Open 17zhangw opened 7 months ago
Are we planning experiments that require multi-machine (i.e., we cannot shard experiments manually ourselves)?
I was looking at cloudlab yesterday, but I'm not sure if it is worth the effort to also take on the responsibility for network setup.
Note that if we are running trials on ray on different machines, we will need to setup some proper cloud/remote storage that is visible to all machines: https://github.com/ray-project/ray/issues/37177.
This behavior is as-of ray 2.5.