NCA-GENL Practice Test 27 — 15 Questions

Question 1

Which activation function is most commonly used in the hidden layers of deep neural networks to mitigate the vanishing gradient problem?

Accepted Answer

ReLU. The ReLU (Rectified Linear Unit) activation function is the industry standard for hidden layers because it avoids the saturation characteristic of sigmoid or tanh functions. By outputting zero for negative inputs and linear values for positive inputs, it maintains a gradient of one during backpropagation for active neurons. This allows gradients to flow through deep networks without shrinking exponentially, which is essential for training modern, deep architectures effectively.

Answer

Sigmoid

Answer

Softmax

Answer

Hyperbolic Tangent (tanh)

Question 2

An AI researcher is fine-tuning a large language model and wants to minimize GPU memory consumption during training without altering the model's primary weight representations or introducing quantization error during inference. Which technique provides this capability by decomposing weight matrices into low-rank trainable adaptation matrices?

Accepted Answer

Low-Rank Adaptation (LoRA) freezing pre-trained weights and training injected rank decomposition matrices.. Low-Rank Adaptation (LoRA) freezes the original pre-trained model weights and injects trainable rank decomposition matrices into the architecture. This drastically reduces the number of trainable parameters and optimizer states in GPU memory during backpropagation, while allowing weights to be merged back into the original matrices for zero-latency deployment.

Answer

Post-Training Quantization (PTQ) converting weights from FP16 to INT8 integer representations.

Answer

Full-Parameter Fine-Tuning updating all weights across every transformer layer simultaneously using FP32 precision.

Answer

Knowledge Distillation transferring learned representations from a large teacher model to a smaller student network architecture.

Question 3

A machine learning engineer is training a deep neural network for image classification. They notice that the training loss decreases steadily, but the validation loss starts to increase after a few epochs. Which technique is most directly aimed at addressing this issue?

Accepted Answer

Adding dropout layers. The described behavior—training loss decreasing while validation loss increases—is a classic sign of overfitting. Dropout is a regularization method that randomly drops units during training, which reduces the network's capacity to memorize noise and encourages more generalizable representations. Other options either do not address overfitting or would exacerbate the problem.

Answer

Increasing the learning rate

Answer

Removing batch normalization

Answer

Reducing the size of the training set

Question 4

In the context of transformer models, what is the purpose of the 'Attention Mask' during the training process?

Accepted Answer

It prevents the model from attending to future tokens in autoregressive models.. The attention mask is essential for transformer architectures to handle sequences of varying lengths and to enforce causal constraints. In tasks like sequence generation, it prevents the model from 'peeking' at future tokens, ensuring that the prediction at each position depends only on preceding tokens. This mechanism is critical for maintaining the autoregressive nature of models like GPT and ensuring correct model training.

Answer

It serves to reduce the number of parameters in the self-attention layer.

Answer

It optimizes the data movement between the GPU's L1 and L2 cache.

Answer

It performs weight pruning to compress the model size after training.

Question 5

An ML engineer is training a transformer-based language model on a single NVIDIA A100 GPU. They observe that the training loss decreases initially but then becomes NaN after a few hundred steps. The learning rate is 1e-4, and mixed precision with FP16 is enabled. Which action is most likely to stabilize training while preserving the benefits of mixed precision?

Accepted Answer

Enable dynamic loss scaling to automatically adjust the loss scale factor during training.. Dynamic loss scaling is the standard technique to prevent underflow and overflow in FP16 mixed precision training. It scales the loss up to keep gradients in representable range and automatically backs off when NaNs occur. This stabilizes training while retaining the performance and memory advantages of mixed precision, unlike switching to FP32 or altering hyperparameters.

Answer

Switch to FP32 training for all layers.

Answer

Reduce the batch size to 1 to minimize memory usage.

Answer

Increase the learning rate to 1e-3 to escape the NaN region faster.

Question 6

Refer to the exhibit. You are running a multi-node distributed fine-tuning experiment and receive this error. What does this indicate about your experimentation environment?

Accepted Answer

There is an issue with the cluster interconnect.. This NCCL (NVIDIA Collective Communications Library) error signifies a breakdown in inter-node communication. In distributed training, nodes must synchronize gradients; a reset connection implies that the network fabric is failing to maintain the persistent connections required for this synchronization. This is a critical infrastructure issue that prevents the experiment from proceeding, requiring a check of the interconnects, such as InfiniBand or Ethernet switches, before any further experimentation can occur.

Answer

The model weights are corrupted.

Answer

The learning rate is too high.

Answer

The batch size is too large.

Question 7

A developer is writing a Python client for an NVIDIA-hosted LLM endpoint and needs the model to answer every request in a strict JSON schema without extra prose. Where should the formatting contract be expressed so it applies consistently across all requests from the service?

Accepted Answer

In a system-role message describing the required schema and the rule against extra prose.. A service-wide formatting contract belongs in the system-role message, which the endpoint treats as standing instructions for the whole conversation and which the client can reuse on every call. User-turn reminders, assistant-role content, and custom headers either dilute the instruction, misrepresent the turn, or are simply ignored.

Answer

In the assistant-role message, since the assistant produces the JSON output.

Answer

In an HTTP header on the request, such as X-Response-Schema.

Answer

Appended to each user message as a trailing reminder sentence.

Question 8

Which approach is most effective for visualizing 'attention heads' in a Transformer model to debug why the model ignores specific information?

