NCA-GENL Practice Test 16 — 15 Questions

Question 1

A machine learning engineer is training a large language model on a cluster of NVIDIA GPUs. During training, she observes that the loss occasionally spikes to NaN, causing the training to fail. She suspects that the issue is related to the numerical precision of the computations. Which technique is most appropriate to mitigate this issue while maintaining training stability?

Choose an answer to begin — your selection is scored in the full session.

Accepted Answer

Use gradient clipping. Gradient clipping is a widely used technique to prevent exploding gradients, which are a common cause of NaN losses in deep learning, especially when training large models with mixed precision. By capping gradient norms, it ensures that parameter updates remain stable, allowing training to proceed without numerical overflow.

Answer

Switch to double precision

Answer

Increase the learning rate

Answer

Reduce the batch size

Question 2

A team is preparing a dashboard to monitor an LLM inference service in production. They want visualizations that surface latency problems and resource saturation before users are affected. Which two visualizations are most appropriate for this goal? (Choose two.)

Accepted Answer

A gauge or time-series chart of GPU memory utilization and KV cache occupancy. Effective production monitoring pairs a latency view with a resource view. Percentile time series expose tail latency changes that averages conceal, while GPU memory and KV cache occupancy show the underlying pressure that causes those changes. Together they let the team act on leading indicators. Traffic-share pies, token word clouds, and input-output scatter plots describe workload characteristics rather than service health, so they do not support preemptive detection of degradation.

Answer

A word cloud of the most frequent prompt tokens received today

Answer

A scatter plot of prompt length versus generated response length for a random sample

Answer

A pie chart of total requests grouped by model version for the current day

Question 3

A team is analyzing the latency of an LLM inference service. They have per-request latency data for 10,000 requests and want to visualize the distribution to identify whether there is a long tail that could violate a service-level objective. Which visualization is most appropriate for this purpose?

Accepted Answer

A histogram of request latencies with a logarithmic x-axis.. To detect a long tail in latency, the full distribution must be visualized. A histogram with a logarithmic x-axis shows all latencies and keeps extreme values visible, making tail behavior clear. The other options aggregate away the distribution, focus on the median, or examine a different relationship, none of which reveal tail latency.

Answer

A scatter plot of request latency versus request payload size.

Answer

A line chart of the 50th percentile latency over time.

Answer

A bar chart of average latency per hour over the data collection period.

Question 4

A machine learning engineer is evaluating a large language model (LLM) on a text generation task. They observe that the model produces coherent and fluent sentences, but the content is factually incorrect and sometimes contradicts known facts. Which term best describes this phenomenon?

Accepted Answer

Hallucination. The phenomenon described—fluent but factually incorrect text generation—is known as hallucination. It occurs because LLMs are trained to predict likely sequences of words, not to verify facts. Hallucination is a significant challenge for deploying LLMs in applications requiring factual accuracy, and it can be mitigated with retrieval-augmented generation or factual consistency checks.

Answer

Catastrophic forgetting

Answer

Overfitting

Answer

Mode collapse

Question 5

A data scientist is analyzing token length distribution across a 12-million-document pretraining corpus destined for an NVIDIA NCA-GENL pipeline. The histogram is heavily right-skewed with a long tail beyond 8,192 tokens. Which visualization should be produced NEXT to decide a safe max_sequence_length without discarding most of the corpus?

Accepted Answer

A cumulative distribution function (CDF) plot of token lengths with a vertical marker at each candidate max_sequence_length.. Choosing max_sequence_length requires knowing the fraction of the corpus that would be truncated at each candidate value. The cumulative distribution function of token lengths gives that fraction directly, so markers at 2,048, 4,096, and 8,192 tokens reveal the exact data-loss cost of each setting. The other charts describe shape, spread, or content but never quantify cumulative truncation.

Answer

A pie chart of documents bucketed into 1K-token bins.

Answer

A word cloud of the most frequent tokens in the longest documents.

Answer

A box plot of token lengths grouped by document source.

Question 6

A research team is training a transformer-based language model and wants to reduce the computational cost of the self-attention mechanism for very long input sequences. They are considering replacing the standard scaled dot-product attention with an approximation. Which statement accurately describes a trade-off of using an approximate attention method?

