將 Fairseq WMT19 翻譯系統移植至 Hugging Face Transformers

In the field of natural language processing (NLP), machine translation has always been a core challenge. Facebook AI Research (FAIR) achieved outstanding results in the 2019 Machine Translation Competition (WMT19) with their Transformer-based translation system, particularly in bidirectional English-German and English-Russian translation. However, these models were originally developed using Facebook's own `fairseq` sequence-to-sequence toolkit. While `fairseq` is powerful and flexible, its installation, configuration, and inference APIs are relatively complex for many developers accustomed to the Hugging Face `transformers` ecosystem, and it is difficult to integrate into other NLP workflows.