The Task

The Translation task fine-tunes the model to translate text from one language to another.


Currently supports the WMT16 dataset or custom input text files.

Input Text (English): "The ground is black, the sky is blue and the car is red."
Model Output (German): "Der Boden ist schwarz, der Himmel ist blau und das Auto ist rot."


To use this task, select a Seq2Seq Encoder/Decoder based model, such as multi-lingual T5 or BART. Conventional models like GPT/BERT will not work as they are encoder only. In addition, you also need a tokenizer that has been created on multi-lingual text. This is true for mt5 and mbart.

python task=nlp/translation dataset=nlp/translation/wmt16 backbone.pretrained_model_name_or_path=google/mt5-base

By default we fine-tune to translate from English to Romanian. This can be changed by specifying the source/target languages (see more in conf/dataset/nlp/translation/wmt16.yaml):

python task=nlp/translation dataset=nlp/translation/wmt16 backbone.pretrained_model_name_or_path=google/mt5-base dataset.cfg.dataset_config_name=de-en dataset.cfg.source_language=de dataset.cfg.target_language=en

Translation Using Your Own Files

To use custom text files, the files should contain new line delimited json objects within the text files.

    "source": "example source text",
    "target": "example target text"

We override the dataset files, allowing us to still use the data transforms defined with this dataset.

python task=nlp/translation dataset.cfg.train_file=train.json dataset.cfg.validation_file=valid.json

Translation Inference Pipeline (experimental)

By default we use the translation pipeline, which requires a source text string.

For Hydra to correctly parse your input argument, if your input contains any special characters you must either wrap the entire call in single quotes like ‘+x=”my, sentence”’ or escape special characters. See escaped characters in unquoted values.

python task=nlp/translation +checkpoint_path=/path/to/model.ckpt +x="The ground is black, the sky is blue and the car is red."

You can also run prediction using a HuggingFace pre-trained model:

python task=nlp/translation backbone.pretrained_model_name_or_path="Helsinki-NLP/opus-mt-en-de" '+x="The ground is black, the sky is blue and the car is red."'
Read the Docs v: stable
On Read the Docs
Project Home

Free document hosting provided by Read the Docs.