Merge branch 'master' into hf-tgi

nargokul · web-flow · commit ef48c2bad81c · 2025-10-17T10:59:55.000-07:00
diff --git a/CHANGELOG.md b/CHANGELOG.md
@@ -1,5 +1,30 @@
 # Changelog
 
+## v2.253.1 (2025-10-14)
+
+### Bug Fixes and Other Changes
+
+ * Update instance type regex to also include hyphens
+ * Revert the change "Add Numpy 2.0 support"
+ * [hf-tei] add image uri to utils
+ * add TEI 1.8.2
+
+## v2.253.0 (2025-10-10)
+
+### Features
+
+ * Added condition to allow eval recipe.
+ * add model_type hyperparameter support for Nova recipes
+
+### Bug Fixes and Other Changes
+
+ * Fix for a failed slow test: numpy fix
+ * Add numpy 2.0 support
+ * chore: domain support for eu-isoe-west-1
+ * Adding default identity implementations to InferenceSpec
+ * djl regions fixes #5273
+ * Fix flaky integ test
+
 ## v2.252.0 (2025-09-29)
 
 ### Features
diff --git a/VERSION b/VERSION
@@ -1 +1 @@
-2.252.1.dev0
+2.253.2.dev0
diff --git a/src/sagemaker/estimator.py b/src/sagemaker/estimator.py
@@ -2119,7 +2119,7 @@ def _get_instance_type(self):
             instance_type = instance_group.instance_type
             if is_pipeline_variable(instance_type):
                 continue
-            match = re.match(r"^ml[\._]([a-z\d]+)\.?\w*$", instance_type)
+            match = re.match(r"^ml[\._]([a-z\d\-]+)\.?\w*$", instance_type)
 
