NCA-GENL Core Machine Learning and AI Knowledge Practice Question
A developer is using a pretrained large language model for a text summarization task. They want to adapt the model to a domain-specific corpus of legal documents but have limited GPU memory and a small labeled dataset. Which fine-tuning approach is most parameter-efficient and suitable for this scenario?
⚠ Common exam trap
The trap here is equating parameter efficiency with simply freezing most layers, when methods like LoRA adapt internal representations with minimal added parameters.
Answer choices
Why each option matters
Answer the question above first, then reveal the full breakdown to understand why each option is right or wrong.
Correct answer & explanation
✓
Low-Rank Adaptation (LoRA)
LoRA is a parameter-efficient fine-tuning method that freezes the base model and trains small rank-decomposition matrices. It drastically cuts memory usage and trainable parameters, making it feasible on limited hardware and small datasets. Full fine-tuning is resource-heavy, training from scratch is impractical, and training only the head under-adapts to the specialized legal domain.
Answer analysis
Option-by-option breakdown
For each option: why learners choose it and why it is or isn't the right answer here.
- ✓
Low-Rank Adaptation (LoRA)
Why this is correct
LoRA injects trainable low-rank matrices into existing layers while freezing the original weights, dramatically reducing the number of trainable parameters and memory footprint. This makes it ideal for limited GPU memory and small datasets, as only the adapters are updated. It also preserves pretrained knowledge, reducing overfitting risk and enabling efficient domain adaptation for legal summarization.
- ✗
Training the model from scratch on the legal corpus
Why it's wrong here
Training from scratch demands massive computational resources and a large corpus, far beyond what the developer has. It also discards the general language understanding of the pretrained model, which is valuable even for domain-specific tasks. With a small labeled dataset, this approach would almost certainly underperform and is not feasible given the memory constraints.
- ✗
Freezing all layers and training only the output head
Why it's wrong here
Training only the output head limits adaptation to the new domain because the internal representations remain fixed to the pretraining distribution. For legal text, which differs significantly from general web text, this yields subpar summarization quality. While parameter-efficient, it does not provide the depth of adaptation needed and is less effective than methods that adjust internal representations.
- ✗
Full fine-tuning of all model parameters
Why it's wrong here
Full fine-tuning updates every weight in the model, requiring substantial GPU memory to store gradients and optimizer states, which exceeds the developer's limited resources. It also risks catastrophic forgetting and overfitting on a small labeled dataset. While it can yield strong performance, it is not parameter-efficient and is impractical under the stated constraints.
About these practice questions
One of 367 original NCA-GENL practice questions on Courseiva, each with a full explanation and wrong-answer analysis — not exam dumps or protected exam content. Learn why practice questions differ from exam dumps →
JA
Written and reviewed by Johnson Ajibi, MSc IT Security
Senior Network & Security Engineer · founder of Courseiva
Last reviewed September 2026 · checked against the official NVIDIA exam blueprint
This NCA-GENL practice question is part of Courseiva's free NVIDIA certification practice question bank. Courseiva provides original exam-style practice questions with explanations, topic-based practice, mock exams, readiness tracking, and study analytics to help learners prepare for the NCA-GENL exam.