How to run many current Ray Tune runs without taxing head node

Ray Libraries (Data, Train, Tune, Serve)
1

In our organization, people want to run up to 20 Tune runs concurrently, but each Tune run client connection uses 0.5-1 gb of ram on the head node. Is there any way to scale this up besides paying for a larger instance head node?

2

are you using Ray client?
can you try Ray jobs instead? Ray Jobs Overview — Ray 2.4.0