Free NVIDIA-Certified Professional: Generative AI LLMs (NCP-GENL) Practice Questions

Sample NCP-GENL Practice Questions

Browse all 20 free NVIDIA-Certified Professional: Generative AI LLMs practice questions below.

1. Your team is training a 175B parameter model across 64 DGX A100 nodes. The current configuration uses only data parallelism, but you're experiencing suboptimal GPU utilization and high communication overhead. Which parallelism strategy would best optimize training throughput?

   • Use PTD-P: tensor parallelism within each DGX node, pipeline parallelism across nodes, and data parallelism for scale
   • Switch to pure data parallelism but increase batch size to reduce communication frequency
   • Use tensor parallelism across all 64 nodes to maximize model distribution
   • Implement pipeline parallelism only, splitting the model layers across all nodes

2. When deploying a 70B parameter LLM for production inference with strict latency requirements (<100ms) and limited GPU memory, which TensorRT-LLM optimization technique provides the best balance of speed and model quality?

   • Deploy the full FP32 model and rely on batch processing to amortize latency
   • Implement model pruning to remove 50% of parameters before deployment
   • Use FP8 quantization with speculative decoding to balance quality and latency
   • Apply aggressive INT4 post-training quantization (PTQ) to all layers

3. Your data preparation pipeline for training a domain-specific LLM processes 500GB of raw text. The current CPU-based tokenization is taking 36 hours. Which NVIDIA technology would most effectively accelerate this preprocessing step?

   • Use TensorRT-LLM to accelerate tokenization on GPU
   • Switch to NeMo Framework's built-in inference pipeline
   • Distribute the tokenization across multiple DGX nodes using pipeline parallelism
   • Implement RAPIDS cuDF for GPU-accelerated text processing and tokenization

4. You're implementing few-shot learning for a classification task and getting inconsistent results across multiple runs with the same prompt. The model uses greedy decoding (temperature=0). Which prompt engineering technique would most effectively improve output consistency?

   • Add more examples to the prompt until consistency improves
   • Switch from few-shot to zero-shot prompting to reduce complexity
   • Switch to self-consistency sampling with temperature > 0 and select the most frequent answer
   • Increase temperature to 1.5 to get more creative outputs

5. A long-context application processes 32K token sequences but faces GPU memory constraints during inference. Profiling shows KV cache consuming 80% of available memory. Which optimization technique provides the best memory reduction while maintaining accuracy?

   • Use Grouped Query Attention (GQA) to share key-value heads across multiple query heads
   • Implement Multi-Head Latent Attention (MLA) with low-rank projection to compress KV cache to a shared latent space
   • Reduce the maximum sequence length to 8K tokens to fit within memory constraints
   • Apply FlashAttention to restructure attention computation and reduce memory overhead

6. Your team is deploying a 70B parameter LLM on a multi-GPU DGX system and needs to minimize inference latency for real-time applications. Which quantization technique in TensorRT-LLM would provide the best balance of speed and accuracy for small batch sizes?

   • FP8 quantization for both weights and activations
   • FP16 mixed precision without quantization
   • INT8 SmoothQuant with both weight and activation quantization
   • INT4 GPTQ weight-only quantization

7. When using TensorRT-LLM's Model Optimizer for post-training quantization (PTQ), which calibration technique is specifically designed to address activation outliers that can degrade quantization quality?

   • SmoothQuant
   • GPTQ
   • AutoQuantize
   • AWQ (Activation-Aware Weight Quantization)

8. Your production LLM deployment uses TensorRT-LLM on Hopper GPUs and experiences high memory pressure with long context lengths. What is the recommended approach for KV cache optimization?

   • Use INT8 KV cache quantization for maximum compression
   • Use FP8 KV cache quantization
   • Reduce batch size to fit the full precision KV cache
   • Disable KV cache to save memory

9. A model optimization engineer is comparing quantization strategies for a 175B parameter model deployed on Blackwell architecture GPUs. Which quantization format provides specialized kernel support and high compression for this architecture?

   • FP8 quantization
   • INT4 AWQ
   • NVFP4 (4-bit floating-point)
   • INT8 SmoothQuant

10. When deploying models quantized with TensorRT-LLM's Model Optimizer PTQ framework, which inference frameworks are natively supported for deployment?

