Merge branch 'master' into feature-release

makungaj1 · web-flow · commit a2630674cd37 · 2024-07-08T00:32:01.000-07:00
diff --git a/CHANGELOG.md b/CHANGELOG.md
@@ -1,5 +1,25 @@
 # Changelog
 
+## v2.224.4 (2024-07-04)
+
+### Bug Fixes and Other Changes
+
+ * allow for inf spec and server override to be passed
+
+## v2.224.3 (2024-07-03)
+
+### Bug Fixes and Other Changes
+
+ * Upgrade local dependencies
+ * Improve docstrings for estimator tags
+
+## v2.224.2 (2024-06-27)
+
+### Bug Fixes and Other Changes
+
+ * Update DJLModel class for latest container releases
+ * list_models() for python3.8
+
 ## v2.224.1 (2024-06-21)
 
 ### Bug Fixes and Other Changes
diff --git a/VERSION b/VERSION
@@ -1 +1 @@
-2.224.2.dev0
+2.224.5.dev0
diff --git a/requirements/extras/local_requirements.txt b/requirements/extras/local_requirements.txt
@@ -1,3 +1,3 @@
 urllib3>=1.26.8,<3.0.0
-docker>=5.0.2,<7.0.0
+docker>=5.0.2,<8.0.0
 PyYAML>=5.4.1,<7
diff --git a/requirements/extras/test_requirements.txt b/requirements/extras/test_requirements.txt
@@ -38,4 +38,4 @@ accelerate>=0.24.1,<=0.27.0
 schema==0.7.5
 tensorflow>=2.1,<=2.16
 mlflow>=2.12.2,<2.13
-huggingface_hub>=0.23.4
+huggingface_hub>=0.23.4
diff --git a/src/sagemaker/estimator.py b/src/sagemaker/estimator.py
@@ -274,7 +274,10 @@ def __init__(
                 AWS services needed. If not specified, the estimator creates one
                 using the default AWS configuration chain.
             tags (Optional[Tags]):
-                Tags for labeling a training job. For more, see
+                Tags for labeling a training job. These won't be propagated to Models,
+                Endpoints during :meth:`~sagemaker.estimator.EstimatorBase.deploy`. The
+                :meth:`~sagemaker.estimator.EstimatorBase.deploy` takes in a seperate
+                tags parameter. For more on tags, see
                 https://docs.aws.amazon.com/sagemaker/latest/dg/API_Tag.html.
             subnets (list[str] or list[PipelineVariable]): List of subnet ids. If not
                 specified training job will be created without VPC config.
diff --git a/src/sagemaker/image_uri_config/pytorch.json b/src/sagemaker/image_uri_config/pytorch.json
@@ -1107,7 +1107,8 @@
             "1.12": "1.12.1",
             "2.0": "2.0.1",
             "2.1": "2.1.0",
-            "2.2": "2.2.1"
+            "2.2": "2.2.1",
+            "2.3": "2.3.0"
         },
         "versions": {
             "1.12.1": {
@@ -1331,6 +1332,50 @@
                     "us-west-2": "763104351884"
                 },
                 "repository": "pytorch-inference-graviton"
+            },
+            "2.3.0": {
+                "container_version": {
+                    "cpu": "ubuntu20.04"
+                },
+                "py_versions": [
+                    "py311"
+                ],
+                "registries": {
+                    "af-south-1": "626614931356",
+                    "ap-east-1": "871362719292",
+                    "ap-northeast-1": "763104351884",
+                    "ap-northeast-2": "763104351884",
+                    "ap-northeast-3": "364406365360",
+                    "ap-south-1": "763104351884",
+                    "ap-south-2": "772153158452",
+                    "ap-southeast-1": "763104351884",
+                    "ap-southeast-2": "763104351884",
+                    "ap-southeast-3": "907027046896",
+                    "ap-southeast-4": "457447274322",
+                    "ca-central-1": "763104351884",
+                    "ca-west-1": "204538143572",
+                    "cn-north-1": "727897471807",
+                    "cn-northwest-1": "727897471807",
+                    "eu-central-1": "763104351884",
+                    "eu-central-2": "380420809688",
+                    "eu-north-1": "763104351884",
+                    "eu-south-1": "692866216735",
+                    "eu-south-2": "503227376785",
+                    "eu-west-1": "763104351884",
+                    "eu-west-2": "763104351884",
+                    "eu-west-3": "763104351884",
+                    "il-central-1": "780543022126",
+                    "me-central-1": "914824155844",
+                    "me-south-1": "217643126080",
+                    "sa-east-1": "763104351884",
+                    "us-east-1": "763104351884",
+                    "us-east-2": "763104351884",
+                    "us-gov-east-1": "446045086412",
+                    "us-gov-west-1": "442386744353",
+                    "us-west-1": "763104351884",
+                    "us-west-2": "763104351884"
+                },
+                "repository": "pytorch-inference-graviton"
             }
         }
     },
diff --git a/src/sagemaker/jumpstart/constants.py b/src/sagemaker/jumpstart/constants.py
@@ -206,6 +206,8 @@
 
