fix: Revert "change: add a check to prevent launching a modelparallel job on CPU only instances" (#3280)

Author: navinsoni

src/sagemaker/fw_utils.py

@@ -835,52 +835,6 @@ def validate_pytorch_distribution(
         raise ValueError(err_msg)
 
 
-def validate_distribution_instance(sagemaker_session, distribution, instance_type):
-    """Check to prevent launching a modelparallel job on CPU only instances.
-
-    Args:
-        sagemaker_session (sagemaker.session.Session): Session object which
-            manages interactions with Amazon SageMaker APIs and any other
-            AWS services needed.
-        distribution (dict): A dictionary with information to enable distributed training.
-            distribution = {
-                "smdistributed": {
-                    "modelparallel": {
-                        "enabled": True,
-                        "parameters": {
-                            ...
-                        },
-                    },
-                },
-                ...
-            }
-        instance_type (str): A string representing the type of training instance selected.
-
-    Raises:
-        ValueError: when modelparallel is enabled, if the instance_type does not support GPU.
-    """
-    if "smdistributed" not in distribution:
-        # Distribution strategy other than smdistributed is selected
-        return
-
-    if "modelparallel" not in distribution["smdistributed"]:
-        # Strategy other than modelparallel is selected
-        return
-
-    if not distribution["smdistributed"]["modelparallel"]["enabled"]:
-        # Strategy modelparallel is not enabled
-        return
-
-    instance_desc = sagemaker_session.boto_session.client("ec2").describe_instance_types(
-        InstanceTypes=[f"{instance_type}"]
-    )
-    if "GpuInfo" not in instance_desc["InstanceTypes"][0]:
-        raise ValueError(
-            f"modelparallel only runs on GPU-enabled instances. "
-            f"{instance_type} does not support GPU."
-        )
-
-
 def python_deprecation_warning(framework, latest_supported_version):
     """Placeholder docstring"""
     return PYTHON_2_DEPRECATION_WARNING.format(

src/sagemaker/pytorch/estimator.py

@@ -25,7 +25,6 @@
     python_deprecation_warning,
     validate_version_or_image_args,
     validate_distribution,
-    validate_distribution_instance,
 )
 from sagemaker.pytorch import defaults
 from sagemaker.pytorch.model import PyTorchModel
@@ -221,12 +220,6 @@ def __init__(
             entry_point, source_dir, hyperparameters, image_uri=image_uri, **kwargs
         )
         if distribution is not None:
-            instance_type = self._get_instance_type()
-            # remove "ml." prefix
-            if instance_type[:3] == "ml.":
-                instance_type = instance_type[3:]
-            validate_distribution_instance(self.sagemaker_session, distribution, instance_type)
-
             distribution = validate_distribution(
                 distribution,
                 self.instance_groups,

tests/unit/test_fw_utils.py

@@ -49,15 +49,6 @@ def sagemaker_session():
     session_mock.sagemaker_client.describe_training_job = Mock(
         return_value={"ModelArtifacts": {"S3ModelArtifacts": "s3://m/m.tar.gz"}}
     )
-    session_mock.boto_session.client("ec2").describe_instance_types = Mock(
-        return_value={
-            "InstanceTypes": [
-                {
-                    "CpuInfo": {},
-                },
-            ],
-        }
-    )
     return session_mock
 
 
@@ -742,31 +733,6 @@ def test_validate_smdistributed_not_raises():
             )
 
 
-def test_validate_distribution_instance_no_smdistributed(sagemaker_session):
-    distribution = {}
-    instance_type = "mock_type"
-    fw_utils.validate_distribution_instance(sagemaker_session, distribution, instance_type)
-
-
-def test_validate_distribution_instance_no_modelparallel(sagemaker_session):
-    distribution = {"smdistributed": {}}
-    instance_type = "mock_type"
-    fw_utils.validate_distribution_instance(sagemaker_session, distribution, instance_type)
-
-
-def test_validate_distribution_instance_disabled_modelparallel(sagemaker_session):
-    distribution = {"smdistributed": {"modelparallel": {"enabled": False}}}
-    instance_type = "mock_type"
-    fw_utils.validate_distribution_instance(sagemaker_session, distribution, instance_type)
-
-
-def test_validate_distribution_instance_raise(sagemaker_session):
-    distribution = {"smdistributed": {"modelparallel": {"enabled": True}}}
-    instance_type = "mock_type"
-    with pytest.raises(ValueError):
-        fw_utils.validate_distribution_instance(sagemaker_session, distribution, instance_type)
-
-
 def test_validate_smdistributed_raises():
     bad_args = [
         {"smdistributed": "dummy"},