Hi @<1569496075083976704:profile|SweetShells3> , I would suggest cloning experiments. This way you have full history of everything
Yes, I know, but it is possible to save history for restarted tasks?
When we use dynamic MIG allocation on GPU and meet the limits for project (or another project with higher rank runs task that requires resources already allocated by low-rank project), it stops the experiment and then run another one. The first one runs after resources are freed up. As far as I understand, it can't clone experiment, only restart :(