Question Answering on XQuAD
94.1F1 (de)ORACLE
Evaluation Results
| Method | Links | ||||||||||||||||||||||||||||||||||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| ORACLE2026.04 | 94.1 | — | — | — | — | — | — | — | 92 | — | 89.8 | — | — | — | — | — | — | — | — | — | 91 | — | 91.7 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| ORACLEStrategy=Upper bound selection, Backbone=DeepSeek-v3.12026.04 | 94.1 | — | — | — | — | — | — | — | 92 | — | 89.8 | — | — | — | — | — | — | — | — | — | 91 | — | 91.7 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| TRANSLATE2026.04 | 90.6 | — | — | — | — | — | — | — | 88.2 | — | 82.5 | — | — | — | — | — | — | — | — | — | 87.2 | — | 87.1 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| TRANSLATEStrategy=Translate-then-inference, Backbone=DeepSeek-v3.12026.04 | 90.6 | — | — | — | — | — | — | — | 88.2 | — | 82.5 | — | — | — | — | — | — | — | — | — | 87.2 | — | 87.1 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| NATIVE2026.04 | 89.2 | — | — | — | — | — | — | — | 87.2 | — | 83.6 | — | — | — | — | — | — | — | — | — | 86.6 | — | 86.7 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| XGBOOST2026.04 | 89.2 | — | — | — | — | — | — | — | 87.9 | — | 84.7 | — | — | — | — | — | — | — | — | — | 88.6 | — | 87.6 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| MLP2026.04 | 89.2 | — | — | — | — | — | — | — | 88 | — | 83.3 | — | — | — | — | — | — | — | — | — | 86.5 | — | 86.7 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| NATIVEStrategy=Native inference, Backbone=DeepSeek-v3.12026.04 | 89.2 | — | — | — | — | — | — | — | 87.2 | — | 83.6 | — | — | — | — | — | — | — | — | — | 86.6 | — | 86.7 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| CLASSIFIER (Ours)Strategy=Learned Routing (XGBoost), Backbone=DeepSeek-v3.12026.04 | 89.2 | — | — | — | — | — | — | — | 87.9 | — | 84.7 | — | — | — | — | — | — | — | — | — | 88.6 | — | 87.6 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| LTSFT2023.06 | 80.5 | — | — | 73 | — | — | 78.6 | — | 80.6 | — | 74.3 | — | 77.8 | — | 69.7 | — | 72.2 | — | 76.5 | — | 68.9 | — | 75.9 | — | — | — | 82.4 | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| XLMRImplementation=Standard2021.10 | 80.4 | 86.5 | 75.7 | — | — | 63.4 | 79.8 | 61.7 | 82 | 63.9 | — | — | — | — | — | — | — | — | — | — | — | — | 82.2 | 66.2 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| BISTIL-TFLRF=22023.06 | 77.4 | — | — | 69.4 | — | — | 73.8 | — | 77.6 | — | 69.7 | — | 75 | — | 66.7 | — | 68.8 | — | 72.8 | — | 64.5 | — | 72.3 | — | — | — | 79.1 | -3.6 | — | — | — | — | — | — | — | — | — | — | — | — | |
| MINILMv22023.06 | 75.5 | — | — | 66.4 | — | — | 72.4 | — | 76.6 | — | 69.6 | — | 74 | — | 63.8 | — | 67.6 | — | 73.3 | — | 64.6 | — | 71.1 | — | — | — | 78.3 | -4.8 | — | — | — | — | — | — | — | — | — | — | — | — | |
| XLM-R BaseBackbone=XLM-R Base, Transfer=Zero-shot from English2020.04 | 74.4 | 83.6 | 72.1 | 66.8 | 49.1 | 60.1 | 73 | 55.7 | 76.4 | 58.3 | 68.2 | 51.7 | 74.3 | 58.1 | 66.5 | 56.7 | 68.3 | 52.8 | 73.7 | 53.8 | 51.3 | 42 | 70.6 | 55.5 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| XLM-R Base MLM-TRGBackbone=XLM-R Base, Target Language Adaptation=Masked Language Modeling (MLM), Transfer=Zero-shot from English2020.04 | 73.7 | 84.7 | 72.6 | 67 | 49.2 | 58.8 | 73.2 | 55.7 | 76.6 | 58.3 | 69.8 | 53.6 | 74.3 | 57.9 | 67 | 55.8 | 68.6 | 53 | 75.5 | 54.9 | 52.2 | 43.1 | 71.1 | 55.7 