NVIDIA-Certified Associate: Generative AI LLMs

Exam practice questions

15 / 140 questions

Free preview · Answers and explanations includedPage 1 of 2

Data AnalysisNCA-GENL practice

Most documents in a corpus are short, but a small number are extremely long. The team wants to describe a typical document and estimate the long-document processing burden. Which summary is most informative?

Select one answerAReport the median together with upper length percentilesBReport only the minimum document lengthCReport the total document count as if it were a typical lengthDReport only the alphabetical order of file names

ExperimentationNCA-GENL practice

During fine-tuning, the loss repeatedly jumps after large parameter updates. Data and batch configuration are unchanged, and the team suspects the optimization step is too aggressive. Which first experiment directly tests that hypothesis?

Select one answerAReplace the validation set with the training setBIncrease the learning rate substantially without changing any other settingCIncrease the number of output categories unrelated to the taskDReduce the learning rate and compare the training behavior

Trustworthy AINCA-GENL practice

A team must explain which model release produced an assessment and what evidence supported its approval for use. Which TWO records support this audit need? Select TWO.

Select two answersAOnly the number of users who opened the applicationBOnly a generic statement that the model is modernCModel and deployment-version provenanceDEvaluation results tied to the approved use case and release

Software DevelopmentNCA-GENL practice

A team enables dynamic batching for an interactive inference service. Which TWO actions help improve capacity while preserving its response-time requirement? Select TWO.

Select two answersAAllow requests to wait indefinitely for the largest possible batchBValidate only one isolated request with no competing trafficCSet an appropriate upper bound on time spent waiting to form batchesDLoad-test the configuration at representative request concurrency

ExperimentationNCA-GENL practice

A fine-tuning run continues to reduce training loss after validation loss has begun to rise. Which TWO practices help select a model that generalizes better? Select TWO.

Select two answersAAlways choose the final checkpoint regardless of validation resultsBRetain the checkpoint with the best relevant validation performanceCMonitor performance on a held-out validation set during trainingDMerge validation labels into the training batches after each evaluation

Data AnalysisNCA-GENL practice

Before analyzing a newly collected support dataset, an analyst notices that some rows may be incomplete and others may have been imported twice. Which first step best establishes the condition of the data?

Select one answerAProfile missing values and duplicate records using documented checksBTrain the largest available language model before inspecting any rowsCDelete every row containing a short message without further analysisDAssume the ingestion job guarantees that all records are complete

Core Machine Learning and AI KnowledgeNCA-GENL practice

A product team is testing a streaming assistant. Users complain about waiting before any text appears, although the complete answer finishes within the total response-time target. Which measurement most directly investigates their complaint?

Select one answerAAverage document embedding timeBTime to first tokenCTraining steps per secondDTotal output token count

ExperimentationNCA-GENL practice

A team wants to adapt an LLM to domain-specific writing while limiting the memory required for trainable parameters and optimizer state. Which technique is designed for this purpose?

Select one answerAFull fine-tuning of every model parameter with separate optimizer stateBIncreasing the context window without changing the training methodCLoRA, which learns low-rank parameter updates while freezing the base weightsDDuplicating the complete model once for every training example

Trustworthy AINCA-GENL practice

An organization uses an AI assistant to help reviewers assess job applications. The organization remains responsible for hiring decisions. Which design best supports that accountability?

Select one answerAAssume a high overall accuracy score removes the need for oversightBRemove records of how the assistant was used in each assessmentCAssign responsible human reviewers, document the system's role, and provide a process for reviewing decisionsDTreat the model's output as an unchallengeable final decision

Core Machine Learning and AI KnowledgeNCA-GENL practice

A support assistant must complete at least 99% of requests successfully and keep slow responses within an agreed limit. Which TWO measurements directly assess these service requirements? Select TWO.

Select two answersANumber of training epochsBHigh-percentile response latencyCModel vocabulary sizeDRequest failure rate

Software DevelopmentNCA-GENL practice

A platform team needs a model-serving system with support for multiple inference backends and a managed model repository. Which NVIDIA component is designed for this role?

Select one answerAA model-training loop without a serving interfaceBNVIDIA Triton Inference ServerCNVIDIA TensorRT used only to build an inference engineDNVIDIA NeMo Curator used only for dataset preparation

Core Machine Learning and AI KnowledgeNCA-GENL practice

An analyst has already assigned a reviewed error category to each failed assistant response. The lead asks which three categories account for the largest number of failures this week. What should the analyst do next?

Select one answerASelect the three categories with the longest descriptionsBCompute a new embedding for each category name and rank vector lengthsCFine-tune the assistant using the weekly failure records immediatelyDGroup the weekly records by category and count failures in each group

Core Machine Learning and AI KnowledgeNCA-GENL practice

A chatbot works well for one tester but becomes slow when many employees connect. The team wants evidence about capacity before expanding access. Which evaluation is most useful?

