sagemaker.train.defaults

sagemaker.train.defaults#

This module contains logic for setting defaults in ModelTrainer.

Classes

`JumpStartTrainDefaults`()	Class for the JumpStart Defaults.
`TrainDefaults`()	Class to set the base default values for ModelTrainer.

class sagemaker.train.defaults.JumpStartTrainDefaults[source]#

Bases: object

Class for the JumpStart Defaults.

get_base_job_name(base_job_name: str | None = None) → str[source]#: Get the default base job name for JumpStart.

static get_compute(jumpstart_config: JumpStartConfig, compute: Compute | None = None, sagemaker_session: Session | None = None) → Compute[source]#: Get the default compute for JumpStart.

get_enviornment(compute: Compute, environment: Dict[str, str] | None = None, sagemaker_session: Session | None = None) → Dict[str, str][source]#: Get the default environment for JumpStart.

get_hyperparameters(compute: Compute, hyperparameters: Dict[str, Any] | None = None, environment: Dict[str, str] | None = None, sagemaker_session: Session | None = None) → Dict[str, Any][source]#: Get the default hyperparameters for JumpStart.

get_model_artifact_input(compute: Compute, input_data_config: List[Channel | InputData] | None = None, environment: Dict[str, str] | None = None, sagemaker_session: Session | None = None) → List[Channel | InputData][source]#: Get the default model artifact input for JumpStart.

get_networking(networking: Networking | None = None, sagemaker_session: Session | None = None) → Networking[source]#: Get the default networking for JumpStart.

get_output_data_config(base_job_name: str, output_data_config: OutputDataConfig | None = None, sagemaker_session: Session | None = None) → OutputDataConfig[source]#

get_source_code(source_code: SourceCode | None = None, sagemaker_session: Session | None = None) → SourceCode[source]#: Get the default source code for JumpStart.

get_tags(tags: List[Tag] | None = None, sagemaker_session: Session | None = None) → List[Tag][source]#: Get the default tags for JumpStart.

get_training_dataset_input(input_data_config: List[Channel | InputData] | None = None, sagemaker_session: Session | None = None) → List[Channel | InputData][source]#: Get the default training dataset input for JumpStart.

get_training_image(compute: Compute, training_image: str | None = None, sagemaker_session: Session | None = None) → str[source]#: Get the default training image for JumpStart.

class sagemaker.train.defaults.TrainDefaults[source]#

Bases: object

Class to set the base default values for ModelTrainer.

static get_base_job_name(base_job_name: str | None = None, algorithm_name: str | None = None, training_image: str | None = None) → str[source]#: Get the default base job name.

static get_compute(compute: Compute | None = None) → Compute[source]#: Get the default compute.

static get_output_data_config(base_job_name: str, output_data_config: OutputDataConfig | None = None, sagemaker_session: Session | None = None) → OutputDataConfig[source]#: Get the default output data config.

static get_role(role: str | None = None, sagemaker_session: Session | None = None) → str[source]#

Get and validate the training execution role.

Resolves the explicitly provided role (or the caller’s own identity if none is given) and validates it has the permissions/trust required for training. Never creates an IAM role: if validation fails, a RoleValidationError is raised explaining what to grant or how to create a dedicated role via IamRoleResolver().create_execution_role.

static get_sagemaker_session(sagemaker_session: Session | None = None) → Session[source]#: Get the default SageMaker session.

static get_stopping_condition(stopping_condition: StoppingCondition | None = None) → StoppingCondition[source]#: Get the default stopping condition.

static verify_hyperpod_caller_permissions(sagemaker_session: Session | None = None, cluster_name: str | None = None) → bool | None[source]#

Verify the caller can drive the HyperPod CLI (warn, non-blocking).

HyperPod jobs are submitted by the HyperPod CLI running as the caller’s own identity, so — unlike serverless/SMTJ training, which resolves an execution role via get_role() — there is no execution role for the SDK to create here. This checks the caller’s cluster-connect permissions and logs a warning if any are missing.

Returns the verdict from verify_hyperpod_connect_permissions() (True/False/None); it never raises on a missing permission.

sagemaker.train.defaults

Contents

sagemaker.train.defaults#