Is there a way to share knowledge between two ray.tune runs?

If I have multiple ray.tune run on a cluster with a shared file system, is there a way to have the searchers fit on each other’s result?

Unfortunately we don’t have a recommended architecture for this type of service. Maybe try using something like SigOpt if that’s still available?

You can also probably integrate with Vizier (GitHub - google/vizier: Python-based research interface for blackbox and hyperparameter optimization, based on the internal Google Vizier Service.) with ray tune, though it would require some work on your side