Transformer Assemble(PART III)

写在前面 本文首发于公众号:NewBeeNLP 这一期魔改Transformers主要关注对原始模型中位置信息的讨论与优化, Self-Attention with RPR from Google,NAACL2018 Self-Attention with SPR from Tencent,EMNLP 2019 TENER from FDU Encoding Word Order in Complex Embedding,ICLR2020 1、Self-Attention with Relative Position Representations 一篇短文不是很难理解,文章要解决的