Rui

@rajarsheem During training, we can apply the output_layer after all time steps finished [here](https://github.com/tensorflow/nmt/blob/master/nmt/model.py#L407) because we have the word ids in target language. So the outputs [here](https://github.com/tensorflow/nmt/blob/master/nmt/model.py#L391) contains rnn outputs...

@rajarsheem we don't feed hidden state argmax because we have the target ids. See how the TrainingHelper is created https://github.com/tensorflow/nmt/blob/master/nmt/model.py#L373.

@rajarsheem Yes, the code you referenced in the last comment is only for teacher forcing during training, so that's why the output_layer is not being used.

@rajarsheem Yes, I think you can implement a custom GreedyEmebddingHelper (which accepts an output layer), so you don't need to pass-in the output layer to the BasicDecoder. For example, You...

Can you provide a command that can regenerate the issue with this codebase? Also python version and tensorflow version may also help.

@bota7070 It means you have input sequence with zero length, which is an empty line.

@aleSuglia To make attention works correctly, we don't want allocate any weights to the paddings, and the sum of all weights must be 1. Therefore, a sequence of all paddings...

@playma In addition to the --num_train_steps, you can get some epoch info [here](https://github.com/tensorflow/nmt/blob/master/nmt/train.py#L358).

@zxu7 The code doesn't have a character level embedding option. However, you may tokenize data at character level, and prepare a character level vocab file to train a character model...

You need to make sure your code split the sentence into characters instead of words. By default, the code will split sentences by space, which will cause the problem for...