Accepted Answer

Visualization of attention head weights as heat maps.. Attention maps (often visualized as heat maps or dependency graphs) allow researchers to see which tokens the model focuses on during inference. By visualizing these weights, developers can identify if the model is failing to attend to critical context, which explains why it might ignore provided information in a RAG pipeline. This visibility is essential for understanding the internal logic of the model and fixing grounding issues in generative AI workflows.

Answer

Global average loss across the training run.

Answer

A bar chart of token frequencies in the dataset.

Answer

A histogram of the output sequence length.

Question 9

Which NVIDIA SDK is specifically optimized for high-performance deep learning inference and supports the deployment of quantized models?

Accepted Answer

NVIDIA TensorRT.. TensorRT is the industry-standard SDK for optimizing deep learning inference on NVIDIA GPUs. It provides advanced techniques like layer fusion, precision calibration (FP8, INT8), and kernel auto-tuning. For developers, mastering TensorRT is essential to transition from research code to high-speed, scalable production deployments, ensuring that models operate at peak efficiency while respecting the strict latency requirements of modern enterprise applications.

Answer

NVIDIA CUDA Toolkit.

Answer

NVIDIA cuDNN.

Answer

NVIDIA NCCL.

Question 10

An enterprise deployment of an LLM is exhibiting signs of hallucination where the model generates plausible but factually incorrect technical documentation. Which strategy is most effective for improving factual grounding within the NVIDIA NeMo framework?

Accepted Answer

Implement a RAG pipeline to inject context from a verified vector database.. Retrieval-Augmented Generation (RAG) is the standard architectural approach to ground LLMs by providing external, verified documentation at inference time. By fetching relevant chunks from a trusted vector database, the model reduces reliance on parametric memory, which is prone to hallucinations. This methodology is essential in high-stakes enterprise environments where accuracy is critical for compliance and operational reliability, ensuring the generated output remains strictly bound to the provided source material.

Answer

Increase the temperature parameter to 1.5 to maximize output diversity.

Answer

Fine-tune the model exclusively on a large, uncurated corpus of internet text.

Answer

Remove all system prompts to prevent model bias during the inference phase.

Question 11

In the experimentation loop, what is the role of a 'validation split' during model fine-tuning?

Accepted Answer

To provide an unbiased evaluation of generalization. A validation split is used to monitor performance on unseen data during the training process, providing a metric for generalization. Unlike the training set, which the model directly optimizes, the validation set acts as an objective check. This prevents developers from making decisions based on overfitting, ensuring that the model maintains its utility on real-world data and identifying when to stop the training process.

Answer

To increase the total amount of training data

Answer

To speed up the backpropagation process

Answer

To serve as the final test set for deployment

Question 12

In an experiment comparing different fine-tuning methods (LoRA vs. Full Fine-tuning), which metric is most useful for determining the efficiency of the experimentation process itself?

Accepted Answer

The compute hours required to reach a target validation score.. When comparing fine-tuning techniques, efficiency metrics like 'Time-to-convergence' or 'Compute-efficiency-per-epoch' are vital. These metrics quantify the resource cost of achieving a target accuracy, allowing researchers to choose the most cost-effective approach for their specific hardware. This is essential in an industrial setting where GPU hours and time-to-market are significant constraints for project viability.

Answer

The total number of parameters in the model.

Answer

The final training loss at the end of the experiment.

Answer

The frequency of the model checkpointing during the run.

Question 13

A data science team is building a model to predict whether a customer will churn based on historical account activity. They have a large dataset with labeled outcomes (churned or not churned). Which type of machine learning is most appropriate for this task?

Accepted Answer

Supervised learning with a classification algorithm. The task is to predict a binary outcome (churn or no churn) using historical data where the outcome is known. This is a classic supervised learning problem, and classification algorithms are designed to learn from labeled examples to predict discrete categories. Unsupervised, reinforcement, and semi-supervised methods do not directly leverage the available labels for prediction.

Answer

Unsupervised learning with clustering

Answer

Reinforcement learning with a reward function

Answer

Semi-supervised learning with a small labeled set

Question 14

A team's NeMo fine-tuning experiment runs on a fixed compute budget and they must choose how to allocate it between searching hyperparameters and training the final model. Their hyperparameter search space is large and each trial is expensive. Which allocation strategy best balances finding a strong configuration against producing a well-trained final model?

Accepted Answer

Use a budget-aware search such as successive halving or Bayesian optimization, then spend the remaining budget training the best configuration to completion.. With a fixed budget and expensive trials, the efficient path is to let early results prune weak candidates and concentrate resources on promising ones, then commit remaining budget to fully training the winner. This avoids both the waste of exhaustive grid search and the risk of delivering an under-trained final model.

Answer

Spend the entire budget on a dense grid search over every hyperparameter combination.

Answer

Run many short trials with identical settings to reduce measurement noise, then pick any configuration.

Answer

Train one configuration to completion and skip hyperparameter search entirely.

Question 15

Which TWO of the following are primary benefits of using NVIDIA Triton Inference Server for deploying generative AI models?

Accepted Answer

It provides native support for dynamic batching of inference requests.. NVIDIA Triton simplifies the deployment of complex AI models by supporting multiple frameworks and optimizing inference. Its ability to handle concurrent model execution and dynamic batching is critical for maximizing GPU utilization in production environments. Mastering these features allows developers to build scalable, high-performance pipelines that serve multiple models efficiently without needing custom backend infrastructure for every specific model architecture.

Answer

It enforces a strict requirement to use only PyTorch-based model backends.

Answer

It automatically converts all models to the proprietary CUDA-Graph format.

Answer

It removes the need for any containerization technology.