Update model

Author: samruds

src/sagemaker/serve/builder/transformers_builder.py

@@ -94,7 +94,7 @@ def _create_transformers_model(self) -> Type[Model]:
         )
         hf_config = image_uris.config_for_framework("huggingface").get("inference")
         config = hf_config["versions"]
-        base_hf_version = sorted(config.keys(), key=lambda v: Version(v))[0]
+        base_hf_version = sorted(config.keys(), key=lambda v: Version(v), reverse=True)[0]
 
         if hf_model_md is None:
             raise ValueError("Could not fetch HF metadata")
@@ -269,7 +269,7 @@ def _get_supported_version(self, hf_config, hugging_face_version, base_fw):
                 if len(hugging_face_version.split(".")) == 2:
                     base_fw_version = ".".join(base_fw_version.split(".")[:-1])
                 versions_to_return.append(base_fw_version)
-        return sorted(versions_to_return)[0]
+        return sorted(versions_to_return, reverse=True)[0]
 
     def _build_for_transformers(self):
         """Method that triggers model build

tests/integ/sagemaker/serve/test_schema_builder.py

@@ -208,9 +208,9 @@ def test_model_builder_happy_path_with_task_provided_remote_schema_mode(
 )
 @pytest.mark.parametrize(
     "model_id, task_provided, instance_type_provided",
-    [("openai/whisper-large-v3", "automatic-speech-recognition", "ml.m5.xlarge")],
+    [("openai/whisper-tiny.en", "automatic-speech-recognition", "ml.m5.4xlarge")],
 )
-def test_model_builder_happy_path_with_task_provided_remote_schema_mode_asr(
+def test_model_builder_with_task_provided_remote_schema_mode_asr(
     model_id, task_provided, sagemaker_session, instance_type_provided
 ):
     model_builder = ModelBuilder(
@@ -228,34 +228,6 @@ def test_model_builder_happy_path_with_task_provided_remote_schema_mode_asr(
     assert model_builder.schema_builder.sample_input == inputs
     assert model_builder.schema_builder.sample_output == outputs
 
-    with timeout(minutes=SERVE_SAGEMAKER_ENDPOINT_TIMEOUT):
-        caught_ex = None
-        try:
-            iam_client = sagemaker_session.boto_session.client("iam")
-            role_arn = iam_client.get_role(RoleName="SageMakerRole")["Role"]["Arn"]
-
-            logger.info("Deploying and predicting in SAGEMAKER_ENDPOINT mode...")
-            predictor = model.deploy(
-                role=role_arn, instance_count=1, instance_type=instance_type_provided
-            )
-
-            predicted_outputs = predictor.predict(inputs)
-            assert predicted_outputs is not None
-
-        except Exception as e:
-            caught_ex = e
-        finally:
-            cleanup_model_resources(
-                sagemaker_session=model_builder.sagemaker_session,
-                model_name=model.name,
-                endpoint_name=model.endpoint_name,
-            )
-            if caught_ex:
-                logger.exception(caught_ex)
-                assert (
-                    False
-                ), f"{caught_ex} was thrown when running transformers sagemaker endpoint test"
-
 
 def test_model_builder_negative_path_with_invalid_task(sagemaker_session):
     model_builder = ModelBuilder(