Multi-turn Dialogue Evaluation on MT-Eval (Expansion, Follow-up, Recollection, Refinement)
7.34Expansion ScoreSFT
Evaluation Results
| Method | Links | |||||
|---|---|---|---|---|---|---|
| SFTBackbone=Qwen2.5-14B2024.08 | 7.34 | 8.1 | 6.24 | 5.68 | 6.84 | |
| ISMBackbone=Qwen2.5-14B2024.08 | 7.03 | 8.37 | 6.63 | 5.96 | 7 | |
| ISMBackbone=LLaMA3.1-8B2024.08 | 6.8 | 7.15 | 5.55 | 5.17 | 6.17 | |
| SFTBackbone=Qwen2.5-7B2024.08 | 6.63 | 7.71 | 5.63 | 5.29 | 6.32 | |
| ISMBackbone=Qwen2.5-7B2024.08 | 6.57 | 7.72 | 5.95 | 5.36 | 6.4 | |
| DirectBackbone=Qwen2.5-14B2024.08 | 6.43 | 7.55 | 5.32 | 4.31 | 5.9 | |
| SFTBackbone=LLaMA3.1-8B2024.08 | 6.21 | 7.1 | 5.13 | 5.03 | 5.87 | |
| DirectBackbone=Qwen2.5-7B2024.08 | 4.7 | 6.03 | 3.84 | 2.92 | 4.37 | |
| DirectBackbone=LLaMA3.1-8B2024.08 | 3.66 | 4.23 | 1.39 | 2.09 | 2.84 |