I have a severe space limitation on my deployment device that mean I cannot fit the entire "Keras" package.
Is there a way to get a prediction from a trained keras model without the entire keras package?
I have noted that the Keras docker is ~1GB. I would need it to be < 650MB