I am not the person you're replying to, but in my experience / opinion if you want to do distributed high-performance numerical computing, executing on heteroge

I am not the person you're replying to, but in my experience / opinion if you want to do distributed high-performance numerical computing, executing on heterogeneous hardware, C++ is really far and away the best language to get this done. If you're CPU-only, maybe Fortran is a better option but when GPUs enter the mix I don't see how you're going to have an easier time getting the requisite performance with anything else. I mean, like, you can interface with TensorFlow via Python but the big-kid parts of the implementation are in C++ / CUDA.