top of page

NCA-GENL practice exam and study guide

NVIDIA-Certified Associate Generative AI LLM

Timed practice exams for NCA-GENL on the real exam's 60 minute limit. NVIDIA states 50 to 60 questions, and a full practice exam on the site serves 60, the top of that range. Every question is original and every option is explained. Your score is broken down by the sections NVIDIA publishes, so you see what to study next. The question bank holds hundreds of practice questions.

Exam facts

  • Questions: 50 to 60 multiple choice [1]
  • Time limit: 1 hour [1]
  • Scoring: Pass or fail, with no score reported [2]
  • Level: Associate [1]
  • Exam fee: $125 [1]
  • Exam languages (real exam): English [1]
  • Validity: Two years from issuance [1]

Tokens: 1 for $4.99, 5 for $14.99, 10 for $24.99 or 25 for $49.99. One token runs a full timed NCA-GENL practice exam, and every shorter session on that exam is then free. The Unlimited pass is $149.99 for a year of every practice session on every exam.

Get tokens

Exam sections and weights

NVIDIA publishes these sections and objectives for NCA-GENL, quoted here as NVIDIA words them. [1]

Section 1: Core Machine Learning and AI Knowledge, 30% of the exam

  • "Assist in deployment and evaluation of model scalability, performance, and reliability under the supervision of senior team members."
  • "Awareness of the process of extracting insights from large datasets using data mining, data visualization, and similar techniques."
  • "Build LLM use cases such as retrieval-augmented generation (RAG), chatbots, and summarizers."
  • "Curate and embed content datasets for RAGs."
  • "Familiarity with the fundamentals of machine learning (e.g., feature engineering, model comparison, cross validation)."
  • "Familiarity with the capabilities of Python natural language packages (spaCy, NumPy, vector databases, etc.)."
  • "Read research papers (articles, conference papers, etc.) to identify emerging LLM trends and technologies."
  • "Select and use models to create text embeddings."
  • "Use prompt engineering principles to create prompts to achieve desired results."
  • "Use Python packages (spaCy, NumPy, Keras, etc.) to implement specific traditional machine learning analyses."

Section 2: Data Analysis, 14% of the exam

  • "Awareness of the process of extracting insights from large datasets using data mining, data visualization, and similar techniques."
  • "Compare models using statistical performance metrics, such as loss functions or proportion of explained variance."
  • "Conduct data analysis under the supervision of a senior team member."
  • "Create graphs, charts, or other visualizations to convey the results of data analysis using specialized software."
  • "Identify relationships and trends or any factors that could affect the results of research."

Section 3: Experimentation, 22% of the exam

  • "Assist in model training and training optimization under the supervision of a senior team member."
  • "Assist in preparing (e.g., scraping, tokenization) large datasets for pretraining, fine-tuning, and RLHF."
  • "Assist in the design and conduct of hardware or software tests for LLM applications."
  • "Assist in the evaluation of current or emerging technologies to consider factors such as cost, portability, compatibility, or usability."
  • "Awareness of, and/or participation in, data collection from human subjects (e.g., RLHF)."
  • "Evaluate and refine existing models / benchmarking."
  • "Executes experimentation to evaluate models and pipelines."

Section 4: Software Development, 24% of the exam

  • "Assist in the deployment and evaluations of model scalability, performance, and reliability under the supervision of a senior team member."
  • "Build LLM use cases such as RAGs, chatbots, and summarizers."
  • "Familiarity with the capabilities of Python natural language packages (spaCy, NumPy, vector databases, etc.)."
  • "Identify system data, hardware, or software components required to meet user needs."
  • "Monitor functioning of data collection, experiments, and other software processes."
  • "Use Python packages (spaCy, NumPy, Keras, etc.) to implement specific traditional machine learning analyses."
  • "Write software components or scripts under the supervision of a senior team member."

Section 5: Trustworthy AI, 10% of the exam

  • "Describe the ethical principles of trustworthy AI."
  • "Describe the balance between data privacy and the importance of data consent."
  • "Describe how to use NVIDIA and other technologies to improve AI trustworthiness."
  • "Describe how to minimize bias in AI systems."

