def __init__(
self,
backend_executor: Union[BackendExecutor, ActorWrapper],
backend_config: BackendConfig,
train_func: Union[Callable[[], T], Callable[[Dict[str, Any]], T]],
dataset_spec: RayDatasetSpec,
checkpoint_manager: CheckpointManager,
checkpoint: Optional[Union[Dict, str, Path, Checkpoint]],
checkpoint_strategy: Optional[CheckpointConfig],
run_dir: Optional[Path] = None,
):
self._backend_executor = backend_executor
self._backend = backend_config.backend_cls()
self._train_func = train_func
self._dataset_spec = dataset_spec
self._run_dir = run_dir
self._checkpoint_manager = checkpoint_manager
self._checkpoint_strategy = checkpoint_strategy
self._start_training(
train_func=train_func,
run_dir=run_dir,
dataset_spec=self._dataset_spec,
checkpoint=checkpoint,
checkpoint_strategy=checkpoint_strategy,
)
self._final_results = None
self._finished_training = False
def __init__(
self,
backend_config: BackendConfig,
# TODO(xwjiang): Legacy Ray Train trainer clean up!
trial_info: Optional[TrialInfo] = None,
num_workers: int = 1,
num_cpus_per_worker: float = 1,
num_gpus_per_worker: float = 0,
additional_resources_per_worker: Optional[Dict[str, float]] = None,
max_retries: int = 3,
):
self._backend_config = backend_config
self._backend = backend_config.backend_cls()
self._num_workers = num_workers
self._num_cpus_per_worker = num_cpus_per_worker
self._num_gpus_per_worker = num_gpus_per_worker
self._additional_resources_per_worker = additional_resources_per_worker
self._max_failures = max_retries
if self._max_failures < 0:
self._max_failures = float("inf")
self._num_failures = 0
self._initialization_hook = None
self._placement_group = None
self._trial_info = trial_info
self.worker_group = InactiveWorkerGroup()
self.dataset_shards = None
def __init__(
self,
backend_executor_actor: ActorHandle,
backend_config: BackendConfig,
train_func: Union[Callable[[], T], Callable[[Dict[str, Any]], T]],
run_dir: Path,
dataset: Optional[Union[RayDataset, Dict[str, RayDataset]]],
checkpoint_manager: CheckpointManager,
checkpoint: Optional[Union[Dict, str, Path]],
checkpoint_strategy: Optional[CheckpointStrategy],
):
self._backend_executor_actor = backend_executor_actor
self._backend = backend_config.backend_cls()
self._train_func = train_func
self._dataset = dataset
self._run_dir = run_dir
self._checkpoint_manager = checkpoint_manager
self._checkpoint_strategy = checkpoint_strategy
self._start_training(
train_func=train_func,
run_dir=run_dir,
dataset=dataset,
checkpoint=checkpoint,
checkpoint_strategy=checkpoint_strategy,
)
self._final_results = None
self._finished_training = False