Fix VolumeSizeInGB missing from v3 deploy for models with inference_v… (#5847)

* Fix VolumeSizeInGB missing from v3 deploy for models with inference_volume_size

from_jumpstart_config() retrieves deploy kwargs (which includes volume_size)
but never persisted it on the ModelBuilder instance. Then _deploy_core_endpoint
read volume_size from kwargs with no self.* fallback, so it was always None.

Two-line fix in model_builder.py:
1. from_jumpstart_config(): persist volume_size from deploy_kwargs (same as
   model_data_download_timeout and container_startup_health_check_timeout)
2. _deploy_core_endpoint(): fall back to self.volume_size when not in kwargs

Same pattern as EnableNetworkIsolation fix in v3.10.1.

* Fix integ test: add accept_eula=True for Llama model

---------

Co-authored-by: Liam Neal Reilly <lnealrei@amazon.com>

Author: lhnealreilly

sagemaker-serve/src/sagemaker/serve/model_builder.py

@@ -2774,7 +2774,7 @@ def _deploy_core_endpoint(self, **kwargs):
             self._deserializer = deserializer
 
         data_capture_config = kwargs.get("data_capture_config", None)
-        volume_size = kwargs.get("volume_size", None)
+        volume_size = kwargs.get("volume_size", getattr(self, "volume_size", None))
         inference_recommendation_id = kwargs.get("inference_recommendation_id", None)
         explainer_config = kwargs.get("explainer_config", None)
         endpoint_logging = kwargs.get("endpoint_logging", False)
@@ -3620,6 +3620,7 @@ def from_jumpstart_config(
             "container_startup_health_check_timeout"
         )
         mb_instance.inference_ami_version = deploy_kwargs.get("inference_ami_version")
+        mb_instance.volume_size = deploy_kwargs.get("volume_size")
 
         # Apply network isolation from JumpStart model spec if not set by user via network param
         if not mb_instance._enable_network_isolation and deploy_kwargs.get(

…nteg/test_jumpstart_network_isolation.py …ts/integ/test_jumpstart_deploy_parity.pysagemaker-serve/tests/integ/test_jumpstart_network_isolation.py renamed to sagemaker-serve/tests/integ/test_jumpstart_deploy_parity.py sagemaker-serve/tests/integ/test_jumpstart_network_isolation.py renamed to sagemaker-serve/tests/integ/test_jumpstart_deploy_parity.py

@@ -65,3 +65,53 @@ def test_jumpstart_build_enables_network_isolation():
     finally:
         core_model.delete()
         logger.info("Model deleted.")
+
+
+VOLUME_SIZE_MODEL_ID = "meta-textgenerationneuron-llama-2-7b"
+VOLUME_SIZE_INSTANCE_TYPE = "ml.inf2.xlarge"
+
+
+@pytest.mark.slow_test
+def test_jumpstart_build_sets_volume_size():
+    """Integration test verifying volume_size from model specs is propagated.
+
+    JumpStart model specs define inference_volume_size for models that need
+    large EBS volumes for model weights. This test validates that ModelBuilder
+    propagates volume_size through both from_jumpstart_config() and build() paths,
+    matching v2 behavior where VolumeSizeInGB appears in CreateEndpointConfig.
+    """
+    logger.info("Starting JumpStart volume_size integration test...")
+
+    # Test from_jumpstart_config path
+    compute = Compute(instance_type=VOLUME_SIZE_INSTANCE_TYPE)
+    jumpstart_config = JumpStartConfig(model_id=VOLUME_SIZE_MODEL_ID, accept_eula=True)
+    model_builder = ModelBuilder.from_jumpstart_config(
+        jumpstart_config=jumpstart_config, compute=compute
+    )
+
+    assert getattr(model_builder, "volume_size", None) is not None, (
+        f"ModelBuilder.volume_size should be set after from_jumpstart_config() "
+        f"for model {VOLUME_SIZE_MODEL_ID} on {VOLUME_SIZE_INSTANCE_TYPE}, got None"
+    )
+    logger.info(f"from_jumpstart_config set volume_size={model_builder.volume_size}")
+
+    # Test build path (also sets volume_size via _build_for_jumpstart)
+    unique_id = str(uuid.uuid4())[:8]
+    core_model = model_builder.build(model_name=f"js-volsize-test-{unique_id}")
+    logger.info(f"Model created: {core_model.model_name}")
+
+    try:
+        assert getattr(model_builder, "volume_size", None) is not None, (
+            f"ModelBuilder.volume_size should persist after build() "
+            f"for model {VOLUME_SIZE_MODEL_ID}, got None"
+        )
+        assert model_builder.volume_size >= 256, (
+            f"volume_size should be >= 256, "
+            f"got {model_builder.volume_size}"
+        )
+        logger.info(
+            f"✅ volume_size={model_builder.volume_size} correctly set"
+        )
+    finally:
+        core_model.delete()
+        logger.info("Model deleted.")

