Machine Translation on En-De 2017 (test)
38.9BLEUmBART + MT* w/ adapters
Evaluation results
| Method | Links | ||
|---|---|---|---|
| mBART + MT* w/ adaptersObjectives=NMT + MLM, trainable_params=12.6M2022.12 | 38.9 | 0.664 | |
| VGAMTObjectives=MMT + VMLM, trainable_params=13.2M2022.12 | 38.3 | 0.653 | |
| mBART + MT*Objectives=NMT, trainable_params=-2022.12 | 32.3 | 0.506 | |
| TLM + MT*Objectives=NMT, trainable_params=42M2022.12 | 31.5 | 0.341 | |
| VTLM + MMT*Objectives=MMT, trainable_params=44M2022.12 | 30.7 | 0.322 | |
| Vanilla MT*Objectives=NMT, trainable_params=4.1M2022.12 | 30.3 | 0.259 | |
| Gated Fusion*Objectives=MMT, trainable_params=2.8M2022.12 | 29.5 | 0.236 | |
| Graph-MMT*Objectives=MMT, trainable_params=4.1M2022.12 | 29 | 0.226 |