Databricks Generative AI Engineer Associate Question 6
Select 3A company wants to design a generative AI application using Databricks that generates product descriptions based on customer reviews. The application must scale for high user demand, ensure low latency, and maintain cost efficiency. Which combination of strategies would be most appropriate for designing this application?
- A
Use a pre-trained large language model (LLM) hosted on Databricks and fine-tune it with domain-specific data.
- B
Deploy the model on a single-node cluster to minimize infrastructure costs.
- C
Leverage Databricks AutoML to perform hyperparameter tuning for the fine-tuned model.
- D
Implement model serving on Databricks to enable real-time inference with scalable endpoints.
- E
Store customer reviews in a Delta Lake table and integrate it with Databricks Feature Store for streamlined data preprocessing.
Show answer and explanation
Correct answers: A, D, E
Explanation
The correct strategies involve fine-tuning a pre-trained LLM to adapt it to the domain, using Databricks model serving for scalable real-time inference, and leveraging Delta Lake and Databricks Feature Store for efficient data preparation. Together, these approaches address the requirements of scalability, low latency, and cost efficiency.
- A. Correct.
Using a pre-trained LLM and fine-tuning it with domain-specific data allows the model to better understand the product domain while saving time and resources compared to training from scratch. This is a best practice in generative AI application design.
- B. Incorrect.
Deploying the model on a single-node cluster may reduce costs but will fail to meet the application's scaling and low-latency requirements. It is not suitable for high user demand.
- C. Incorrect.
While Databricks AutoML is useful for tasks like classification or regression, in this case, it is not necessary for fine-tuning a pre-trained LLM. Fine-tuning can be done using other specialized libraries and techniques.
- D. Correct.
Model serving on Databricks provides scalable endpoints for real-time inference, which is critical for handling high user demand and ensuring low latency.
- E. Correct.
Using Delta Lake to store customer reviews and integrating with Databricks Feature Store allows for efficient data preprocessing and feature engineering, which is essential for a robust generative AI application.