Yeah that makes sense, I mean it will probably be a bit more than that per month when it's up but half when it's down (just fyi, when AWS instances are down you still pay for the EBS storage).
If you are trying o save a buck here, double check on that otherwise you will end at the same cost level but after spending resource on migrating.
If you want a good hack you can always download the data and then just store it locally (i.e. half the migration job) and just reduce the number of users when the account is not being used, probably less work and cheaper 🙂
Hi StickyShrimp60
The best way is through APIs, you can query all the Tasks and then one by one use task.export_task with task.get_reported_scalars , task.get_reported_plots, task.get_reported_console_output, to get the details, after that you can recreate thee Task with import_task, and manually report the scalars/plots/console
btw: is self hosted server cheaper than the 15$ a month hosted service, or is it just a convenience
Hi AgitatedDove14 - thanks a lot for the input, I will have a look at the API
btw: is self hosted server cheaper than the 15$ a month hosted service, or is it just a convenience
ClearML is quite cheap for sure, but our monthly runs to something like 200$ (we have 4 users and some 14GB of metrics store that we want to maintain). The main point for us is that we do not use ClearML on a daily basis. It is maybe 6 months since we last used it and we are probably going to start using it again for new experiments in the coming 6 months, but by self-hosting we can just stop the aws instance when not needed and pay a fraction of the 200$/month fors torrage.