[Feature] Separate concurrency setting for remote ML

[seekified]

I have searched the existing feature requests to make sure this is not a duplicate request.

• Yes

The feature

My NAS can only handle ~2 concurrent ML tasks, but my PC can do 16 (and that's without acceleration since the ROCm support isn't ready yet!). My PC won't always be available though, so for this reason I would like to see a separate concurrency setting that is applied only for remote ML jobs.

Platform

• Server
• Web
• Mobile

[seekified]

I did see this comment discussing something similar:
#11862 (reply in thread)

Perhaps it's difficult to implement in the queue manager, but I don't appear to be alone in wishing for this flexibility.