Hi all,
So from the other posts that I have read about pruning, I’ve learned that the pytorch pruning module doesn’t actually speed up the model or reduce memory because it just inserts zeros. I read in one post that a potential solution would be to create an entirely new network with the unpruned weights of the original network. Would this be possible to do that with a CNN that has been filter pruned (zeroing out entire kernels)? If so, how would you go about it?