Databricks Generative AI Engineer Associate Question 89
Select 3You are developing a generative AI application on Databricks using a pre-trained Large Language Model (LLM). The application must support user queries, manage input/output token limits efficiently, and log user interactions for auditing purposes. Which components or practices should you implement to meet these requirements?
- A
Implement a prompt engineering strategy to structure user queries effectively.
- B
Use the Databricks Delta table to store and log user interactions.
- C
Set a token limit for both input and output using the LLM API's parameters.
- D
Deploy the application without any rate-limiting to maximize throughput.
- E
Perform fine-tuning of the pre-trained LLM even if the base model already supports your use case.
Show answer and explanation
Correct answers: A, B, C
Explanation
To meet the requirements of supporting user queries, managing token limits, and logging user interactions for auditing purposes, you must implement prompt engineering for better query structuring, set token limits when interacting with the LLM API, and use a robust logging mechanism such as Databricks Delta tables. These practices ensure the application is functional, efficient, and auditable.
- A. Correct.
Implementing a prompt engineering strategy ensures the user queries are structured in a way that the LLM can generate accurate and relevant responses, addressing the requirement of supporting user queries.
- B. Correct.
Using Databricks Delta tables to log user interactions provides an auditable, scalable, and efficient way to manage and store logs, meeting the auditing requirement.
- C. Correct.
Setting a token limit for both input and output using the LLM API parameters ensures the application manages token limits efficiently, preventing errors related to exceeding the model's token capacity.
- D. Incorrect.
Deploying without rate-limiting is not advisable as it can lead to overloading the system, causing performance degradation and potential outages. This does not align with the application's requirements.
- E. Incorrect.
Fine-tuning the pre-trained LLM is unnecessary if the base model already supports the use case. Fine-tuning is only needed when the pre-trained model lacks domain-specific knowledge or behavior.