Free NCA-AMAD Practice Test: 600 Questions for the NVIDIA AI Model and Application Development Exam

Correct answer: C. The agent is goal-directed and plans and executes multi-step tasks using tools

C is correct because traditional generative AI apps follow a request-and-respond pattern, while agents coordinate models with external tools to plan and carry out multi-step work toward a goal. A is incorrect because agents are defined by their use of tools, APIs and data sources, the opposite of answering from training data alone. B is incorrect because a fixed decision tree describes a rule-based system; agents reason and choose their own steps. D is incorrect because model size does not define an agent; agents can run on small open models or frontier models alike.

Correct answer: D. An agent harness

D is correct because NVIDIA describes the harness as the scaffolding that gives the LLM the ability to act: it connects short- and long-term memory, knows which tools are available and can create new skills if allowed. A is incorrect because the secure runtime is the policy-enforced sandbox in which the agent executes, and the stem says the prototype already has one. B is incorrect because the LLM already reasons well; a bigger model still could not reach memory or tools without the scaffolding around it. C is incorrect because a planning module breaks complex tasks into steps; it does not connect the LLM to memory or tools.

Correct answer: B. Replaces each category with a target statistic computed for that category

B is correct because target encoding is a groupby aggregation where the aggregated column is the target, such as the mean price for each brand, so a category becomes a number the model can use. A is incorrect because that is binning, which reduces the precision of a numeric column rather than encoding a category with the target. C is incorrect because that describes one-hot encoding, which creates one column per category and never looks at the target. D is incorrect because that is standardization, a scaling step for numeric features that uses neither the target nor the categories.

Correct answer: A. Exploratory data analysis

A is correct because exploratory data analysis (EDA) examines the data's quality, distributions and relationships before modeling, and these checks are its basic steps. B is incorrect because ensembling combines the predictions of several trained models, such as by hill climbing, so it needs trained models first. C is incorrect because hyperparameter tuning searches model settings after a model and a validation scheme already exist. D is incorrect because inference optimization speeds up a deployed model and does not examine the training data.

Correct answer: C. A distributed layer for disaggregated serving, KV-aware routing and autoscaling

C is correct because Dynamo leaves the single-GPU forward pass to the engine and adds the system layer around it: disaggregated prefill and decode, KV cache-aware routing, KV cache offloading across memory tiers and SLA-based autoscaling. A is incorrect because post-training quantization is done with TensorRT Model Optimizer; Dynamo serves models but does not quantize them. B is incorrect because Dynamo is an inference serving framework; distributed training is the job of tools such as Megatron-LM and the NeMo framework. D is incorrect because Dynamo is engine-agnostic and runs on top of vLLM, SGLang and TensorRT-LLM rather than replacing them.

Correct answer: C. SLA-based planning that scales prefill and decode workers from live traffic

C is correct because Dynamo's planner takes TTFT and time-per-output-token targets, watches live traffic and adds or removes prefill and decode workers, so capacity follows peaks that a static offline configuration cannot handle. A is incorrect because routing improves cache hits among the existing workers; it cannot add capacity when every worker is saturated. B is incorrect because offloading enlarges the cache to avoid recomputing prompts; it does not change how many workers serve the load. D is incorrect because an offline estimate is what produced the static setup; it does not react to traffic once the model is in production.

Correct answer: B. It avoids the extra inference latency of managing a separate adapter

B is correct because merging adds the low-rank update into the pretrained weights, giving a model that is structurally the same as the base model, so no separate adapter has to be handled at inference time. A is incorrect because the merged weights have the same shape and size as the original weight matrices; merging adds no compression. C is incorrect because once the adapter is merged it is part of the weights; swapping adapters at runtime is what dynamic loading provides. D is incorrect because a merged model becomes bespoke and serves only the task it was tuned for; mixed-task batching is the benefit of loading adapters dynamically.

Correct answer: A. One NIM with the base model that loads the right adapter for each request

A is correct because dynamic LoRA serving keeps one copy of the base model and applies the requested adapter per request, so requests for different tasks can share batches without hosting separate custom models. B is incorrect because three merged models triple the base-model memory and cannot batch requests across tasks; merging is recommended only when a deployment serves a single task. C is incorrect because merging produces a model specialized for one task; adding three task updates into the same weights does not keep each task's behavior separate. D is incorrect because this discards the fine-tuning the team already did; the adapters exist because prompting the base model was not accurate enough.