             if match:
                 family = match[1]
diff --git a/src/sagemaker/fw_utils.py b/src/sagemaker/fw_utils.py
@@ -962,7 +962,7 @@ def validate_distribution_for_instance_type(instance_type, distribution):
     """
     err_msg = ""
     if isinstance(instance_type, str):
-        match = re.match(r"^ml[\._]([a-z\d]+)\.?\w*$", instance_type)
+        match = re.match(r"^ml[\._]([a-z\d\-]+)\.?\w*$", instance_type)
         if match and match[1].startswith("trn"):
             keys = list(distribution.keys())
             if len(keys) == 0:
@@ -1083,7 +1083,7 @@ def _is_gpu_instance(instance_type):
         bool: Whether or not the instance_type supports GPU
     """
     if isinstance(instance_type, str):
-        match = re.match(r"^ml[\._]([a-z\d]+)\.?\w*$", instance_type)
+        match = re.match(r"^ml[\._]([a-z\d\-]+)\.?\w*$", instance_type)
         if match:
             if match[1].startswith("p") or match[1].startswith("g"):
                 return True
@@ -1102,7 +1102,7 @@ def _is_trainium_instance(instance_type):
         bool: Whether or not the instance_type is a Trainium instance
     """
     if isinstance(instance_type, str):
-        match = re.match(r"^ml[\._]([a-z\d]+)\.?\w*$", instance_type)
+        match = re.match(r"^ml[\._]([a-z\d\-]+)\.?\w*$", instance_type)
         if match and match[1].startswith("trn"):
             return True
     return False
@@ -1149,7 +1149,7 @@ def _instance_type_supports_profiler(instance_type):
         bool: Whether or not the region supports Amazon SageMaker Debugger profiling feature.
     """
     if isinstance(instance_type, str):
-        match = re.match(r"^ml[\._]([a-z\d]+)\.?\w*$", instance_type)
+        match = re.match(r"^ml[\._]([a-z\d\-]+)\.?\w*$", instance_type)
         if match and match[1].startswith("trn"):
             return True
     return False
diff --git a/src/sagemaker/image_uri_config/huggingface-tei-cpu.json b/src/sagemaker/image_uri_config/huggingface-tei-cpu.json
@@ -7,7 +7,8 @@
             "1.2": "1.2.3",
             "1.4": "1.4.0",
             "1.6": "1.6.0",
-            "1.7": "1.7.0"
+            "1.7": "1.7.0",
+            "1.8": "1.8.2"
         },
         "versions": {
             "1.2.3": {
@@ -244,7 +245,54 @@
                 "container_version": {
                     "cpu": "ubuntu22.04"
                 }
+            },
+            "1.8.2":{
+                "py_versions": [
+                    "py310"
+                ],
+                "registries": {
+                    "af-south-1": "510948584623",
+                    "ap-east-1": "651117190479",
+                    "ap-northeast-1": "354813040037",
+                    "ap-northeast-2": "366743142698",
+                    "ap-northeast-3": "867004704886",
+                    "ap-south-1": "720646828776",
+                    "ap-south-2": "628508329040",
+                    "ap-southeast-1": "121021644041",
+                    "ap-southeast-2": "783357654285",
+                    "ap-southeast-3": "951798379941",
+                    "ap-southeast-4": "106583098589",
+                    "ca-central-1": "341280168497",
+                    "ca-west-1": "190319476487",
+                    "cn-north-1": "450853457545",
+                    "cn-northwest-1": "451049120500",
+                    "eu-central-1": "492215442770",
+                    "eu-central-2": "680994064768",
+                    "eu-north-1": "662702820516",
+                    "eu-south-1": "978288397137",
+                    "eu-south-2": "104374241257",
+                    "eu-west-1": "141502667606",
+                    "eu-west-2": "764974769150",
+                    "eu-west-3": "659782779980",
+                    "il-central-1": "898809789911",
+                    "me-central-1": "272398656194",
+                    "me-south-1": "801668240914",
+                    "sa-east-1": "737474898029",
+                    "us-east-1": "683313688378",
+                    "us-east-2": "257758044811",
+                    "us-gov-east-1": "237065988967",
+                    "us-gov-west-1": "414596584902",
+                    "us-iso-east-1": "833128469047",
+                    "us-isob-east-1": "281123927165",
+                    "us-west-1": "746614075791",
+                    "us-west-2": "246618743249"
+                },
+                "tag_prefix": "2.0.1-tei1.8.2",
+                "repository": "tei-cpu",
+                "container_version": {
+                    "cpu": "ubuntu22.04"
+                }
             }
         }
     }
-}
+}
diff --git a/src/sagemaker/image_uri_config/huggingface-tei.json b/src/sagemaker/image_uri_config/huggingface-tei.json
@@ -7,7 +7,8 @@
             "1.2": "1.2.3",
             "1.4": "1.4.0",
             "1.6": "1.6.0",
-            "1.7": "1.7.0"
+            "1.7": "1.7.0",
