i have no idea about the encoder_state

[peinbill]

for layer_id in range(bi_layer_size):
encoder_state.append(bi_encoder_state[0][layer_id])
encoder_state.append(bi_encoder_state[1][layer_id])
encoder_state = tuple(encoder_state)

and i do not know why we should change the order of encoder_state? And i think the attention model just use the last output about encoder.