feature: pytorch 1.3.1 eia

Author: Deng

src/sagemaker/fw_utils.py

@@ -53,7 +53,14 @@
 )
 
 VALID_PY_VERSIONS = ["py2", "py3"]
-VALID_EIA_FRAMEWORKS = ["tensorflow", "tensorflow-serving", "mxnet", "mxnet-serving"]
+VALID_EIA_FRAMEWORKS = [
+    "tensorflow",
+    "tensorflow-serving",
+    "mxnet",
+    "mxnet-serving",
+    "pytorch",
+    "pytorch-serving",
+]
 VALID_ACCOUNTS_BY_REGION = {"us-gov-west-1": "246785580436", "us-iso-east-1": "744548109606"}
 ASIMOV_VALID_ACCOUNTS_BY_REGION = {"us-iso-east-1": "886529160074"}
 OPT_IN_ACCOUNTS_BY_REGION = {"ap-east-1": "057415533634", "me-south-1": "724002660598"}
@@ -71,6 +78,7 @@
     "mxnet-serving-eia": "mxnet-inference-eia",
     "pytorch": "pytorch-training",
     "pytorch-serving": "pytorch-inference",
+    "pytorch-serving-eia": "pytorch-inference-eia",
 }
 
 MERGED_FRAMEWORKS_LOWEST_VERSIONS = {
@@ -82,6 +90,7 @@
     "mxnet-serving-eia": [1, 4, 1],
     "pytorch": [1, 2, 0],
     "pytorch-serving": [1, 2, 0],
+    "pytorch-serving-eia": {"py3": [1, 3, 1]},
 }
 
 DEBUGGER_UNSUPPORTED_REGIONS = ["us-gov-west-1", "us-iso-east-1"]
@@ -117,6 +126,8 @@ def _is_dlc_version(framework, framework_version, py_version):
     """
     lowest_version_list = MERGED_FRAMEWORKS_LOWEST_VERSIONS.get(framework)
     if isinstance(lowest_version_list, dict):
+        if py_version not in lowest_version_list:
+            raise ValueError("{} is not supported in {}.".format(framework, py_version))
         lowest_version_list = lowest_version_list[py_version]
 
     if lowest_version_list:

src/sagemaker/pytorch/README.rst

@@ -6,6 +6,8 @@ With PyTorch Estimators and Models, you can train and host PyTorch models on Ama
 
 Supported versions of PyTorch: ``0.4.0``, ``1.0.0``, ``1.1.0``, ``1.2.0``, ``1.3.1``, ``1.4.0``.
 
+Supported versions of TensorFlow for Elastic Inference: ``1.3.1``.
+
 We recommend that you use the latest supported version, because that's where we focus most of our development efforts.
 
 You can visit the PyTorch repository at https://github.com/pytorch/pytorch.

tests/data/pytorch_eia/mnist.py

@@ -0,0 +1,13 @@
+# Copyright 2020 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.
+# This file is intentionally left blank to invoke default model_fn and predict_fn

tests/data/pytorch_eia/model_mnist.tar.gz

@@ -27,6 +27,10 @@
 MNIST_DIR = os.path.join(DATA_DIR, "pytorch_mnist")
 MNIST_SCRIPT = os.path.join(MNIST_DIR, "mnist.py")
 
+EIA_DIR = os.path.join(DATA_DIR, "pytorch_eia")
+EIA_MODEL = os.path.join(EIA_DIR, "model_mnist.tar.gz")
+EIA_SCRIPT = os.path.join(EIA_DIR, "mnist.py")
+
 
 @pytest.fixture(scope="module", name="pytorch_training_job")
 def fixture_training_job(sagemaker_session, pytorch_full_version, cpu_instance_type):
@@ -115,6 +119,35 @@ def test_deploy_model(pytorch_training_job, sagemaker_session, cpu_instance_type
         assert output.shape == (batch_size, 10)
 
 
+@pytest.mark.skipif(
+    PYTHON_VERSION == "py2",
+    reason="PyTorch EIA does not support Python 2.",
+)
+def test_deploy_model_with_accelerator(sagemaker_session, cpu_instance_type):
+    endpoint_name = "test-pytorch-deploy-eia-{}".format(sagemaker_timestamp())
+    model_data = sagemaker_session.upload_data(path=EIA_MODEL)
+    pytorch = PyTorchModel(
+        model_data,
+        "SageMakerRole",
+        framework_version="1.3.1",
+        entry_point=EIA_SCRIPT,
+        sagemaker_session=sagemaker_session,
+    )
+    with timeout_and_delete_endpoint_by_name(endpoint_name, sagemaker_session):
+        predictor = pytorch.deploy(
+            initial_instance_count=1,
+            instance_type=cpu_instance_type,
+            accelerator_type="ml.eia1.large",
+            endpoint_name=endpoint_name,
+        )
+
+        batch_size = 100
+        data = numpy.random.rand(batch_size, 1, 28, 28).astype(numpy.float32)
+        output = predictor.predict(data)
+
+        assert output.shape == (batch_size, 10)
+
+
 def _upload_training_data(pytorch):
     return pytorch.sagemaker_session.upload_data(
         path=os.path.join(MNIST_DIR, "training"),

tests/integ/test_pytorch_train.py

@@ -311,6 +311,37 @@ def test_mxnet_eia_images():
     )
 
 
+def test_pytorch_eia_images():
+    image_uri = fw_utils.create_image_uri(
+        "us-east-1",
+        "pytorch-serving",
+        "ml.c4.2xlarge",
+        "1.3.1",
+        "py3",
+        accelerator_type="ml.eia1.large",
+    )
+    assert (
+        image_uri
+        == "{}.dkr.ecr.us-east-1.amazonaws.com/pytorch-inference-eia:1.3.1-cpu-py3".format(
+            fw_utils.ASIMOV_PROD_ACCOUNT
+        )
+    )
+
+
+def test_pytorch_eia_py2_error():
+    error_message = "pytorch-serving-eia is not supported in py2."
+    with pytest.raises(ValueError) as error:
+        fw_utils.create_image_uri(
+            "us-east-1",
+            "pytorch-serving",
+            "ml.c4.2xlarge",
+            "1.3.1",
+            "py2",
+            accelerator_type="ml.eia1.large",
+        )
+    assert error_message in str(error)
+
+
 def test_create_image_uri_override_account():
     image_uri = fw_utils.create_image_uri(
         "us-west-1", MOCK_FRAMEWORK, "ml.p3.2xlarge", "1.0rc", "py3", account="fake"

tests/unit/test_fw_utils.py

@@ -348,8 +348,8 @@ def test_model_image_accelerator(sagemaker_session):
     model = PyTorchModel(
         MODEL_DATA, role=ROLE, entry_point=SCRIPT_PATH, sagemaker_session=sagemaker_session
     )
-    with pytest.raises(ValueError):
-        model.prepare_container_def(INSTANCE_TYPE, accelerator_type=ACCELERATOR_TYPE)
+    predictor = model.deploy(1, CPU)
+    assert isinstance(predictor, PyTorchPredictor)
 
 
 def test_train_image_default(sagemaker_session):