Free NVIDIA-Certified Associate Generative AI LLMs Certification Exam Practice Questions

Sample Practice Questions

Try these 10 sample questions from the NVIDIA-Certified Associate Generative AI LLMs Certification Exam practice bank. Each includes the answer and explanation. Continue with the interactive practice above to review the full question set.

1. 1A model does not support batching in Triton. What value must its max_batch_size configuration use?

   1. A.0Correct answer: A
   2. B.1
   3. C.4
   4. D.8

   Explanation: The guide states: "For models that do not support batching, or do not support batching in the specific ways described above, max_batch_size must be set to zero." The model does not support batching, so zero is required.

   Reference: Triton Inference Server User Guide: Model Configuration — Maximum Batch Size

2. 2In the standard k-means iteration, how are the new centroids calculated after assigning samples?

   1. A.From the farthest sample in each cluster
   2. B.From the mean of samples assigned to each centroidCorrect answer: B
   3. C.From the median of all cluster centroids
   4. D.From the variance of the entire dataset

   Explanation: The documentation says: "The second step creates new centroids by taking the mean value of all of the samples assigned to each previous centroid." This update follows assignment to the nearest centroid.

   Reference: scikit-learn User Guide: Clustering — 2.3.2. K-means

3. 3In the Keras EarlyStopping callback, what does patience control during training?

   1. A.The number of epochs before monitoring begins
   2. B.The number of epochs without improvement before stoppingCorrect answer: B
   3. C.The minimum change required to count as improvement
   4. D.The number of epochs to pause monitoring after an improvement

   Explanation: Keras defines patience as the "Number of epochs with no improvement after which training will be stopped." It controls waiting before stopping.

   Reference: Keras 3 API: EarlyStopping — EarlyStopping class > Arguments

4. 4A deployed Triton model is stateless and receives many independent inference requests. What does the dynamic batcher do?

   1. A.Combines requests into batches dynamicallyCorrect answer: A
   2. B.Routes every request to a designated instance to preserve order
   3. C.Combines requests only when their input text is identical
   4. D.Queues requests and sends them one at a time without combining them

   Explanation: The guide states: "Dynamic batching is a feature of Triton that allows inference requests to be combined by the server". It should be used for stateless models.

   Reference: Triton Inference Server User Guide: Batchers — Dynamic Batcher

5. 5How does the NIST AI RMF describe transparency for an AI system?

   1. A.Availability of information about the system and its outputsCorrect answer: A
   2. B.The meaning of outputs in the context of their intended use
   3. C.The mechanisms used to produce the system's outputs
   4. D.The ability to maintain performance under changing conditions

   Explanation: NIST states: "Transparency reflects the extent to which information about an AI system and its outputs is available to individuals interacting with such a system". Transparency alone does not guarantee accuracy or fairness.

   Reference: NIST AI Risk Management Framework 1.0 — 3.4 Accountable and Transparent, p. 15

6. 6In Triton, where can instance_group be configured to support parallel execution for an ensemble?

   1. A.On the model repository root
   2. B.On each composing modelCorrect answer: B
   3. C.On the ensemble model itself
   4. D.On each inference request

   Explanation: The guide states: "each composing model that makes up an ensemble can specify instance_group in its config file and individually support parallel execution". The ensemble itself has no physical instance and cannot specify this field.

   Reference: Triton Inference Server User Guide: Model Configuration — Ensemble Model Instance Groups

7. 7Why might an analyst run k-means several times with different centroid initializations?

   1. A.Because initialization affects only labels rather than cluster assignments
   2. B.Because the first solution is necessarily the global minimum
   3. C.Because convergence can depend on initialization and reach a local minimumCorrect answer: C
   4. D.Because repeated starts determine the number of clusters

   Explanation: The guide says: "K-means will always converge, however this may be to a local minimum. This is highly dependent on the initialization of the centroids." Different initializations address that sensitivity.

   Reference: scikit-learn User Guide: Clustering — 2.3.2. K-means

8. 8In Keras EarlyStopping, what does restore_best_weights=True select when training ends?

   1. A.Weights from the epoch with the best monitored quantityCorrect answer: A
   2. B.Weights from the final epoch regardless of the monitored value
   3. C.Weights from the first epoch regardless of the monitored value
   4. D.Weights from the epoch with the largest change in the monitored value

   Explanation: The reference defines restore_best_weights as whether to "restore model weights from the epoch with the best value of the monitored quantity." The default is False.

   Reference: Keras 3 API: EarlyStopping — EarlyStopping class > Arguments

9. 9A stateful Triton model needs all requests in each sequence routed to the same model instance. Which configuration meets this requirement?

   1. A.Dynamic batching alone that preserves affinity for every sequence
   2. B.Sequence batching that routes each sequence to the same instanceCorrect answer: B
   3. C.Sequence batching that routes each request to any instance
   4. D.Dynamic batching that groups independent requests without sequence routing

   Explanation: The guide states that "the sequence batcher should be used for stateful models where a sequence of inference requests must be routed to the same model instance." This condition matches the deployment.

   Reference: Triton Inference Server User Guide: Batchers — Sequence Batcher

10. 10In the NIST AI RMF, what does explainability describe?

    1. A.The availability of information about the system and its outputs
    2. B.The mechanisms underlying an AI system's operationCorrect answer: B
    3. C.The meaning of system output in its functional context
    4. D.The ability to maintain performance under varied circumstances

    Explanation: NIST defines explainability as "a representation of the mechanisms underlying AI systems’ operation". It distinguishes this from interpretability of output in context.

    Reference: NIST AI Risk Management Framework 1.0 — 3.5 Explainable and Interpretable, p. 16