Pytorch implement of Attention Is All You Need
you can rewrite the file config.yaml to configure your training process or translation process.
Using Multi30k dataset from torchtext, about how to use the dataset, this will help you.
all details about implementation are in model.ipynb.
- It takes too long to translate long sentences. Optimize the beam search algorithm in Model/translator, the method
translate_batch
. - Surrport cuda