    • Custom inference engines only
    • Only TensorRT-LLM
    • TensorRT-LLM, vLLM, and SGLang
    • Only vLLM and SGLang

11. When preparing a large web-scraped dataset for LLM pre-training, which quality filtering step is most critical to prevent memorization and improve generalization?

    • Converting all text to lowercase
    • Removing all numbers from the text
    • Exact and near-duplicate detection and removal
    • Limiting each document to exactly 512 tokens

12. What is the primary advantage of using SentencePiece for tokenization compared to word-level tokenization?

    • It handles multiple languages and rare words through language-agnostic subword segmentation
    • It eliminates the need for any preprocessing or normalization
    • It only works with English text
    • It always produces smaller vocabulary sizes than BPE

13. When using NVIDIA RAPIDS cuDF for LLM data preprocessing, what is the primary performance benefit?

    • GPU-accelerated dataframe operations providing 10-100x speedup over CPU-based pandas
    • It only works with text data, not numerical data
    • It generates synthetic training data
    • Automatic data cleaning without any configuration

14. What is the purpose of vocabulary size in tokenization, and what tradeoff does it represent?

    • It determines the maximum number of training examples
    • Smaller vocabularies always train faster
    • Larger vocabulary always improves model quality with no downsides
    • Vocabulary size balances sequence length versus embedding table size and token coverage

15. In the context of LLM training data, what does perplexity measure when used as a data quality filter?

    • The number of spelling errors per 1000 words
    • The number of unique words in the document
    • The reading difficulty level of the text
    • The model's uncertainty about the text - high perplexity text may be low quality or out-of-distribution

16. Your team is building a customer service chatbot that needs to follow a specific troubleshooting workflow. The LLM sometimes skips steps or hallucinates solutions. Which prompt engineering technique would most effectively ensure the model follows the required step-by-step reasoning process?

    • Increase the maximum token limit to give the model more space for responses
    • Use zero-shot prompting with detailed instructions in the system message
    • Use temperature=0 to make the model deterministic and prevent hallucinations
    • Implement few-shot Chain-of-Thought prompting with examples showing the complete troubleshooting reasoning process

17. You're deploying a NVIDIA NIM-based API that must return customer data in strict JSON format with specific field validation (e.g., phone numbers matching regex patterns, dates in ISO format). Which structured generation approach provides the most reliable output formatting?

    • Post-process the output with regex to fix formatting issues after generation
    • Use few-shot prompting with JSON examples and ask the model to follow the format
    • Fine-tune the model on thousands of correctly formatted JSON examples
    • Use NVIDIA NIM's constrained decoding with guided_json parameter and JSON schema validation

18. Your ReAct agent built with NVIDIA NeMo Agent Toolkit is performing poorly on complex multi-step tasks, often choosing wrong tools or getting stuck in loops. What is the most effective optimization strategy?

    • Reduce the number of available tools to simplify the decision space
    • Engineer better prompts with clear tool descriptions, add few-shot examples of successful multi-step reasoning, and tune the thought-action-observation format
    • Increase temperature to introduce more randomness in tool selection
    • Switch to a larger LLM model to improve reasoning capabilities

19. You're using NVIDIA's MIPROv2 to optimize prompts for a classification task. After several iterations, the optimizer suggests adding 8 few-shot examples, but this pushes your prompt to 3,500 tokens. What's the best approach to balance performance and cost?

    • Fine-tune the model on the examples instead of using few-shot prompting
    • Keep all 8 examples since MIPROv2 identified them as optimal for performance
    • Test a reduced set of 3-4 strategically selected examples and measure if performance degradation is acceptable for the cost savings
    • Move examples to a separate context and use retrieval to fetch only the most relevant 2-3 per query

20. Your application needs to generate SQL queries from natural language. The LLM occasionally produces syntactically invalid SQL or uses wrong table names. Which combination of techniques provides the most robust solution?

    • Post-process generated SQL with a validation library and retry on errors
    • Increase temperature to explore more SQL syntax variations
    • Combine constrained decoding with SQL grammar, Chain-of-Thought reasoning for query planning, and provide schema context in the prompt
    • Use few-shot prompting with SQL examples only