Zero Shot Cross Lingual Transfer On Xtreme
评估指标
Avg
Question Answering
Sentence Retrieval
Sentence-pair Classification
Structured Prediction
评测结果
各个模型在此基准测试上的表现结果
| Paper Title | Repository | ||||||
|---|---|---|---|---|---|---|---|
| Turing ULR v6 | 85.5 | 77.1 | 94.4 | 91.0 | 83.8 | XLM-E: Cross-lingual Language Model Pre-training via ELECTRA | |
| MShenNonG+TDT | 85.0 | 76.3 | 94.4 | 90.4 | 83.1 | - | - |
| MShenNonG | 85.0 | 76.3 | 94.4 | 90.4 | 83.1 | - | - |
| Turing ULR v5 | 84.5 | 76.3 | 93.7 | 90.3 | 81.7 | XLM-E: Cross-lingual Language Model Pre-training via ELECTRA | |
| InfoXLM-XFT | 82.2 | 75.2 | 92.4 | 89.3 | 75.5 | - | - |
| Ensemble-Distil-XFT (ED-XFT) | 82.0 | 75.2 | 92.4 | 89.2 | 74.6 | - | - |
| CoFe | 84.1 | 75.0 | 94.2 | 90.1 | 81.4 | - | - |
| Turing ULR v5 (XLM-E) | 83.7 | 74.3 | 93.7 | 90.0 | 81.4 | - | - |
| mT5 | 40.9 | 73.6 | NA | 89.8 | NA | mT5: A massively multilingual pre-trained text-to-text transformer | |
| VECO | 82.0 | 73.4 | 93.3 | 89.0 | 76.7 | - | - |
| VECO + HICTL | 82.0 | 73.4 | 93.3 | 89.0 | 76.7 | - | - |
| T-ULRv2 + StableTune | 80.7 | 72.9 | 89.3 | 88.8 | 75.4 | InfoXLM: An Information-Theoretic Framework for Cross-Lingual Language Model Pre-Training | |
| Unicoder+ZCode | 81.6 | 72.5 | 93.7 | 88.4 | 76.2 | - | - |
| Unicoder + ZCode | 81.6 | 72.5 | 93.7 | 88.4 | 76.2 | - | - |
| ERNIE-M | 80.9 | 72.3 | 91.9 | 87.9 | 75.6 | ERNIE-M: Enhanced Multilingual Representation by Aligning Cross-lingual Semantics with Monolingual Corpora | |
| Polyglot | 81.7 | 71.9 | 90.8 | 88.3 | 80.6 | - | - |
| HiCTL | 80.8 | 71.9 | 92.6 | 89.0 | 74.4 | - | - |
| Anonymous3 | 79.9 | 71.7 | 89.0 | 88.2 | 74.6 | - | - |
| RemBERT | 56.1 | 68.6 | NA | 84.1 | 73.3 | - | - |
| FILTER | 77.0 | 68.5 | 84.4 | 87.5 | 71.9 | FILTER: An Enhanced Fusion Method for Cross-lingual Language Understanding |
0 of 25 row(s) selected.