Google Cloud Generative AI Leader Generative AI Leader practice test

A data scientist is trying to get online predictions from a Vertex AI endpoint but receives the error shown. What is the most likely cause?

Exhibit

Refer to the exhibit.

```
ERROR: Prediction failed: model 'projects/my-project/locations/us-central1/models/123' is not deployed to endpoint 'projects/my-project/locations/us-central1/endpoints/456'. Deploy the model to the endpoint before sending prediction requests.
```

Trap 1: The region in the request does not match the endpoint region

The region matches; the error would be different if regions mismatched.

Trap 2: The endpoint ID is incorrect

The endpoint ID is correctly referenced; the error is about the model not being deployed.

Trap 3: The model ID is incorrect

The model ID is correctly referenced; the error is about deployment status.

A
The region in the request does not match the endpoint region
Why wrong: The region matches; the error would be different if regions mismatched.
B
The model has not been deployed to the specified endpoint
The error message directly states the model is not deployed to the endpoint.
C
The endpoint ID is incorrect
Why wrong: The endpoint ID is correctly referenced; the error is about the model not being deployed.
D
The model ID is incorrect
Why wrong: The model ID is correctly referenced; the error is about deployment status.

Question 2mediummultiple choice

A data scientist notices that a text generation model deployed on Vertex AI returns repetitive outputs after a few turns in a chat application. What is the most likely cause and the best parameter adjustment?

Trap 1: The max_output_tokens is too low; increase it to allow more diverse…

Max tokens controls length, not repetition.

Trap 2: The model is overfitted; switch to a smaller model.

Overfitting is unlikely in pre-trained models; repetition is a decoding issue.

Trap 3: The temperature is too low; increase temperature to add randomness.

Low temperature makes output more deterministic, increasing repetition.

A
The max_output_tokens is too low; increase it to allow more diverse output.
Why wrong: Max tokens controls length, not repetition.
B
The top_p value is too high; reduce top_p to limit token sampling.
Reducing top_p narrows the token pool, reducing repetition.
C
The model is overfitted; switch to a smaller model.
Why wrong: Overfitting is unlikely in pre-trained models; repetition is a decoding issue.
D
The temperature is too low; increase temperature to add randomness.
Why wrong: Low temperature makes output more deterministic, increasing repetition.

Question 3mediummulti select

A company is deploying a generative AI model for medical diagnosis support. Which THREE considerations are critical for responsible AI?

Trap 1: Maximize model throughput to handle high volumes.

Throughput is secondary to safety.

Trap 2: Use the cheapest model to reduce costs.

Cost should not compromise quality or safety.

A
Ensure the training data is diverse and representative.
Diverse data reduces bias.
B
Maximize model throughput to handle high volumes.
Why wrong: Throughput is secondary to safety.
C
Implement human oversight for all diagnostic suggestions.
Human in the loop ensures safety.
D
Provide clear disclaimers about the model's limitations.
Transparency is essential for responsible AI.
E
Use the cheapest model to reduce costs.
Why wrong: Cost should not compromise quality or safety.

Question 4hardmulti select

Which THREE considerations are critical when deploying a generative AI model using Vertex AI Endpoints for a latency-sensitive application? (Choose THREE.)

Trap 1: Number of model versions

Model versioning does not directly affect latency.

Trap 2: Number of model instances

While important, autoscaling handles instance count dynamically.

A
Model size and architecture
Larger models introduce higher latency.
B
Number of model versions
Why wrong: Model versioning does not directly affect latency.
C
GPU type and number
GPU selection impacts inference speed.
D
Autoscaling configuration
Proper autoscaling ensures low latency under varying load.
E
Number of model instances
Why wrong: While important, autoscaling handles instance count dynamically.

Read the full NAT/PAT explanation →

Question 5hardmultiple choice

A company is deploying a generative AI model for customer support. They want to reduce hallucinations while maintaining fluency. They have a large dataset of previous support conversations. Which strategy should they prioritize?

Trap 1: Increase the beam search width to 10.

Wider beam search improves fluency but not factual accuracy.

Trap 2: Fine-tune the model on the conversation dataset.

Fine-tuning may help but doesn't guarantee factual grounding for new queries.

Trap 3: Set the temperature to 0.1.

Low temperature reduces creativity but doesn't add factual grounding.

A
Increase the beam search width to 10.
Why wrong: Wider beam search improves fluency but not factual accuracy.
B
Implement retrieval-augmented generation (RAG) using the conversation dataset as a knowledge base.
RAG retrieves relevant facts from the dataset, reducing hallucinations.
C
Fine-tune the model on the conversation dataset.
Why wrong: Fine-tuning may help but doesn't guarantee factual grounding for new queries.
D
Set the temperature to 0.1.
Why wrong: Low temperature reduces creativity but doesn't add factual grounding.

Question 6easymulti select

Which TWO techniques are commonly used to control the style and tone of a generative model's output?

Trap 1: Adjusting the temperature

Temperature controls randomness, not style.

Trap 2: Modifying the top_k value

Top_k affects token selection diversity.

Trap 3: Changing the top_p value

Top_p controls nucleus sampling.

A
Adjusting the temperature
Why wrong: Temperature controls randomness, not style.
B
Modifying the top_k value
Why wrong: Top_k affects token selection diversity.
C
Fine-tuning on a dataset with desired style
Fine-tuning adapts the model to a specific style.
D
Prompt engineering with style instructions
Prompts can specify desired style.
E
Changing the top_p value
Why wrong: Top_p controls nucleus sampling.