The Description of the Project

Owing to only some of the entries in embedding layers updating during each batch training, it is implerative that we can just store the updating entries in the memory and outsourced the left data to other places.

The Intuitive Solution (no idea)