Accepted Answer

It reduces memory and compute complexity but may sacrifice some model quality compared to full attention.. Standard self-attention scales quadratically with sequence length, which is costly for long inputs. Approximate attention methods reduce this complexity to near-linear or sub-quadratic by sparsifying or factorizing the attention matrix. The trade-off is that the approximation may not capture all token interactions, potentially lowering model quality or requiring additional training. This balance between efficiency and accuracy is the central consideration.

Answer

It eliminates the need for positional encodings because approximate attention is order-invariant.

Answer

It increases the number of parameters in the model, which always improves accuracy.

Answer

It makes the model immune to the vanishing gradient problem in deep transformer stacks.

Question 7

A hospital's AI governance committee is reviewing a generative model that drafts discharge summaries. They require a documented, auditable record showing which source documents, consent forms, and preprocessing steps produced each training example. Which Trustworthy AI practice does this requirement describe?

Accepted Answer

Data lineage tracking. The committee wants to trace each training example back to its source documents, consent forms, and preprocessing operations, which is precisely what data lineage tracking captures and preserves for audit. Privacy-enhancing techniques such as differential privacy or federated learning change how data is used but do not document its origin, and quantization is purely an inference optimization. Only lineage tracking yields the traceable, reviewable record the governance process demands.

Answer

Federated learning

Answer

Differential privacy

Answer

Model quantization

Question 8

During a fine-tuning run you observe that the training loss decreases smoothly, but validation loss begins rising after epoch 3. You want a single visualization that makes this divergence and the resulting overfitting point immediately obvious to reviewers. Which plot should you produce?

Accepted Answer

A line chart of training and validation loss versus epoch on the same axes. Overfitting is a temporal phenomenon: the two loss curves move together, then separate. A line chart with epoch on the x-axis and both curves on the y-axis preserves that trajectory and makes the inflection where validation loss turns upward easy to locate. Aggregating to a single epoch, plotting noisy batch points, or collapsing across epochs all remove the temporal evidence needed to identify the divergence.

Answer

A histogram of validation loss values across all epochs

Answer

A bar chart comparing final training loss and final validation loss at the last epoch

Answer

A scatter plot of individual batch losses colored by epoch

Question 9

A healthcare company is deploying an LLM to assist with clinical documentation. To ensure Trustworthy AI, they must implement measures to detect and mitigate hallucinations that could lead to incorrect medical records. Which two actions should they take? (Choose two.)

Accepted Answer

Deploy a fact-checking module that cross-references generated statements against a trusted medical database.. To mitigate hallucinations in clinical documentation, grounding responses in verified guidelines via RAG and adding a fact-checking module against a trusted database are effective. These actions ensure outputs are based on authoritative sources and verified before use. Other options like larger models, fine-tuning, or higher temperature do not reliably reduce hallucinations and may introduce new risks.

Answer

Set the model's temperature to a higher value to encourage more diverse outputs.

Answer

Use a larger LLM with more parameters to improve factual accuracy.

Answer

Fine-tune the LLM on the company's historical clinical notes.

Question 10

A team is building a dashboard to monitor an LLM evaluation pipeline that scores model outputs against a reference dataset. They want the dashboard to support rapid diagnosis when a new model checkpoint regresses. Which TWO visualization practices best support that goal? (Choose two.)

Accepted Answer

Plot the metric distribution for the new checkpoint against the previous checkpoint on the same axis, with confidence intervals.. Rapid regression diagnosis requires two things: seeing whether a change exceeds noise, and knowing where the change occurred. Overlaying checkpoint distributions with confidence intervals addresses the first by separating real shifts from variation, while a per-slice breakdown addresses the second by localizing the degradation to a category or prompt type. Aggregate tiles, raw output tables, and 3D scatter plots either hide the distribution, overwhelm the viewer, or distort comparison.

Answer

Display raw model outputs for every test example in a single scrollable table as the primary view.

Answer

Use a 3D rotating scatter plot of all metrics to maximize the amount of information shown at once.

Answer

Show only the single aggregate score per checkpoint in a large numeric tile.

Question 11

A team is building a dashboard to monitor an LLM training run and wants to detect data-quality problems early. They have access to per-batch training loss, per-batch gradient norm, input sequence length statistics, and token frequency counts. Which two visualizations are most appropriate for surfacing data-quality issues rather than hardware or throughput issues? (Choose two.)

