sagemaker-hyperpod 3.0.0__py3-none-any.whl

sagemaker/hyperpod/cli/commands/training.py

@@ -0,0 +1,347 @@
+import click
+import logging
+import os
+import yaml
+import shutil
+import subprocess
+from pathlib import Path
+from sagemaker.hyperpod.training.hyperpod_pytorch_job import HyperPodPytorchJob
+from sagemaker.hyperpod.common.config import Metadata
+import tempfile
+from typing import List, Dict, Any, Optional, Callable, get_args, get_origin, Literal
+from sagemaker.hyperpod.cli.training_utils import generate_click_command
+from importlib.metadata import entry_points
+from hyperpod_pytorch_job_template.registry import SCHEMA_REGISTRY
+
+
+@click.command("hyp-pytorch-job")
+@click.option("--version", default="1.0", help="Schema version to use")
+@click.option("--debug", default=False, help="Enable debug mode")
+@generate_click_command(
+    schema_pkg="hyperpod_pytorch_job_template",
+    registry=SCHEMA_REGISTRY,
+)
+def pytorch_create(version, debug, config):
+    """Create a PyTorch job"""
+    try:
+        click.echo(f"Using version: {version}")
+        job_name = config.get("name")
+        namespace = config.get("namespace")
+        spec = config.get("spec")
+
+        # Prepare metadata
+        metadata_kwargs = {"name": job_name}
+        if namespace:
+            metadata_kwargs["namespace"] = namespace
+
+        # Prepare job kwargs
+        job_kwargs = {
+            "metadata": Metadata(**metadata_kwargs),
+            "replica_specs": spec.get("replica_specs"),
+        }
+
+        # Add nproc_per_node if present
+        if "nproc_per_node" in spec:
+            job_kwargs["nproc_per_node"] = spec.get("nproc_per_node")
+
+        # Add run_policy if present
+        if "run_policy" in spec:
+            job_kwargs["run_policy"] = spec.get("run_policy")
+
+        # Create job
+        job = HyperPodPytorchJob(**job_kwargs)
+        job.create(debug=debug)
+
+    except Exception as e:
+        raise click.UsageError(f"Failed to create job: {str(e)}")
+
+
+@click.command("hyp-pytorch-job")
+@click.option(
+    "--namespace",
+    "-n",
+    default="default",
+    help="Optional. The namespace to list jobs from. Defaults to 'default' namespace.",
+)
+def list_jobs(namespace: str):
+    """List all HyperPod PyTorch jobs"""
+    try:
+        jobs = HyperPodPytorchJob.list(namespace=namespace)
+
+        if not jobs:
+            click.echo("No jobs found.")
+            return
+
+        # Define headers and widths
+        headers = ["NAME", "NAMESPACE", "STATUS", "AGE"]
+        widths = [30, 20, 15, 15]
+
+        # Print header
+        header = "".join(f"{h:<{w}}" for h, w in zip(headers, widths))
+        click.echo("\n" + header)
+        click.echo("-" * sum(widths))
+
+        # Print each job
+        for job in jobs:
+            # Get status from conditions
+            status = "Unknown"
+            age = "N/A"
+            if job.status and job.status.conditions:
+                for condition in reversed(job.status.conditions):
+                    if condition.status == "True":
+                        status = condition.type
+                        break
+
+                # Calculate age
+                if job.status and job.status.conditions:
+                    # Find the 'Created' condition to get the start time
+                    created_condition = next(
+                        (c for c in job.status.conditions if c.type == "Created"), None
+                    )
+                    if created_condition and created_condition.lastTransitionTime:
+                        from datetime import datetime, timezone
+
+                        start_time = datetime.fromisoformat(
+                            created_condition.lastTransitionTime.replace("Z", "+00:00")
+                        )
+                        now = datetime.now(timezone.utc)
+                        delta = now - start_time
+                        if delta.days > 0:
+                            age = f"{delta.days}d"
+                        else:
+                            hours = delta.seconds // 3600
+                            if hours > 0:
+                                age = f"{hours}h"
