search

stefan-it / nmt-en-vi

Neural Machine Translation system for English to Vietnamese (IWSLT'15 English-Vietnamese data)
59 stars 14 forks source link

New model #4

Closed stefan-it closed 6 years ago

stefan-it commented 6 years ago

This PR updates the readme file and introduces a new model, trained with tensor2tensor in version 1.9.0 on a RTX 2080 TI.

Checkpoint averaging was performed.

Minor corrections for new tensor2tensor version are also included (fixes #3)

Model changelog:

Model BLEU-score
Transformer (Base) - old model 28.12 (cased)
Transformer (Base) - old model 28.97 (uncased)
Transformer (Base) - new model 28.43 (cased)
Transformer (Base) - new model 29.31 (uncased)