To build a conda pytorch package with GPU support

* Build the required Docker containers:

    Dockerfiles can be found at: https://github.com/jjhelmus/docker-images

    CUDA 9.0: pkg_build_cos6_cuda90
    CUDA 9.2: pkg_build_cos6_cuda92
    CUDA 10.0: pkg_build_cos6_cuda100

* Start the docker container using:

    ```
    sudo nvidia-docker run -v `pwd`:/io -it pkg_build_cos6_cuda90
    ```

* Update conda and conda-build, and navigate to the recipe root folder.

    Modify conda_build_config.yaml in this directory to specify the
    CUDA, CuDNN, and python versions.

    To start a build use:

    ```
    conda build .
    ```
