Merge branch 'master' into amtviz

Author: chad119

CHANGELOG.md

@@ -1,5 +1,15 @@
 # Changelog
 
+## v2.239.2 (2025-02-18)
+
+### Bug Fixes and Other Changes
+
+ * Add warning about not supporting torch.nn.SyncBatchNorm
+ * pass in inference_ami_version to model_based endpoint type
+ * Fix hyperparameter strategy docs
+ * Add framework_version to all TensorFlowModel examples
+ * Move RecordSerializer and RecordDeserializer to sagemaker.serializers and sagemaker.deserialzers
+
 ## v2.239.1 (2025-02-14)
 
 ### Bug Fixes and Other Changes

CONTRIBUTING.md

@@ -61,6 +61,10 @@ Before sending us a pull request, please ensure that:
    1. Follow the instructions at [Modifying an EBS Volume Using Elastic Volumes (Console)](https://docs.aws.amazon.com/AWSEC2/latest/UserGuide/requesting-ebs-volume-modifications.html#modify-ebs-volume) to increase the EBS volume size associated with the newly created EC2 instance.
    1. Wait 5-10min for the new EBS volume increase to finalize.
    1. Allow EC2 to claim the additional space by stopping and then starting your EC2 host.
+2. Set up a venv to manage dependencies:
+   1. `python -m venv ~/.venv/myproject-env` to create the venv
+   2. `source ~/.venv/myproject-env/bin/activate` to activate the venv
+   3. `deactivate` to exit the venv
 
 
 ### Pull Down the Code
@@ -74,8 +78,8 @@ Before sending us a pull request, please ensure that:
 ### Run the Unit Tests
 
 1. Install tox using `pip install tox`
-1. Install coverage using `pip install .[test]`
-1. cd into the sagemaker-python-sdk folder: `cd sagemaker-python-sdk` or `cd /environment/sagemaker-python-sdk`
+1. cd into the github project sagemaker-python-sdk folder: `cd sagemaker-python-sdk` or `cd /environment/sagemaker-python-sdk`
+1. Install coverage using `pip install '.[test]'`
 1. Run the following tox command and verify that all code checks and unit tests pass: `tox tests/unit`
 1. You can also run a single test with the following command: `tox -e py310 -- -s -vv <path_to_file><file_name>::<test_function_name>`
 1. You can run coverage via runcvoerage env : `tox -e runcoverage -- tests/unit` or `tox -e py310 -- tests/unit --cov=sagemaker --cov-append --cov-report xml`

VERSION

@@ -1 +1 @@
-2.239.2.dev0
+2.239.3.dev0

doc/frameworks/pytorch/using_pytorch.rst

@@ -375,6 +375,9 @@ To initialize distributed training in your script, call
 `torch.distributed.init_process_group
 <https://pytorch.org/docs/master/distributed.html#torch.distributed.init_process_group>`_
 with the desired backend and the rank of the current host.
+Warning: Some torch features, such as (and likely not limited to) ``torch.nn.SyncBatchNorm``
+is not supported and its existence in ``init_process_group`` will cause an exception during
+distributed training.
 
 .. code:: python
 

doc/frameworks/tensorflow/deploying_tensorflow_serving.rst

@@ -64,7 +64,7 @@ If you already have existing model artifacts in S3, you can skip training and de
 
   from sagemaker.tensorflow import TensorFlowModel
 
-  model = TensorFlowModel(model_data='s3://mybucket/model.tar.gz', role='MySageMakerRole')
+  model = TensorFlowModel(model_data='s3://mybucket/model.tar.gz', role='MySageMakerRole', framework_version='x.x.x')
 
   predictor = model.deploy(initial_instance_count=1, instance_type='ml.c5.xlarge')
 
@@ -74,7 +74,7 @@ Python-based TensorFlow serving on SageMaker has support for `Elastic Inference
 
     from sagemaker.tensorflow import TensorFlowModel
 
-    model = TensorFlowModel(model_data='s3://mybucket/model.tar.gz', role='MySageMakerRole')
+    model = TensorFlowModel(model_data='s3://mybucket/model.tar.gz', role='MySageMakerRole', framework_version='x.x.x')
 
     predictor = model.deploy(initial_instance_count=1, instance_type='ml.c5.xlarge', accelerator_type='ml.eia1.medium')
 

doc/frameworks/tensorflow/using_tf.rst

@@ -468,7 +468,7 @@ If you already have existing model artifacts in S3, you can skip training and de
 
   from sagemaker.tensorflow import TensorFlowModel
 
-  model = TensorFlowModel(model_data='s3://mybucket/model.tar.gz', role='MySageMakerRole')
+  model = TensorFlowModel(model_data='s3://mybucket/model.tar.gz', role='MySageMakerRole', framework_version='x.x.x')
 
   predictor = model.deploy(initial_instance_count=1, instance_type='ml.c5.xlarge')
 
