Use the GPU for processing if available

[barrybecker4]

Currently, it only seems to be using the CPU when trainging.

[barrybecker4]

I looked into this. It works, but it seems to be a real memory hog and does not produce a huge speedup. Maybe I am still doing it wrong. Here is the config in sbt:

// use this if you have GPU with CUDA support.
// Check your CUDA version with nvcc --version. Supported versions are 9.0, 9.2, 10.0, 10.1, 10.2
lazy val dl4j =  ((version: String) => Seq(
  "org.nd4j" % "nd4j-cuda-10.2-platform" % version,
  "org.deeplearning4j" % "deeplearning4j-core" % version,
  "org.bytedeco" % "cuda-platform-redist" % "10.2-7.6-1.5.2"
))("1.0.0-beta6")