Hello,
I don't get why you need to set the session at all, this mixes keras and tf.keras and will bring problems. You don't need to do that at all to use the GPU.
About multi-GPU, there is the keras.utils.multi_gpu to train using multiple GPUs. See example at https://keras.io/utils/#multi_gpu_model
There is no way to select "portions" as far as I know.