AI-300 Practice Question 15: MLflow No Code Deployment
Microsoft Machine Learning Operations Engineer Associate (AI-300) · Test 2 · question 5 of 10
A scikit-learn model was logged with mlflow.sklearn.log_model() and registered in the workspace with type mlflow_model. An engineer new to the team has written a score.py and a conda environment file, assuming both are needed, and is about to deploy the model to a managed online endpoint. What should you tell them to do?
Show answer and explanation
Answer:
"When you deploy your MLflow model to an online endpoint, you don't need to specify a scoring script or an environment; this functionality is known as no-code deployment." Azure Machine Learning supplies an MLflow base image containing the inference server and a scoring script, and installs the packages listed in the model's conda.yaml at container runtime (option B). The engineer's two files are simply unnecessary work for a PyFunc-flavored MLflow model.
- A is wrong because that rule describes a custom model deployment. The whole point of registering the model as mlflow_model is that the MLmodel metadata lets Azure Machine Learning build the serving stack itself. - C and D are both wrong because they split a pair that Microsoft keeps together: the scoring script and the environment are generated together for an MLflow model, and conversely "if you specify a scoring script for an MLflow model deployment, you also need to specify the environment for the deployment". You cannot supply one and have the other generated.
Reference: Deploy MLflow models to real-time endpoints - Azure Machine Learning