Fix ModelBuilder so that it doesn't assume every ModelPackage was from JumpStart (#5719)

Author: bobbywlindsey

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

@@ -2331,7 +2331,9 @@ def _build_single_modelbuilder(
                 )
             model_package = self._fetch_model_package()
             # Fetch recipe config first to set image_uri, instance_type, env_vars, and s3_upload_path
-            self._fetch_and_cache_recipe_config()
+            base_model = model_package.inference_specification.containers[0].base_model
+            if base_model is not None:
+                self._fetch_and_cache_recipe_config()
 
             # Nova models use a completely different deployment architecture
             if self._is_nova_model():

sagemaker-serve/tests/unit/test_model_builder.py

@@ -462,6 +462,47 @@ def test_build_single_modelbuilder_with_model_customization(self, mock_is_1p, mo
         mock_model_class.create.assert_called_once()
         self.assertEqual(result, mock_created_model)
 
+    @patch('sagemaker.serve.model_builder.Model')
+    @patch('sagemaker.serve.model_builder.is_1p_image_uri')
+    def test_build_single_modelbuilder_with_model_customization_no_jumpstart(self, mock_is_1p, mock_model_class):
+        """Test _build_single_modelbuilder skips _fetch_and_cache_recipe_config when base_model is None."""
+        mock_is_1p.return_value = True
+
+        # Setup mock model package with base_model = None (custom model package, not JumpStart)
+        mock_model_package = Mock()
+        mock_container = Mock()
+        mock_container.base_model = None
+        mock_container.model_data_source.s3_data_source.s3_uri = "s3://bucket/model"
+        mock_model_package.inference_specification.containers = [mock_container]
+
+        # Setup training job with model_package_config
+        self.mock_training_job.model_package_config = Mock()
+        self.mock_training_job.model_package_config.source_model_package_arn = (
+            "arn:aws:sagemaker:us-east-1:123456789012:model-package/source"
+        )
+
+        mock_created_model = Mock()
+        mock_model_class.create.return_value = mock_created_model
+
+        builder = ModelBuilder(
+            model=self.mock_training_job,
+            role_arn="arn:aws:iam::123456789012:role/SageMakerRole",
+            sagemaker_session=self.mock_session,
+            image_uri="test-image:latest",
+            instance_type="ml.g5.2xlarge"
+        )
+
+        with patch.object(builder, '_fetch_model_package', return_value=mock_model_package):
+            with patch.object(builder, '_fetch_and_cache_recipe_config') as mock_recipe:
+                with patch.object(builder, '_get_serve_setting', return_value=Mock()):
+                    with patch.object(builder, '_is_nova_model', return_value=False):
+                        with patch.object(builder, '_fetch_peft', return_value=None):
+                            result = builder._build_single_modelbuilder()
+
+        mock_recipe.assert_not_called()
+        mock_model_class.create.assert_called_once()
+        self.assertEqual(result, mock_created_model)
+
     def test_deploy_model_customization_new_endpoint(self):
         """Test _deploy_model_customization for new endpoint creation."""
         from sagemaker.core.shapes import InferenceComponentComputeResourceRequirements