+                            else:
+                                minutes = (delta.seconds % 3600) // 60
+                                age = f"{minutes}m"
+
+            # Format row
+            row = "".join(
+                [
+                    f"{job.metadata.name:<{widths[0]}}",
+                    f"{job.metadata.namespace:<{widths[1]}}",
+                    f"{status:<{widths[2]}}",
+                    f"{age:<{widths[3]}}",
+                ]
+            )
+            click.echo(row)
+
+            click.echo()  # Add empty line at the end
+
+    except Exception as e:
+        raise click.UsageError(f"Failed to list jobs: {str(e)}")
+
+
+@click.command("hyp-pytorch-job")
+@click.option(
+    "--job-name", required=True, help="Required. The name of the job to describe"
+)
+@click.option(
+    "--namespace",
+    "-n",
+    default="default",
+    help="Optional. The namespace of the job. Defaults to 'default' namespace.",
+)
+def pytorch_describe(job_name: str, namespace: str):
+    """Describe a HyperPod PyTorch job"""
+    try:
+        job = HyperPodPytorchJob.get(name=job_name, namespace=namespace)
+
+        if job is None:
+            raise click.UsageError(f"Job {job_name} not found in namespace {namespace}")
+
+        # Print basic info
+        click.echo("\nJob Details:")
+        click.echo("=" * 80)
+        click.echo(f"Name:           {job.metadata.name}")
+        click.echo(f"Namespace:      {job.metadata.namespace}")
+
+        # Print Spec details
+        click.echo("\nSpec:")
+        click.echo("-" * 80)
+        click.echo(f"Processes per Node: {getattr(job, 'nprocPerNode', 'N/A')}")
+
+        # Print Replica Specs
+        for replica in job.replicaSpecs:
+            click.echo(f"\nReplica Spec:")
+            click.echo(f"  Name:     {getattr(replica, 'name', 'N/A')}")
+            click.echo(f"  Replicas: {getattr(replica, 'replicas', 'N/A')}")
+            click.echo(f"  Spares:   {getattr(replica, 'spares', 'N/A')}")
+
+            # Container details
+            if (
+                hasattr(replica, "template")
+                and hasattr(replica.template, "spec")
+                and hasattr(replica.template.spec, "containers")
+            ):
+                for container in replica.template.spec.containers:
+                    click.echo("\n  Container:")
+                    click.echo(
+                        f"    Name:            {getattr(container, 'name', 'N/A')}"
+                    )
+                    click.echo(
+                        f"    Image:           {getattr(container, 'image', 'N/A')}"
+                    )
+                    click.echo(
+                        f"    Image Pull Policy: {getattr(container, 'imagePullPolicy', 'N/A')}"
+                    )
+                    if container.resources:
+                        click.echo("    Resources:")
+                        if container.resources.limits:
+                            click.echo(f"      Limits:   {container.resources.limits}")
+                        if container.resources.requests:
+                            click.echo(
+                                f"      Requests: {container.resources.requests}"
+                            )
+
+        # Print Run Policy
+        click.echo("\nRun Policy:")
+        click.echo("-" * 80)
+        if hasattr(job, "runPolicy"):
+            click.echo(
+                f"Clean Pod Policy:          {getattr(job.runPolicy, 'cleanPodPolicy', 'N/A')}"
+            )
+            click.echo(
+                f"TTL Seconds After Finished: {getattr(job.runPolicy, 'ttlSecondsAfterFinished', 'N/A')}"
+            )
+        else:
+            click.echo("Run Policy: N/A")
+
+        # Print Status
+        click.echo("\nStatus:")
+        click.echo("-" * 80)
+        if job.status:
+            if job.status.conditions:
+                click.echo("Conditions:")
+                for condition in job.status.conditions:
+                    click.echo(
+                        f"  Type:               {getattr(condition, 'type', 'N/A')}"
+                    )
+                    click.echo(
