flytekitplugins.awssagemaker.SagemakerTrainingJobConfig#
- class flytekitplugins.awssagemaker.SagemakerTrainingJobConfig(training_job_resource_config, algorithm_specification, should_persist_output=<function SagemakerTrainingJobConfig.<lambda>>)[source]#
Configuration for Running Training Jobs on Sagemaker. This config can be used to run either the built-in algorithms or custom algorithms.
- Parameters
training_job_resource_config (flytekitplugins.awssagemaker.models.training_job.TrainingJobResourceConfig) – Configuration for Resources to use during the training
algorithm_specification (flytekitplugins.awssagemaker.models.training_job.AlgorithmSpecification) – Specification of the algorithm to use
should_persist_output (Callable[[flytekitplugins.awssagemaker.distributed_training.DistributedTrainingContext], bool]) – This method will be invoked and will decide if the generated model should be persisted as the output.
NOTE: Useful only for distributed training
default: single node training - always persist output
default: distributed training - always persist output on node with rank-0
- Return type
None
Methods
- should_persist_output()#
Attributes
- training_job_resource_config: flytekitplugins.awssagemaker.models.training_job.TrainingJobResourceConfig
- algorithm_specification: flytekitplugins.awssagemaker.models.training_job.AlgorithmSpecification