Proactive Dialogue Generation on DuRecDial OOD 1.0 (test)
1.45Perplexity (PPL)Qwen
Evaluation results
| Method | Links | ||||||||
|---|---|---|---|---|---|---|---|---|---|
| QwenModel Size=32B, Evaluation Protocol=Prompt-based2026.05 | 1.45 | 30.36 | 22.2 | 15.4 | 0.9 | 8 | 40.5 | 1 | |
| QwenModel Size=14B, Evaluation Protocol=Prompt-based2026.05 | 1.52 | 29.06 | 21 | 14.3 | 0.9 | 8.9 | 43.29 | 1.75 | |
| LLaMAModel Size=8B, Evaluation Protocol=Prompt-based2026.05 | 1.76 | 27.93 | 22.4 | 15.6 | 1.2 | 10.6 | 36.78 | 22.69 | |
| LLaMAModel Size=3B, Evaluation Protocol=Fine-tuning2026.05 | 4.23 | 46.68 | 43 | 36.2 | 1.9 | 12.4 | 60.96 | 34.41 | |
| QwenModel Size=3B, Evaluation Protocol=Fine-tuning2026.05 | 4.57 | 47.29 | 43.9 | 36.7 | 2.1 | 14.7 | 62.93 | 27.43 | |
| LLaMAModel Size=1B, Evaluation Protocol=Fine-tuning2026.05 | 4.58 | 44.89 | 40.8 | 34 | 1.7 | 10.7 | 57.25 | 38.9 | |
| Our0.3B (soft)Inference Mode=Soft2026.05 | 5.34 | 44.33 | 43.2 | 34.8 | 1.2 | 6.5 | 47.83 | 20.2 | |
| Our0.3B (hard)Inference Mode=Hard2026.05 | 5.36 | 44.2 | 43.1 | 34.7 | 1.2 | 6.6 | 47.83 | 19.7 |