Problem $5[10$ points$]$

Before the invention of Transformer and LLMs$,$ the RNN$-$based Seq$2$Seq model was a very popular architecture for tasks with sequences as both input and output, such as machine translations, summarization, etc. Read the Seq$2$Seq paper and answer the following questions.

$5\mathrm{.}1[2$ points$]$ How does the encoder pass all the information of the input sequence to the decoder?

$5\mathrm{.}2[2$ points$]$ How does the decoder know when to stop generation?

$5\mathrm{.}3[6$ points$]$ What are the benefits of the encoder$-$decoder Transformer compared to the Seq$2$Seq model? Explain at least two benefits.$$