metadata
license: mit
language:
- de
- en
tags:
- translation
- pytorch
datasets:
- multi30k
metrics:
- bleu
model-index:
- name: multi30k
results:
- task:
type: translation
dataset:
type: multi30k
name: multi30k-de-en
metrics:
- type: bleu
value: 33.468
name: Test BLEU
args: n_gram=4
Seq2seq + Attention
Pytorch implementation of Neural Machine Translation by Jointly Learning to Align and Translate. Trained on the Multi30k-de-en dataset with sentencepiece as the tokenizer.
Here's the attention heatmap of a random sample from the test set: