lstm_cell.state_size is an Integer or TensorShape, what does "for dim in lstm_cell.state_size" mean? And I don't know how "state_update_op = tf.assign(init_states[i], new_state)" works, since you assign init_states using "tf.zeros([self._max_batch_size, dim]" in L548.
https://github.com/allenai/bilm-tf/blob/7cffee2b0986be51f5e2a747244836e1047657f4/bilm/model.py#L547
lstm_cell.state_size is an Integer or TensorShape, what does "for dim in lstm_cell.state_size" mean? And I don't know how "state_update_op = tf.assign(init_states[i], new_state)" works, since you assign init_states using "tf.zeros([self._max_batch_size, dim]" in L548.