pass accelerator_type for tfs model

CHANGELOG.rst

@@ -8,6 +8,7 @@ CHANGELOG
 * doc-fix: Remove incorrect parameter for EI TFS Python README
 * feature: ``Predictor``: delete SageMaker model
 * feature: ``Pipeline``: delete SageMaker model
+* bug-fix: pass accelerator_type in ``deploy`` for REST API TFS ``Model``
 
 1.18.3.post1
 ============

src/sagemaker/fw_utils.py

@@ -106,7 +106,7 @@ def _accelerator_type_valid_for_framework(framework, accelerator_type=None, opti
 
     if framework not in VALID_EIA_FRAMEWORKS:
         raise ValueError('{} is not supported with Amazon Elastic Inference. Currently only '
-                         'TensorFlow and MXNet are supported for SageMaker.'.format(framework))
+                         'Python-based TensorFlow and MXNet are supported for SageMaker.'.format(framework))
 
     if optimized_families:
         raise ValueError('Neo does not support Amazon Elastic Inference.')

src/sagemaker/tensorflow/deploying_python.rst

@@ -25,7 +25,7 @@ like this:
 
 The code block above deploys a SageMaker Endpoint with one instance of the type 'ml.c4.xlarge'.
 
-TensorFlow serving on SageMaker has support for `Elastic Inference <https://docs.aws.amazon.com/sagemaker/latest/dg/ei.html>`_, which allows for inference acceleration to a hosted endpoint for a fraction of the cost of using a full GPU instance. In order to attach an Elastic Inference accelerator to your endpoint provide the accelerator type to ``accelerator_type`` to your ``deploy`` call.
+Python-based TensorFlow serving on SageMaker has support for `Elastic Inference <https://docs.aws.amazon.com/sagemaker/latest/dg/ei.html>`_, which allows for inference acceleration to a hosted endpoint for a fraction of the cost of using a full GPU instance. In order to attach an Elastic Inference accelerator to your endpoint provide the accelerator type to ``accelerator_type`` to your ``deploy`` call.
 
 .. code:: python
 

src/sagemaker/tensorflow/deploying_tensorflow_serving.rst

@@ -34,6 +34,8 @@ estimator object to create a SageMaker Endpoint:
 
 The code block above deploys a SageMaker Endpoint with one instance of the type 'ml.c5.xlarge'.
 
+As of now, only the Python based TensorFlow serving endpoints support Elastic Inference. For more information, see `Deploying to Python-based Endpoints <https://github.com/aws/sagemaker-python-sdk/blob/master/src/sagemaker/tensorflow/deploying_python.rst#deploying-to-python-based-endpoints>`_.
+
 What happens when deploy is called
 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
 

src/sagemaker/tensorflow/serving.py

@@ -123,7 +123,7 @@ def __init__(self, model_data, role, image=None, framework_version=TF_VERSION,
         self._container_log_level = container_log_level
 
     def prepare_container_def(self, instance_type, accelerator_type=None):
-        image = self._get_image_uri(instance_type)
+        image = self._get_image_uri(instance_type, accelerator_type)
         env = self._get_container_env()
         return sagemaker.container_def(image, self.model_data, env)
 
@@ -139,10 +139,10 @@ def _get_container_env(self):
         env[Model.LOG_LEVEL_PARAM_NAME] = Model.LOG_LEVEL_MAP[self._container_log_level]
         return env
 
-    def _get_image_uri(self, instance_type):
+    def _get_image_uri(self, instance_type, accelerator_type=None):
         if self.image:
             return self.image
 
         region_name = self.sagemaker_session.boto_region_name
         return create_image_uri(region_name, Model.FRAMEWORK_NAME, instance_type,
-                                self._framework_version)
+                                self._framework_version, accelerator_type=accelerator_type)

tests/unit/test_tfs.py

@@ -25,6 +25,7 @@
 CSV_CONTENT_TYPE = 'text/csv'
 INSTANCE_COUNT = 1
 INSTANCE_TYPE = 'ml.c4.4xlarge'
+ACCELERATOR_TYPE = 'ml.eia.medium'
 ROLE = 'Dummy'
 REGION = 'us-west-2'
 PREDICT_INPUT = {'instances': [1.0, 2.0, 5.0]}
@@ -75,6 +76,13 @@ def test_tfs_model(sagemaker_session, tf_version):
     assert isinstance(predictor, Predictor)
 
 
+def test_tfs_model_image_accelerator(sagemaker_session, tf_version):
+    model = Model("s3://some/data.tar.gz", role=ROLE, framework_version=tf_version,
+                  sagemaker_session=sagemaker_session)
+    with pytest.raises(ValueError):
+        model.prepare_container_def(INSTANCE_TYPE, accelerator_type=ACCELERATOR_TYPE)
+
+
 def test_tfs_model_with_log_level(sagemaker_session, tf_version):
     model = Model("s3://some/data.tar.gz", role=ROLE, framework_version=tf_version,
                   container_log_level=logging.INFO,