I would like to prevent all runs in a grid search running in parallel. The only way I found is to use staged runs and the queue option for a sequential execution with wait-for-running=yes.
Is there another way to achieve that goal?
Does the queue option only work with GPUs? I get an error as if guildai cannot find a tensorflow version, though the “non-staged” run works perfectly.
Guild only ever runs operations in series. The only way to run operations in parallel is to use queues (or otherwise start Guild commands independently, e.g. separate terminals, background jobs, etc.)
What are you seeing where Guild runs operations in parallel?