Closed nashid closed 4 years ago
Well, now that Keras has been integrated into TF, it's all a little bit convoluted. NMT-Keras uses an extended version of keras
(NOT tf.keras
).
OpenNMT-tf
makes use of tf.keras
, as part of the TF ecosystem.
https://github.com/keras-team/keras/blob/master/examples/lstm_seq2seq.py is a simple Seq2Seq model, that lacks several features, such as attention mechanisms or beam search.
I am a bit confused when to adopt which framework with keras? Also what is difference between opennmt-tf (https://github.com/OpenNMT/OpenNMT-tf) vs nmt-keras?
Also how does it differ than simply using https://github.com/keras-team/keras/blob/master/examples/lstm_seq2seq.py