AI-102 Question 469
Select 4You are tasked with optimizing the performance and cost of a generative AI model deployed on Azure. The model is being used to generate text for customer support responses. Which actions should you take to optimize the AI solution without compromising response quality?
- A
Enable Azure OpenAI fine-tuning to specialize the model for customer support-specific scenarios.
- B
Use a smaller, more efficient model variant if it meets the required accuracy for customer responses.
- C
Deploy the model using an Azure Kubernetes Service (AKS) cluster with autoscaling enabled.
- D
Disable logging and monitoring to reduce operational overhead.
- E
Implement token limits to restrict the length of generated responses.
Show answer and explanation
Correct answers: A, B, C, E
Explanation
Optimizing generative AI solutions in Azure requires a combination of strategies, including fine-tuning models for domain-specific tasks, selecting efficient model variants, leveraging scalable infrastructure, and implementing operational controls like token limits. These actions help balance performance, cost, and quality. Disabling logging and monitoring is not a valid optimization strategy as it hinders operational visibility.
- A. Correct.
Fine-tuning the Azure OpenAI model allows it to specialize for the specific domain of customer support, making it more efficient and accurate for the task.
- B. Correct.
Using a smaller, more efficient model variant can reduce computational cost and improve performance, provided it meets the required accuracy standards.
- C. Correct.
Deploying the model on Azure Kubernetes Service (AKS) with autoscaling ensures that resources are dynamically allocated based on demand, optimizing costs and performance.
- D. Incorrect.
Disabling logging and monitoring is not a recommended optimization strategy, as it can compromise the ability to diagnose issues and track performance.
- E. Correct.
Implementing token limits to restrict response length controls computational resources and ensures responses remain concise, reducing unnecessary processing.