Memory Requirements
Larger and more complicated models will require more memory
Model Loading Time
Models that require more memory will take a longer time to load
Inference Time
If your models take a long time for inference, then you might as well stick to the cloud
Uses two global hyperparameters
Helps to carry out tasks in a timely fashion on a computationally limited platform
It has a low latency model
It is built primarily from depthwise separable convolutions initially introduced and then which are subsequently used in the Inception Model