Dialogue Generation on ByteDance Proactive Sale (test)
47.49BLEUMORE
Evaluation Results
| Method | Links | |||||||
|---|---|---|---|---|---|---|---|---|
| MOREBackbone=Qwen3-8B2026.06 | 47.49 | 64.13 | 1.65 | 1.55 | 1.28 | 1.57 | 0.95 | |
| C-DynaOptBackbone=Qwen3-8B2026.06 | 46.77 | 62.83 | 1.22 | 1.44 | 1.21 | 1.52 | 0.84 | |
| SFTBackbone=Qwen3-8B2026.06 | 46.55 | 61.42 | 0.77 | 1.2 | 1.13 | 1.43 | 0.46 | |
| EMORLBackbone=Qwen3-8B2026.06 | 46.51 | 62.14 | 1.29 | 1.42 | 1.23 | 1.51 | 0.88 | |
| CoTBackbone=Qwen3-8B2026.06 | 46.24 | 62.7 | 0.79 | 1.41 | 1.1 | 1.45 | 0.77 | |
| RLHFBackbone=Qwen3-8B2026.06 | 44.14 | 61.77 | 0.96 | 1.37 | 1.14 | 1.48 | 0.8 | |
| KTOBackbone=Qwen3-8B2026.06 | 43.99 | 60.17 | 1.04 | 1.38 | 1.12 | 1.47 | 0.71 | |
| DPOBackbone=Qwen3-8B2026.06 | 43.35 | 58.34 | 1.16 | 1.37 | 1.15 | 1.49 | 0.75 |