How to copy a node from a gmodule to another gmodule?

[royal-feng]

I want to do something like copy a node from a gmodule called net_A to another gmodule called net_B,so I do such operations
netB.modules[i] = netA.modules[counter]
and when the forward in netB, it failed.And the error imformation like that:
torch/install/bin/luajit: ...torch/install/share/lua/5.1/cudnn/SpatialConvolution.lua:32: Only Cuda supported duh!
stack traceback:
[C]: in function 'assert'
...torch/install/share/lua/5.1/cudnn/SpatialConvolution.lua:32: in function 'resetWeightDescriptors'
...torch/install/share/lua/5.1/cudnn/SpatialConvolution.lua:96: in function 'checkInputChanged'
...torch/install/share/lua/5.1/cudnn/SpatialConvolution.lua:120: in function 'createIODescriptors'
...torch/install/share/lua/5.1/cudnn/SpatialConvolution.lua:188: in function 'func'
/home/zsf/torch/install/share/lua/5.1/nngraph/gmodule.lua:345: in function 'neteval'
/home/zsf/torch/install/share/lua/5.1/nngraph/gmodule.lua:380: in function 'forward'
test3.lua:335: in main chunk
[C]: in function 'dofile'
.../zsf/torch/install/lib/luarocks/rocks/trepl/scm-1/bin/th:150: in main chunk
[C]: at 0x00406670

So I try to copy the node.weight from netA to netB,like that
netB.modules[i].weight=netA.modules[counter].weight
netB.modules[i].bias=netA.modules[counter].bias
But the result after netB:forward(input) is not so good. Is my copy operation something wrong?
Thank you for answer!

[arthitag]

Hi,
I don't have an answer to your question but I am facing a similar issue.
I am trying to copy weights from a trained nngraph model to a new nngraph model (with almost same structure) that I created from file. When I do a forward pass results are all wrong and I can't figure out why. For copying weights I used
params0, gp0 = model0:parameters()
params1, gp1= model1:parameters()
and then serially copied all weight/bias tensors in wt to the params tensors on model 1 using the tensor1:copy(tensor2).
I later verified if the weights of two models are the same, and if the inputs are getting passed correctly.
I don't know why but some of the intermediate layer activations in the forward pass are growing really large and then going to all 0's.