NCP-GENL mock exam

Question 1, Model OptimizationYou need a model to fit in less GPU memory and run faster, and you accept a small quality loss. What do you apply?

• AQuantization
• BRaising the temperature
• CMore attention heads
• DA larger context window

Question 2, GPU Acceleration and OptimizationThroughput is low because requests are processed one at a time. Which serving technique gives the biggest gain?

• AUsing a larger tokenizer
• BDisabling the KV cache
• CLowering max tokens to one
• DDynamic or continuous batching

Question 3, GPU Acceleration and OptimizationWhy does a KV cache speed up token generation?

• AIt moves the token embeddings to disk and loads them back on each step
• BIt reduces the vocabulary to the tokens seen earlier in the same request
• CIt stores past attention keys and values, so they are not recomputed for each new token
• DIt compresses the model weights so more of them fit in GPU memory at once

Question 4, Prompt EngineeringAn extraction task keeps returning inconsistent formats, even with clear instructions. What usually helps most?

• AFew-shot examples of the exact output
• BA GPU with more memory for longer outputs
• CA model with a later knowledge cutoff
• DA higher temperature on each request

Question 5, Fine-TuningYou need domain behavior from a large base model, with limited GPU budget. Which approach fits?

• ARetrain the full model from scratch
• BDelete the base model's layers
• CEnsemble ten base models
• DTrain a LoRA adapter on the base model

Question 6, Model DeploymentYou need optimized serving of an LLM on NVIDIA GPUs with an API you can deploy quickly. What is the NVIDIA product for this?

• ANVIDIA RAPIDS libraries
• BNVIDIA NIM microservices
• CNVIDIA Omniverse platform
• DNVIDIA Base Command Manager

Question 7, Production Monitoring and ReliabilityAfter deployment, p99 latency climbs slowly while GPU utilization stays flat. What do you examine first?

• ARequest queue depth and batch behavior
• BThe model's license and usage terms
• CThe tokenizer's vocabulary size
• DThe fine-tuning parameters of the model

Question 8, Safety, Ethics, and ComplianceYou must block unsafe prompts and keep responses on topic, with rules you control. Which tool is designed for this?

• ANVIDIA NeMo Guardrails
• BNVIDIA TensorRT-LLM
• CNVIDIA RAPIDS cuDF
• DNVIDIA Base Command Platform