Merge branch 'master' into patch-1

Author: shreyapandit

src/sagemaker/debugger/__init__.py

@@ -15,6 +15,7 @@
 
 from sagemaker.debugger.debugger import (  # noqa: F401
     CollectionConfig,
+    DEBUGGER_FLAG,
     DebuggerHookConfig,
     framework_name,
     get_default_profiler_rule,

src/sagemaker/debugger/debugger.py

@@ -32,6 +32,7 @@
 from sagemaker.utils import build_dict
 
 framework_name = "debugger"
+DEBUGGER_FLAG = "USE_SMDEBUG"
 
 
 def get_rule_container_image_uri(region):

src/sagemaker/estimator.py

@@ -29,6 +29,7 @@
 from sagemaker.analytics import TrainingJobAnalytics
 from sagemaker.debugger import TensorBoardOutputConfig  # noqa: F401 # pylint: disable=unused-import
 from sagemaker.debugger import (
+    DEBUGGER_FLAG,
     DebuggerHookConfig,
     FrameworkProfile,
     get_default_profiler_rule,
@@ -2269,6 +2270,11 @@ def _validate_and_set_debugger_configs(self):
                     )
                     self.debugger_hook_config = False
 
+        if self.debugger_hook_config is False:
+            if self.environment is None:
+                self.environment = {}
+            self.environment[DEBUGGER_FLAG] = "0"
+
     def _stage_user_code_in_s3(self):
         """Upload the user training script to s3 and return the location.
 

src/sagemaker/workflow/step_collections.py

@@ -112,8 +112,8 @@ def __init__(
         if "entry_point" in kwargs:
             repack_model = True
             entry_point = kwargs.pop("entry_point", None)
-            source_dir = kwargs.get("source_dir")
-            dependencies = kwargs.get("dependencies")
+            source_dir = kwargs.pop("source_dir", None)
+            dependencies = kwargs.pop("dependencies", None)
             kwargs = dict(**kwargs, output_kms_key=kwargs.pop("model_kms_key", None))
 
             repack_model_step = _RepackModelStep(
@@ -130,13 +130,10 @@ def __init__(
             steps.append(repack_model_step)
             model_data = repack_model_step.properties.ModelArtifacts.S3ModelArtifacts
 
-        # remove kwargs consumed by model repacking step
-        kwargs.pop("entry_point", None)
-        kwargs.pop("source_dir", None)
-        kwargs.pop("dependencies", None)
-        kwargs.pop("output_kms_key", None)
+            # remove kwargs consumed by model repacking step
+            kwargs.pop("output_kms_key", None)
 
-        if model is not None:
+        elif model is not None:
             if isinstance(model, PipelineModel):
                 self.model_list = model.models
                 self.container_def_list = model.pipeline_container_def(inference_instances[0])
@@ -156,7 +153,9 @@ def __init__(
                     entry_point = model_entity.entry_point
                     source_dir = model_entity.source_dir
                     dependencies = model_entity.dependencies
+                    kwargs = dict(**kwargs, output_kms_key=model_entity.model_kms_key)
                     name = model_entity.name or model_entity._framework_name
+
                     repack_model_step = _RepackModelStep(
                         name=f"{name}RepackModel",
                         depends_on=depends_on,
@@ -166,12 +165,16 @@ def __init__(
                         entry_point=entry_point,
                         source_dir=source_dir,
                         dependencies=dependencies,
+                        **kwargs,
                     )
                     steps.append(repack_model_step)
                     model_entity.model_data = (
                         repack_model_step.properties.ModelArtifacts.S3ModelArtifacts
                     )
 
+                    # remove kwargs consumed by model repacking step
+                    kwargs.pop("output_kms_key", None)
+
         register_model_step = _RegisterModelStep(
             name=name,
             estimator=estimator,

tests/integ/test_debugger.py

@@ -18,6 +18,7 @@
 import pytest
 
 from sagemaker.debugger.debugger import (
+    DEBUGGER_FLAG,
     DebuggerHookConfig,
     Rule,
     rule_configs,
@@ -748,6 +749,7 @@ def test_mxnet_with_debugger_hook_config_disabled(
         job_description = mx.latest_training_job.describe()
 
         assert job_description.get("DebugHookConfig") is None
+        assert job_description.get("Environment", {}).get(DEBUGGER_FLAG) == "0"
 
 
 def _get_rule_evaluation_statuses(job_description):

tests/unit/sagemaker/workflow/test_steps.py

@@ -23,7 +23,7 @@
     patch,
 )
 
-from sagemaker.debugger import ProfilerConfig
+from sagemaker.debugger import DEBUGGER_FLAG, ProfilerConfig
 from sagemaker.estimator import Estimator
 from sagemaker.tensorflow import TensorFlow
 from sagemaker.inputs import TrainingInput, TransformInput, CreateModelInput
@@ -275,6 +275,7 @@ def test_training_step_tensorflow(sagemaker_session):
                 "sagemaker_distributed_dataparallel_custom_mpi_options": '""',
             },
             "ProfilerConfig": {"S3OutputPath": "s3://my-bucket/"},
+            "Environment": {DEBUGGER_FLAG: "0"},
         },
         "CacheConfig": {"Enabled": True, "ExpireAfter": "PT1H"},
     }