When using torch.nn.Linear, suppose the input for the layer is X, where X is a 
matrix, and the weight matrix for this layer is W, which is a 
matrix, then the output for the layer is
%5E%7B%5Cmathrm%7BT%7D%7D%20--%5C%5C%20--(Wx_%7B2%7D%5E%7B%5Cmathrm%7BT%7D%7D)%5E%7B%5Cmathrm%7BT%7D%7D%20--%5C%5C%20%5Cvdots%20%5C%5C%20--(Wx_%7BN%7D%5E%7B%5Cmathrm%7BT%7D%7D)%5E%7B%5Cmathrm%7BT%7D%7D%20--%20%5Cend%7Bbmatrix%7D)
If X is a matrix with shape 
, all the dimensions except 
will be treated as part of the batch dimension.