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| WordOTImplementation=Ours2021.10 | 73.6 | 84.2 | 72.4 | — | — | 57.8 | 65.6 | 48.5 | 75.5 | 57.1 | — | — | — | — | — | — | — | — | — | — | — | — | 74.7 | 59 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| MAD-X BaseBackbone=XLM-R Base, Invertible Adapters=true, Transfer=Zero-shot from English2020.04 | 72.9 | 83.5 | 72.6 | 65.5 | 48.2 | 56 | 72.9 | 54.6 | 75.9 | 56.9 | 68.2 | 51.3 | 73.1 | 56.7 | 67.8 | 55.9 | 67 | 49.8 | 73.7 | 53.3 | 52.7 | 42.8 | 70.3 | 54.4 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| MAD-XbaseImplementation=Standard2021.10 | 72.9 | 83.5 | 72.6 | — | — | 56 | 72.9 | 54.6 | 75.9 | 56.9 | — | — | — | — | — | — | — | — | — | — | — | — | 76.3 | 60 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| MAD-XbaseImplementation=Internal Benchmarking2021.10 | 72.1 | 82 | 71.1 | — | — | 54.5 | 71.7 | 53.7 | 74.3 | 55.7 | — | — | — | — | — | — | — | — | — | — | — | — | 75 | 58.8 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| MAD-X Base-INVBackbone=XLM-R Base, Invertible Adapters=false, Transfer=Zero-shot from English2020.04 | 72 | 83.3 | 72.1 | 64 | 47.1 | 55.8 | 71 | 52.9 | 74.6 | 55.5 | 67.3 | 51 | 72.1 | 55.1 | 64.1 | 51.8 | 66.2 | 49.6 | 73 | 53.6 | 50.9 | 40.6 | 67 | 53.2 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| mBERTImplementation=Internal Benchmarking2021.10 | 72 | 83.7 | 72.4 | — | — | 55.9 | 62.3 | 45.3 | 75.6 | 57.5 | — | — | — | — | — | — | — | — | — | — | — | — | 73.4 | 57.8 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| BISTIL-TFLRF=32023.06 | 70.7 | — | — | 62.4 | — | — | 63.3 | — | 74.7 | — | 61.4 | — | 68.7 | — | 54.3 | — | 62.9 | — | 63 | — | 60.4 | — | 65 | — | — | — | 73.4 | -10.9 | — | — | — | — | — | — | — | — | — | — | — | — | |
| mBERTImplementation=Standard2021.10 | 70.6 | 83.5 | 72.2 | — | — | 54 | 62.6 | 44.9 | 75.5 | 56.9 | — | — | — | — | — | — | — | — | — | — | — | — | 73.1 | 57 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| MINIJOINTMethod category=Proposed, Train cost (EFLOPs)=21.1, Train cost (days)=8.12022.12 | 70.1 | — | — | 53.8 | — | — | 69.9 | — | 73.7 | — | 51.8 | — | 68.6 | — | 56.5 | — | 53.2 | — | 64.9 | — | 52.5 | — | 61.5 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| BL_BASEMethod category=Standard, Train cost (EFLOPs)=54.1, Train cost (days)=20.92022.12 | 69.2 | — | — | 52.4 | — | — | 70 | — | 74.8 | — | 53.8 | — | 68.7 | — | 57.1 | — | 57.2 | — | 65.9 | — | 53.7 | — | 62.3 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| MINIPOSTMethod category=Proposed, Train cost (EFLOPs)=29.3, Train cost (days)=11.32022.12 | 68.8 | — | — | 52.4 | — | — | 69.9 | — | 75.1 | — | 55.3 | — | 68.2 | — | 56.9 | — | 58.6 | — | 65.7 | — | 53.5 | — | 62.4 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| MAD-XmBERTImplementation=Internal Benchmarking2021.10 | 68.6 | 81.7 | 69.7 | — | — | 52.1 | 58.6 | 41.3 | 71.8 | 51.9 | — | — | — | — | — | — | — | — | — | — | — | — | 70.2 | 53.8 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| L2Implementation=Internal Benchmarking2021.10 | 67.5 | 81.4 | 69.4 | — | — | 51.3 | 56.6 | 40.3 | 66.2 | 45.4 | — | — | — | — | — | — | — | — | — | — | — | — | 68 | 51.6 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| XLMImplementation=Standard2021.10 | 66 | 74.2 | 62.1 | — | — | 49.7 | 57.5 | 39.1 | 68.2 | 49.8 | — | — | — | — | — | — | — | — | — | — | — | — | 66.5 | 50.2 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| BL_SMALLMethod category=Standard, Train cost (EFLOPs)=21.1, Train cost (days)=8.12022.12 | 61.2 | — | — | 48.2 | — | — | 63.3 | — | 65.8 | — | 46.2 | — | 61 | — | 50.4 | — | 52.1 | — | 60.3 | — | 46.2 | — | 55.5 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| BL_BASE2022.12 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | 2.3 | 1.1 | 1.7 | 0.8 | 3.3 | 1.9 | 2.1 | 2.2 | 1.2 | 1.5 | — | — | |
