AIF-C01 Fundamentals of Generative AI Practice Question

A data scientist is fine-tuning a large language model on Amazon SageMaker for a text summarization task. The training loss decreases steadily but the validation loss starts increasing after a few epochs. What should the scientist do to address this issue?

⚠ Common exam trap

The AIF-C01 exam often tests the distinction between overfitting and underfitting; the trap here is that candidates may mistakenly think increasing epochs (Option C) always improves performance, ignoring the validation loss divergence that signals overfitting.

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

✓

Use early stopping based on validation loss

The validation loss increasing while training loss decreases is a classic sign of overfitting. Early stopping based on validation loss halts training when the validation loss stops improving, preventing overfitting and saving computational resources. This is a standard technique in SageMaker's built-in training algorithms and custom training scripts.

Answer analysis

Option-by-option breakdown

For each option: why learners choose it and why it is or isn't the right answer here.

✗
Reduce the batch size
Why it's wrong here
Reducing batch size does not address the divergence between falling training loss and rising validation loss, which signals overfitting; it alters gradient noise and step count, not the model's capacity to memorise. It is tempting because smaller batches can regularise training, and would suit scenarios where generalisation is poor due to sharp minima.
✗
Increase the learning rate
Why it's wrong here
Raising the learning rate worsens the divergence between training and validation loss, since the model is already overfitting and larger steps amplify weight updates away from the generalisable minimum. Learning-rate increases suit underfitting, where both losses remain high and converge too slowly.
✗
Increase the number of training epochs
Why it's wrong here
Increasing epochs worsens the divergence: training loss keeps falling while validation loss climbs, which is overfitting, so further passes fit noise rather than generalisable patterns. It tempts because extra epochs help underfitting, where both losses remain high and still decreasing — the opposite symptom to the one described here.
✓
Use early stopping based on validation loss
Why this is correct
Early stopping halts training once validation loss stops improving, restoring the checkpoint with the lowest validation loss and preventing further overfitting. This satisfies the requirement to counter rising validation loss while training loss still falls.

Go deeper

AIF-C01 Fundamentals of Generative AI Practice Question

Why each option matters

Option-by-option breakdown

Related to this question