Improve existing checkpoint in Pytorch

My goal is to improve the existing checkpoint mechanism. Please share your suggestions or opinions

What do you think is missing or not working properly at the moment? :slight_smile:

I am referring the paper “On Efficient Constructions of Checkpoints”

I am not sure this is implemented or not, my implementation is based on this paper.
If code for this research project is implemented, could you please point me or share the link.

I’m not aware on an implementation of this algorithm.