+                        f"  Status:             {getattr(condition, 'status', 'N/A')}"
+                    )
+                    click.echo(
+                        f"  Last Transition:    {getattr(condition, 'lastTransitionTime', 'N/A')}"
+                    )
+                    if condition.message:
+                        click.echo(f"  Message:            {condition.message}")
+                    click.echo()
+        else:
+            click.echo("No status information available")
+
+    except Exception as e:
+        raise click.UsageError(f"Failed to describe job: {str(e)}")
+
+
+@click.command("hyp-pytorch-job")
+@click.option(
+    "--job-name", required=True, help="Required. The name of the job to delete"
+)
+@click.option(
+    "--namespace",
+    "-n",
+    default="default",
+    help="Optional. The namespace of the job. Defaults to 'default' namespace.",
+)
+def pytorch_delete(job_name: str, namespace: str):
+    """Delete a HyperPod PyTorch job"""
+    try:
+        job = HyperPodPytorchJob.get(name=job_name, namespace=namespace)
+        job.delete()
+
+        if job is None:
+            raise click.UsageError(f"Job {job_name} not found in namespace {namespace}")
+
+    except Exception as e:
+        raise click.UsageError(f"Failed to describe job: {str(e)}")
+
+
+@click.command("hyp-pytorch-job")
+@click.option(
+    "--job-name",
+    required=True,
+    help="Required. Specify the job name to list its associated pods.",
+)
+@click.option(
+    "--namespace",
+    "-n",
+    default="default",
+    help="Optional. The namespace of the job. Defaults to 'default' namespace.",
+)
+def pytorch_list_pods(job_name: str, namespace: str):
+    """List all HyperPod PyTorch pods corresponding to the job"""
+    try:
+        job = HyperPodPytorchJob.get(name=job_name, namespace=namespace)
+        pods = job.list_pods()
+
+        if not pods:
+            click.echo(f"\nNo pods found for job: {job_name}")
+            return
+
+        # Define headers and widths
+        headers = ["POD NAME", "NAMESPACE"]
+        widths = [50, 20]
+
+        # Print header
+        click.echo(f"\nPods for job: {job_name}")
+        header = "".join(f"{h:<{w}}" for h, w in zip(headers, widths))
+        click.echo("\n" + header)
+        click.echo("-" * sum(widths))
+
+        # Print each pod
+        for pod in pods:
+            row = "".join([f"{pod:<{widths[0]}}", f"{namespace:<{widths[1]}}"])
+            click.echo(row)
+
+        click.echo()
+
+    except Exception as e:
+        raise click.UsageError(f"Failed to list jobs: {str(e)}")
+
+
+@click.command("hyp-pytorch-job")
+@click.option(
+    "--job-name",
+    required=True,
+    help="Required. Specify the job name for pod log retrieval.",
+)
+@click.option(
+    "--pod-name", required=True, help="Required. The name of the pod to get logs from."
+)
+@click.option(
+    "--namespace",
+    "-n",
+    default="default",
+    help="Optional. The namespace of the job. Defaults to 'default' namespace.",
+)
+def pytorch_get_logs(job_name: str, pod_name: str, namespace: str):
+    """Get specific logs from pod corresponding to the job"""
+    try:
+        click.echo("Listing logs for pod: " + pod_name)
+        job = HyperPodPytorchJob.get(name=job_name, namespace=namespace)
+        logs = job.get_logs_from_pod(pod_name=pod_name)
+
+        if not logs:
+            click.echo("No logs available.")
+            return
+
+        # Split logs into lines and display them
+        log_lines = logs.split("\n")
+        for line in log_lines:
+            if line.strip():  # Skip empty lines
+                # Color coding based on log level
+                if "ERROR" in line.upper():
+                    click.secho(line, fg="red")
+                elif "WARNING" in line.upper():
+                    click.secho(line, fg="yellow")
+                elif "INFO" in line.upper():
+                    click.secho(line, fg="green")
+                else:
+                    click.echo(line)
+
+        click.echo("\nEnd of logs")
+        click.echo("=" * 80)
+
+    except Exception as e:
+        raise click.UsageError(f"Failed to list jobs: {str(e)}")

