Compositional Multi-tasking Summarization, Tone Adjustment, Translation (test)
26.96R-L ScoreLearnable Calibration++
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| Learnable Calibration++Efficient?=true2025.07 | 26.96 | 34.34 | |
| Learnable CalibrationEfficient?=true2025.07 | 24.58 | 28.89 | |
| Shared Learnable Calibration++Efficient?=true2025.07 | 22.68 | 15.54 | |
| Learnable Calibration++variant=++2025.07 | 21.42 | 39.16 | |
| Shared Learnable CalibrationEfficient?=true2025.07 | 20.98 | 16.86 | |
| Multi-step LoRA usageEfficient?=false2025.07 | 20.23 | 34.32 | |
| Learnable Calibrationvariant=base2025.07 | 19.19 | 28.66 | |
| Joint-expert LoRAEfficient?=false2025.07 | 19.08 | 16.14 | |
| In-context learningEfficient?=true2025.07 | 16.96 | 24.12 | |
| Multi-step LoRA usagemode=Multi-step2025.07 | 16.5 | 43.82 | |
| Main-task LoRAEfficient?=true2025.07 | 16.43 | 4.18 | |
| LoraHub mergeEfficient?=true2025.07 | 16.13 | 3.03 | |
| DAM/ZipLoRA mergeEfficient?=true2025.07 | 15.68 | 2.78 | |
| Concat mergeEfficient?=true2025.07 | 15.27 | 2.76 | |
| Linear mergeEfficient?=true2025.07 | 15.26 | 2.74 | |
| TIES mergeEfficient?=true2025.07 | 14.95 | 6.06 | |
| LM-Cocktail mergeEfficient?=true2025.07 | 14.88 | 5.7 | |
| Slerp mergeEfficient?=true2025.07 | 14.87 | 4.87 | |
| DARE mergeEfficient?=true2025.07 | 14.51 | 6.34 | |
| Joint-expert LoRAmode=Joint-expert2025.07 | 14.35 | 12.75 | |
| Auxiliary-task LoRAEfficient?=true2025.07 | 14.33 | 5.81 | |
| Zero-shotEfficient?=true2025.07 | 13.93 | 6.52 | |
| Translation LoRAtask_specific=Translation2025.07 | 10.95 | 0 | |
| Linear mergemode=Linear merge2025.07 | 9.89 | 0.01 | |
| Summarization LoRAtask_specific=Summarization2025.07 | 9.41 | 0 | |
| Tone adj. LoRAtask_specific=Tone adjustment2025.07 | 8.39 | 0.01 | |
| Zero-shotmode=Zero-shot2025.07 | 7.35 | 0.03 |