README.md

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: