Configuring GPUs/cluster size at runtime

[jvmncs]

Just a short snippet demoing how you'd do this. Other runtime configuration is best passed to the clustered function itself (run_benchmark in this example) via remote or spawn.