What's the rationale of using Variable to index nn.Embedding

1

Hi everyone, since the indices are all constant and do not require to compute the gradient, why do we have to pass a Variable type to index the entries in nn.Embedding instead of simply using a torch.LongTensor type? Thanks.

2

For consistency with other modules, which all take Variables as inputs

3

@colesbury Thanks for reply, it does make sense.