Bugfix a pytorch deep learning model for running with CUDA

I am working on a new deep learning project with a special model and loss function. But I am not familiar with pytorch. The Details of implementation in pytorch are to difficult and time consuming for me. The project is hosted on gitHub: [login to view URL] I need the code in at least two versions or three. First version runs only and completely with a CPU. The second version runs with a single GPU via CUDA (from nvidia). (and [login to view URL]) An optional third version runs distributed on multiple machines with multiple gpu's. (with [login to view URL]) This does not work currently. I can't get a version that runs completely on the CPU because I always get the error: "RuntimeError: Expected all tensors to be on the same device, but found at least two devices, cpu and cuda:0! (when checking argument for argument target in method wrapper_nll_loss_forward)" The single GPU version also does not work because the source code is not 100% CUDA compatible: "RuntimeError: Input type ([login to view URL]) and weight type ([login to view URL]) should be the same or input should be a MKLDNN tensor and weight is a dense tensor" The main goal is to fix all issues of the CUDA Version of my code so I finaly can run it. There is a google colab notebook for testing, understanding and debugging every piece of code in runtime order: [login to view URL]