Summarization on XL-Sum Arabic (test)
22.32ROUGE-LGRASP LoRA Merge + pruning
Evaluation results
| Method | Links | ||||||
|---|---|---|---|---|---|---|---|
| GRASP LoRA Merge + pruningAdaptation Method=GRASP LoRA Merge, Training Languages=Arabic+English, Pruning Strategy=Learned prune ratio2026.01 | 22.32 | 67.49 | 75.84 | 7.93 | — | — | |
| Adaptive LoRA Merge + pruningAdaptation Method=Adaptive LoRA Merge, Training Languages=Arabic+English, Pruning Strategy=Best grid search over [10% – 80%]2026.01 | 20.19 | 70 | 74.96 | 6.18 | — | — | |
| LoRAAdaptation Method=LoRA, Training Languages=Arabic, Pruning Strategy=Without Pruning2026.01 | 18.17 | — | 74.33 | 4.46 | — | — | |
| Adaptive LoRA MergeAdaptation Method=Adaptive LoRA Merge, Training Languages=Arabic+English, Pruning Strategy=Without Pruning2026.01 | 18.15 | — | 74.32 | 4.46 | — | — | |
| LoRAHubAdaptation Method=LoRAHub, Training Languages=Arabic+English, Pruning Strategy=Without Pruning2026.01 | 16.85 | — | 73.76 | 5.1 | — | — | |
| Llama3 8B-I + LATBAlignment Method=Language-Aware Token Boosting2026.06 | 12.45 | — | — | — | 20.44 | 8.02 | |
| TIES mergeAdaptation Method=TIES merge, Training Languages=Arabic+English, Pruning Strategy=Without Pruning2026.01 | 11.99 | — | 69.59 | 2.19 | — | — | |
| Llama3 8B-I + Adaptive LATBAlignment Method=Adaptive Language-Aware Token Boosting2026.06 | 11.84 | — | — | — | 19.79 | 7.62 | |
| Suzume 8B-Multilingual2026.06 | 11.69 | — | — | — | 19.6 | 7.09 | |
| Llama3 8B-IPrompting Strategy=Strict Prompt2026.06 | 11.66 | — | — | — | 19.22 | 7.46 | |
| Zero shotAdaptation Method=Zero-shot, Training Languages=Arabic, Pruning Strategy=Without Pruning2026.01 | 10.6 | — | 69.08 | 2.05 | — | — | |
| Llama3 8B-IPrompting Strategy=Default2026.06 | 1.42 | — | — | — | 1.54 | 0.19 |