Sequence to sequence modelling on discrete values

I have a sequence of length 10, and I wish to predict the other half given the first half. I followed the tutorial of Seq2Seq modeling for English-French translation, and removed all Softmax layers and added relu activations instead. Then I changed the negative log likelihood loss to MSE loss. This is where i do something wrong, the decoder outputs a set of 5 values. and the target is 1 value (using teacher forcing), so obviously I get the error:
RuntimeError: input and target shapes do not match: input [1 x 5], target [1]

Could someone please guide me, on how to fix this?