Merge pull request #1 from aws/master

merging from main repo to fork before re-submitting horovod PR again

Author: ddavydenko

README.rst

@@ -6,14 +6,6 @@
 SageMaker Python SDK
 ====================
 
-.. image:: https://travis-ci.org/aws/sagemaker-python-sdk.svg?branch=master
-   :target: https://travis-ci.org/aws/sagemaker-python-sdk
-   :alt: Build Status
-
-.. image:: https://codecov.io/gh/aws/sagemaker-python-sdk/branch/master/graph/badge.svg
-   :target: https://codecov.io/gh/aws/sagemaker-python-sdk
-   :alt: CodeCov
-
 .. image:: https://img.shields.io/pypi/v/sagemaker.svg
    :target: https://pypi.python.org/pypi/sagemaker
    :alt: Latest Version

doc/using_mxnet.rst

@@ -91,7 +91,7 @@ is good practice.
 Note that SageMaker doesn't support argparse actions.
 For example, if you want to use a boolean hyperparameter, specify ``type`` as ``bool`` in your script and provide an explicit ``True`` or ``False`` value for this hyperparameter when you create the TensorFlow estimator.
 
-For a complete example of a TensorFlow training script, see `mnist.py <https://github.com/awslabs/amazon-sagemaker-examples/blob/master/sagemaker-python-sdk/tensorflow_distributed_mnist/mnist.py>`__.
+For a complete example of a TensorFlow training script, see `mnist.py <https://github.com/awslabs/amazon-sagemaker-examples/blob/master/sagemaker-python-sdk/tensorflow_script_mode_training_and_serving/mnist.py>`__.
 
 
 Adapting your local TensorFlow script
@@ -360,7 +360,7 @@ Training with MKL-DNN disabled
 SageMaker TensorFlow CPU images use TensorFlow built with Intel® MKL-DNN optimization.
 
 In certain cases you might be able to get a better performance by disabling this optimization
-(`for example when using small models <https://github.com/awslabs/amazon-sagemaker-examples/blob/d88d1c19861fb7733941969f5a68821d9da2982e/sagemaker-python-sdk/tensorflow_iris_dnn_classifier_using_estimators/iris_dnn_classifier.py#L7-L9>`_)
+(for example when using small models).
 
 You can disable MKL-DNN optimization for TensorFlow ``1.8.0`` and above by setting two following environment variables:
 

doc/using_tf.rst

@@ -341,6 +341,7 @@ def invoke_endpoint(
         ContentType=None,
         Accept=None,
         CustomAttributes=None,
+        TargetModel=None,
     ):
         """
 
@@ -365,6 +366,9 @@ def invoke_endpoint(
         if CustomAttributes is not None:
             headers["X-Amzn-SageMaker-Custom-Attributes"] = CustomAttributes
 
+        if TargetModel is not None:
+            headers["X-Amzn-SageMaker-Target-Model"] = TargetModel
+
         r = self.http.request("POST", url, body=Body, preload_content=False, headers=headers)
 
         return {"Body": r, "ContentType": Accept}

src/sagemaker/local/local_session.py

@@ -17,7 +17,7 @@
 from six.moves.urllib.parse import urlparse
 
 import sagemaker
-from sagemaker import s3
+from sagemaker import local, s3
 from sagemaker.model import Model
 from sagemaker.session import Session
 
@@ -210,6 +210,9 @@ def deploy(
         if role is None:
             raise ValueError("Role can not be null for deploying a model")
 
+        if instance_type == "local" and not isinstance(self.sagemaker_session, local.LocalSession):
+            self.sagemaker_session = local.LocalSession()
+
         container_def = self.prepare_container_def(instance_type, accelerator_type=accelerator_type)
         self.sagemaker_session.create_model(
             self.name,

src/sagemaker/multidatamodel.py

@@ -534,7 +534,8 @@ def _create_or_update_code_dir(
         if os.path.isdir(dependency):
             shutil.copytree(dependency, os.path.join(lib_dir, os.path.basename(dependency)))
         else:
-            os.mkdir(lib_dir)
+            if not os.path.exists(lib_dir):
+                os.mkdir(lib_dir)
             shutil.copy2(dependency, lib_dir)
 
 

src/sagemaker/utils.py

@@ -177,6 +177,67 @@ def test_multi_data_model_deploy_pretrained_models(
         assert "Could not find endpoint" in str(exception.value)
 
