NCA-GENL Data Analysis and Visualization Practice Question
A team is comparing two LLM fine-tuning runs on the same dataset. Run A used a cosine learning-rate schedule, and Run B used a constant learning rate. They plot validation loss versus training step for both runs on the same axes. Run A's curve is smooth, while Run B's curve shows a sharp upward spike around step 800 and then recovers. The team wants to determine whether the spike in Run B indicates a data-order artifact or a genuine optimization instability. Which additional visualization is most useful for that diagnosis?
⚠ Common exam trap
The trap here is treating a validation-loss spike as purely an optimization problem and ignoring the role of data ordering within the epoch.
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
✓
A scatter plot of per-batch training loss for Run B colored by batch index modulo epoch length
Plotting per-batch training loss with color encoding the batch's position modulo the epoch length exposes whether the spike recurs at a fixed data position. A recurring pattern implies a data-order artifact, such as a hard shard or mislabeled examples, while a random pattern implies optimization instability. Summary charts and hardware metrics do not preserve the temporal and data-order information needed for this diagnosis.
Answer analysis
Option-by-option breakdown
For each option: why learners choose it and why it is or isn't the right answer here.
- ✗
A bar chart of final validation loss for Run A and Run B
Why it's wrong here
A bar chart of final validation loss only compares end states and discards the temporal context around step 800. It cannot distinguish whether the spike came from a specific batch, a learning-rate interaction, or random noise. Since the question is about diagnosing the cause of a transient spike, a summary bar chart provides no information about what happened at that step.
- ✓
A scatter plot of per-batch training loss for Run B colored by batch index modulo epoch length
Why this is correct
Coloring per-batch training loss by the batch's position within the epoch reveals whether the spike aligns with a particular data shard or ordering pattern. If high-loss batches cluster at the same modulo position, that points to a data-order artifact such as a difficult shard. If the spike appears at random positions, it suggests optimization instability instead. This directly addresses the diagnostic goal.
- ✗
A heatmap of attention weights for one validation example
Why it's wrong here
Attention weights from a single validation example are too local to explain a training-time spike at step 800. They show how the model routes information for that example, not which training batch caused the loss jump. Without linking the example to the spike step, this visualization cannot separate a data-order artifact from optimization instability.
- ✗
A line chart of GPU utilization for Run A and Run B
Why it's wrong here
GPU utilization shows hardware efficiency, not the cause of a validation-loss spike. A dip in utilization might correlate with a data-loading stall, but it would not tell the team whether the spike is a data-order artifact or an optimization issue. This metric is useful for throughput tuning, not for diagnosing the source of a transient loss increase.
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.