 JUMPSTART_MODEL_HUB_NAME = "SageMakerPublicHub"
 
+JUMPSTART_MODEL_HUB_NAME = "SageMakerPublicHub"
+
 JUMPSTART_DEFAULT_MANIFEST_FILE_S3_KEY = "models_manifest.json"
 JUMPSTART_DEFAULT_PROPRIETARY_MANIFEST_KEY = "proprietary-sdk-manifest.json"
 
diff --git a/src/sagemaker/jumpstart/enums.py b/src/sagemaker/jumpstart/enums.py
@@ -93,10 +93,8 @@ class JumpStartTag(str, Enum):
     MODEL_ID = "sagemaker-sdk:jumpstart-model-id"
     MODEL_VERSION = "sagemaker-sdk:jumpstart-model-version"
     MODEL_TYPE = "sagemaker-sdk:jumpstart-model-type"
-
     INFERENCE_CONFIG_NAME = "sagemaker-sdk:jumpstart-inference-config-name"
     TRAINING_CONFIG_NAME = "sagemaker-sdk:jumpstart-training-config-name"
-
     HUB_CONTENT_ARN = "sagemaker-sdk:hub-content-arn"
 
 
diff --git a/src/sagemaker/jumpstart/factory/estimator.py b/src/sagemaker/jumpstart/factory/estimator.py
@@ -62,6 +62,7 @@
 from sagemaker.jumpstart.utils import (
     add_hub_content_arn_tags,
     add_jumpstart_model_info_tags,
+    add_jumpstart_model_id_version_tags,
     get_eula_message,
     get_default_jumpstart_session_with_user_agent_suffix,
     update_dict_if_key_not_present,
diff --git a/src/sagemaker/jumpstart/factory/model.py b/src/sagemaker/jumpstart/factory/model.py
@@ -49,6 +49,7 @@
 from sagemaker.jumpstart.utils import (
     add_hub_content_arn_tags,
     add_jumpstart_model_info_tags,
+    add_jumpstart_model_id_version_tags,
     get_default_jumpstart_session_with_user_agent_suffix,
     get_neo_content_bucket,
     update_dict_if_key_not_present,
diff --git a/src/sagemaker/jumpstart/hub/interfaces.py b/src/sagemaker/jumpstart/hub/interfaces.py
@@ -585,7 +585,7 @@ def from_json(self, json_obj: Dict[str, Any]) -> None:
         )
         self.hosting_eula_uri: Optional[str] = json_obj.get("HostingEulaUri")
         self.hosting_model_package_arn: Optional[str] = json_obj.get("HostingModelPackageArn")
-
+          
         self.inference_config_rankings = self._get_config_rankings(json_obj)
         self.inference_config_components = self._get_config_components(json_obj)
         self.inference_configs = self._get_configs(json_obj)
diff --git a/src/sagemaker/jumpstart/types.py b/src/sagemaker/jumpstart/types.py
@@ -867,10 +867,8 @@ def _get_regional_property(
 
         if regional_property_alias is None and regional_property_value is None:
             instance_type_family = get_instance_type_family(instance_type)
-
             if instance_type_family in {"", None}:
                 return None
-
             if self.regional_aliases:
                 regional_property_alias = (
                     self.variants.get(instance_type_family, {})
@@ -1976,17 +1974,15 @@ def extract_region_from_arn(arn: str) -> Optional[str]:
         hub_region = None
         if match:
             hub_region = match.group(2)
-
             return hub_region
-
+          
         match = re.match(HUB_ARN_REGEX, arn)
         if match:
             hub_region = match.group(2)
             return hub_region
 
         return hub_region
 
-
 class JumpStartCachedContentValue(JumpStartDataHolderType):
     """Data class for the s3 cached content values."""
 
diff --git a/src/sagemaker/jumpstart/utils.py b/src/sagemaker/jumpstart/utils.py
@@ -446,6 +446,21 @@ def add_hub_content_arn_tags(
     return tags
 
 
+def add_hub_content_arn_tags(
+    tags: Optional[List[TagsDict]],
+    hub_arn: str,
+) -> Optional[List[TagsDict]]:
+    """Adds custom Hub arn tag to JumpStart related resources."""
+
+    tags = add_single_jumpstart_tag(
+        hub_arn,
+        enums.JumpStartTag.HUB_CONTENT_ARN,
+        tags,
+        is_uri=False,
+    )
+    return tags
+
+
 def add_jumpstart_uri_tags(
     tags: Optional[List[TagsDict]] = None,
     inference_model_uri: Optional[Union[str, dict]] = None,
diff --git a/src/sagemaker/serve/builder/model_builder.py b/src/sagemaker/serve/builder/model_builder.py
@@ -848,6 +848,8 @@ def build(  # pylint: disable=R0911
         if self.env_vars.get("HUGGING_FACE_HUB_TOKEN") and not self.env_vars.get("HF_TOKEN"):
             self.env_vars["HF_TOKEN"] = self.env_vars.get("HUGGING_FACE_HUB_TOKEN")
 
