I saw many Pytorch examples using flatten_parameters in the forward function of the RNN
self.rnn.flatten_parameters()
I saw this RNNBase and it is written that it
Resets parameter data pointer so that they can use faster code paths
What does that mean?