Zeqiang Lai

torch2.0 should work. This is a problem caused by the stylegan cuda extension. Check this out https://github.com/Zeqiang-Lai/DragGAN/issues/12

Yes，there are many issues similar as yours on windows，it seems some of them solve the problem without conda by installing visual studio. I will try to simplify the installation later...

Maybe you could update the code to the latest version. I made some changes to provide a fallback option when the custom op could not be build, which no longer...

Do you mean you want to uninstall ninja ? If it is, I don't know if it would cause problems. So I would recommend keep it. What I mean is...

this is warning not errors， looks like you are running correctly，simply enter the url into the browser，you will see the interface

Thanks for the feedback. By default, we only need one of them to build the husky ckpt. Did you have those two llama folders created but without the actual ckpt?...

@tinaYA524 Due to the license issue, we could not provide the original checkpoint of LLAMA. Therefore, you need to fill a form to request a url from facebook. Please refer...

> > @tinaYA524 Due to the license issue, we could not provide the original checkpoint of LLAMA. Therefore, you need to fill a form to request a url from facebook....

It is normal one step take 1s. The results could fail in some cases. These are the limitations of this algorithm, which are also discussed in the original paper.

The weights are obtained from here. You could use their training pipelines. https://github.com/NVlabs/stylegan2-ada-pytorch https://github.com/rosinality/stylegan2-pytorch