Databricks Machine Learning Associate Question 632
Select 2You are tasked with deploying a custom machine learning model to a Databricks model serving endpoint. The model has been trained in a different platform and saved in a format that is not natively supported by Databricks MLflow. Which of the following steps is required to ensure successful deployment?
- A
Register the model in the MLflow Model Registry before deploying it to an endpoint.
- B
Write a custom Python wrapper or inference function to handle the model's prediction logic.
- C
Ensure the model is converted to the ONNX format to be compatible with Databricks model endpoints.
- D
Configure the model endpoint with the correct Conda or pip dependencies required for the model.
- E
Deploy the model directly without any modifications, as Databricks automatically supports all model formats.
Show answer and explanation
Correct answers: B, D
Explanation
To deploy a model in a non-native format to a Databricks model serving endpoint, you need to define an inference function (e.g., a Python wrapper) to handle prediction logic. Additionally, you must specify the correct dependencies in the model's runtime environment to ensure it works as expected. These steps ensure compatibility and successful deployment.
- A. Incorrect.
Registering the model in the MLflow Model Registry is not mandatory for deploying a custom model to a Databricks endpoint, especially if the model is not in MLflow-native format.
- B. Correct.
If the model comes from a non-native format, a custom Python wrapper is necessary to define the inference logic for handling predictions.
- C. Incorrect.
While ONNX is a common format, Databricks does not require models to be in ONNX format for deployment. Custom formats can work as long as the appropriate dependencies and logic are provided.
- D. Correct.
Configuring the correct Conda or pip dependencies is essential to ensure the model's runtime environment includes all necessary libraries for successful inference.
- E. Incorrect.
Databricks does not automatically support all model formats. Custom formats require additional steps such as writing a wrapper function and ensuring dependencies are satisfied.