Dialogue Generation on ByteDance Customer Service (test)
38.45BLEUMORE
Evaluation Results
| Method | Links | |||||||
|---|---|---|---|---|---|---|---|---|
| MOREBackbone=Qwen2.5-32B2026.06 | 38.45 | 57.22 | 1.35 | 1.4 | 1.39 | 1.46 | 0.92 | |
| EMORLBackbone=Qwen2.5-32B2026.06 | 37.85 | 56.8 | 1.2 | 1.28 | 1.25 | 1.4 | 0.9 | |
| RLHFBackbone=Qwen2.5-32B2026.06 | 37.71 | 56.7 | 1.03 | 1.28 | 1.3 | 1.35 | 0.84 | |
| C-DynaOptBackbone=Qwen2.5-32B2026.06 | 37.62 | 56.59 | 1.14 | 1.27 | 1.32 | 1.36 | 0.85 | |
| CoTBackbone=Qwen2.5-32B2026.06 | 37.58 | 56.61 | 0.89 | 1.31 | 1.28 | 1.38 | 0.81 | |
| SFTBackbone=Qwen2.5-32B2026.06 | 36.97 | 54.55 | 0.78 | 1.18 | 1.21 | 1.36 | 0.52 | |
| KTOBackbone=Qwen2.5-32B2026.06 | 36.64 | 55.24 | 0.94 | 1.25 | 1.22 | 1.41 | 0.73 | |
| DPOBackbone=Qwen2.5-32B2026.06 | 35.31 | 53.19 | 0.97 | 1.29 | 1.17 | 1.42 | 0.77 |