Correct answer: A. It learns features from raw data through many stacked layers

A is correct because deep learning stacks layers of nonlinear transformations that learn features such as edges and shapes directly from raw data, so the features do not have to be engineered by hand first. B is incorrect because classical models such as gradient-boosted trees often match or beat neural networks on small tabular datasets. C is incorrect because neural networks are commonly trained on labeled data; whether a task is supervised is a choice of problem, not what separates them from classical models. D is incorrect because a neural network's behavior comes from weights adjusted during training, not from fixed hand-written rules.

Correct answer: D. Backpropagation passes the error back through the layers to adjust weights

D is correct because with backpropagation inside gradient descent, the error between prediction and target flows back through the layers and each weight is adjusted to reduce it, repeated over many iterations. A is incorrect because the architecture is fixed before training; training changes the weights, not the number of layers. B is incorrect because hand-written rules describe rule-based systems; a neural network learns by updating its weights from data. C is incorrect because fitting trees to residual errors describes gradient boosting, a classical ensemble method, not neural network training.

Correct answer: B. It downloads model files ahead of time and keeps them on a persistent volume

B is correct because a NIMCache pulls the model profiles from NGC or another source and persists them on a PVC, so NIM pods can start from the cached copy instead of downloading at startup. A is incorrect because scaling pods is the job of a horizontal pod autoscaler, which the NIMService resource can create. C is incorrect because the cache stores model files on a volume; it does not build container images. D is incorrect because request distribution is handled by the Kubernetes service or an ingress, not by the cache resource.

Correct answer: D. A packaged, configurable set of Kubernetes manifests that installs with one command

D is correct because Helm bundles the Deployments, Services and other Kubernetes objects an application needs, with values the team can override, so the whole stack installs and upgrades as one release. A is incorrect because the container image is what the chart deploys; the chart itself is a set of templated Kubernetes manifests. B is incorrect because GPU drivers are managed separately, for example by the GPU Operator; a Helm chart describes application resources. C is incorrect because performance figures come from tools such as genai-perf, not from the deployment package.

Correct answer: C. A Jetson AGX Thor onboard computer

C is correct because the third computer is the robot's own runtime computer, Jetson AGX Thor, which performs inference on the robot's sensor data. A is incorrect because a cloud instance is not mounted on the robot and cannot provide real-time onboard inference. B is incorrect because Omniverse and Cosmos on RTX PRO Servers form the second computer, used for simulation and synthetic data. D is incorrect because DGX is the first computer, used to train the robot's AI models.

Correct answer: B. Physical AI acts in the physical world; agentic AI works in digital environments

B is correct because NVIDIA describes physical AI as end-to-end models that perceive, reason, interact with and navigate the physical world, whereas agentic AI operates in digital environments. A is incorrect because both kinds of AI run on accelerated GPU computing; the difference is where the system acts, not the processor type. C is incorrect because image and video generation are still digital outputs; physical AI is defined by perceiving and acting in the real world. D is incorrect because physical AI needs large, diverse training data too, which is exactly why it relies on simulation and synthetic data.

Correct answer: D. It evaluates real usage data and triggers retraining when quality slips

D is correct because the flywheel captures real-world usage, evaluates model performance and triggers retraining or fine-tuning as needed, which keeps the model aligned with current users and context. A is incorrect because a calendar swap ignores whether quality actually dropped and skips evaluation on the team's own data. B is incorrect because frozen weights are exactly what lets a model drift out of date as data and user behavior change. C is incorrect because temperature only changes output randomness; it does not teach the model anything new.

Correct answer: D. NeMo Datastore

D is correct because NeMo Datastore handles the actual files behind datasets and models, including uploading, downloading and versioning, while NeMo Entity Store tracks the entities themselves. A is incorrect because NeMo Retriever extracts, embeds and reranks content for retrieval; it is not a dataset and model file store. B is incorrect because NeMo Guardrails enforces safety and topic policies on inputs and outputs; it does not store or version files. C is incorrect because NeMo Evaluator runs benchmarks and custom evaluations; it reads datasets but does not manage their versions.