sagemaker/hyperpod/cli/constants/command_constants.py

@@ -0,0 +1,105 @@
+# Copyright Amazon.com, Inc. or its affiliates. All Rights Reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License"). You
+# may not use this file except in compliance with the License. A copy of
+# the License is located at
+#
+#     http://aws.amazon.com/apache2.0/
+#
+# or in the "license" file accompanying this file. This file is
+# distributed on an "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF
+# ANY KIND, either express or implied. See the License for the specific
+# language governing permissions and limitations under the License.
+from enum import Enum
+from pathlib import Path
+
+GENERATED_LAUNCHER_CONFIG_FILE_PATH = "/tmp/"
+HYPERPOD_KUBERNETES_JOB_PREFIX = "hyperpod-k8s-job"
+HYPERPOD_CLUSTER_CONTEXT_FILE_NAME = "hyperpod_current_context.json"
+NODE_AFFINITY_DICT = {
+    "required": {"sagemaker.amazonaws.com/node-health-status": ["Schedulable"]},
+    "preferred": {"sagemaker.amazonaws.com/deep-health-check-status": ["Passed"]},
+    "weights": [100],
+}
+DEEP_HEALTH_CHECK_PASSED_ONLY_NODE_AFFINITY_DICT = {
+    "required": {
+        "sagemaker.amazonaws.com/deep-health-check-status": ["Passed"],
+    },
+}
+KUEUE_QUEUE_NAME_LABEL_KEY = "kueue.x-k8s.io/queue-name"
+KUEUE_WORKLOAD_PRIORITY_CLASS_LABEL_KEY = "kueue.x-k8s.io/priority-class"
+KUEUE_JOB_UID_LABEL_KEY = "kueue.x-k8s.io/job-uid"
+HYPERPOD_AUTO_RESUME_ANNOTATION_KEY = "sagemaker.amazonaws.com/enable-job-auto-resume"
+HYPERPOD_MAX_RETRY_ANNOTATION_KEY = "sagemaker.amazonaws.com/job-max-retry-count"
+ENV_VARS_DICT = {"NCCL_DEBUG": "INFO"}
+SAGEMAKER_HYPERPOD_NAME_LABEL = "sagemaker.amazonaws.com/cluster-name"
+HP_HEALTH_STATUS_LABEL = "sagemaker.amazonaws.com/node-health-status"
+INSTANCE_TYPE_LABEL = "node.kubernetes.io/instance-type"
+DEEP_HEALTH_CHECK_STATUS_LABEL = "sagemaker.amazonaws.com/deep-health-check-status"
+SAGEMAKER_MANAGED_QUEUE_LABEL= "sagemaker.amazonaws.com/sagemaker-managed-queue"
+SAGEMAKER_QUOTA_ALLOCATION_LABEL = "sagemaker.amazonaws.com/quota-allocation-id"
+TEMP_KUBE_CONFIG_FILE = "/tmp/kubeconfig"
+HYPERPOD_NAMESPACE_PREFIX = "hyperpod-ns-"
+SAGEMAKER_MANAGED_LOCAL_QUEUE_SUFFIX = "-localqueue"
+SAGEMAKER_MANAGED_CLUSTER_QUEUE_SUFFIX = "-clusterqueue"
+SAGEMAKER_TRAINING_LAUNCHER_DIR = str(Path(__file__).parent.parent / "sagemaker_hyperpod_recipes")
+NVIDIA_GPU_RESOURCE_LIMIT_KEY = "nvidia.com/gpu"
+AVAILABLE_ACCELERATOR_DEVICES_KEY = "AvailableAcceleratorDevices"
+TOTAL_ACCELERATOR_DEVICES_KEY = "TotalAcceleratorDevices"
+USER_NAME_LABEL_KEY = "sagemaker.user/created-by"
+
+class PullPolicy(Enum):
+    ALWAYS = "Always"
+    IF_NOT_PRESENT = "IfNotPresent"
+    NEVER = "Never"
+
+
+class RestartPolicy(Enum):
+    ALWAYS = "Always"
+    ON_FAILURE = "OnFailure"
+    NEVER = "Never"
+    EXIT_CODE = "ExitCode"
+
+
+class OutputFormat(Enum):
+    JSON = "json"
+    TABLE = "table"
+
+
+class PersistentVolumeClaim:
+    claim_name: str
+    mount_path: str
+
+    def __init__(self, claim_name, mount_path):
+        self.claim_name = claim_name
+        self.mount_path = mount_path
+
+
+class SchedulerType(Enum):
+    KUEUE = "Kueue"
+    SAGEMAKER = "SageMaker"
+    NONE = "None"
+
+    def get_values():
+        return [scheduler_type.value for scheduler_type in SchedulerType]
+
+    def get_default():
+        return SchedulerType.SAGEMAKER
+
+class JobPatchType(Enum):
+    SUSPEND = "suspend"
+    UNSUSPEND = "unsuspend"
+
+    def get_values():
+        return [type.value for type in JobPatchType]
+
+
+class Volume:
+    volume_name: str
+    host_path: str
+    mount_path: str
+
+    def __init__(self, volume_name, host_path, mount_path):
+        self.host_path = host_path
+        self.mount_path = mount_path
+        self.volume_name = volume_name

sagemaker/hyperpod/cli/constants/exception_constants.py

@@ -0,0 +1,13 @@
+# Copyright Amazon.com, Inc. or its affiliates. All Rights Reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License"). You
+# may not use this file except in compliance with the License. A copy of
+# the License is located at
+#
+#     http://aws.amazon.com/apache2.0/
+#
+# or in the "license" file accompanying this file. This file is
+# distributed on an "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF
+# ANY KIND, either express or implied. See the License for the specific
+# language governing permissions and limitations under the License.
+RESOURCE_NOT_FOUND_CODE = 404

