(Question) VP vs VE Pre-trained model

[thibautissenhuth]

Hello,

Congrats for this nice work! It is much appreciated to be able to fine-tune a CM in 1 GPU-hour.

I have one question about the pre-trained model you rely on, and think I am missing something obvious. You use the model edm-cifar10-32x32-uncond-vp.pkl which corresponds to the training with Variance-Preserving process. However in your code, you use the Variance-Exploding process. Why don't you use the base model as edm-cifar10-32x32-uncond-ve.pkl ?