o `۷i@sddlmZmZddlmZmZddlmZddl m Z m Z m Z e ddGdddeZ e Gd d d e Ze Gd d d e Zd S))ABCabstractmethod)AnyDict)get_train_context) Deprecated DeveloperAPI PublicAPIstable) stabilityc@seZdZdZedeeeffddZedefddZ edefddZ ed d Z edefd d Z e defd dZe defddZe defddZe defddZe defddZe defddZee ddZdS) TrainContextz(Abstract interface for training context.returncCsddlm}t|)zB[Deprecated] User metadata dict passed to the Trainer constructor.r)!_GET_METADATA_DEPRECATION_MESSAGE)ray.train.contextrDeprecationWarning)selfrrN/home/ubuntu/vllm_env/lib/python3.10/site-packages/ray/train/v2/api/context.py get_metadatas zTrainContext.get_metadatacCddlm}t|d)z4[Deprecated] Trial name for the corresponding trial.r*_TUNE_SPECIFIC_CONTEXT_DEPRECATION_MESSAGEget_trial_namerrrformatrrrrrr zTrainContext.get_trial_namecCr)z2[Deprecated] Trial id for the corresponding trial.rr get_trial_idrrrrrrrzTrainContext.get_trial_idcCr)z9[Deprecated] Trial resources for the corresponding trial.rrget_trial_resourcesrrrrrr'rz TrainContext.get_trial_resourcescCr)a[Deprecated] Log directory corresponding to the trial directory for a Tune session. This is deprecated for Ray Train and should no longer be called in Ray Train workers. If this directory is needed, please pass it into the `train_loop_config` directly. rr get_trial_dirrrrrrr0s zTrainContext.get_trial_dircCdS)z,Experiment name for the corresponding trial.Nrrrrrget_experiment_name=sz TrainContext.get_experiment_namecCr )aGet the current world size (i.e. total number of workers) for this run. .. testcode:: import ray.train from ray.train.torch import TorchTrainer NUM_WORKERS = 2 def train_fn_per_worker(config): assert ray.train.get_context().get_world_size() == NUM_WORKERS trainer = TorchTrainer( train_fn_per_worker, scaling_config=ray.train.ScalingConfig(num_workers=NUM_WORKERS), ) trainer.fit() Nrr!rrrget_world_sizeBzTrainContext.get_world_sizecCr )aGet the world rank of this worker. .. testcode:: import ray.train from ray.train.torch import TorchTrainer def train_fn_per_worker(config): if ray.train.get_context().get_world_rank() == 0: print("Worker 0") trainer = TorchTrainer( train_fn_per_worker, scaling_config=ray.train.ScalingConfig(num_workers=2), ) trainer.fit() Nrr!rrrget_world_rankYzTrainContext.get_world_rankcCr )aGet the local rank of this worker (rank of the worker on its node). .. testcode:: import ray.train from ray.train.torch import TorchTrainer def train_fn_per_worker(config): if ray.train.get_context().get_local_rank() == 0: print("Local rank 0 worker") trainer = TorchTrainer( train_fn_per_worker, scaling_config=ray.train.ScalingConfig(num_workers=2), ) trainer.fit() Nrr!rrrget_local_rankor&zTrainContext.get_local_rankcCr )a!Get the local world size of this node (i.e. number of workers on this node). Example: .. testcode:: import ray.train from ray.train.torch import TorchTrainer def train_fn_per_worker(): print(ray.train.get_context().get_local_world_size()) trainer = TorchTrainer( train_fn_per_worker, scaling_config=ray.train.ScalingConfig(num_workers=2), ) trainer.fit() Nrr!rrrget_local_world_sizer$z!TrainContext.get_local_world_sizecCr )aGet the rank of this node. Example: .. testcode:: import ray.train from ray.train.torch import TorchTrainer def train_fn_per_worker(): print(ray.train.get_context().get_node_rank()) trainer = TorchTrainer( train_fn_per_worker, scaling_config=ray.train.ScalingConfig(num_workers=1), ) trainer.fit() Nrr!rrr get_node_rankr$zTrainContext.get_node_rankcCr )aEReturns the :class:`~ray.train._internal.storage.StorageContext` storage context which gives advanced access to the filesystem and paths configured through `RunConfig`. NOTE: This is a DeveloperAPI, and the `StorageContext` interface may change without notice between minor versions. Nrr!rrr get_storages zTrainContext.get_storageN)__name__ __module__ __qualname____doc__rrstrrrrrrrrr"intr#r%r'r(r)rr*rrrrr s6  r c@sleZdZdZdefddZdefddZdefddZdefd d Z defd d Z defd dZ ddZ dS)DistributedTrainContextz4Implementation of TrainContext for distributed mode.r cC tSN)get_internal_train_contextr"r!rrrr" z+DistributedTrainContext.get_experiment_namecCr2r3)r4r#r!rrrr#r5z&DistributedTrainContext.get_world_sizecCr2r3)r4r%r!rrrr%r5z&DistributedTrainContext.get_world_rankcCr2r3)r4r'r!rrrr'r5z&DistributedTrainContext.get_local_rankcCr2r3)r4r(r!rrrr(r5z,DistributedTrainContext.get_local_world_sizecCr2r3)r4r)r!rrrr)r5z%DistributedTrainContext.get_node_rankcCr2r3)r4r*r!rrrr*r5z#DistributedTrainContext.get_storageN) r+r,r-r.r/r"r0r#r%r'r(r)r*rrrrr1s r1c @seZdZdZ     ddededededed ef d d Zd efd dZd efddZd efddZ d efddZ d efddZ d efddZ ddZ dS)LocalTrainContextz.Implementation of TrainContext for local mode.rexperiment_name world_size world_rank local_ranklocal_world_size node_rankcCs(||_||_||_||_||_||_dSr3)r8r9r:r;r<r=)rr8r9r:r;r<r=rrr__init__s   zLocalTrainContext.__init__r cC|jSr3)r8r!rrrr"z%LocalTrainContext.get_experiment_namecCr?r3)r9r!rrrr#r@z LocalTrainContext.get_world_sizecCr?r3)r:r!rrrr%r@z LocalTrainContext.get_world_rankcCr?r3)r;r!rrrr'r@z LocalTrainContext.get_local_rankcCr?r3)r<r!rrrr(r@z&LocalTrainContext.get_local_world_sizecCr?r3)r=r!rrrr)r@zLocalTrainContext.get_node_rankcCstd)Nz+Local storage context not yet implemented. )NotImplementedErrorr!rrrr*szLocalTrainContext.get_storageN)r7rrr7r)r+r,r-r.r/r0r>r"r#r%r'r(r)r*rrrrr6s6  r6N)abcrrtypingrr(ray.train.v2._internal.execution.contextrr4ray.util.annotationsrrr r r1r6rrrrs 6