Sample questions

Three questions from the NCA-GENL practice bank, with the answer and why each option is right or wrong.

Question 1

Which requirement applies when chunked context, also called chunked prefill, is turned on in TensorRT-LLM?

  • A. The FMHA paged KV cache is on, and each chunk except the last is a multiple of the KV cache block size.
  • B. Every chunk has the same size, including the last one, so the prompt is padded out to a full chunk.
  • C. The prompt still has to fit within one pass, because chunking keeps the original limit on input length.
  • D. The paged KV cache is off, because each context chunk is stored in its own contiguous buffer.

Answer: A. Chunked context processes a long prompt in several pieces where TensorRT-LLM once handled all context tokens at once. Two conditions come with it: the FMHA paged KV cache, and chunk sizes tied to the KV cache block size. [3]

  • A. Right. TensorRT-LLM needs the FMHA paged KV cache enabled for chunked context, and every chunk before the last must be an integer multiple of the KV cache block size.
  • B. Wrong. The size rule exempts the last chunk. Only the chunks before it must be a whole multiple of the KV cache block size.
  • C. Wrong. TensorRT-LLM notes that chunking contexts also removes constraints on input length, so a long prompt is no longer held to one pass.
  • D. Wrong. Chunked context depends on the paged KV cache. The FMHA paged KV cache must be enabled for the feature to work.

Question 2

Which limit of NVIDIA cuDF DataFrame.nlargest, compared with pandas, affects ranking claims by amount and then by date?

  • A. cuDF orders in ascending order only, so one call returns the smallest amounts first.
  • B. cuDF drops the columns not used for ordering, so the date column is missing from the result.
  • C. cuDF orders by a single column only, so one call cannot rank by amount and then by date.
  • D. cuDF keeps every tied row past n, so the result can hold more rows than requested.

Answer: C. nlargest gives the top rows by one column in descending order. With duplicate values, keep decides whether the first or the last occurrence is preferred. [4]

  • A. Wrong. nlargest returns the first n rows with the largest values in columns, in descending order.
  • B. Wrong. The columns that are not specified are returned as well. They are only left out of the ordering.
  • C. Right. cuDF's pandas compatibility note for nlargest says only a single column is supported in columns, so ranking on a second column needs another approach.
  • D. Wrong. keep accepts 'first' or 'last' to decide which duplicate wins, and the call returns n rows.

Question 3

Which num_query_groups value gives grouped-query attention to a Megatron Bridge GPT model with 32 attention heads?

  • A. Set num_query_groups to 1 in the GPTModelProvider model configuration.
  • B. Set num_query_groups to 32 in the GPTModelProvider model configuration.
  • C. Set num_query_groups to 8 in the GPTModelProvider model configuration.
  • D. Set num_query_groups to 12 in the GPTModelProvider model configuration.

Answer: C. One model setting, num_query_groups, selects among multi-query, grouped-query and regular multihead attention in Megatron Bridge. [5]

  • A. Wrong. Setting num_query_groups to 1 treats all attention heads as a single group, giving multi-query attention.
  • B. Wrong. Matching num_query_groups to the head count, or leaving it as None, gives regular multihead attention with one group per head.
  • C. Right. For grouped-query attention, num_query_groups is set to a divisor of the total head count that is more than one and less than the total, so 8 groups of 32 heads gives 4 heads per group.
  • D. Wrong. 12 does not divide 32, and the setting for grouped-query attention must be a divisor of the total number of attention heads.

Practice for NCA-GENL on ExamSens.ai

ExamSens.ai is not authorized, endorsed or sponsored by NVIDIA.

NVIDIA is a trademark and/or registered trademark of NVIDIA Corporation in the U.S. and other countries.

References

[1] https://www.nvidia.com/en-us/learn/certification/generative-ai-llm-associate/

[2] https://www.nvidia.com/en-us/learn/certification/#faq

[3] https://nvidia.github.io/TensorRT-LLM/features/paged-attention-ifb-scheduler.html

[4] https://docs.nvidia.com/cudf/latest/cudf/api_docs/api/cudf.DataFrame.nlargest/

[5] https://docs.nvidia.com/nemo/megatron-bridge/latest/training/attention-optimizations.html

bottom of page