no code implementations • 5 Mar 2022 • Alessandro Generale, Till Blume, Michael Cochez
The amount of gradient information that needs to be stored during training for real-world graphs is often too large for the amount of memory available on most GPUs.