Machine Translation on WMT EN-DE 2014 (test) with Efficiency Tracking
28.3BLEU ScoreTransformer w/ LABO
Evaluation results
| Method | Links | ||
|---|---|---|---|
| Transformer w/ LABObeam size=5, Backbone=6-layer encoder-decoder transformer2023.05 | 28.3 | — | |
| Transformer w/ LSbeam size=5, Backbone=6-layer encoder-decoder transformer2023.05 | 27.7 | — | |
| Transformer w/ AFLbeam size=5, Backbone=6-layer encoder-decoder transformer2023.05 | 27.5 | — | |
| Transformer w/ CPbeam size=5, Backbone=6-layer encoder-decoder transformer2023.05 | 27.4 | — | |
| Knee scheduleBudget=30 epochs, Model=Transformer2020.03 | 27.28 | 80 | |
| Transformerbeam size=5, Backbone=6-layer encoder-decoder transformer2023.05 | 27.1 | — | |
| Cosine DecayBudget=30 epochs, Model=Transformer2020.03 | 26.95 | — | |
| One-CycleBudget=30 epochs, Model=Transformer2020.03 | 26.8 | — | |
| Linear DecayBudget=30 epochs, Model=Transformer2020.03 | 26.77 | — | |
| Transformer w/ FLbeam size=5, Backbone=6-layer encoder-decoder transformer2023.05 | 26.6 | — |