sagemaker-serve/tests/unit/test_model_builder_coverage_boost.py

@@ -12,6 +12,9 @@
 from sagemaker.serve.mode.function_pointers import Mode
 from sagemaker.serve.utils.types import ModelServer
 from sagemaker.core.training.configs import Compute, Networking
+from sagemaker.core.jumpstart.configs import JumpStartConfig
+from sagemaker.core.inference_config import AsyncInferenceConfig
+from botocore.exceptions import ClientError
 
 
 class TestModelBuilderInit(unittest.TestCase):
@@ -189,7 +192,6 @@ class TestBuildDefaultAsyncInferenceConfig(unittest.TestCase):
 
     def test_build_default_async_config(self):
         """Test building default async inference config."""
-        from sagemaker.core.inference_config import AsyncInferenceConfig
 
         mb = ModelBuilder(model=Mock())
         mb.model_name = "test-model"
@@ -256,7 +258,6 @@ def test_does_ic_exist_true(self):
 
     def test_does_ic_exist_false(self):
         """Test IC doesn't exist."""
-        from botocore.exceptions import ClientError
 
         mb = ModelBuilder(model=Mock())
         mb.sagemaker_session = Mock()
@@ -366,7 +367,6 @@ class TestFromJumpStartConfig(unittest.TestCase):
 
     def test_from_jumpstart_config_basic(self):
         """Test creating ModelBuilder from JumpStart config."""
-        from sagemaker.core.jumpstart.configs import JumpStartConfig
 
         js_config = JumpStartConfig(
             model_id="test-model",
@@ -384,8 +384,6 @@ def test_from_jumpstart_config_basic(self):
     @patch("sagemaker.serve.model_builder._retrieve_model_deploy_kwargs")
     def test_from_jumpstart_config_applies_network_isolation(self, mock_deploy_kwargs):
         """Test that enable_network_isolation from deploy kwargs is applied."""
-        from sagemaker.core.jumpstart.configs import JumpStartConfig
-        from sagemaker.core.training.configs import Compute
 
         mock_deploy_kwargs.return_value = {
             "model_data_download_timeout": 600,
@@ -409,6 +407,132 @@ def test_from_jumpstart_config_applies_network_isolation(self, mock_deploy_kwarg
 
         self.assertTrue(mb._enable_network_isolation)
 
