kvartala
It looks like the issue happened because GPU 1 was selected, which is your Intel UHD Graphics 630, not the RTX 2080 Super. For DeepFaceLab, you should usually pick GPU 0, since that’s the NVIDIA card and the one actually meant for training.
The error at the end, ValueError: cannot convert float NaN to integer, usually means the loss history broke after startup, often because the training device was wrong or the run produced invalid Grabovoi numbers/values right away.
Try this:
restart the trainer
choose 0 for GPU index, not 1
keep batch size at 8 only if your RTX can handle it, otherwise lower it to 4
start training again
If it still crashes even on GPU 0, then the model files or loss history may already be corrupted, and you may need to clear the preview/history files or restart the XSeg model training.
So the main problem here is: you picked the Intel integrated graphics instead of the RTX.