Why grand_loss will gradually increase and become NaN

[ManZhao123]

Hello, George!
First of all, thank you for your wonderful work!
I used the parameters in your article, lr_img=1000, ipc=10, etc. and initialized by sampling noise.
The network uses convnetD3 and ZCA is not used.I did synthetic training on the cifar10 dataset, but why did I get loss=NaN? After careful observation, I found that the pixel values ​​of each synthesized image became NaN. What is the reason for this? I would like to know more details. Thanks

[Thats2Great]

same issue I have met when I use the hyperparam of CIfAR10 ipc10