NCA-GENL Question 92
Select 3You are tasked with analyzing a dataset used to fine-tune a Large Language Model (LLM). Under the supervision of a senior team member, you are asked to identify potential issues in the dataset that could negatively impact the model’s performance. Which of the following actions should you take?
- A
Check for imbalanced data distribution across different categories.
- B
Ensure that the dataset includes only the largest text samples to maximize training efficiency.
- C
Identify and remove any duplicate records in the dataset.
- D
Analyze the dataset for biased or harmful content that could influence the model’s responses.
- E
Focus only on data formatting issues, as content quality is handled automatically by the LLM.
Show answer and explanation
Correct answers: A, C, D
Explanation
To ensure the quality and reliability of an LLM, identifying and addressing issues like data imbalance, duplication, and biased content is crucial. These issues can negatively impact the model's generalization and fairness. Focusing only on formatting or large text samples overlooks essential aspects of dataset preparation, which must be carefully analyzed under supervision.
- A. Correct.
This is correct because imbalanced data distribution can lead to biased model behavior, especially in multi-category tasks.
- B. Incorrect.
This is incorrect because focusing solely on large text samples might eliminate valuable data diversity, which is critical for training LLMs.
- C. Correct.
This is correct because duplicate records can lead to overfitting or redundancy during training, reducing model performance.
- D. Correct.
This is correct because biased or harmful content in the dataset can propagate unintended behavior in the model’s responses.
- E. Incorrect.
This is incorrect because content quality is not automatically handled by the LLM. Reviewing content quality is a critical part of dataset preparation.