The Hugging Face blog post describes how to port the fairseq WMT'19 translation system to use the Transformers library. The original system was a state-of-the-art machine translation model that used a combination of attention mechanisms and recurrent neural networks. By adapting it to use the pre-trained transformer models available in the Transformers library, users can take advantage of these pre-trained models without having to train their own from scratch. --- Why it matters: This matters because it allows researchers and developers to easily leverage the strengths of pre-trained transformer models in their own translation systems, potentially improving performance and reducing training time. Source: https://huggingface.co/blog/porting-fsmt