https://mpost.io/vi/perfusion-nvidia-introduces-a-compact-100-kb-neural-network-with-efficient-training-time/