+            "1.8": "1.8.2"
         },
         "versions": {
             "1.2.3": {
@@ -244,7 +245,54 @@
                 "container_version": {
                     "gpu": "cu122-ubuntu22.04"
                 }
+            },
+            "1.8.2": {
+                "py_versions": [
+                    "py310"
+                ],
+                "registries": {
+                    "af-south-1": "510948584623",
+                    "ap-east-1": "651117190479",
+                    "ap-northeast-1": "354813040037",
+                    "ap-northeast-2": "366743142698",
+                    "ap-northeast-3": "867004704886",
+                    "ap-south-1": "720646828776",
+                    "ap-south-2": "628508329040",
+                    "ap-southeast-1": "121021644041",
+                    "ap-southeast-2": "783357654285",
+                    "ap-southeast-3": "951798379941",
+                    "ap-southeast-4": "106583098589",
+                    "ca-central-1": "341280168497",
+                    "ca-west-1": "190319476487",
+                    "cn-north-1": "450853457545",
+                    "cn-northwest-1": "451049120500",
+                    "eu-central-1": "492215442770",
+                    "eu-central-2": "680994064768",
+                    "eu-north-1": "662702820516",
+                    "eu-south-1": "978288397137",
+                    "eu-south-2": "104374241257",
+                    "eu-west-1": "141502667606",
+                    "eu-west-2": "764974769150",
+                    "eu-west-3": "659782779980",
+                    "il-central-1": "898809789911",
+                    "me-central-1": "272398656194",
+                    "me-south-1": "801668240914",
+                    "sa-east-1": "737474898029",
+                    "us-east-1": "683313688378",
+                    "us-east-2": "257758044811",
+                    "us-gov-east-1": "237065988967",
+                    "us-gov-west-1": "414596584902",
+                    "us-iso-east-1": "833128469047",
+                    "us-isob-east-1": "281123927165",
+                    "us-west-1": "746614075791",
+                    "us-west-2": "246618743249"
+                },
+                "tag_prefix": "2.0.1-tei1.8.2",
+                "repository": "tei",
+                "container_version": {
+                    "gpu": "cu122-ubuntu22.04"
+                }
             }
         }
     }
-}
+}
diff --git a/src/sagemaker/serve/utils/optimize_utils.py b/src/sagemaker/serve/utils/optimize_utils.py
@@ -38,7 +38,7 @@ def _is_inferentia_or_trainium(instance_type: Optional[str]) -> bool:
         bool: Whether the given instance type is Inferentia or Trainium.
     """
     if isinstance(instance_type, str):
-        match = re.match(r"^ml[\._]([a-z\d]+)\.?\w*$", instance_type)
+        match = re.match(r"^ml[\._]([a-z\d\-]+)\.?\w*$", instance_type)
         if match:
             if match[1].startswith("inf") or match[1].startswith("trn"):
                 return True
diff --git a/src/sagemaker/utils.py b/src/sagemaker/utils.py
@@ -1529,7 +1529,7 @@ def get_instance_type_family(instance_type: str) -> str:
     """
     instance_type_family = ""
     if isinstance(instance_type, str):
-        match = re.match(r"^ml[\._]([a-z\d]+)\.?\w*$", instance_type)
+        match = re.match(r"^ml[\._]([a-z\d\-]+)\.?\w*$", instance_type)
         if match is not None:
             instance_type_family = match[1]
     return instance_type_family
diff --git a/tests/data/serve_resources/mlflow/pytorch/MLmodel b/tests/data/serve_resources/mlflow/pytorch/MLmodel
@@ -14,7 +14,7 @@ flavors:
     code: null
     model_data: data
     pytorch_version: 2.0.1+cu117
-mlflow_version: 2.10.2
+mlflow_version: 2.11.1
 model_size_bytes: 4971001
 model_uuid: 2d85043bbf504b1e9950e124c46a1719
 run_id: 98b8d2e2c0e74ab59f4c26f7cb3de233
diff --git a/tests/data/serve_resources/mlflow/tensorflow/MLmodel b/tests/data/serve_resources/mlflow/tensorflow/MLmodel
@@ -10,7 +10,7 @@ flavors:
     code: null
     model_type: tf2-module
     saved_model_dir: tf2model
-mlflow_version: 2.11.1
+mlflow_version: 2.20.3
 model_size_bytes: 23823
 model_uuid: 40d2323944294fce898d8693455f60e8
 run_id: 592132312fb84935b201de2c027c54c6
diff --git a/tests/integ/sagemaker/serve/test_serve_mlflow_xgboost_flavor_happy.py b/tests/integ/sagemaker/serve/test_serve_mlflow_xgboost_flavor_happy.py
@@ -28,7 +28,7 @@
     XGBOOST_MLFLOW_RESOURCE_DIR,
     SERVE_SAGEMAKER_ENDPOINT_TIMEOUT,
     # SERVE_LOCAL_CONTAINER_TIMEOUT,
-    PYTHON_VERSION_IS_NOT_310,
+    # PYTHON_VERSION_IS_NOT_310,
 )
 from tests.integ.timeout import timeout
 from tests.integ.utils import cleanup_model_resources
@@ -147,16 +147,17 @@ def model_builder(request):
 #                 ), f"{caught_ex} was thrown when running pytorch squeezenet local container test"
 
