Forward pass of RNN on list of tensors only returns first pass

Question

I want to make multiple forward passes of a recurrent neural network in one call, where my input is sequences of variable length. Consider the following network and my attem