As per https://pytorch.org/cppdocs/#autograd, the torch::Tensor classes maintain autograd info. This might be necessary for us, or might be wasteful. This issue is to determine if at::empty and related constructors give us our desired behaviour, and if so, to benchmark any change in speed.
As per https://pytorch.org/cppdocs/#autograd, the
torch::Tensorclasses maintain autograd info. This might be necessary for us, or might be wasteful. This issue is to determine ifat::emptyand related constructors give us our desired behaviour, and if so, to benchmark any change in speed.