CPU Mem vs Self CPU Mem in pytorch profiler

Question

I'm trying to profile my deep learning model using pytorch profiler (torch.autograd.profiler). There are two kinds of memory consumption, CPU and Self CPU. When I searched about it, I found this in pytorch document.

Note the difference between self cpu time and cpu time - operators can call other operators, self cpu time exludes time spent in children operator calls, while total cpu time includes it.

I understand this but my result seemed nonsense. There are some weird value that Self CPU mem exceeds CPU mem. I want to know what is differences between Self CPU mem and CPU mem. Thanks.

enter image description here