This enables a working metal and Cuda jit compilation of the graph to enable efficient use of GPU resources.