GPU consumption in new PyTorch repo with Google Colab

[ghost]

I'm using your SytleGAN2-ADA Pytorch repo, I noticed that only 12-13 GB GPU ram is used with your 11gb gpu cfg. In the stylegan2-ada repo the graphics card always uses 15+ GB. Is it possible to optimize this?

Workflow Question:
Since Google Colab always interrupts my training after 12 hrs max, I'm doing this right? I specify next training kimg and initial strength?

eg. last training kimg 44.0 and augment 0.321 in the new training I give these values to continue learning?

Set Initial Augmentation Strength
init_strength = 0.0321

Set Initial Kimg count
nkimg_count = 44

And:
There is a grid_vid.py in your repro, how do I use that? I could not find a description anywhere.

Thanks a lot for your effort. For people like me who can't really program, it's wonderful thanks to you to still create art with AI.