How do you optimize a PyTorch model for inference on a hardware-constrained edge device at Allen Control Systems?
