When I use 2 different devices to run the pix2pix training part , one can smoothly finish the training part but another leads to 'nan' in loss function since the begining as the figure shows. The environments and dataset(facades) are quite the same.
Do you mean the same training runs well on one device, and it produces NaN on the other device? Or did you try to do multi-gpu training? In the case of former, yes, it's likely a GPU issue...
When I use 2 different devices to run the pix2pix training part , one can smoothly finish the training part but another leads to 'nan' in loss function since the begining as the figure shows. The environments and dataset(facades) are quite the same.