Torch.eig() seems really unstable, can anyone explain this result?

Now I found a way somehow to make conda install the right version. The hack I did was:

conda uninstall pytorch
conda install pytorch torchvision cuda80 -c soumith
conda install pytorch torchvision -c pytorch

I have no idea why this is working though. However, now I have the right version:

pytorch:     0.2.0-py27hc03bea1_4cu80 soumith [cuda80] --> 0.3.0-py27_cuda8.0.61_cudnn7.0.3hf383a3f_4 pytorch
torchvision: 0.1.9-py27hdb88a65_1     soumith --> 0.2.0-py27hfb27419_1                       pytorch

I will test if the eig() function in this version work for me and update later. Thanks a lot!