Merge branch 'master' into fix/skip-none-hyperparameters

Author: Harsh270519

sagemaker-train/src/sagemaker/train/common_utils/finetune_utils.py

@@ -15,6 +15,7 @@
 from sagemaker.core.shapes import ServerlessJobConfig, Channel, DataSource, ModelPackageConfig, MlflowConfig
 from sagemaker.train.configs import InputData, OutputDataConfig
 from sagemaker.train.defaults import TrainDefaults
+from sagemaker.train.constants import get_sagemaker_hub_name
 
 logger = logging.getLogger(__name__)
 
@@ -317,13 +318,15 @@ def _resolve_model_package_arn(model_package) -> Optional[str]:
 
 
 def _get_fine_tuning_options_and_model_arn(model_name: str, customization_technique: str, training_type, sagemaker_session,
-                                         hub_name: str = "SageMakerPublicHub") -> tuple:
+                                         hub_name: Optional[str] = None) -> tuple:
     """Get fine-tuning options and model ARN for given customization technique.
     
     Returns:
         tuple: (FineTuningOptions, model_arn, is_gated_model)
     """
-    
+    if hub_name is None:
+        hub_name = get_sagemaker_hub_name()
+
     try:
 
         hub_content = _get_hub_content_metadata(

sagemaker-train/src/sagemaker/train/common_utils/model_resolution.py

@@ -14,6 +14,7 @@
 from enum import Enum
 import re
 from sagemaker.train.base_trainer import BaseTrainer
+from sagemaker.train.constants import get_sagemaker_hub_name
 from sagemaker.core.utils.utils import Unassigned
 
 
@@ -52,8 +53,6 @@ class _ModelResolver:
     and fine-tuned ModelPackage objects/ARNs.
     """
 
-    DEFAULT_HUB_NAME = "SageMakerPublicHub"
-    
     def __init__(self, sagemaker_session=None):
         """
         Initialize the resolver.
@@ -89,7 +88,7 @@ def resolve_model_info(
             if base_model.startswith("arn:aws:sagemaker:") and ":model-package/" in base_model:
                 return self._resolve_model_package_arn(base_model)
             else:
-                return self._resolve_jumpstart_model(base_model, hub_name or self.DEFAULT_HUB_NAME)
+                return self._resolve_jumpstart_model(base_model, hub_name or get_sagemaker_hub_name())
         # Handle BaseTrainer type
         elif isinstance(base_model, BaseTrainer):
             if hasattr(base_model, '_latest_training_job') and hasattr(base_model._latest_training_job,

sagemaker-train/src/sagemaker/train/constants.py

@@ -40,7 +40,13 @@
     + f"&& {SM_DRIVERS_CONTAINER_PATH}/{TRAIN_SCRIPT}",
 ]
 
-HUB_NAME = "SageMakerPublicHub"
+def get_sagemaker_hub_name() -> str:
+    """Return the SageMaker Hub name, honoring SAGEMAKER_HUB_NAME env var override.
+
+    Resolved at call time so tests and dev workflows can override the hub
+    without re-importing this module. Defaults to ``"SageMakerPublicHub"``.
+    """
+    return os.environ.get("SAGEMAKER_HUB_NAME", "SageMakerPublicHub")
 
 # Allowed reward model IDs for RLAIF trainer with region restrictions
 _ALLOWED_REWARD_MODEL_IDS = {

sagemaker-train/src/sagemaker/train/dpo_trainer.py

@@ -23,7 +23,7 @@
 )
 from sagemaker.core.telemetry.telemetry_logging import _telemetry_emitter
 from sagemaker.core.telemetry.constants import Feature
-from sagemaker.train.constants import HUB_NAME
+from sagemaker.train.constants import get_sagemaker_hub_name
 
 logger = logging.getLogger(__name__)
 logger.setLevel(logging.INFO)
@@ -244,7 +244,7 @@ def train(self,
         )
 
         vpc_config = self.networking if self.networking else None
-        tags = _get_studio_tags(self._model_name, HUB_NAME)
+        tags = _get_studio_tags(self._model_name, get_sagemaker_hub_name())
 
         # Build TrainingJob.create() arguments
         create_args = {

sagemaker-train/src/sagemaker/train/evaluate/benchmark_evaluator.py

@@ -21,6 +21,7 @@
 from .execution import EvaluationPipelineExecution
 from sagemaker.core.telemetry.telemetry_logging import _telemetry_emitter
 from sagemaker.core.telemetry.constants import Feature
+from sagemaker.train.constants import get_sagemaker_hub_name
 
 _logger = logging.getLogger(__name__)
 
