CUDA is reinitialized for every subprocess which trains a keras model

Question

I am using CUDA/CUDNN to sequentially train many tensorflow keras models on my GPU (for an evolutionary algorithm to optimize hyperparameters). Initially, the program would