Fairseq is a sequence modeling toolkit created by Facebook AI Research. It allows to train and serve custom models for translation, summarization, language modeling and other text generation tasks. It also provides reference implementations of various sequence-to-sequence models.
In the following examples, we will show how to integrate Fairseq into Amazon SageMaker by creating your own container and using it to train and serve predictions.
fairseq_sagemaker_pretrained_en2fr.ipynb: example of using a pre-trained English-French model to serve predictions and test the inference experiencefairseq_sagemaker_translate_en2fr.ipynb: end-to-end example of training an English-French translation model
The examples are using Fairseq v0.6.0.