@@ -478,7 +478,7 @@ Python-based TensorFlow serving on SageMaker has support for `Elastic Inference
 
     from sagemaker.tensorflow import TensorFlowModel
 
-    model = TensorFlowModel(model_data='s3://mybucket/model.tar.gz', role='MySageMakerRole')
+    model = TensorFlowModel(model_data='s3://mybucket/model.tar.gz', role='MySageMakerRole', framework_version='x.x.x')
 
     predictor = model.deploy(initial_instance_count=1, instance_type='ml.c5.xlarge', accelerator_type='ml.eia1.medium')
 
@@ -767,7 +767,8 @@ This customized Python code must be named ``inference.py`` and is specified thro
 
     model = TensorFlowModel(entry_point='inference.py',
                             model_data='s3://mybucket/model.tar.gz',
-                            role='MySageMakerRole')
+                            role='MySageMakerRole',
+                            framework_version='x.x.x')
 
 In the example above, ``inference.py`` is assumed to be a file inside ``model.tar.gz``. If you want to use a local file instead, you must add the ``source_dir`` argument. See the documentation on `TensorFlowModel <https://sagemaker.readthedocs.io/en/stable/frameworks/tensorflow/sagemaker.tensorflow.html#sagemaker.tensorflow.model.TensorFlowModel>`_.
 
@@ -923,7 +924,8 @@ processing. There are 2 ways to do this:
     model = TensorFlowModel(entry_point='inference.py',
                             dependencies=['requirements.txt'],
                             model_data='s3://mybucket/model.tar.gz',
-                            role='MySageMakerRole')
+                            role='MySageMakerRole',
+                            framework_version='x.x.x')
 
 
 2. If you are working in a network-isolation situation or if you don't
@@ -941,7 +943,8 @@ processing. There are 2 ways to do this:
     model = TensorFlowModel(entry_point='inference.py',
                            dependencies=['/path/to/folder/named/lib'],
                            model_data='s3://mybucket/model.tar.gz',
-                           role='MySageMakerRole')
+                           role='MySageMakerRole',
+                           framework_version='x.x.x')
 
 For more information, see: https://github.com/aws/sagemaker-tensorflow-serving-container#prepost-processing
 

doc/v2.rst

@@ -324,9 +324,9 @@ The follow serializer/deserializer classes have been renamed and/or moved:
 +--------------------------------------------------------+-------------------------------------------------------+
 | ``sagemaker.predictor._NPYSerializer``                 | ``sagemaker.serializers.NumpySerializer``             |
 +--------------------------------------------------------+-------------------------------------------------------+
-| ``sagemaker.amazon.common.numpy_to_record_serializer`` | ``sagemaker.amazon.common.RecordSerializer``          |
+| ``sagemaker.amazon.common.numpy_to_record_serializer`` | ``sagemaker.serializers.RecordSerializer``            |
 +--------------------------------------------------------+-------------------------------------------------------+
-| ``sagemaker.amazon.common.record_deserializer``        | ``sagemaker.amazon.common.RecordDeserializer``        |
+| ``sagemaker.amazon.common.record_deserializer``        | ``sagemaker.deserializers.RecordDeserializer``        |
 +--------------------------------------------------------+-------------------------------------------------------+
 | ``sagemaker.predictor._JsonDeserializer``              | ``sagemaker.deserializers.JSONDeserializer``          |
 +--------------------------------------------------------+-------------------------------------------------------+

src/sagemaker/amazon/common.py

@@ -13,84 +13,16 @@
 """Placeholder docstring"""
 from __future__ import absolute_import
 
-import io
 import logging
 import struct
 import sys
 
 import numpy as np
 
 from sagemaker.amazon.record_pb2 import Record
-from sagemaker.deprecations import deprecated_class
-from sagemaker.deserializers import SimpleBaseDeserializer
-from sagemaker.serializers import SimpleBaseSerializer
 from sagemaker.utils import DeferredError
 
