Change Convolution padding when testing

Is there a way to change the padding of torch::nn::Conv2d objects stored in a torch::nn::Sequential after training so that the test network will allow to get the same size as the input ?
Here is a code snippet:

auto Fast=TestNetwork->getFastSequential();
size_t Sz=Fast->size();
size_t cc=0;
for (cc=0;cc<Sz;cc++)
if (Fast->named_children()[cc].key()==std::string(“conv”)+std::to_string(cc))
{ Fast->named_children()[cc].options.padding=1;}

It complains about AnyModule object not having a member named options.