attention seq2seq need

[superhy]

No description provided.

[zsdonghao]

i mark this issue as help wanted, if any one used TF's dynamic RNN encoder and attention seq2seq, feel free to contribute ~

[superhy]

if I want to input the embedding matrix which trained by myself with the other API(like Gensim), I don't want use embedding layer in TL even TF, I want call the clear attention seq2seq or Peekly seq2seq directly.

[superhy]

In the other words, EmbeddingAttentionSeq2seqWrapper function which like the project "easy_seq2seq" is too low in TF-version