MazeLocalLauncher

class hydra_plugins.maze_local_launcher.MazeLocalLauncher(*args: Any, **kwargs: Any)

Custom Hydra launcher distributing the jobs in separate processes on the local machine.

The implementation is based on https://github.com/facebookresearch/hydra/blob/master/examples/plugins/example_launcher_plugin/hydra_plugins/example_launcher_plugin/example_launcher.py.

Parameters

n_jobs – Maximum number of parallel jobs. If -1, all CPUs are used.

launch(job_overrides: Sequence[Sequence[str]], initial_job_idx: int) → Sequence[hydra.core.utils.JobReturn]

Implementation of Launcher.launch

Parameters
  • job_overrides – a List of List<String>, where each inner list is the arguments for one job run.

  • initial_job_idx – Initial job idx in batch.

Returns

an array of return values from run_job with indexes corresponding to the input list indexes.

setup(config: omegaconf.DictConfig, config_loader: hydra.core.config_loader.ConfigLoader, task_function: hydra.TaskFunction)None

Implementation of Launcher.setup, called before the launch

Parameters
  • config – The master config

  • config_loader – The config loader, used to derive the job configurations from the sweep run.

  • task_function – The job entry point as function object. This is not used at all, as it is much simpler to call the same command from bash than to serialize the function object, transfer it to the pod and calling the deserialized method there.