Dialogue Response Generation on ConvAI2
22.89F1MoCoRP LLM
Evaluation results
| Method | Links | ||||||
|---|---|---|---|---|---|---|---|
| MoCoRP LLMBackbone=Mistral v0.1, Training Protocol=SFT+DPO, Inference Mode=posterior2025.12 | 22.89 | 19.27 | 8.44 | 23.65 | 21.77 | 18.62 | |
| Mistral v0.1Training Protocol=SFT+DPO, Inference Mode=prior2025.12 | 22.84 | 19.24 | 8.42 | 23.65 | 21.74 | 18.57 | |
| MoCoRP LLMBackbone=Mistral v0.1, Training Protocol=SFT, Inference Mode=posterior2025.12 | 22.51 | 18.74 | 8.13 | 22.82 | 21.07 | 17.47 | |
| Mistral v0.1Training Protocol=SFT, Inference Mode=prior2025.12 | 22.47 | 18.74 | 8.14 | 22.75 | 21 | 17.36 | |
| MoCoRP LLMBackbone=LLaMA 3 it, Training Protocol=SFT+DPO, Inference Mode=posterior2025.12 | 21.25 | 18.05 | 7.27 | 22.55 | 20.56 | 17.78 | |
| LLaMA 3 itTraining Protocol=SFT+DPO, Inference Mode=prior2025.12 | 21.18 | 17.9 | 7.3 | 22.22 | 20.38 | 17.35 | |
| MoCoRP LLMBackbone=Qwen2, Training Protocol=SFT+DPO, Inference Mode=posterior2025.12 | 21.09 | 18.15 | 7.3 | 22.68 | 20.76 | 20.7 | |
| MoCoRP LLMBackbone=LLaMA 3 it, Training Protocol=SFT, Inference Mode=posterior2025.12 | 21.03 | 17.54 | 7.03 | 21.74 | 19.92 | 16.57 | |
| Qwen2Training Protocol=SFT+DPO, Inference Mode=prior2025.12 | 21.01 | 18.16 | 7.21 | 22.73 | 20.86 | 21.2 | |
| LLaMA 3 itTraining Protocol=SFT, Inference Mode=prior2025.12 | 20.91 | 17.4 | 6.92 | 21.52 | 19.81 | 16.46 | |
| MoCoRP LLMBackbone=Qwen2, Training Protocol=SFT, Inference Mode=posterior2025.12 | 20.23 | 16.71 | 6.67 | 20.42 | 18.91 | 18.95 | |
| Qwen2Training Protocol=SFT, Inference Mode=prior2025.12 | 20.22 | 16.68 | 6.57 | 20.39 | 18.87 | 18.83 |