 
+@pytest.mark.local_mode
+def test_multi_data_model_deploy_pretrained_models_local_mode(container_image, sagemaker_session):
+    timestamp = sagemaker_timestamp()
+    endpoint_name = "test-multimodel-endpoint-{}".format(timestamp)
+    model_name = "test-multimodel-{}".format(timestamp)
+
+    # Define pretrained model local path
+    pretrained_model_data_local_path = os.path.join(DATA_DIR, "sparkml_model", "mleap_model.tar.gz")
+
+    with timeout(minutes=30):
+        model_data_prefix = os.path.join(
+            "s3://", sagemaker_session.default_bucket(), "multimodel-{}/".format(timestamp)
+        )
+        multi_data_model = MultiDataModel(
+            name=model_name,
+            model_data_prefix=model_data_prefix,
+            image=container_image,
+            role=ROLE,
+            sagemaker_session=sagemaker_session,
+        )
+
+        # Add model before deploy
+        multi_data_model.add_model(pretrained_model_data_local_path, PRETRAINED_MODEL_PATH_1)
+        # Deploy model to an endpoint
+        multi_data_model.deploy(1, "local", endpoint_name=endpoint_name)
+        # Add models after deploy
+        multi_data_model.add_model(pretrained_model_data_local_path, PRETRAINED_MODEL_PATH_2)
+
+        endpoint_models = []
+        for model_path in multi_data_model.list_models():
+            endpoint_models.append(model_path)
+        assert PRETRAINED_MODEL_PATH_1 in endpoint_models
+        assert PRETRAINED_MODEL_PATH_2 in endpoint_models
+
+        predictor = RealTimePredictor(
+            endpoint=endpoint_name,
+            sagemaker_session=multi_data_model.sagemaker_session,
+            serializer=npy_serializer,
+            deserializer=string_deserializer,
+        )
+
+        data = numpy.zeros(shape=(1, 1, 28, 28))
+        result = predictor.predict(data, target_model=PRETRAINED_MODEL_PATH_1)
+        assert result == "Invoked model: {}".format(PRETRAINED_MODEL_PATH_1)
+
+        result = predictor.predict(data, target_model=PRETRAINED_MODEL_PATH_2)
+        assert result == "Invoked model: {}".format(PRETRAINED_MODEL_PATH_2)
+
+        # Cleanup
+        multi_data_model.sagemaker_session.sagemaker_client.delete_endpoint_config(
+            EndpointConfigName=endpoint_name
+        )
+        multi_data_model.sagemaker_session.delete_endpoint(endpoint_name)
+        multi_data_model.delete_model()
+    with pytest.raises(Exception) as exception:
+        sagemaker_session.sagemaker_client.describe_model(ModelName=multi_data_model.name)
+        assert "Could not find model" in str(exception.value)
+        sagemaker_session.sagemaker_client.describe_endpoint_config(name=endpoint_name)
+        assert "Could not find endpoint" in str(exception.value)
+
+
 def test_multi_data_model_deploy_trained_model_from_framework_estimator(
     container_image, sagemaker_session, cpu_instance_type
 ):

tests/integ/test_multidatamodel.py

@@ -19,4 +19,4 @@ echo "set SAGEMAKER_ROLE_ARN=$SAGEMAKER_ROLE_ARN"
 --region us-west-2 \
 --lifecycle-config-name install-python-sdk \
 --notebook-instance-role-arn $SAGEMAKER_ROLE_ARN \
-./amazon-sagemaker-examples/sagemaker-python-sdk/tensorflow_distributed_mnist/tensorflow_batch_transform_mnist.ipynb
+./amazon-sagemaker-examples/sagemaker-python-sdk/pytorch_cnn_cifar10/pytorch_local_mode_cifar10.ipynb

tests/scripts/run-notebook-test.sh

@@ -444,6 +444,8 @@ def test_repack_model_without_source_dir(tmp, fake_s3):
             "model-dir/model",
             "dependencies/a",
             "dependencies/some/dir/b",
+            "aa",
+            "bb",
             "source-dir/inference.py",
             "source-dir/this-file-should-not-be-included.py",
         ],
@@ -457,6 +459,8 @@ def test_repack_model_without_source_dir(tmp, fake_s3):
         dependencies=[
             os.path.join(tmp, "dependencies/a"),
             os.path.join(tmp, "dependencies/some/dir"),
+            os.path.join(tmp, "aa"),
+            os.path.join(tmp, "bb"),
         ],
         model_uri="s3://fake/location",
         repacked_model_uri="s3://destination-bucket/model.tar.gz",
@@ -466,6 +470,8 @@ def test_repack_model_without_source_dir(tmp, fake_s3):
     assert list_tar_files(fake_s3.fake_upload_path, tmp) == {
         "/model",
         "/code/lib/a",
+        "/code/lib/aa",
+        "/code/lib/bb",
         "/code/lib/dir/b",
         "/code/inference.py",
     }