Sharing the checkpoint

[gwang-kim]

Hi, thanks for your great implementation.

Can you share the checkpoint trained on FFHQ dataset?

Thanks.

[rosinality]

I'm newly training the model on ffhq. It will need some time...

[gwang-kim]

Thanks for your reply. I'll be happy if you share the model when training is finished!

[rosinality]

I have finished the training. You can find checkpoints in https://www.dropbox.com/s/rpoou152x8pt1ag/ckpt-2400k.pt?dl=0.

[gwang-kim]

@rosinality Thanks a lot!

[YoelShoshan]

@rosinality Thank you for sharing! What is the proper way to init the model for the checkpoint that you provided? Unet(... ? )

[rosinality]

@YoelShoshan You can use the parameters from https://github.com/rosinality/denoising-diffusion-pytorch/blob/master/config/diffusion.conf

or like this:

from tensorfn import load_config

from diffusion import GaussianDiffusion

conf = load_config(DiffusionConfig, 'config/diffusion.conf')
ckpt = torch.load('ckpt-2400k.pt')
model = conf.model.make()
model.load_state_dict(ckpt['ema'])
model = model.to('cuda')
betas = conf.diffusion.beta_schedule.make()
diffusion = GaussianDiffusion(betas).to('cuda')

[YoelShoshan]

@rosinality Thanks!

[GaParmar]

@rosinality could you share the FFHQ pretrained model again? I think the dropbox link from the previous comment was deleted

[rosinality]

@GaParmar Hello, I have re-uploaded the checkpoint. https://www.dropbox.com/s/9z5yr95og67z78p/ddpm-2400k.pt?dl=0