ray.air.ScalingConfig#

class ray.air.ScalingConfig(trainer_resources: Optional[Union[Dict, Domain, Dict[str, List]]] = None, num_workers: Optional[Union[int, Domain, Dict[str, List]]] = None, use_gpu: Union[bool, Domain, Dict[str, List]] = False, resources_per_worker: Optional[Union[Dict, Domain, Dict[str, List]]] = None, placement_strategy: Union[str, Domain, Dict[str, List]] = 'PACK', _max_cpu_fraction_per_node: Optional[Union[float, Domain, Dict[str, List]]] = None)[source]#

Bases: object

Configuration for scaling training.

Parameters

  • trainer_resources – Resources to allocate for the trainer. If None is provided, will default to 1 CPU.

  • num_workers – The number of workers (Ray actors) to launch. Each worker will reserve 1 CPU by default. The number of CPUs reserved by each worker can be overridden with the resources_per_worker argument.

  • use_gpu – If True, training will be done on GPUs (1 per worker). Defaults to False. The number of GPUs reserved by each worker can be overridden with the resources_per_worker argument.

  • resources_per_worker – If specified, the resources defined in this Dict is reserved for each worker. Define the "CPU" and "GPU" keys (case-sensitive) to override the number of CPU or GPUs used by each worker.

  • placement_strategy – The placement strategy to use for the placement group of the Ray actors. See Placement Group Strategies for the possible options.

  • _max_cpu_fraction_per_node – [Experimental] The max fraction of CPUs per node that Train will use for scheduling training actors. The remaining CPUs can be used for dataset tasks. It is highly recommended that you set this to less than 1.0 (e.g., 0.8) when passing datasets to trainers, to avoid hangs / CPU starvation of dataset tasks. Warning: this feature is experimental and is not recommended for use with autoscaling (scale-up will not trigger properly).

PublicAPI (beta): This API is in beta and may change before becoming stable.

Methods

as_placement_group_factory()

Returns a PlacementGroupFactory to specify resources for Tune.

from_placement_group_factory(pgf)

Create a ScalingConfig from a Tune's PlacementGroupFactory

Attributes

additional_resources_per_worker

Resources per worker, not including CPU or GPU resources.

num_cpus_per_worker

The number of CPUs to set per worker.

num_gpus_per_worker

The number of GPUs to set per worker.

num_workers

placement_strategy

resources_per_worker

total_resources

Map of total resources required for the trainer.

trainer_resources

use_gpu