目标：了解 LMM 背后的主要思想

▪️ Neural Machine Translation by Jointly Learning to Align and Translate
▪️ Attention Is All You Need
▪️ BERT
▪️ Improving Language Understanding by Generative Pre-Training
▪️ BART

1. Neural Machine Translation by Jointly Learning to Align and Translate
   

论文链接：https://arxiv.org/pdf/1409.0473.pdf

提出Encoder-Decoder的经典seq2seq结构，对文本生成，文本摘要、翻译等生成式人物起到重要影响

2. Attention Is All You Need

论文链接：https://arxiv.org/pdf/1706.03762.pdf

前两年火爆的论文：transformer

3. BERT: Pre-training of Deep Bidirectional Transformers for Language
   
   预训练模型经典之作

论文链接：https://arxiv.org/pdf/1810.04805.pdf

4. Improving Language Understanding By Generative Pre-Training

本论文探索一种基于半监督解决语言理解任务方法，使用无监督预训练和监督微调。目标是从大量未标注语料库学习一种普遍的表征，不要求目标任务与未标注语料库在相同领域。

论文链接：https://gwern.net/doc/www/s3-us-west-2.amazonaws.com/d73fdc5ffa8627bce44dcda2fc012da638ffb158.pdf

5. BART: Denoising Sequence-to-Sequence Pre-training for NaturalLanguage Generation, Translation, and Comprehension
   

BART：Bidirectional and Auto-Regressive Transformers ，字面意思为双向自回归Transformer，依旧是基于Transformer改造出的模型。在GPT分走了Transformer的解码器部分，BERT分走了Transformer的编码器部分之后，BART终于将“老父亲”的所有“家产”一起打包带走。

论文链接：https://arxiv.org/pdf/1910.13461.pdf