A simple GRU decoder cell for seq2seq model in Keras.
from keras.layers import RNN
from AttnDecoder import AttnDecoderCell
cell = AttnDecoderCell(output_dims)
decoder_layer = RNN(cell, return_sequences=True)
decoder_outputs = decoder_layer(input_tensor, initial_state=init_state, constants=[attn_sequence])where init_state is the first hidden state of decoder, and attn_sequence is the sequence that needs attention mechanism.