Hey there,
I have a quick question. I am using an Embedding space in my model and my final step is to get a score for each object in the embedding space. So my approach is to just multiply with the transposed weights:
self.embed = nn.Embedding(..)
...
scores = out.mm(self.embed.weight.t())
Is that fine in terms of autograd? Are the embedding weights updated in terms of the matrix multiplication?
Thanks,
Magnus