Machine Translation on IWSLT Turkish-English (tst2014)
24.54BLEUDenseNMT-8L-2
Evaluation results
| Method | Links | |
|---|---|---|
| DenseNMT-8L-2Number of layers=8, Growth rate=2, Embedding size=256, Hidden size=1282018.06 | 24.54 | |
| DenseNMT-8L-2Number of layers=8, Growth rate=22018.06 | 23.68 | |
| BASE2018.06 | 22.59 | |
| deep fusion (Gülçehre et al., 2015)2015.11 | 20.6 | |
| Gigawordsynthtraining data=parallel/Gigawordsynth, instances=8.4m/8.4m2015.11 | 20.4 | |
| parallelsynthtraining data=parallel/parallelsynth, instances=6m/6m2015.11 | 20 | |
| baseline (Gülçehre et al., 2015)2015.11 | 18.7 | |
| RNN2018.06 | 18.64 | |
| baselinetraining data=parallel, instances=7.2m2015.11 | 18.3 | |
| Gigawordmonotraining data=parallel/Gigawordmono, instances=7.6m/7.6m2015.11 | 18.2 |