Reply Generation + Translation on Compositional Multi-tasking (test)
14.99W-RJoint-expert LoRA
Evaluation Results
| Method | Links | |
|---|---|---|
| Joint-expert LoRA2025.07 | 14.99 | |
| Learnable Calibration++Parameter Sharing=Separate2025.07 | 12.14 | |
| Multi-step LoRA usage2025.07 | 10.04 | |
| Learnable Calibration++Parameter Sharing=Shared2025.07 | 9.99 | |
| Learnable CalibrationParameter Sharing=Separate2025.07 | 8.85 | |
| Learnable CalibrationParameter Sharing=Shared2025.07 | 6.63 | |
| DAM/ZipLoRA merge2025.07 | 4.62 | |
| Auxiliary-task LoRA2025.07 | 4.03 | |
| Concat merge2025.07 | 3.97 | |
| Linear merge2025.07 | 3.94 | |
| In-context learning2025.07 | 3.93 | |
| Slerp merge2025.07 | 3.73 | |
| TIES merge2025.07 | 3.53 | |
| LoraHub merge2025.07 | 3.26 | |
| LM-Cocktail merge2025.07 | 3.24 | |
| DARE merge2025.07 | 2.95 | |
| Zero-shot2025.07 | 2.03 | |
| Main-task LoRA2025.07 | 1.25 |