Deployment and installation of pre-trained parameters works fine. Training obviously not finishing due to missing GPU and no CPU support for 3d deconvolutions.