Summarization + Translation on Compositional Multi-tasking (test)
35.57ROUGE-1 ScoreLearnable Calibration++
Evaluation results
| Method | Links | |||
|---|---|---|---|---|
| Learnable Calibration++Parameter Sharing=Separate2025.07 | 35.57 | 12.11 | 28.64 | |
| Learnable CalibrationParameter Sharing=Separate2025.07 | 31.83 | 9.53 | 25.4 | |
| Multi-step LoRA usage2025.07 | 27.2 | 7.92 | 21.25 | |
| Joint-expert LoRA2025.07 | 26.77 | 7.05 | 21.36 | |
| Learnable Calibration++Parameter Sharing=Shared2025.07 | 23.91 | 6.25 | 19.23 | |
| Learnable CalibrationParameter Sharing=Shared2025.07 | 21.15 | 5.3 | 17.04 | |
| DAM/ZipLoRA merge2025.07 | 20.38 | 4.38 | 16.19 | |
| In-context learning2025.07 | 18.05 | 3.76 | 14.46 | |
| Concat merge2025.07 | 17.04 | 3.08 | 14.39 | |
| Linear merge2025.07 | 16.9 | 3.02 | 14.27 | |
| Auxiliary-task LoRA2025.07 | 16.77 | 2.98 | 13.99 | |
| LoraHub merge2025.07 | 16.51 | 2.8 | 13.78 | |
| Slerp merge2025.07 | 16.4 | 2.97 | 13.96 | |
| Main-task LoRA2025.07 | 16.33 | 2.73 | 13.39 | |
| LM-Cocktail merge2025.07 | 15.97 | 2.88 | 13.62 | |
| TIES merge2025.07 | 14.35 | 2.65 | 12.25 | |
| DARE merge2025.07 | 10.78 | 1.96 | 9.27 | |
| Zero-shot2025.07 | 8.68 | 1.45 | 7.49 |