Troubleshooting practice questions

Which THREE components are essential for a production-grade generative AI deployment on OCI? (Select THREE)

Trap 1: OCI Vault for secrets

While good practice, not strictly essential if secrets are managed elsewhere.

Trap 2: OCI Data Flow for data processing

Data Flow is for big data processing, not required for inference.

A
OCI Logging for audit
Logging is critical for monitoring and compliance.
B
OCI Vault for secrets
Why wrong: While good practice, not strictly essential if secrets are managed elsewhere.
C
OCI Data Flow for data processing
Why wrong: Data Flow is for big data processing, not required for inference.
D
Dedicated AI cluster
Essential for hosting the model with predictable performance.
E
OCI IAM policies for access control
IAM policies are fundamental for securing the deployment.

Question 2easymulti select

A developer is troubleshooting an OCI Generative AI inference request that returns a 400 Bad Request error. Which three common causes could result in this error? (Choose three.)

Trap 1: Invalid API key in the request header

An invalid API key results in a 401 Unauthorized error.

Trap 2: Network connectivity issues

Connectivity issues lead to timeouts or 5xx errors, not 400.

A
Incorrect endpoint URL
A wrong URL may cause the request to be malformed or routed incorrectly, resulting in a 400.
B
Invalid API key in the request header
Why wrong: An invalid API key results in a 401 Unauthorized error.
C
Missing required parameters in the request body
Required fields like 'modelId' or 'prompt' are omitted.
D
Exceeding the model's maximum token limit
The input or output exceeds the model's token limit, causing a bad request.
E
Network connectivity issues
Why wrong: Connectivity issues lead to timeouts or 5xx errors, not 400.

Question 3mediummulti select

A DevOps engineer is setting up monitoring and logging for a generative AI inference endpoint. Which three resources should they enable? (Select THREE.)

Trap 1: OCI VCN flow logs for network traffic

Incorrect: Flow logs are for network security analysis, not application monitoring.

Trap 2: OCI Audit logs for all API calls

Incorrect: Audit logs capture control plane operations, not inference data.

A
OCI VCN flow logs for network traffic
Why wrong: Incorrect: Flow logs are for network security analysis, not application monitoring.
B
OCI Logging for inference requests and responses
Correct: Logging allows auditing and debugging of inference calls.
C
OCI Monitoring metrics for endpoint latency and error rates
Correct: Metrics drive performance dashboards and alerts.
D
OCI Application Performance Monitoring (APM) for tracing inference requests
Correct: APM provides distributed tracing to diagnose performance bottlenecks.
E
OCI Audit logs for all API calls
Why wrong: Incorrect: Audit logs capture control plane operations, not inference data.

Question 4hardmulti select

A team is troubleshooting a chatbot that sometimes outputs harmful content despite having a system prompt with safety instructions. Which THREE measures should they implement to reduce the risk?

Trap 1: Increase temperature to make output more random

Higher temperature increases randomness, which could increase risk of harmful outputs.

Trap 2: Add more few-shot examples of safe responses

Few-shot examples may help but are not a direct safety control; they can be overridden.

A
Reinforce the system prompt with explicit safety constraints and periodic reminders
Strengthening the system prompt with detailed safety rules reduces the chance of harmful outputs.
B
Increase temperature to make output more random
Why wrong: Higher temperature increases randomness, which could increase risk of harmful outputs.
C
Add more few-shot examples of safe responses
Why wrong: Few-shot examples may help but are not a direct safety control; they can be overridden.
D
Use output filtering to scan and block harmful responses before showing to users
Output filtering is a critical safety layer to catch unintended content.
E
Implement input filtering to detect and block malicious prompt injection attempts
Input filtering helps prevent prompt injection that could bypass safety instructions.

Question 5hardmulti select

An OCI user is troubleshooting a prompt that sometimes produces outputs containing offensive language. The prompt uses a system prompt to set a professional tone. Which THREE steps should the user take to mitigate this issue? (Select three.)