Accepted Answer

A line chart of per-batch training loss over steps. Per-batch training loss and input sequence length histograms both reflect the model's interaction with the data. Loss spikes or shifts can indicate corrupted or mislabeled batches, while sequence-length histograms reveal truncation or padding anomalies. GPU utilization, memory usage, and network throughput are infrastructure metrics that describe hardware behavior, not data correctness, so they do not surface data-quality issues.

Answer

A line chart of network throughput between nodes

Answer

A line chart of GPU utilization over time

Answer

A bar chart of memory usage per GPU

Question 12

A financial services company is deploying an NVIDIA NIM microservice that answers questions about internal loan policies. Compliance requires that every response be traceable to the exact source paragraph, and that unsupported claims never reach the user. Which approach best enforces this requirement at inference time?

Accepted Answer

Enable retrieval-augmented generation with citations and reject any answer whose claims are not grounded in the retrieved passages.. Traceability to an exact source paragraph and prevention of unsupported claims require grounding each answer in retrieved documents and validating that grounding before the response is released. Retrieval-augmented generation with citations supplies the provenance, while a grounding check enforces it. The other approaches either increase variability, embed knowledge without citations, or only record outputs after the fact, none of which meet both compliance conditions simultaneously.

Answer

Fine-tune the base LLM on the full internal loan policy corpus and rely on the fine-tuned weights to reproduce policy text accurately.

Answer

Deploy the model behind an API gateway that logs every prompt and response, then have compliance staff review the logs weekly.

Answer

Increase the model's temperature so that responses draw on a wider range of internal knowledge, improving coverage of loan policy topics.

Question 13

A financial services company is deploying an NVIDIA NIM microservice for a customer-facing loan advisory chatbot. The compliance team requires that every response be traceable to a verified source document, and that any response not grounded in those documents be suppressed. Which approach best satisfies this requirement?

Accepted Answer

Implement retrieval-augmented generation with NeMo Guardrails to enforce grounding and block unverified responses.. The requirement is twofold: responses must be traceable to verified documents, and ungrounded responses must be suppressed. Retrieval-augmented generation provides the grounding by injecting verified source content into the prompt, while NeMo Guardrails enforces runtime output rails that can block or rewrite unsupported answers. This combination directly delivers auditable, source-anchored behavior for a regulated advisory use case.

Answer

Fine-tune the base model on the company's historical loan documents and deploy it without additional runtime controls.

Answer

Enable NVIDIA TensorRT-LLM quantization to reduce latency so agents can manually review every response.

Answer

Increase the model temperature to encourage more creative and comprehensive answers.

Question 14

A financial services company is deploying an NVIDIA NIM microservice for an internal LLM assistant that summarizes earnings call transcripts. The security team wants to ensure that the model cannot be coerced via prompt injection into revealing confidential merger discussions embedded in prior context. Which NVIDIA-developed safety mechanism should be integrated directly into the inference pipeline to evaluate prompts and responses against a defined policy at runtime?

Accepted Answer

NVIDIA NeMo Guardrails. NeMo Guardrails is purpose-built to enforce safety policies at runtime by intercepting LLM inputs and outputs. It can detect and block prompt injection attempts that try to extract confidential context, which is exactly the risk described. The other options are performance or preprocessing tools that do not evaluate content against a safety policy, so they cannot prevent the leakage scenario.

Answer

NVIDIA Triton Inference Server model ensembles

Answer

NVIDIA TensorRT-LLM quantization

Answer

NVIDIA DALI preprocessing pipelines

Question 15

A financial services company has deployed an NVIDIA NIM microservice hosting a Llama 3 70B model for internal document summarization. The security team wants to ensure that the model's outputs cannot be used to exfiltrate sensitive customer data that may have been memorized during pretraining. Which NVIDIA AI Enterprise feature should be implemented to detect and filter such memorized content in real time?

Accepted Answer

NVIDIA NeMo Guardrails with a custom output rail using a sensitivity classifier. NeMo Guardrails is purpose-built for adding programmable guardrails to LLM applications, including output filtering. A custom output rail can run a sensitivity classifier to detect and block memorized sensitive data before it is returned to the user. This provides a real-time, configurable safeguard that directly addresses the risk of data exfiltration from a deployed NIM microservice.

Answer

NVIDIA Triton Inference Server with dynamic batching and model ensemble

Answer

NVIDIA Morpheus with a pre-trained sensitive information detection model

Answer

NVIDIA TensorRT-LLM with INT8 quantization and kernel fusion