Merge branch 'aws:master' into master

Author: tlian25

.pylintrc

@@ -384,7 +384,7 @@ max-returns=6
 max-branches=12
 
 # Maximum number of statements in function / method body
-max-statements=100
+max-statements=105
 
 # Maximum number of parents for a class (see R0901).
 max-parents=7

CHANGELOG.md

@@ -1,5 +1,33 @@
 # Changelog
 
+## v2.231.0 (2024-08-30)
+
+### Features
+
+ * Add SageMaker Core to the dependency
+
+### Bug Fixes and Other Changes
+
+ * Disable test_mnist_async
+ * SMP v2.5
+ * update image_uri_configs  08-29-2024 07:17:59 PST
+
+## v2.230.0 (2024-08-28)
+
+### Features
+
+ * FastAPI integration for In_Process Mode (2/2)
+
+### Bug Fixes and Other Changes
+
+ * chore: add HF LLM neuronx 0.0.24 image
+ * TF-2.16 test modification and handling
+ * fix test fail
+ * Add troubleshooting links to exceptions
+ * cross account private hub model fine-tuning
+ * chore: cleanup jumpstart factory
+ * disable failing integration tests
+
 ## v2.229.0 (2024-08-15)
 
 ### Features

VERSION

@@ -1 +1 @@
-2.229.1.dev0
+2.231.1.dev0

pyproject.toml

@@ -47,6 +47,7 @@ dependencies = [
   "psutil",
   "PyYAML~=6.0",
   "requests",
+  "sagemaker-core>=1.0.0,<2.0.0",
   "schema",
   "smdebug_rulesconfig==1.0.1",
   "tblib>=1.7.0,<4",

requirements/extras/huggingface_requirements.txt

@@ -1,2 +1,5 @@
 accelerate>=0.24.1,<=0.27.0
 sagemaker_schema_inference_artifacts>=0.0.5
+uvicorn>=0.30.1
+fastapi>=0.111.0
+nest-asyncio

requirements/extras/test_requirements.txt

@@ -1,4 +1,5 @@
 tox==3.24.5
+numpy>=1.24.0
 build[virtualenv]==1.2.1
 flake8==4.0.1
 pytest==6.2.5
@@ -40,3 +41,6 @@ schema==0.7.5
 tensorflow>=2.1,<=2.16
 mlflow>=2.12.2,<2.13
 huggingface_hub>=0.23.4
+uvicorn>=0.30.1
+fastapi>=0.111.0
+nest-asyncio

src/sagemaker/algorithm.py

@@ -157,6 +157,20 @@ def __init__(
                 available (default: ``None``).
             **kwargs: Additional kwargs. This is unused. It's only added for AlgorithmEstimator
                 to ignore the irrelevant arguments.
+
+        Raises:
+            ValueError:
+            - If an AWS IAM Role is not provided.
+            - Bad value for instance type.
+            RuntimeError:
+            - When setting up custom VPC, both subnets and security_group_ids are not provided
+            - If instance_count > 1 (distributed training) with instance type local or local gpu
+            - If LocalSession is not used with instance type local or local gpu
+            - file:// output path used outside of local mode
+            botocore.exceptions.ClientError:
+            - algorithm arn is incorrect
+            - insufficient permission to access/ describe algorithm
+            - algorithm is in a different region
         """
         self.algorithm_arn = algorithm_arn
         super(AlgorithmEstimator, self).__init__(

src/sagemaker/base_predictor.py

@@ -430,6 +430,8 @@ def update_endpoint(
                 - If ``initial_instance_count``, ``instance_type``, or ``accelerator_type`` is
                   specified and either ``model_name`` is ``None`` or there are multiple models
                   associated with the endpoint.
+            botocore.exceptions.ClientError: If SageMaker throws an error while creating
+            endpoint config, describing endpoint or updating endpoint
         """
         production_variants = None
         current_model_names = self._get_model_names()

src/sagemaker/environment_variables.py

@@ -20,7 +20,7 @@
 from sagemaker.jumpstart import utils as jumpstart_utils
 from sagemaker.jumpstart import artifacts
 from sagemaker.jumpstart.constants import DEFAULT_JUMPSTART_SAGEMAKER_SESSION
-from sagemaker.jumpstart.enums import JumpStartScriptScope
+from sagemaker.jumpstart.enums import JumpStartModelType, JumpStartScriptScope
 from sagemaker.session import Session
 
 logger = logging.getLogger(__name__)
@@ -38,6 +38,7 @@ def retrieve_default(
     instance_type: Optional[str] = None,
     script: JumpStartScriptScope = JumpStartScriptScope.INFERENCE,
     config_name: Optional[str] = None,
+    model_type: JumpStartModelType = JumpStartModelType.OPEN_WEIGHTS,
 ) -> Dict[str, str]:
     """Retrieves the default container environment variables for the model matching the arguments.
 
