A Transformer implementation based on an implementation from Harvard

The Levenshtein transformer implementation is built on top of the Transformer above, with many elements borrowed from Fairseq

1.先运行prepare.sh

2.再运行get_xsum_data.py

3.运行levenshtein_en_de.py