when do torch.zeros(1).cuda(), out of memory.
When I install pytorch, and I do torch.zeros(1).cuda(), it works. But after I install matcaffe, it goes wrong.
What exactly is the error message you’re getting?
Also, could you look at the output of nvidia-smi? Maybe after you installed matcafe all of your GPU memory is being taken
Right! The GPU has been taken… And now it can be used. Thanks~