1.16.26

Kubeflow PyTorch

Directory

Classes

Class Description
flytekitplugins.kfpytorch.task.CleanPodPolicy CleanPodPolicy describes how to deal with pods when the job is finished.
flytekitplugins.kfpytorch.task.Elastic Configuration for [torch elastic training](https://pytorch.
flytekitplugins.kfpytorch.task.ElasticWorkerResult A named tuple representing the result of a torch elastic worker process.
flytekitplugins.kfpytorch.task.Master Configuration for master replica group.
flytekitplugins.kfpytorch.task.PyTorch Configuration for an executable [PyTorch Job](https://github.
flytekitplugins.kfpytorch.task.PyTorchFunctionTask Plugin that submits a PyTorchJob (see https://github.
flytekitplugins.kfpytorch.task.PytorchElasticFunctionTask Plugin for distributed training with torch elastic/torchrun (see.
flytekitplugins.kfpytorch.task.RestartPolicy RestartPolicy describes how the replicas should be restarted.
flytekitplugins.kfpytorch.task.RunPolicy RunPolicy describes some policy to apply to the execution of a kubeflow job.
flytekitplugins.kfpytorch.task.Worker

Functions

Function Description
flytekitplugins.kfpytorch.error_handling.create_recoverable_error_file() Create a file to signal to the agent process that an exception in the worker process is recoverable.
flytekitplugins.kfpytorch.error_handling.is_recoverable_worker_error() Check if the error in the worker process is recoverable.
flytekitplugins.kfpytorch.pod_template.add_shared_mem_volume_to_pod_template() Add shared memory volume and volume mount to the pod template.
flytekitplugins.kfpytorch.task.spawn_helper() Help to spawn worker processes.

Packages

Package Description
flytekitplugins.kfpytorch.error_handling Handle errors in elastic training jobs.
flytekitplugins.kfpytorch.pod_template
flytekitplugins.kfpytorch.task This Plugin adds the capability of running distributed pytorch training to Flyte using backend plugins, natively on.