in Advanced guide (http://opennmt.net/Advanced/#training-from-pre-trained-parameters):
Note that the model topology and dropout value can not be changed during a retraining.
What is the specific structure of the model topology? layer size + rnn size + ?
Sorry, I don’t understand the question.
sorry for my carelessness, I have changed my question. Thank you!
It is everything that impacts dimensions or the graph architecture. It is mostly the options listed in the **Model options** section in th train.lua -h.