Error on using Tenser flow.

CUDA is being reinitialized for each sub processes which is training a keras model

I was pursuing a machine learning certification course and entered a problem and could not resolve it. I am using CUDA to train one after one on many tensorflow keras models on my Graphic processing unit. At first the program started to crash stating that the system is out of memory. I asked my course instructor and he informed that using a new sub-process for every model would clear the GPU memory automatically and it did work. But the thing is every each process reinitialize CUDA making it very time consuming. Although I asked my course instructor about the same, but thought of asking here, anybody know about it?

[Joss Knight]

Are you asking a question about python and TensorFlow in a MATLAB forum?