Reply Generation + Tone Adjustment on Compositional Multi-tasking (test)
14.33W-R ScoreJoint-expert LoRA
Evaluation Results
| Method | Links | |
|---|---|---|
| Joint-expert LoRA2025.07 | 14.33 | |
| Learnable Calibration++Parameter Sharing=Separate2025.07 | 13.54 | |
| Learnable Calibration++Parameter Sharing=Shared2025.07 | 11.06 | |
| Learnable CalibrationParameter Sharing=Separate2025.07 | 10.86 | |
| Learnable CalibrationParameter Sharing=Shared2025.07 | 9.23 | |
| Main-task LoRA2025.07 | 9.12 | |
| LoraHub merge2025.07 | 8.69 | |
| Multi-step LoRA usage2025.07 | 8.09 | |
| DAM/ZipLoRA merge2025.07 | 8 | |
| Linear merge2025.07 | 7.72 | |
| Concat merge2025.07 | 7.65 | |
| LM-Cocktail merge2025.07 | 6.67 | |
| Slerp merge2025.07 | 6.57 | |
| TIES merge2025.07 | 6.47 | |
| In-context learning2025.07 | 4.66 | |
| Auxiliary-task LoRA2025.07 | 4.59 | |
| DARE merge2025.07 | 4.56 | |
| Zero-shot2025.07 | 3.53 |