In Neural Machine translation assignment- Iterate for Ty steps

[arvind]

In the modelf part of the assignment we have to iterate Ty steps to get the information for each timestamp, is it possible to implement the modelf without using the loop in tensorflow, because I have seen various RNN or sequential models without implementing loop in their models and still getting result ?

Please provide the detail explanation based on your response, it will be helpful .

[jonaslalin]

You would need to define a new RNN cell which encapsulates the post attention LSTM and attention calculation. Then you can use the RNN layer with your custom cell:

tf.keras.layers.RNN  |  TensorFlow Core v2.6.0