+        self.sagemaker_session.settings._local_download_dir = self.model_path
+
         # https://github.com/boto/botocore/blob/develop/botocore/useragent.py#L258
         # decorate to_string() due to
         # https://github.com/boto/botocore/blob/develop/botocore/client.py#L1014-L1015
@@ -903,6 +905,20 @@ def build(  # pylint: disable=R0911
 
         raise ValueError("%s model server is not supported" % self.model_server)
 
+    def _build_validations(self):
+        """Validations needed for model server overrides, or auto-detection or fallback"""
+        if self.mode == Mode.IN_PROCESS:
+            raise ValueError("IN_PROCESS mode is not supported yet!")
+
+        if self.inference_spec and self.model:
+            raise ValueError("Can only set one of the following: model, inference_spec.")
+
+        if self.image_uri and not is_1p_image_uri(self.image_uri) and self.model_server is None:
+            raise ValueError(
+                "Model_server must be set when non-first-party image_uri is set. "
+                + "Supported model servers: %s" % supported_model_servers
+            )
+
     def _build_for_model_server(self):  # pylint: disable=R0911, R1710
         """Model server overrides"""
         if self.model_server not in supported_model_servers:
@@ -915,8 +931,8 @@ def _build_for_model_server(self):  # pylint: disable=R0911, R1710
         if self.model_metadata:
             mlflow_path = self.model_metadata.get(MLFLOW_MODEL_PATH)
 
-        if not self.model and not mlflow_path:
-            raise ValueError("Missing required parameter `model` or 'ml_flow' path")
+        if not self.model and not mlflow_path and not self.inference_spec:
+            raise ValueError("Missing required parameter `model` or 'ml_flow' path or inf_spec")
 
         if self.model_server == ModelServer.TORCHSERVE:
             return self._build_for_torchserve()
diff --git a/src/sagemaker/session.py b/src/sagemaker/session.py
@@ -680,7 +680,6 @@ def general_bucket_check_if_user_has_permission(
             s3 (str): S3 object from boto session
             region (str): The region in which to create the bucket.
             bucket_creation_date_none (bool):Indicating whether S3 bucket already exists or not
-
         """
         try:
             s3.meta.client.head_bucket(Bucket=bucket_name)
diff --git a/tests/unit/sagemaker/jumpstart/constants.py b/tests/unit/sagemaker/jumpstart/constants.py
@@ -9876,8 +9876,31 @@
                         },
                     ]
                 },
+                "p4de": {"properties": {"image_uri": "$gpu_ecr_uri_1"}},
+                "p5": {"properties": {"image_uri": "$gpu_ecr_uri_1"}},
             },
         },
+        "ContextualHelp": {
+            "HubFormatTrainData": [
+                "A train and an optional validation directories. Each directory contains a CSV/JSON/TXT. ",
+                "- For CSV/JSON files, the text data is used from the column called 'text' or the first column if no column called 'text' is found",  # noqa: E501
+                "- The number of files under train and validation (if provided) should equal to one, respectively.",
+                " [Learn how to setup an AWS S3 bucket.](https://docs.aws.amazon.com/AmazonS3/latest/dev/UsingBucket.html)",  # noqa: E501
+            ],
+            "HubDefaultTrainData": [
+                "Dataset: [SEC](https://www.sec.gov/edgar/searchedgar/companysearch)",
+                "SEC filing contains regulatory documents that companies and issuers of securities must submit to the Securities and Exchange Commission (SEC) on a regular basis.",  # noqa: E501
+                "License: [CC BY-SA 4.0](https://creativecommons.org/licenses/by-sa/4.0/legalcode)",
+            ],
+        },
+        "ModelDataDownloadTimeout": 1200,
+        "ContainerStartupHealthCheckTimeout": 1200,
+        "EncryptInterContainerTraffic": True,
+        "DisableOutputCompression": True,
+        "MaxRuntimeInSeconds": 360000,
+        "DynamicContainerDeploymentSupported": True,
+        "TrainingModelPackageArtifactUri": None,
+        "Dependencies": [],
     },
     "meta-textgeneration-llama-2-70b": {
         "Url": "https://ai.meta.com/resources/models-and-libraries/llama-downloads/",
diff --git a/tests/unit/sagemaker/jumpstart/model/test_model.py b/tests/unit/sagemaker/jumpstart/model/test_model.py
@@ -1492,6 +1492,7 @@ def test_attach(
             model_version="some-version",
             inference_component_name="some-ic-name",
         )
+        
         mock_get_model_info_from_endpoint.assert_not_called()
 
     @mock.patch(
diff --git a/tests/unit/sagemaker/serve/builder/test_model_builder.py b/tests/unit/sagemaker/serve/builder/test_model_builder.py
@@ -2328,8 +2328,6 @@ def test_build_tensorflow_serving_non_mlflow_case(
             mock_session,
         )
 
-        # builder.build(sagemaker_session=mock_session, role_arn=mock_role_arn, mode=Mode.SAGEMAKER_ENDPOINT)
-
     @pytest.mark.skip(reason="Implementation not completed")
     @patch.object(ModelBuilder, "_get_serve_setting", autospec=True)
     @patch("sagemaker.serve.utils.telemetry_logger._send_telemetry")
