fix: respect accept_eula in ModelBuilder LoRA deployment path (#5705)

* Update accept_eula to respect user setup

* Enable EULA acceptance in model customization tests

Set accept_eula to True in model builder to fix tests

* fix: add missing model_path attr in TestLoraAcceptEula

To fix failing unit tests in PR: #5696

* fix(tests): fix TestLoraAcceptEula missing dataclass attrs and patches

---------

Co-authored-by: Molly He <mollyhe@amazon.com>
Co-authored-by: Gokul Anantha Narayanan <166456257+nargokul@users.noreply.github.com>

Author: 3 people

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

@@ -2373,6 +2373,13 @@ def _build_single_modelbuilder(
                         "HostingArtifactUri not found in JumpStart hub metadata. "
                         "Cannot deploy LORA adapter without base model artifacts."
                     )
+                accept_eula = getattr(self, "accept_eula", None)
+                if not accept_eula:
+                    raise ValueError(
+                        "accept_eula must be set to True to deploy this model. "
+                        "Please set accept_eula=True on the ModelBuilder instance to confirm "
+                        "you have read and accepted the end-user license agreement for this model."
+                    )
                 container_def = ContainerDefinition(
                     image=self.image_uri,
                     environment=self.env_vars,
@@ -2381,7 +2388,7 @@ def _build_single_modelbuilder(
                             "s3_uri": hosting_artifact_uri,
                             "s3_data_type": "S3Prefix",
                             "compression_type": "None",
-                            "model_access_config": {"accept_eula": True},
+                            "model_access_config": {"accept_eula": accept_eula},
                         }
                     },
                 )

sagemaker-serve/tests/integ/test_model_customization_deployment.py

@@ -119,6 +119,7 @@ def test_build_from_training_job(self, training_job_name):
 
         training_job = TrainingJob.get(training_job_name=training_job_name)
         model_builder = ModelBuilder(model=training_job)
+        model_builder.accept_eula = True
         model = model_builder.build(model_name=f"test-model-{int(time.time())}-{random.randint(100, 10000)}")
 
         assert model is not None
@@ -139,6 +140,7 @@ def test_deploy_from_training_job(self, training_job_name, endpoint_name, cleanu
 
         training_job = TrainingJob.get(training_job_name=training_job_name)
         model_builder = ModelBuilder(model=training_job, instance_type="ml.g5.4xlarge")
+        model_builder.accept_eula = True
         model_builder.build(model_name=f"test-model-{int(time.time())}-{random.randint(100, 10000)}")
 
         peft_type = model_builder._fetch_peft()
@@ -187,6 +189,7 @@ def test_build_from_model_package(self, model_package_arn):
 
         model_package = ModelPackage.get(model_package_name=model_package_arn)
         model_builder = ModelBuilder(model=model_package)
+        model_builder.accept_eula = True
         model = model_builder.build()
 
         assert model is not None
@@ -201,6 +204,7 @@ def test_deploy_from_model_package(self, model_package_arn, cleanup_endpoints):
         model_package = ModelPackage.get(model_package_name=model_package_arn)
         endpoint_name = f"e2e-{int(time.time())}-{random.randint(100, 10000)}"
         model_builder = ModelBuilder(model=model_package)
+        model_builder.accept_eula = True
         model_builder.build()
         endpoint = model_builder.deploy(endpoint_name=endpoint_name)
 
@@ -220,6 +224,7 @@ def test_instance_type_from_recipe(self, training_job_name):
 
         training_job = TrainingJob.get(training_job_name=training_job_name)
         model_builder = ModelBuilder(model=training_job)
+        model_builder.accept_eula = True
         model_builder.build()
 
         assert model_builder.instance_type is not None

sagemaker-serve/tests/unit/test_model_builder.py

@@ -715,3 +715,96 @@ def test_deploy_passes_inference_config_to_model_customization(self):
                 call_kwargs = mock_deploy_mc.call_args[1]
                 self.assertEqual(call_kwargs['inference_config'], inference_config)
                 self.assertEqual(result, mock_endpoint)
+
+
+class TestLoraAcceptEula(unittest.TestCase):
+    """Tests for accept_eula handling in the LoRA deployment path."""
+
+    def _make_mb(self, accept_eula=None):
+        mb = ModelBuilder.__new__(ModelBuilder)
+        mb.accept_eula = accept_eula
+        mb.image_uri = "some-image-uri"
+        mb.env_vars = {}
+        mb.model_name = None
+        mb.model_path = "/tmp/fake-model-path"
+        mb.role_arn = "arn:aws:iam::123456789012:role/role"
+        mb.model = MagicMock()
+        mb._adapter_s3_uri = None
+        mb.shared_libs = []
+        mb.dependencies = {"auto": True}
+        mb.image_config = None
+        mb.inference_spec = None
+        mb.schema_builder = None
+        mb.modelbuilder_list = None
+        mb.sagemaker_session = None
+        mb.s3_model_data_url = None
+        mb.source_code = None
+        mb.model_server = None
+        mb.model_metadata = None
+        mb.log_level = None
+        mb.content_type = None
+        mb.accept_type = None
+        mb.compute = None
+        mb.network = None
+        mb.instance_type = None
+        mb.mode = None
+        return mb
+
+    def _patch_lora_deps(self, mb, hosting_uri="s3://bucket/hosting/"):
+        """Patch all dependencies needed to reach the LoRA ContainerDefinition block."""
+        patches = [
+            patch.object(mb, "_get_serve_setting", return_value=MagicMock()),
+            patch.object(mb, "_is_model_customization", return_value=True),
+            patch.object(mb, "_fetch_model_package", return_value=MagicMock()),
+            patch.object(mb, "_fetch_and_cache_recipe_config"),
+            patch.object(mb, "_is_nova_model", return_value=False),
+            patch.object(mb, "_fetch_peft", return_value="LORA"),
+            patch.object(mb, "_fetch_hub_document_for_custom_model",
+                         return_value={"HostingArtifactUri": hosting_uri}),
+        ]
+        return patches
+
+    def test_lora_build_raises_when_accept_eula_false(self):
+        mb = self._make_mb(accept_eula=False)
+        patches = self._patch_lora_deps(mb)
+        for p in patches:
+            p.start()
+        try:
+            with self.assertRaises(ValueError) as ctx:
+                mb._build_single_modelbuilder()
+            self.assertIn("accept_eula", str(ctx.exception))
+        finally:
+            for p in patches:
+                p.stop()
+
+    def test_lora_build_raises_when_accept_eula_not_set(self):
+        mb = self._make_mb(accept_eula=None)
+        patches = self._patch_lora_deps(mb)
+        for p in patches:
+            p.start()
+        try:
+            with self.assertRaises(ValueError) as ctx:
+                mb._build_single_modelbuilder()
+            self.assertIn("accept_eula", str(ctx.exception))
+        finally:
+            for p in patches:
+                p.stop()
+
+    @patch("sagemaker.serve.model_builder.ContainerDefinition")
+    @patch("sagemaker.serve.model_builder.Model")
+    def test_lora_build_passes_accept_eula_true(self, mock_model, mock_container_def):
+        mb = self._make_mb(accept_eula=True)
+        mock_model.create.return_value = MagicMock()
+        patches = self._patch_lora_deps(mb)
+        for p in patches:
+            p.start()
+        try:
+            mb._build_single_modelbuilder()
+            call_kwargs = mock_container_def.call_args[1]
+            eula_val = (
+                call_kwargs["model_data_source"]["s3_data_source"]["model_access_config"]["accept_eula"]
+            )
+            self.assertTrue(eula_val)
+        finally:
+            for p in patches:
+                p.stop()