Keras model training is possible but shapes are incompatible for loss function

Question

I have a Keras model which outputs tensors of shape (None, 50, 1). I am fitting it against target tensors of shape (None, 1). The loss is mse