InvalidArgumentError: LossTensor is inf or nan : Tensor had NaN values

[luofangcheng]

There has been an error in the training session. My workstation is: 1, windows 10 64X 2, NVIDIA GeForce RTX 3060 3, cuda 10.1 4, cudnn 7.6

I am trying to fix the bug as this method: https://github.com/DeepLabCut/DeepLabCut/issues/1017, but it didn't work.

can you help me to fix it ?

Thanks