Google Professional Machine Learning Engineer Question 85
Select 3Google Cloud PlatformYou have deployed a TensorFlow model using Vertex AI for online predictions. During testing, you notice that the predictions are slower than expected, and the latency is above your service-level agreement (SLA). Which of the following steps should you take to debug and optimize the model's performance?
- A
Enable model monitoring in Vertex AI to analyze the prediction traffic and distribution.
- B
Use the Explainable AI feature to identify which features are contributing to the latency.
- C
Increase the machine type or add more nodes to the serving endpoint in Vertex AI.
- D
Inspect the input data format and ensure it matches the expected format for the model.
- E
Use TensorFlow Model Optimization Toolkit to quantize or prune the model for improved performance.
Show answer and explanation
Correct answers: C, D, E
Explanation
To debug and optimize the performance of a deployed model, you should first ensure that the input data format is correct, as incorrect inputs can add overhead. Additionally, using tools like the TensorFlow Model Optimization Toolkit to quantize or prune the model can reduce latency by decreasing computational demands. If these steps are insufficient, scaling the infrastructure (e.g., increasing machine type or adding nodes) can provide additional resources to meet the required SLA.
- A. Incorrect.
While enabling model monitoring is a good practice, it is primarily used to track drift and anomalies in the data, not for debugging latency issues.
- B. Incorrect.
The Explainable AI feature is used to understand feature importance and does not directly address performance optimization or latency debugging.
- C. Correct.
Increasing the machine type or adding more nodes can help improve the performance of the deployed model by providing additional computational resources.
- D. Correct.
Ensuring that the input data format matches the model's requirements is a critical step, as incorrect data formatting can cause unnecessary preprocessing overhead and increase latency.
- E. Correct.
Using TensorFlow Model Optimization Toolkit can reduce the size and computational complexity of the model, which can significantly improve the model's inference speed.