention Sink Size=4, # Entropy=508, # Recent=0, ηdecay=12024.05 | 70.7 | -14.65 | |
| RandomLLMBackbone=Vicuna-13b, Attention Sink Size=4, # Entropy=508, # Recent=0, ηdecay=12024.05 | 69.27 | -1.83 | |
| IntervalLLMBackbone=Yi-6b, Attention Sink Size=4, # Entropy=508, # Recent=0, ηdecay=12024.05 | 65.2 | -11.7 | |
| IntervalLLMBackbone=Vicuna-13b, Attention Sink Size=4, # Entropy=508, # Recent=0, ηdecay=12024.05 | 62.05 | -9.05 | |
| SirLLMBackbone=Vicuna-7b, Attention Sink Size=4, # Entropy=508, # Recent=0, ηdecay=0.52024.05 | 59.15 | 1.6 | |
| StreamLLMBackbone=Vicuna-7b, Attention Sink Size=4, # Entropy=0, # Recent=508, ηdecay=12024.05 | 57.55 | — | |
| RandomLLMBackbone=Vicuna-7b, Attention Sink Size=4, # Entropy=508, # Recent=0, ηdecay=12024.05 | 57.48 | -0.13 | |
| IntervalLLMBackbone=Vicuna-7b, Attention Sink Size=4, # Entropy=508, # Recent=0, ηdecay=12024.05 | 54.45 | -3.1 |