Select one answerAMeasure one short request repeatedly with no competing trafficBLoad-test representative prompts at increasing concurrent request levelsCCompare the number of parameters in each candidate modelDEvaluate training accuracy using the deployment dataset

Core Machine Learning and AI KnowledgeNCA-GENL practice

A support team has thousands of unlabeled messages and wants to discover common themes before defining routing categories. Which approach is a reasonable first analysis?

Select one answerACluster message embeddings and inspect representative messages from each clusterBPredict the next token of each message and treat it as its categoryCTrain a supervised classifier using randomly assigned category labelsDSort messages only by their original arrival timestamp

Software DevelopmentNCA-GENL practice

A developer has a trained model and wants to optimize its inference execution on supported NVIDIA GPUs. Which component is primarily intended for building optimized inference engines?

Select one answerANVIDIA NeMo Curator for cleaning training datasetsBNVIDIA Triton scheduling without model-engine optimizationCNVIDIA TensorRTDA vector database for indexing document embeddings

Complete your exam preparation

Continue with the full question bank

Practise free with answers, explanations, and Community discussions. Join Community to unlock the complete question bank and timed mock exams.

Join Community Choose access for your certification

NVIDIA-Certified Associate: Generative AI LLMs

Exam practice questions

15 / 140 questions

Free preview · Answers and explanations includedPage 1 of 2

Data AnalysisNCA-GENL practice

Most documents in a corpus are short, but a small number are extremely long. The team wants to describe a typical document and estimate the long-document processing burden. Which summary is most informative?

ExperimentationNCA-GENL practice

During fine-tuning, the loss repeatedly jumps after large parameter updates. Data and batch configuration are unchanged, and the team suspects the optimization step is too aggressive. Which first experiment directly tests that hypothesis?

Trustworthy AINCA-GENL practice

A team must explain which model release produced an assessment and what evidence supported its approval for use. Which TWO records support this audit need? Select TWO.

Software DevelopmentNCA-GENL practice

A team enables dynamic batching for an interactive inference service. Which TWO actions help improve capacity while preserving its response-time requirement? Select TWO.

ExperimentationNCA-GENL practice

A fine-tuning run continues to reduce training loss after validation loss has begun to rise. Which TWO practices help select a model that generalizes better? Select TWO.

Data AnalysisNCA-GENL practice

Before analyzing a newly collected support dataset, an analyst notices that some rows may be incomplete and others may have been imported twice. Which first step best establishes the condition of the data?

Core Machine Learning and AI KnowledgeNCA-GENL practice

A product team is testing a streaming assistant. Users complain about waiting before any text appears, although the complete answer finishes within the total response-time target. Which measurement most directly investigates their complaint?

Select one answerAAverage document embedding timeBTime to first tokenCTraining steps per secondDTotal output token count

ExperimentationNCA-GENL practice

A team wants to adapt an LLM to domain-specific writing while limiting the memory required for trainable parameters and optimizer state. Which technique is designed for this purpose?

Trustworthy AINCA-GENL practice

An organization uses an AI assistant to help reviewers assess job applications. The organization remains responsible for hiring decisions. Which design best supports that accountability?

Core Machine Learning and AI KnowledgeNCA-GENL practice

A support assistant must complete at least 99% of requests successfully and keep slow responses within an agreed limit. Which TWO measurements directly assess these service requirements? Select TWO.

Select two answersANumber of training epochsBHigh-percentile response latencyCModel vocabulary sizeDRequest failure rate

Software DevelopmentNCA-GENL practice

A platform team needs a model-serving system with support for multiple inference backends and a managed model repository. Which NVIDIA component is designed for this role?

Core Machine Learning and AI KnowledgeNCA-GENL practice

An analyst has already assigned a reviewed error category to each failed assistant response. The lead asks which three categories account for the largest number of failures this week. What should the analyst do next?

Core Machine Learning and AI KnowledgeNCA-GENL practice

A chatbot works well for one tester but becomes slow when many employees connect. The team wants evidence about capacity before expanding access. Which evaluation is most useful?

Core Machine Learning and AI KnowledgeNCA-GENL practice

A support team has thousands of unlabeled messages and wants to discover common themes before defining routing categories. Which approach is a reasonable first analysis?

Software DevelopmentNCA-GENL practice

A developer has a trained model and wants to optimize its inference execution on supported NVIDIA GPUs. Which component is primarily intended for building optimized inference engines?

Select one answerANVIDIA NeMo Curator for cleaning training datasetsBNVIDIA Triton scheduling without model-engine optimizationCNVIDIA TensorRTDA vector database for indexing document embeddings

Complete your exam preparation

Continue with the full question bank

Practise free with answers, explanations, and Community discussions. Join Community to unlock the complete question bank and timed mock exams.

Join Community Choose access for your certification