+    @patch("sagemaker.serve.model_builder._retrieve_model_deploy_kwargs")
+    def test_from_jumpstart_config_applies_volume_size(self, mock_deploy_kwargs):
+        """Test that volume_size from deploy kwargs is applied."""
+
+        mock_deploy_kwargs.return_value = {
+            "model_data_download_timeout": 600,
+            "volume_size": 256,
+        }
+
+        js_config = JumpStartConfig(
+            model_id="meta-textgenerationneuron-llama-2-7b",
+            model_version="1.0.0"
+        )
+
+        mock_session = Mock()
+        mock_session.boto_region_name = "us-west-2"
+
+        mb = ModelBuilder.from_jumpstart_config(
+            jumpstart_config=js_config,
+            role_arn="arn:aws:iam::123456789012:role/SageMakerRole",
+            compute=Compute(instance_type="ml.inf2.xlarge"),
+            sagemaker_session=mock_session,
+        )
+
+        self.assertEqual(mb.volume_size, 256)
+
+    @patch("sagemaker.serve.model_builder.Endpoint.get")
+    @patch("sagemaker.serve.model_builder.session_helper.production_variant")
+    @patch("sagemaker.serve.model_builder._retrieve_model_deploy_kwargs")
+    def test_deploy_passes_volume_size_to_production_variant(
+        self, mock_deploy_kwargs, mock_prod_variant, mock_endpoint_get
+    ):
+        """Test that volume_size kwarg passed to deploy() reaches production_variant."""
+
+        mock_deploy_kwargs.return_value = {"volume_size": 256}
+        mock_prod_variant.return_value = {"VariantName": "AllTraffic"}
+        mock_endpoint_get.return_value = Mock()
+
+        js_config = JumpStartConfig(
+            model_id="meta-textgenerationneuron-llama-2-7b",
+            model_version="1.0.0",
+        )
+
+        mock_session = Mock()
+        mock_session.boto_region_name = "us-west-2"
+        mock_session.endpoint_in_service_or_not = Mock(return_value=False)
+        mock_session.endpoint_from_production_variants = Mock()
+        mock_session.sagemaker_config = {}
+        mock_session.settings = Mock()
+        mock_session.settings.include_jumpstart_tags = False
+        mock_session._append_sagemaker_config_tags = Mock(return_value=[])
+
+        mb = ModelBuilder.from_jumpstart_config(
+            jumpstart_config=js_config,
+            role_arn="arn:aws:iam::123456789012:role/SageMakerRole",
+            compute=Compute(instance_type="ml.inf2.xlarge"),
+            sagemaker_session=mock_session,
+        )
+        mb.built_model = Mock()
+        mb.built_model.model_name = "test-model"
+        mb.model_server = None
+        mb.mode = Mode.SAGEMAKER_ENDPOINT
+
+        # Deploy with explicit volume_size=512 overriding spec's 256
+        mb.deploy(
+            endpoint_name="test-ep",
+            instance_type="ml.inf2.xlarge",
+            initial_instance_count=1,
+            volume_size=512,
+            wait=False,
+        )
+
+        # Verify production_variant was called with user's 512, not spec's 256
+        mock_prod_variant.assert_called_once()
+        call_kwargs = mock_prod_variant.call_args[1]
+        self.assertEqual(call_kwargs["volume_size"], 512)
+
+    @patch("sagemaker.serve.model_builder.Endpoint.get")
+    @patch("sagemaker.serve.model_builder.session_helper.production_variant")
+    @patch("sagemaker.serve.model_builder._retrieve_model_deploy_kwargs")
+    def test_deploy_uses_spec_volume_size_when_not_passed(
+        self, mock_deploy_kwargs, mock_prod_variant, mock_endpoint_get
+    ):
+        """Test that volume_size from spec is used when customer doesn't pass it."""
+
+        mock_deploy_kwargs.return_value = {"volume_size": 256}
+        mock_prod_variant.return_value = {"VariantName": "AllTraffic"}
+        mock_endpoint_get.return_value = Mock()
+
+        js_config = JumpStartConfig(
+            model_id="meta-textgenerationneuron-llama-2-7b",
+            model_version="1.0.0",
+        )
+
+        mock_session = Mock()
+        mock_session.boto_region_name = "us-west-2"
+        mock_session.endpoint_in_service_or_not = Mock(return_value=False)
+        mock_session.endpoint_from_production_variants = Mock()
+        mock_session.sagemaker_config = {}
+        mock_session.settings = Mock()
+        mock_session.settings.include_jumpstart_tags = False
+        mock_session._append_sagemaker_config_tags = Mock(return_value=[])
+
+        mb = ModelBuilder.from_jumpstart_config(
+            jumpstart_config=js_config,
+            role_arn="arn:aws:iam::123456789012:role/SageMakerRole",
+            compute=Compute(instance_type="ml.inf2.xlarge"),
+            sagemaker_session=mock_session,
+        )
+        mb.built_model = Mock()
+        mb.built_model.model_name = "test-model"
+        mb.model_server = None
+        mb.mode = Mode.SAGEMAKER_ENDPOINT
+
+        # Deploy WITHOUT passing volume_size — should use spec's 256
+        mb.deploy(
+            endpoint_name="test-ep",
+            instance_type="ml.inf2.xlarge",
+            initial_instance_count=1,
+            wait=False,
+        )
+
+        mock_prod_variant.assert_called_once()
+        call_kwargs = mock_prod_variant.call_args[1]
+        self.assertEqual(call_kwargs["volume_size"], 256)
+
 
 if __name__ == "__main__":
     unittest.main()