Error to deploy model to an online endpoint

Question

Hello,

I try to deploy a model to an online endpoint. However I have an error when I use local endpoints. I want to deploy a deep learning model for object detection in images.

Here's what I did :

ws = Workspace(subscription_id, resource_group, workspace_name) # connection to the Azure ML Workspace
ml_client = MLClient(
    DefaultAzureCredential(), subscription_id, resource_group, workspace_name
)

# Get a previous training
Id_experiment = "valid-model"
automl_image_run = AutoMLRun(experiment=Experiment(ws,experiment_name), run_id=Id_experiment)

best_child_run = automl_image_run.get_best_child()
model_name = best_child_run.properties['model_name']

model_name = 'model_window_detection'


registered_model = Model(workspace=ws, name=model_name)
model_path = registered_model.download(target_dir='.', exist_ok=True)

# Configure an endpoint
import datetime
endpoint_name = "window-detection-endpoint"
endpoint_name = "endpt-" + datetime.datetime.now().strftime("%m%d%H%M%f")
endpoint = ManagedOnlineEndpoint(
    name = endpoint_name, 
    description="this is a sample endpoint",
    auth_mode="key"
)
# Configure a deployment
# model is registered_model
existing_env_name = "azureml_py38" #Name of my existing virtual environment

# Create the environment object using the existing environment
env = Environment(name=existing_env_name)
code_config = CodeConfiguration(
        code="./",
        scoring_script="score.py"
    )
# Create the deployment using the specified environment
deployment = ManagedOnlineDeployment(
    name="blue",
    endpoint_name=endpoint_name,
    model=registered_model,
    environment=env,  # Use the existing Conda environment
    code_configuration=code_config,
    instance_type="Standard_NC6s_v3",
    instance_count=1
)

ml_client.online_endpoints.begin_create_or_update(endpoint, local=True) #Works


ml_client.online_deployments.begin_create_or_update(
    deployment=deployment, local=True
) #This is the line that generates an error

The error :

{
	"name": "Exception",
	"message": "['Non-string passed to RegistryStr for model', 'Non-string passed to ArmStr for model', \"'Model' object has no attribute '_intellectual_property'\"]",
	"stack": "---------------------------------------------------------------------------
ValidationError                           Traceback (most recent call last)
File /anaconda/envs/azureml_py38/lib/python3.8/site-packages/azure/ai/ml/operations/_local_deployment_helper.py:82, in _LocalDeploymentHelper.create_or_update(self, deployment, local_endpoint_mode, local_enable_gpu)
     80     operation_message = \"Creating local deployment\"
---> 82 deployment_metadata = json.dumps(deployment._to_dict())
     83 endpoint_metadata = (
     84     endpoint_metadata
     85     if endpoint_metadata
     86     else _get_stubbed_endpoint_metadata(endpoint_name=deployment.endpoint_name)
     87 )

File /anaconda/envs/azureml_py38/lib/python3.8/site-packages/azure/ai/ml/entities/_deployment/online_deployment.py:748, in ManagedOnlineDeployment._to_dict(self)
    747 def _to_dict(self) -> Dict:
--> 748     return ManagedOnlineDeploymentSchema(context={BASE_PATH_CONTEXT_KEY: \"./\"}).dump(self)

File /anaconda/envs/azureml_py38/lib/python3.8/site-packages/marshmallow/schema.py:549, in Schema.dump(self, obj, many)
    547     processed_obj = obj
--> 549 result = self._serialize(processed_obj, many=many)
    551 if self._has_processors(POST_DUMP):

File /anaconda/envs/azureml_py38/lib/python3.8/site-packages/marshmallow/schema.py:517, in Schema._serialize(self, obj, many)
    516 for attr_name, field_obj in self.dump_fields.items():
--> 517     value = field_obj.serialize(attr_name, obj, accessor=self.get_attribute)
    518     if value is missing:



File /anaconda/envs/azureml_py38/lib/python3.8/site-packages/azure/ai/ml/_exception_helper.py:337, in log_and_raise_error(error, debug, yaml_operation)
    334 else:
    335     raise error
--> 337 raise Exception(formatted_error)

Exception: ['Non-string passed to RegistryStr for model', 'Non-string passed to ArmStr for model', \"'Model' object has no attribute '_intellectual_property'\"]"
}

I can also give you my script score.py :

# ---------------------------------------------------------
# Copyright (c) Microsoft Corporation. All rights reserved.
# ---------------------------------------------------------
import os

from azureml.automl.core.shared import logging_utilities
from azureml.automl.dnn.vision.object_detection.writers.score import _score_with_model
from azureml.automl.dnn.vision.common.logging_utils import get_logger
from azureml.automl.dnn.vision.common.model_export_utils import (
    load_model, run_inference_helper)
from azureml.automl.dnn.vision.common.utils import _set_logging_parameters
from azureml.contrib.services.aml_request import rawhttp
from azureml.contrib.services.aml_response import AMLResponse
from azureml.core.model import Model

TASK_TYPE = 'image-object-detection'
logger = get_logger('azureml.automl.core.scoring_script_images')


def init():
    global model

    # Set up logging
    _set_logging_parameters(TASK_TYPE, {})

    model_path = os.path.join(os.getenv('AZUREML_MODEL_DIR'), 'model.pt')
    if not os.path.exists(model_path):
        model_path = Model.get_model_path(model_name='AutoML322455a460')

    try:
        logger.info("Loading model from path: {}.".format(model_path))
        model_settings = {"min_size": 600, "max_size": 1333, "box_score_thresh": 0.3, "nms_iou_thresh": 0.5, "box_detections_per_img": 100, 
                          "tile_grid_size": None, "tile_overlap_ratio": 0.25, "tile_predictions_nms_thresh": 0.25}
        model = load_model(TASK_TYPE, model_path, **model_settings)
        logger.info("Loading successful.")
    except Exception as e:
        logging_utilities.log_traceback(e, logger)
        raise


@rawhttp
def run(request):
    logger.info("Request: [{0}]".format(request))
    if request.method == "GET":
        response_body = str.encode(request.full_path)
        return AMLResponse(response_body, 200)

    elif request.method == "POST":
        request_body = request.get_data()
        logger.info("Running inference.")
        result = run_inference_helper(model, request_body, _score_with_model, TASK_TYPE)
        logger.info("Finished inferencing.")
        return AMLResponse(result, 200)
    else:
        return AMLResponse("bad request", 500)