NCA-GENM Question 118
Single answerYou are working alongside a senior team member to deploy a multimodal generative AI model designed for image and text generation. During deployment, the model is taking significantly longer to generate results when handling large datasets. What is the most appropriate action to evaluate the model's scalability and address potential bottlenecks?
- A
Analyze the model's performance using profiling tools to identify bottlenecks in computation or memory usage.
- B
Deploy the model on additional GPU resources without analyzing current performance metrics.
- C
Reduce the size of the dataset being processed to improve generation speed temporarily.
- D
Evaluate the model's inference latency and throughput under different batch sizes.
Show answer and explanation
Correct answer: A
Explanation
When evaluating the scalability and performance of a generative AI model, it is crucial to identify specific bottlenecks through profiling tools before making adjustments to resources or configurations. Profiling allows you to pinpoint inefficiencies in computation or memory usage, enabling targeted optimizations to enhance the model's performance.
- A. Correct.
Correct. Profiling tools help identify specific bottlenecks in the model's computation or memory usage, which is critical for understanding and improving scalability issues.
- B. Incorrect.
Incorrect. Adding more GPU resources without understanding the root cause could lead to inefficient resource usage and may not resolve the performance bottlenecks.
- C. Incorrect.
Incorrect. Reducing the dataset size only addresses the symptom of the issue temporarily, not the underlying scalability problem.
- D. Incorrect.
Incorrect. While evaluating latency and throughput under different batch sizes is helpful, it is better to first identify bottlenecks using profiling tools before making batch size adjustments.