NCA-GENM Question 169
Single answerYou are assisting in training a generative AI multimodal model under the guidance of a senior team member. The model is underperforming during validation with unusually high loss values. You suspect the issue might be related to the learning rate. What is the BEST course of action to take?
- A
Decrease the learning rate gradually and re-evaluate the model's performance.
- B
Increase the learning rate significantly to speed up convergence.
- C
Switch to a completely different optimizer without informing the senior team member.
- D
Analyze the training logs and discuss adjustments to the learning rate with the senior team member.
Show answer and explanation
Correct answer: D
Explanation
When assisting in model training and optimization, it is critical to work collaboratively and rely on evidence-based decisions. Analyzing training logs can provide insights into the root cause of high loss values, and discussing changes with the senior team member ensures alignment and proper guidance. This approach minimizes risks and promotes best practices in training optimization.
- A. Incorrect.
Decreasing the learning rate might help, but this step should only be taken after consulting with the senior team member and analyzing the logs to confirm the issue.
- B. Incorrect.
Increasing the learning rate significantly could lead to instability during training and worsen the model's performance.
- C. Incorrect.
Switching to a different optimizer without informing the senior team member is not a recommended practice as it goes against collaborative workflows and might introduce new issues.
- D. Correct.
Analyzing the training logs and discussing adjustments with the senior team member ensures an informed and collaborative approach, which aligns with the role of assisting in model training under supervision.