Multi-turn Dialogue Evaluation on MT-Bench (test)
6.26MT-Bench ScoreDiaBlo
Evaluation results
| Method | Links | |
|---|---|---|
| DiaBlobackbone=Llama2-7B, r/N=N = 320, #Params=0.3%2025.06 | 6.26 | |
| LoRA-GAbackbone=Llama2-7B, r/N=r = 128, #Params=4.7%2025.06 | 6.13 | |
| Dorabackbone=Llama2-7B, r/N=r = 8, #Params=0.3%2025.06 | 5.97 | |
| LoRA-GAbackbone=Llama2-7B, r/N=r = 8, #Params=0.3%2025.06 | 5.95 | |
| LoRAbackbone=Llama2-7B, r/N=r = 8, #Params=0.3%2025.06 | 5.61 | |
| PiSSAbackbone=Llama2-7B, r/N=r = 8, #Params=0.3%2025.06 | 5.3 |