sagemaker/hyperpod/cli/constants/hyperpod_instance_types.py

@@ -0,0 +1,100 @@
+# Copyright Amazon.com, Inc. or its affiliates. All Rights Reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License"). You
+# may not use this file except in compliance with the License. A copy of
+# the License is located at
+#
+#     http://aws.amazon.com/apache2.0/
+#
+# or in the "license" file accompanying this file. This file is
+# distributed on an "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF
+# ANY KIND, either express or implied. See the License for the specific
+# language governing permissions and limitations under the License.
+from enum import Enum
+
+
+class HyperpodInstanceType(Enum):
+    ML_C5_LARGE = "ml.c5.large"
+    ML_C5_XLARGE = "ml.c5.xlarge"
+    ML_C5_2XLARGE = "ml.c5.2xlarge"
+    ML_C5_4XLARGE = "ml.c5.4xlarge"
+    ML_C5_9XLARGE = "ml.c5.9xlarge"
+    ML_C5_12XLARGE = "ml.c5.12xlarge"
+    ML_C5_18XLARGE = "ml.c5.18xlarge"
+    ML_C5_24XLARGE = "ml.c5.24xlarge"
+    ML_C5N_LARGE = "ml.c5n.large"
+    ML_C5N_2XLARGE = "ml.c5n.2xlarge"
+    ML_C5N_4XLARGE = "ml.c5n.4xlarge"
+    ML_C5N_9XLARGE = "ml.c5n.9xlarge"
+    ML_C5N_18XLARGE = "ml.c5n.18xlarge"
+    ML_G5_XLARGE = "ml.g5.xlarge"
+    ML_G5_2XLARGE = "ml.g5.2xlarge"
+    ML_G5_4XLARGE = "ml.g5.4xlarge"
+    ML_G5_8XLARGE = "ml.g5.8xlarge"
+    ML_G5_12XLARGE = "ml.g5.12xlarge"
+    ML_G5_16XLARGE = "ml.g5.16xlarge"
+    ML_G5_24XLARGE = "ml.g5.24xlarge"
+    ML_G5_48XLARGE = "ml.g5.48xlarge"
+    ML_G6_XLARGE = "ml.g6.xlarge"
+    ML_G6_2XLARGE = "ml.g6.2xlarge"
+    ML_G6_4XLARGE = "ml.g6.4xlarge"
+    ML_G6_8XLARGE = "ml.g6.8xlarge"
+    ML_G6_12XLARGE = "ml.g6.12xlarge"
+    ML_G6_16XLARGE = "ml.g6.16xlarge"
+    ML_G6_24XLARGE = "ml.g6.24xlarge"
+    ML_G6_48XLARGE = "ml.g6.48xlarge"
+    ML_G6E_XLARGE = "ml.g6e.xlarge"
+    ML_G6E_2XLARGE = "ml.g6e.2xlarge"
+    ML_G6E_4XLARGE = "ml.g6e.4xlarge"
+    ML_G6E_8XLARGE = "ml.g6e.8xlarge"
+    ML_G6E_12XLARGE = "ml.g6e.12xlarge"
+    ML_G6E_16XLARGE = "ml.g6e.16xlarge"
+    ML_G6E_24XLARGE = "ml.g6e.24xlarge"
+    ML_G6E_48XLARGE = "ml.g6e.48xlarge"
+    ML_GR6_4XLARGE = "ml.gr6.4xlarge"
+    ML_GR6_8XLARGE = "ml.gr6.8xlarge"
+    ML_I3EN_LARGE = "ml.i3en.large"
+    ML_I3EN_XLARGE = "ml.i3en.xlarge"
+    ML_I3EN_2XLARGE = "ml.i3en.2xlarge"
+    ML_I3EN_3XLARGE = "ml.i3en.3xlarge"
+    ML_I3EN_6XLARGE = "ml.i3en.6xlarge"
+    ML_I3EN_12XLARGE = "ml.i3en.12xlarge"
+    ML_I3EN_24XLARGE = "ml.i3en.24xlarge"
+    ML_M5_LARGE = "ml.m5.large"
+    ML_M5_XLARGE = "ml.m5.xlarge"
+    ML_M5_2XLARGE = "ml.m5.2xlarge"
+    ML_M5_4XLARGE = "ml.m5.4xlarge"
+    ML_M5_8XLARGE = "ml.m5.8xlarge"
+    ML_M5_12XLARGE = "ml.m5.12xlarge"
+    ML_M5_16XLARGE = "ml.m5.16xlarge"
+    ML_M5_24XLARGE = "ml.m5.24xlarge"
+    ML_M7I_LARGE = "ml.m7i.large"
+    ML_M7I_XLARGE = "ml.m7i.xlarge"
+    ML_M7I_2XLARGE = "ml.m7i.2xlarge"
+    ML_M7I_4XLARGE = "ml.m7i.4xlarge"
+    ML_M7I_8XLARGE = "ml.m7i.8xlarge"
+    ML_M7I_12XLARGE = "ml.m7i.12xlarge"
+    ML_M7I_16XLARGE = "ml.m7i.16xlarge"
+    ML_M7I_24XLARGE = "ml.m7i.24xlarge"
+    ML_M7I_48XLARGE = "ml.m7i.48xlarge"
+    ML_P4D_24XLARGE = "ml.p4d.24xlarge"
+    ML_P4DE_24XLARGE = "ml.p4de.24xlarge"
+    ML_P5_48XLARGE = "ml.p5.48xlarge"
+    ML_P5E_48XLARGE = "ml.p5e.48xlarge"
+    ML_P5EN_48XLARGE = "ml.p5en.48xlarge"
+    ML_R7I_LARGE = "ml.r7i.large"
+    ML_R7I_XLARGE = "ml.r7i.xlarge"
+    ML_R7I_2XLARGE = "ml.r7i.2xlarge"
+    ML_R7I_4XLARGE = "ml.r7i.4xlarge"
+    ML_R7I_8XLARGE = "ml.r7i.8xlarge"
+    ML_R7I_12XLARGE = "ml.r7i.12xlarge"
+    ML_R7I_16XLARGE = "ml.r7i.16xlarge"
+    ML_R7I_24XLARGE = "ml.r7i.24xlarge"
+    ML_R7I_48XLARGE = "ml.r7i.48xlarge"
+    ML_T3_MEDIUM = "ml.t3.medium"
+    ML_T3_LARGE = "ml.t3.large"
+    ML_T3_XLARGE = "ml.t3.xlarge"
+    ML_T3_2XLARGE = "ml.t3.2xlarge"
+    ML_TRN1_32XLARGE = "ml.trn1.32xlarge"
+    ML_TRN1N_32XLARGE = "ml.trn1n.32xlarge"
+    ML_TRN2_48XLARGE = "ml.trn2.48xlarge"

