Databricks Generative AI Engineer Associate Question 86
Select 3You are developing a machine learning application on Databricks to fine-tune a large language model (LLM) for a customer support chatbot. The model will be deployed as an API for real-time inference. Which of the following steps are essential to ensure scalable and efficient application deployment?
- A
Optimize the model by quantization or distillation before deployment.
- B
Set up AutoML to automate fine-tuning and directly deploy the best-performing model.
- C
Integrate Databricks Model Serving to deploy the model and handle real-time API requests.
- D
Use feature tables in Databricks Feature Store to serve features during inference.
- E
Deploy the model as a notebook job for API-based inference.
Show answer and explanation
Correct answers: A, C, D
Explanation
To ensure scalable and efficient deployment for real-time inference, it is essential to optimize the model (e.g., via quantization or distillation) to reduce computational requirements. Databricks Model Serving is a purpose-built solution for deploying models as APIs and handling real-time requests. Additionally, using the Databricks Feature Store ensures that feature consistency is maintained between training and inference, which is critical for accurate predictions. AutoML and notebook jobs, while useful for other purposes, do not directly address the requirements for scalable, real-time API deployment.
- A. Correct.
Optimizing the model through quantization or distillation reduces its size and computational requirements, making it more efficient for real-time inference. This is an essential step for scalable deployment.
- B. Incorrect.
AutoML aids in model selection and fine-tuning but is not directly related to deployment or API integration for real-time inference.
- C. Correct.
Using Databricks Model Serving is a best practice for deploying models as APIs for real-time inference. It ensures scalability and efficient request handling.
- D. Correct.
Feature tables in Databricks Feature Store are crucial for serving features during inference, ensuring consistency between training and prediction data.
- E. Incorrect.
Deploying the model as a notebook job is not suitable for real-time API-based inference. Notebook jobs are better suited for batch processing or scheduled tasks.