逐行源码阅读中文笔记。
blog解读与复现：https://blog.csdn.net/qq_39388410/article/details/102081253
建议阅读顺序：Transformer.py-->nn.Transformer.py
原paper：
@inproceedings{opennmt,
author = {Guillaume Klein and
Yoon Kim and
Yuntian Deng and
Jean Senellart and
Alexander M. Rush},
title = {OpenNMT: Open-Source Toolkit for Neural Machine Translation},
booktitle = {Proc. ACL},
year = {2017},
url = {https://doi.org/10.18653/v1/P17-4012},
doi = {10.18653/v1/P17-4012}
}
原code（harvardnlp的pytorch版）： http://nlp.seas.harvard.edu/2018/04/03/attention.html
nn.Transformer已经已经出框架了...
所以nn.Transformer.py这个是pytorch官方实例的代码，搭起来很方便。