Dialogue Generation on DailyDialog
9.12Distinct-1BART joint† (D)
Evaluation results
| Method | Links | ||||||||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| BART joint† (D)Memory=Self2023.05 | 9.12 | — | — | 37.05 | — | 36.92 | 32.09 | — | — | — | — | — | — | — | |
| BART jointMemory=Retrieval2023.05 | 6.13 | — | — | 35.65 | — | 36.72 | 31.55 | — | — | — | — | — | — | — | |
| BART jointMemory=Self2023.05 | 5.84 | — | — | 32.16 | — | 39.8 | 32.15 | — | — | — | — | — | — | — | |
| BART-base2022.05 | 4.95 | 0.65 | 13.4 | 19.47 | 0.0551 | — | — | — | — | — | — | — | — | — | |
| BART dualMemory=Retrieval2023.05 | 4.74 | — | — | 26.01 | — | 29.5 | 21.89 | — | — | — | — | — | — | — | |
| BART dualMemory=Self2023.05 | 4.66 | — | — | 26.16 | — | 33.43 | 22.85 | — | — | — | — | — | — | — | |
| DialoFlowMemory=None2023.05 | 4.56 | — | — | 27.12 | — | 36.17 | 27.67 | — | — | — | — | — | — | — | |
| Transformer+KI2022.05 | 4.39 | 6.13 | 17.48 | 21.88 | 0.0053 | — | — | — | — | — | — | — | — | — | |
| BARTMemory=None2023.05 | 3.92 | — | — | 19.44 | — | 20.72 | 11.36 | — | — | — | — | — | — | — | |
| PLATO-2Memory=None2023.05 | 3.54 | — | — | 25.11 | — | 34.8 | 25.12 | — | — | — | — | — | — | — | |
| NCMMemory=None2023.05 | 3 | — | — | 12.8 | — | 33.6 | 26.8 | — | — | — | — | — | — | — | |
| iVAEMemory=None2023.05 | 2.9 | — | — | 25 | — | 30.9 | 24.9 | — | — | — | — | — | — | — | |
| Transformer2022.05 | 1.48 | 2.65 | 12.97 | 5.1 | 0.0714 | — | — | — | — | — | — | — | — | — | |
| Llama2-7B-SPTLearning Rate=0.001, Backbone=Llama2-7B, Variant=Selective Prompt Tuning2024.06 | 0.0797 | — | 0.1587 | 0.3689 | — | — | — | 18.38 | 0.31 | 0.0449 | 0.1437 | 56.95 | 57.21 | 57.73 | |
| Llama2-7B-PTLearning Rate=0.001, Backbone=Llama2-7B, Variant=Prompt Tuning2024.06 | 0.0771 | — | 0.1566 | 0.354 | — | — | — | 18.03 | 0.18 | 0.0441 | 0.1413 | 55.9 | 55.97 | 56.99 | |
| Llama2-7B-SPTLearning Rate=0.01, Backbone=Llama2-7B, Variant=Selective Prompt Tuning2024.06 | 0.0753 | — | 0.1504 | 0.3468 | — | — | — | 17.4 | 0.08 | 0.0457 | 0.1368 | 53.72 | 53.73 | 55.16 | |
| Llama2-7B-PTLearning Rate=0.01, Backbone=Llama2-7B, Variant=Prompt Tuning2024.06 | 0.0741 | — | 0.1446 | 0.3483 | — | — | — | 17.06 | 0.21 | 0.0409 | 0.13 | 54.58 | 54.71 | 55.6 | |
| DialogVED - SamplingDecoding Strategy=Sampling, Latent Size=642022.04 | 0.058 | — | — | 0.372 | — | 0.431 | 0.37 | — | — | — | — | — | — | — | |
| PLATODecoding Strategy=Beam Search, Beam Size=52022.04 | 0.054 | — | — | 0.291 | — | 0.397 | 0.311 | — | — | — | — | — | — | — | |
| PLATO w/o latentDecoding Strategy=Beam Search, Beam Size=52022.04 | 0.046 | — | — | 0.246 | — | 0.405 | 0.322 | — | — | — | — | — | — | — | |
| DialogVED - GreedyDecoding Strategy=Greedy, Latent Size=642022.04 | 0.045 | — | — | 0.265 | — | 0.459 | 0.41 | — | — | — | — | — | — | — | |
| DialogVEDDecoding Strategy=Beam Search, Beam Size=5, Latent Size=642022.04 | 0.042 | — | — | 0.232 | — | 0.481 | 0.421 | — | — | — | — | — | — | — | |
| DialogVED w/o latentDecoding Strategy=Beam Search, Beam Size=5, Latent Variable=None2022.04 | 0.041 | — | — | 0.222 | — | 0.461 | 0.407 | — | — | — | — | — | — | — | |
| ProphetNetDecoding Strategy=Beam Search, Beam Size=52022.04 | 0.039 | — | — | 0.211 | — | 0.443 | 0.392 | — | — | — | — | — | — | — | |
| Seq2SeqDecoding Strategy=Beam Search, Beam Size=52022.04 | 0.03 | — | — | 0.128 | — | 0.336 | 0.238 | — | — | — | — | — | — | — | |
| iVAE_MIDecoding Strategy=Beam Search, Beam Size=52022.04 | 0.029 | — | — | 0.25 | — | 0.309 | 0.249 | — | — | — | — | — | — | — | |
| CriSPO2024.10 | — | — | — | — | — | — | — | — | — | — | 17.4 | — | — | — | |
| Manualshots=0-shot2024.10 | — | — | — | — | — | — | — | — | — | — | 12.6 | — | — | — | |
| Manualshots=3-shot2024.10 | — | — | — | — | — | — | — | — | — | — | 17.1 | — | — | — | |
| OPRO2024.10 | — | — | — | — | — | — | — | — | — | — | 13.3 | — | — | — |