why is there transpose operation in gru part?

[Arch]

`(x_new, state_new) = gru_cell(x[0], state[0])
state_new = tf.transpose(state_new, (1,0)) # here

            for i in range(1, batch_size):
                (x_, state_) = gru_cell(x[i], state[i])  ##input of GRUCell must be 2 rank, not 3 rank
                state_ = tf.transpose(state_, (1,0))   # here
                state_new = tf.concat([state_new, state_], 0)`

I think "here" shoule be no transpose operation?