@@ -466,7 +467,7 @@ def hyperparameters(self):
 
             override_params = _get_evaluation_override_params(
                 hub_content_name=hub_content_name,
-                hub_name="SageMakerPublicHub",
+                hub_name=get_sagemaker_hub_name(),
                 evaluation_type=evaluation_type,
                 region=region,
                 session=boto_session

sagemaker-train/src/sagemaker/train/evaluate/custom_scorer_evaluator.py

@@ -16,6 +16,7 @@
 from .execution import EvaluationPipelineExecution
 from sagemaker.core.telemetry.telemetry_logging import _telemetry_emitter
 from sagemaker.core.telemetry.constants import Feature
+from sagemaker.train.constants import get_sagemaker_hub_name
 
 _logger = logging.getLogger(__name__)
 
@@ -240,7 +241,7 @@ def hyperparameters(self):
 
             override_params = _get_evaluation_override_params(
                 hub_content_name=hub_content_name,
-                hub_name="SageMakerPublicHub",
+                hub_name=get_sagemaker_hub_name(),
                 evaluation_type="DeterministicEvaluation",
                 region=region,
                 session=boto_session
@@ -365,7 +366,7 @@ def _get_inference_params_from_hub(self, region: str) -> dict:
             _logger.info(f"Fetching evaluation recipe override parameters from hub for model: {hub_content_name}")
             override_params = _get_evaluation_override_params(
                 hub_content_name=hub_content_name,
-                hub_name="SageMakerPublicHub",
+                hub_name=get_sagemaker_hub_name(),
                 evaluation_type="DeterministicEvaluation",
                 region=region,
                 session=session

sagemaker-train/src/sagemaker/train/rlaif_trainer.py

@@ -27,7 +27,7 @@
 )
 from sagemaker.core.telemetry.telemetry_logging import _telemetry_emitter
 from sagemaker.core.telemetry.constants import Feature
-from sagemaker.train.constants import HUB_NAME, _ALLOWED_REWARD_MODEL_IDS
+from sagemaker.train.constants import get_sagemaker_hub_name, _ALLOWED_REWARD_MODEL_IDS
 
 logger = logging.getLogger(__name__)
 
@@ -263,7 +263,7 @@ def train(self, training_dataset: Optional[Union[str, DataSet]] = None, validati
         )
 
         vpc_config = self.networking if self.networking else None
-        tags = _get_studio_tags(self._model_name, HUB_NAME)
+        tags = _get_studio_tags(self._model_name, get_sagemaker_hub_name())
 
         # Build TrainingJob.create() arguments
         create_args = {
@@ -358,7 +358,7 @@ def _process_non_builtin_reward_prompt(self):
             sagemaker_session=self.sagemaker_session
         )
                 hub_content = _get_hub_content_metadata(
-                    hub_name=HUB_NAME,
+                    hub_name=get_sagemaker_hub_name(),
                     hub_content_type="JsonDoc",
                     hub_content_name=self.reward_prompt,
                     session=session.boto_session,

sagemaker-train/src/sagemaker/train/rlvr_trainer.py

@@ -25,7 +25,7 @@
 )
 from sagemaker.core.telemetry.telemetry_logging import _telemetry_emitter
 from sagemaker.core.telemetry.constants import Feature
-from sagemaker.train.constants import HUB_NAME
+from sagemaker.train.constants import get_sagemaker_hub_name
 
 logger = logging.getLogger(__name__)
 
@@ -251,7 +251,7 @@ def train(self, training_dataset: Optional[Union[str, DataSet]] = None,
         )
 
         vpc_config = self.networking if self.networking else None
-        tags = _get_studio_tags(self._model_name, HUB_NAME)
+        tags = _get_studio_tags(self._model_name, get_sagemaker_hub_name())
 
         # Build TrainingJob.create() arguments
         create_args = {

sagemaker-train/src/sagemaker/train/sft_trainer.py

@@ -24,7 +24,7 @@
 )
 from sagemaker.core.telemetry.telemetry_logging import _telemetry_emitter
 from sagemaker.core.telemetry.constants import Feature
-from sagemaker.train.constants import HUB_NAME
+from sagemaker.train.constants import get_sagemaker_hub_name
 
 logger = logging.getLogger(__name__)
 logger.setLevel(logging.INFO)
@@ -245,7 +245,7 @@ def train(self, training_dataset: Optional[Union[str, DataSet]] = None, validati
         )
 
         vpc_config = self.networking if self.networking else None
-        tags = _get_studio_tags(self._model_name, HUB_NAME)
+        tags = _get_studio_tags(self._model_name, get_sagemaker_hub_name())
 
         # Build TrainingJob.create() arguments
         create_args = {

sagemaker-train/tests/unit/train/common_utils/test_model_resolution.py

@@ -63,7 +63,6 @@ def test_resolver_initialization(self):
         """Test ModelResolver initialization."""
         resolver = _ModelResolver()
         assert resolver.sagemaker_session is None
-        assert resolver.DEFAULT_HUB_NAME == "SageMakerPublicHub"
 
     def test_resolver_with_session(self):
         """Test ModelResolver with custom session."""