 
-class RecordSerializer(SimpleBaseSerializer):
-    """Serialize a NumPy array for an inference request."""
-
-    def __init__(self, content_type="application/x-recordio-protobuf"):
-        """Initialize a ``RecordSerializer`` instance.
-
-        Args:
-            content_type (str): The MIME type to signal to the inference endpoint when sending
-                request data (default: "application/x-recordio-protobuf").
-        """
-        super(RecordSerializer, self).__init__(content_type=content_type)
-
-    def serialize(self, data):
-        """Serialize a NumPy array into a buffer containing RecordIO records.
-
-        Args:
-            data (numpy.ndarray): The data to serialize.
-
-        Returns:
-            io.BytesIO: A buffer containing the data serialized as records.
-        """
-        if len(data.shape) == 1:
-            data = data.reshape(1, data.shape[0])
-
-        if len(data.shape) != 2:
-            raise ValueError(
-                "Expected a 1D or 2D array, but got a %dD array instead." % len(data.shape)
-            )
-
-        buffer = io.BytesIO()
-        write_numpy_to_dense_tensor(buffer, data)
-        buffer.seek(0)
-
-        return buffer
-
-
-class RecordDeserializer(SimpleBaseDeserializer):
-    """Deserialize RecordIO Protobuf data from an inference endpoint."""
-
-    def __init__(self, accept="application/x-recordio-protobuf"):
-        """Initialize a ``RecordDeserializer`` instance.
-
-        Args:
-            accept (union[str, tuple[str]]): The MIME type (or tuple of allowable MIME types) that
-                is expected from the inference endpoint (default:
-                "application/x-recordio-protobuf").
-        """
-        super(RecordDeserializer, self).__init__(accept=accept)
-
-    def deserialize(self, data, content_type):
-        """Deserialize RecordIO Protobuf data from an inference endpoint.
-
-        Args:
-            data (object): The protobuf message to deserialize.
-            content_type (str): The MIME type of the data.
-        Returns:
-            list: A list of records.
-        """
-        try:
-            return read_records(data)
-        finally:
-            data.close()
-
-
 def _write_feature_tensor(resolved_type, record, vector):
     """Placeholder Docstring"""
     if resolved_type == "Int32":
@@ -288,7 +220,3 @@ def _resolve_type(dtype):
     if dtype == np.dtype("float32"):
         return "Float32"
     raise ValueError("Unsupported dtype {} on array".format(dtype))
-
-
-numpy_to_record_serializer = deprecated_class(RecordSerializer, "numpy_to_record_serializer")
-record_deserializer = deprecated_class(RecordDeserializer, "record_deserializer")

src/sagemaker/amazon/factorization_machines.py

@@ -17,11 +17,12 @@
 
 from sagemaker import image_uris
 from sagemaker.amazon.amazon_estimator import AmazonAlgorithmEstimatorBase
-from sagemaker.amazon.common import RecordSerializer, RecordDeserializer
 from sagemaker.amazon.hyperparameter import Hyperparameter as hp  # noqa
 from sagemaker.amazon.validation import gt, isin, ge
+from sagemaker.deserializers import RecordDeserializer
 from sagemaker.predictor import Predictor
 from sagemaker.model import Model
+from sagemaker.serializers import RecordSerializer
 from sagemaker.session import Session
 from sagemaker.utils import pop_out_unused_kwarg
 from sagemaker.vpc_utils import VPC_CONFIG_DEFAULT

src/sagemaker/amazon/kmeans.py

@@ -17,11 +17,12 @@
 
 from sagemaker import image_uris
 from sagemaker.amazon.amazon_estimator import AmazonAlgorithmEstimatorBase
-from sagemaker.amazon.common import RecordSerializer, RecordDeserializer
 from sagemaker.amazon.hyperparameter import Hyperparameter as hp  # noqa
 from sagemaker.amazon.validation import gt, isin, ge, le
+from sagemaker.deserializers import RecordDeserializer
 from sagemaker.predictor import Predictor
 from sagemaker.model import Model
+from sagemaker.serializers import RecordSerializer
 from sagemaker.session import Session
 from sagemaker.utils import pop_out_unused_kwarg
 from sagemaker.vpc_utils import VPC_CONFIG_DEFAULT

src/sagemaker/amazon/knn.py

@@ -17,11 +17,12 @@
 
 from sagemaker import image_uris
 from sagemaker.amazon.amazon_estimator import AmazonAlgorithmEstimatorBase
-from sagemaker.amazon.common import RecordSerializer, RecordDeserializer
 from sagemaker.amazon.hyperparameter import Hyperparameter as hp  # noqa
 from sagemaker.amazon.validation import ge, isin
+from sagemaker.deserializers import RecordDeserializer
 from sagemaker.predictor import Predictor
 from sagemaker.model import Model
+from sagemaker.serializers import RecordSerializer
 from sagemaker.session import Session
 from sagemaker.utils import pop_out_unused_kwarg
 from sagemaker.vpc_utils import VPC_CONFIG_DEFAULT

src/sagemaker/amazon/lda.py

@@ -18,11 +18,12 @@
 
 from sagemaker import image_uris
 from sagemaker.amazon.amazon_estimator import AmazonAlgorithmEstimatorBase
-from sagemaker.amazon.common import RecordSerializer, RecordDeserializer
+from sagemaker.deserializers import RecordDeserializer
 from sagemaker.amazon.hyperparameter import Hyperparameter as hp  # noqa
 from sagemaker.amazon.validation import gt
 from sagemaker.predictor import Predictor
 from sagemaker.model import Model
+from sagemaker.serializers import RecordSerializer
 from sagemaker.session import Session
 from sagemaker.utils import pop_out_unused_kwarg
 from sagemaker.vpc_utils import VPC_CONFIG_DEFAULT