Trap 1: Increase temperature to 0.9 to dilute the offending outputs

Higher temperature increases randomness and may produce more unpredictable (and possibly offensive) language.

Trap 2: Remove the system prompt to avoid overriding model's safety training

Removing the system prompt removes a layer of control and may make the problem worse.

A
Apply a frequency penalty to discourage repetition of offensive phrases
Frequency penalty reduces token repetition, which can help if offensive language appears repeatedly.
B
Increase temperature to 0.9 to dilute the offending outputs
Why wrong: Higher temperature increases randomness and may produce more unpredictable (and possibly offensive) language.
C
Test the prompt with diverse inputs including adversarial examples
Testing with edge cases helps identify and fix vulnerabilities.
D
Add a constraint in the system prompt: 'Do not use offensive or inappropriate language.'
Explicit constraints directly guide the model's behavior.
E
Remove the system prompt to avoid overriding model's safety training
Why wrong: Removing the system prompt removes a layer of control and may make the problem worse.

Question 6mediummultiple choice

An AI specialist is troubleshooting why a fine-tuned model produces inconsistent results across different inference calls. What is the most likely cause?

Trap 1: The base model is not suitable

Unsuitable base model leads to consistently poor quality, not inconsistency.

Trap 2: The model is overfitted

Overfitting leads to consistent but poor generalization, not inconsistency.

Trap 3: The fine-tuning dataset is too small

Small dataset may cause overfitting, not necessarily inconsistency.

A
The base model is not suitable
Why wrong: Unsuitable base model leads to consistently poor quality, not inconsistency.
B
The temperature is set too high
High temperature increases randomness, causing variable outputs.
C
The model is overfitted
Why wrong: Overfitting leads to consistent but poor generalization, not inconsistency.
D
The fine-tuning dataset is too small
Why wrong: Small dataset may cause overfitting, not necessarily inconsistency.

Question 7hardmulti select

A prompt engineer is troubleshooting a chatbot that consistently fails to follow instructions when the user includes adversarial input. Which two strategies can mitigate prompt injection attacks? (Choose two.)

Trap 1: Increase temperature to make model less predictable

Higher temperature does not prevent injection; it may even increase erratic behavior.

Trap 2: Use a smaller model to reduce capability

Model size does not prevent injection; smaller models may be less capable but still vulnerable.

Trap 3: Add more few-shot examples with safe outputs

Examples do not prevent injection; they only show desired behavior.

A
Increase temperature to make model less predictable
Why wrong: Higher temperature does not prevent injection; it may even increase erratic behavior.
B
Use instruction shielding: clearly separate system instructions from user input
Separating instructions from user input prevents the model from treating user input as instructions.
C
Use a smaller model to reduce capability
Why wrong: Model size does not prevent injection; smaller models may be less capable but still vulnerable.
D
Add more few-shot examples with safe outputs
Why wrong: Examples do not prevent injection; they only show desired behavior.
E
Implement input validation and sanitization to remove adversarial patterns
Sanitizing input removes potential injection strings before they reach the model.

Question 8hardmulti select

A developer is troubleshooting low recall in a vector search. Which THREE factors should be checked? (Choose three.)

Trap 1: The number of results returned (k) in the search

While k affects recall, the question is about underlying factors causing low recall, not tuning parameters.

Trap 2: The LLM's temperature setting

Temperature affects generation randomness, not retrieval recall.

A
Embedding model quality and relevance to domain
A model not trained on similar data may produce poor embeddings.
B
Chunk size and overlap strategy
Improper chunking can cause important information to be missed or poorly represented.
C
Quality of the query embedding generation
If the query is not embedded correctly, the search will not align with semantic intent.
D
The number of results returned (k) in the search
Why wrong: While k affects recall, the question is about underlying factors causing low recall, not tuning parameters.
E
The LLM's temperature setting
Why wrong: Temperature affects generation randomness, not retrieval recall.