Variable sized mini-batches

[pskrunner14]

Model can't handle variable size mini-batches

[pskrunner14]

Add torch.nn.utils.rnn.pad_sequence as preprocessing part and use packing and padding in encoder