src/sagemaker/amazon/linear_learner.py

@@ -18,11 +18,12 @@
 
 from sagemaker import image_uris
 from sagemaker.amazon.amazon_estimator import AmazonAlgorithmEstimatorBase
-from sagemaker.amazon.common import RecordSerializer, RecordDeserializer
+from sagemaker.deserializers import RecordDeserializer
 from sagemaker.amazon.hyperparameter import Hyperparameter as hp  # noqa
 from sagemaker.amazon.validation import isin, gt, lt, ge, le
 from sagemaker.predictor import Predictor
 from sagemaker.model import Model
+from sagemaker.serializers import RecordSerializer
 from sagemaker.session import Session
 from sagemaker.utils import pop_out_unused_kwarg
 from sagemaker.vpc_utils import VPC_CONFIG_DEFAULT

src/sagemaker/amazon/ntm.py

@@ -17,11 +17,12 @@
 
 from sagemaker import image_uris
 from sagemaker.amazon.amazon_estimator import AmazonAlgorithmEstimatorBase
-from sagemaker.amazon.common import RecordSerializer, RecordDeserializer
 from sagemaker.amazon.hyperparameter import Hyperparameter as hp  # noqa
 from sagemaker.amazon.validation import ge, le, isin
+from sagemaker.deserializers import RecordDeserializer
 from sagemaker.predictor import Predictor
 from sagemaker.model import Model
+from sagemaker.serializers import RecordSerializer
 from sagemaker.session import Session
 from sagemaker.utils import pop_out_unused_kwarg
 from sagemaker.vpc_utils import VPC_CONFIG_DEFAULT

src/sagemaker/amazon/pca.py

@@ -17,11 +17,12 @@
 
 from sagemaker import image_uris
 from sagemaker.amazon.amazon_estimator import AmazonAlgorithmEstimatorBase
-from sagemaker.amazon.common import RecordSerializer, RecordDeserializer
 from sagemaker.amazon.hyperparameter import Hyperparameter as hp  # noqa
 from sagemaker.amazon.validation import gt, isin
+from sagemaker.deserializers import RecordDeserializer
 from sagemaker.predictor import Predictor
 from sagemaker.model import Model
+from sagemaker.serializers import RecordSerializer
 from sagemaker.session import Session
 from sagemaker.utils import pop_out_unused_kwarg
 from sagemaker.vpc_utils import VPC_CONFIG_DEFAULT

src/sagemaker/amazon/randomcutforest.py

@@ -17,11 +17,12 @@
 
 from sagemaker import image_uris
 from sagemaker.amazon.amazon_estimator import AmazonAlgorithmEstimatorBase
-from sagemaker.amazon.common import RecordSerializer, RecordDeserializer
 from sagemaker.amazon.hyperparameter import Hyperparameter as hp  # noqa
 from sagemaker.amazon.validation import ge, le
+from sagemaker.deserializers import RecordDeserializer
 from sagemaker.predictor import Predictor
 from sagemaker.model import Model
+from sagemaker.serializers import RecordSerializer
 from sagemaker.session import Session
 from sagemaker.utils import pop_out_unused_kwarg
 from sagemaker.vpc_utils import VPC_CONFIG_DEFAULT

src/sagemaker/base_deserializers.py

@@ -23,6 +23,7 @@
 import numpy as np
 from six import with_metaclass
 
+from sagemaker.amazon.common import read_records
 from sagemaker.utils import DeferredError
 
 try:
@@ -388,3 +389,31 @@ def deserialize(self, stream, content_type="tensor/pt"):
                 "Unable to deserialize your data to torch.Tensor.\
                     Please provide custom deserializer in InferenceSpec."
             )
+
+
+class RecordDeserializer(SimpleBaseDeserializer):
+    """Deserialize RecordIO Protobuf data from an inference endpoint."""
+
+    def __init__(self, accept="application/x-recordio-protobuf"):
+        """Initialize a ``RecordDeserializer`` instance.
+
+        Args:
+            accept (union[str, tuple[str]]): The MIME type (or tuple of allowable MIME types) that
+                is expected from the inference endpoint (default:
+                "application/x-recordio-protobuf").
+        """
+        super(RecordDeserializer, self).__init__(accept=accept)
+
+    def deserialize(self, data, content_type):
+        """Deserialize RecordIO Protobuf data from an inference endpoint.
+
+        Args:
+            data (object): The protobuf message to deserialize.
+            content_type (str): The MIME type of the data.
+        Returns:
+            list: A list of records.
+        """
+        try:
+            return read_records(data)
+        finally:
+            data.close()