Multi-attribute controllable dialogue generation on ConvAI2-CG (test)
69.03P-SIMDCG
Evaluation Results
| Method | Links | ||||||
|---|---|---|---|---|---|---|---|
| DCGAOP=true, TOP=true, DL=true2023.06 | 69.03 | 81.2 | 30.42 | 19.55 | 2.68 | 12.42 | |
| DCG w/o DLAOP=true, TOP=true, DL=false2023.06 | 68.25 | 79 | 28.53 | 18.34 | 2.39 | 11.63 | |
| DCG w/o TOPAOP=true, TOP=false, DL=true2023.06 | 67.35 | 78.5 | 28.44 | 12.18 | 1.05 | 7.61 | |
| Fine-tuningProtocol=Fine-tuning2023.06 | 65.48 | 69.5 | 19.21 | 16.53 | 2.4 | 10.96 | |
| CTRLModel type=Conditional Transformer2023.06 | 65.2 | 77.65 | 26.12 | 18.39 | 3.12 | 12.23 | |
| Prompt-tuningProtocol=Prompt-tuning2023.06 | 64.84 | 74.3 | 24.56 | 17.59 | 2.6 | 11.22 | |
| DCG w/o AOPAOP=false, TOP=true, DL=true, equivalence=Prompt-tuning2023.06 | 64.84 | 74.3 | 24.56 | 17.59 | 2.6 | 11.22 | |
| DialoGPT-OriProtocol=Original2023.06 | 60.16 | 72.47 | 23.12 | 12.33 | 1.54 | 8.95 | |
| PPLMMethod category=Decoding strategy2023.06 | 59.9 | 75.98 | 25.03 | 13.2 | 1.65 | 9.06 |