| ByT5Scale=Small, Evaluation Protocol=Translate-train2021.05 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | 74 | 59.9 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| ByT5Scale=Base, Evaluation Protocol=Translate-train2021.05 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | 78.5 | 64.6 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| ByT5Scale=Large, Evaluation Protocol=Translate-train2021.05 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | 81.4 | 67.4 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| ByT5Scale=XL, Evaluation Protocol=Translate-train2021.05 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | 83.7 | 69.5 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| ByT5Scale=XXL, Evaluation Protocol=Translate-train2021.05 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | 84.1 | 70.2 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| Llama_Bi_NTPmode=instruction fine-tuned2026.04 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | 88.7 | 19.16 | — | — | — | — | — | — | — | — | — | — | — | — | 86.78 | — | |
| Llama_CLmode=instruction fine-tuned2026.04 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | 88.5 | 19.24 | — | — | — | — | — | — | — | — | — | — | — | — | 87.26 | — | |
| Llama_CLImode=instruction fine-tuned2026.04 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | 85.71 | 16.35 | — | — | — | — | — | — | — | — | — | — | — | — | 79.33 | — | |
| Llama_NTPmode=instruction fine-tuned2026.04 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | 87.5 | 18.82 | — | — | — | — | — | — | — | — | — | — | — | — | 86.28 | — | |
| Llama-3-8Bmode=instruction fine-tuned2026.04 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | 86.8 | 17.48 | — | — | — | — | — | — | — | — | — | — | — | — | 80.54 | — | |
| MINIJOINT2022.12 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | 0.8 | 0.4 | 0.6 | 0.5 | 3.1 | 0.6 | 0.6 | — | 0.5 | 0.6 | — | — | |
| MINIPOST2022.12 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | 1.4 | 0.8 | 1 | 0.4 | 1.3 | 1.3 | 1.4 | 1.2 | 0.8 | 1 | — | — | |
| mT5Scale=Small, Evaluation Protocol=Translate-train2021.05 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | 71.3 | 55.7 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| mT5Scale=Base, Evaluation Protocol=Translate-train2021.05 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | 77.6 | 62.2 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| mT5Scale=Large, Evaluation Protocol=Translate-train2021.05 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | 81.3 | 66.5 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| mT5Scale=XL, Evaluation Protocol=Translate-train2021.05 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | 82.7 | 68.1 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| mT5Scale=XXL, Evaluation Protocol=Translate-train2021.05 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | 85.2 | 71.3 | — | — | — | — | — | — | — | — | — | — | — | — | — | — |