on GPU run as
th convertGPUmodelCPU.lua './dir/model.net' to modify your trained network on GPU to run on CPU.
on CPU run as
th convertLinearmodelConvMM.lua './dir/model.net' to modify your trained network with Linear layer. By replacing the linear layers with ConvolutionMM, you can feed the network bigger inputs than the ones network trained with. You need to modify the input tensor based on your training.