@@ -70,6 +71,8 @@ def retrieve_default(
         script (JumpStartScriptScope): The JumpStart script for which to retrieve environment
             variables.
         config_name (Optional[str]): Name of the JumpStart Model config to apply. (Default: None).
+        model_type (JumpStartModelType): The type of the model, can be open weights model
+            or proprietary model. (Default: JumpStartModelType.OPEN_WEIGHTS).
     Returns:
         dict: The variables to use for the model.
 
@@ -94,4 +97,5 @@ def retrieve_default(
         instance_type=instance_type,
         script=script,
         config_name=config_name,
+        model_type=model_type,
     )

src/sagemaker/estimator.py

@@ -590,25 +590,36 @@ def __init__(
         self.dependencies = dependencies or []
         self.uploaded_code: Optional[UploadedCode] = None
 
-        # Check that the user properly sets both subnet and secutiry_groupe_ids
+        # Check that the user properly sets both subnet and security_group_ids
         if (
             subnets is not None
             and security_group_ids is None
             or security_group_ids is not None
             and subnets is None
         ):
+            troubleshooting = (
+                "Refer to this documentation on using custom VPC: "
+                "https://sagemaker.readthedocs.io/en/v2.24.0/overview.html"
+                "#secure-training-and-inference-with-vpc"
+            )
+            logger.error("Check troubleshooting guide for common errors: %s", troubleshooting)
+
             raise RuntimeError(
                 "When setting up custom VPC, both subnets and security_group_ids must be set"
             )
 
         if self.instance_type in ("local", "local_gpu"):
             if self.instance_type == "local_gpu" and self.instance_count > 1:
-                raise RuntimeError("Distributed Training in Local GPU is not supported")
+                raise RuntimeError(
+                    "Distributed Training in Local GPU is not supported."
+                    " Set instance_count to 1."
+                )
             self.sagemaker_session = sagemaker_session or LocalSession()
             if not isinstance(self.sagemaker_session, sagemaker.local.LocalSession):
                 raise RuntimeError(
                     "instance_type local or local_gpu is only supported with an"
-                    "instance of LocalSession"
+                    "instance of LocalSession. More details on local mode: "
+                    "https://sagemaker.readthedocs.io/en/stable/overview.html#local-mode"
                 )
         else:
             self.sagemaker_session = sagemaker_session or Session()
@@ -631,7 +642,11 @@ def __init__(
             and not is_pipeline_variable(output_path)
             and output_path.startswith("file://")
         ):
-            raise RuntimeError("file:// output paths are only supported in Local Mode")
+            raise RuntimeError(
+                "The 'file://' output paths are only supported when using Local Mode. "
+                "To resolve this issue, ensure you're running in Local Mode with a LocalSession, "
+                "or use an 's3://' output path for jobs running on SageMaker instances."
+            )
         self.output_path = output_path
         self.latest_training_job = None
         self.jobs = []
@@ -646,7 +661,12 @@ def __init__(
             # Now we marked that as Optional because we can fetch it from SageMakerConfig
             # Because of marking that parameter as optional, we should validate if it is None, even
             # after fetching the config.
-            raise ValueError("An AWS IAM role is required to create an estimator.")
+            raise ValueError(
+                "An AWS IAM role is required to create an estimator. "
+                "Please provide a valid `role` argument with the ARN of an IAM role"
+                " that has the necessary SageMaker permissions."
+            )
+
         self.output_kms_key = resolve_value_from_config(
             output_kms_key, TRAINING_JOB_KMS_KEY_ID_PATH, sagemaker_session=self.sagemaker_session
         )
@@ -1855,6 +1875,8 @@ def model_data(self):
             if compression_type not in {"GZIP", "NONE"}:
                 raise ValueError(
                     f'Unrecognized training job output data compression type "{compression_type}"'
+                    '. Please specify either "GZIP" or "NONE" as valid options for '
+                    "the compression type."
                 )
             # model data is in uncompressed form NOTE SageMaker Hosting mandates presence of
             # trailing forward slash in S3 model data URI, so append one if necessary.