sagemaker/hyperpod/cli/constants/kueue_constants.py

@@ -0,0 +1,17 @@
+# Copyright Amazon.com, Inc. or its affiliates. All Rights Reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License"). You
+# may not use this file except in compliance with the License. A copy of
+# the License is located at
+#
+#     http://aws.amazon.com/apache2.0/
+#
+# or in the "license" file accompanying this file. This file is
+# distributed on an "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF
+# ANY KIND, either express or implied. See the License for the specific
+# language governing permissions and limitations under the License.
+KUEUE_CUSTOM_OBJECT_GROUP = "kueue.x-k8s.io"
+WORKLOAD_CUSTOM_OBJECT_PLURAL = "workloads"
+KUEUE_CUSTOM_OBJECT_VERSION = "v1beta1"
+WORKLOAD_PRIORITY_CLASS_CUSTOM_OBJECT_PLURAL = "workloadpriorityclasses"
+CLUSTER_QUEUE_PRIORITY_CLASS_CUSTOM_OBJECT_PLURAL = "clusterqueues"

sagemaker/hyperpod/cli/constants/pytorch_constants.py

@@ -0,0 +1,15 @@
+# Copyright Amazon.com, Inc. or its affiliates. All Rights Reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License"). You
+# may not use this file except in compliance with the License. A copy of
+# the License is located at
+#
+#     http://aws.amazon.com/apache2.0/
+#
+# or in the "license" file accompanying this file. This file is
+# distributed on an "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF
+# ANY KIND, either express or implied. See the License for the specific
+# language governing permissions and limitations under the License.
+PYTORCH_CUSTOM_OBJECT_GROUP = "kubeflow.org"
+PYTORCH_CUSTOM_OBJECT_PLURAL = "pytorchjobs"
+PYTORCH_CUSTOM_OBJECT_VERSION = "v1"