 
-@pytest.mark.skipif(
-    PYTHON_VERSION_IS_NOT_310,  # or NOT_RUNNING_ON_INF_EXP_DEV_PIPELINE,
-    reason="The goal of these test are to test the serving components of our feature",
+@pytest.mark.skip(
+    reason="Skipping it temporarily as we have bug with latest version of XGBoost image \
+           that is numpy 2.0 compatible.",
 )
 def test_happy_xgboost_sagemaker_endpoint_with_torch_serve(
     sagemaker_session,
     xgboost_schema,
     cpu_instance_type,
     test_data,
 ):
+    # TODO: Enable this test once the issue with latest XGBoost image is fixed.
     logger.info("Running in SAGEMAKER_ENDPOINT mode...")
     caught_ex = None
 
diff --git a/tests/unit/sagemaker/image_uris/test_huggingface_llm.py b/tests/unit/sagemaker/image_uris/test_huggingface_llm.py
@@ -26,13 +26,15 @@
         "1.6.0": "2.0.1-tei1.6.0-gpu-py310-cu122-ubuntu22.04",
         "1.7.0": "2.0.1-tei1.7.0-gpu-py310-cu122-ubuntu22.04",
         "1.8.0": "2.0.1-tei1.8.0-gpu-py310-cu122-ubuntu22.04",
+        "1.8.2": "2.0.1-tei1.8.2-gpu-py310-cu122-ubuntu22.04",
     },
     "cpu": {
         "1.2.3": "2.0.1-tei1.2.3-cpu-py310-ubuntu22.04",
         "1.4.0": "2.0.1-tei1.4.0-cpu-py310-ubuntu22.04",
         "1.6.0": "2.0.1-tei1.6.0-cpu-py310-ubuntu22.04",
         "1.7.0": "2.0.1-tei1.7.0-cpu-py310-ubuntu22.04",
         "1.8.0": "2.0.1-tei1.8.0-cpu-py310-ubuntu22.04",
+        "1.8.2": "2.0.1-tei1.8.2-cpu-py310-ubuntu22.04",
     },
 }
 HF_VERSIONS_MAPPING = {
diff --git a/tests/unit/sagemaker/serve/utils/test_optimize_utils.py b/tests/unit/sagemaker/serve/utils/test_optimize_utils.py
@@ -95,6 +95,8 @@
     [
         ("ml.trn1.2xlarge", True),
         ("ml.inf2.xlarge", True),
+        ("ml.trn1-n.2xlarge", True),
+        ("ml.inf2-b.xlarge", True),
         ("ml.c7gd.4xlarge", False),
     ],
 )
diff --git a/tests/unit/test_estimator.py b/tests/unit/test_estimator.py
@@ -2246,6 +2246,21 @@ def test_get_instance_type_gpu(sagemaker_session):
     assert "ml.p3.16xlarge" == estimator._get_instance_type()
 
 
+def test_get_instance_type_gpu_with_hyphens(sagemaker_session):
+    estimator = Estimator(
+        image_uri="some-image",
+        role="some_image",
+        instance_groups=[
+            InstanceGroup("group1", "ml.c4.xlarge", 1),
+            InstanceGroup("group2", "ml.p6-b200.48xlarge", 2),
+        ],
+        sagemaker_session=sagemaker_session,
+        base_job_name="base_job_name",
+    )
+
+    assert "ml.p6-b200.48xlarge" == estimator._get_instance_type()
+
+
 def test_estimator_with_output_compression_disabled(sagemaker_session):
     estimator = Estimator(
         image_uri="some-image",
diff --git a/tests/unit/test_fw_utils.py b/tests/unit/test_fw_utils.py
@@ -1065,6 +1065,13 @@ def test_validate_unsupported_distributions_trainium_raises():
             instance_type="ml.trn1.32xlarge",
         )
 
+    with pytest.raises(ValueError):
+        mpi_enabled = {"mpi": {"enabled": True}}
+        fw_utils.validate_distribution_for_instance_type(
+            distribution=mpi_enabled,
+            instance_type="ml.trn1-n.2xlarge",
+        )
+
     with pytest.raises(ValueError):
         pytorch_ddp_enabled = {"pytorch_ddp": {"enabled": True}}
         fw_utils.validate_distribution_for_instance_type(
@@ -1082,6 +1089,7 @@ def test_validate_unsupported_distributions_trainium_raises():
 
 def test_instance_type_supports_profiler():
     assert fw_utils._instance_type_supports_profiler("ml.trn1.xlarge") is True
+    assert fw_utils._instance_type_supports_profiler("ml.trn1-n.xlarge") is True
     assert fw_utils._instance_type_supports_profiler("ml.m4.xlarge") is False
     assert fw_utils._instance_type_supports_profiler("local") is False
 
@@ -1097,6 +1105,8 @@ def test_is_gpu_instance():
         "ml.g4dn.xlarge",
         "ml.g5.xlarge",
         "ml.g5.48xlarge",
+        "ml.p6-b200.48xlarge",
+        "ml.g6e-12xlarge.xlarge",
         "local_gpu",
     ]
     non_gpu_instance_types = [
@@ -1116,6 +1126,7 @@ def test_is_trainium_instance():
     trainium_instance_types = [
         "ml.trn1.2xlarge",
         "ml.trn1.32xlarge",
+        "ml.trn1-n.2xlarge",
     ]
     non_trainum_instance_types = [
         "ml.t3.xlarge",
diff --git a/tests/unit/test_utils.py b/tests/unit/test_utils.py
@@ -1844,6 +1844,7 @@ def test_instance_family_from_full_instance_type(self):
             "ml.afbsadjfbasfb.sdkjfnsa": "afbsadjfbasfb",
             "ml_fdsfsdf.xlarge": "fdsfsdf",
             "ml_c2.4xlarge": "c2",
+            "ml.p6-b200.48xlarge": "p6-b200",
             "sdfasfdda": "",
             "local": "",
             "c2.xlarge": "",
