1.It seems like that if a pytorch process is stopped just by crtl+c,not killed,it will stay in memory all the time.
2.Using nvidia-smi, I find there are no pytorch program using GPU, the pytorch process still exist.
Do you use multi-processing e.g. in a DataLoader?
If so, some child processed might still be alive.
You would have to find them and kill them with their pid.
As far as I know it’s a knows Python issue.
Thank you very much!
I will check my code again.