Closed cvampal closed 3 months ago
The number of trainable parameters are very similar to those of the original EDM model, since we majorly modify the first conv layer. You could call print('Number of Parameters: {sum(p.numel() for p in model.parameters()):,}")
this easily to get the model size.
Can you please share the number of trainable parameters for each model (CelebA, ImageNet, LSUN-Bedroom, and LSUN-Church)?