This will work too, ill try.
My experiment has validation every 40k iterations so thats the case ;p.

Quartiles

Trying this gives me expected TypeError: TypeError: Expected iter of type (<class 'int'>,), got float

Mean max min

Well this is common thing to do for validation metrics. Training metrics are calculated before validation so it gives better visualisation.
I found this information there: https://twitter.com/aureliengeron/status/1110839480024338432?s=20

So ideally I should setup one machine with storage for datasets for clearml-server and one GPU machine to be clearml-agent with storage enough to fetch data?