You have fine-tuned an Azure OpenAI Service model by using the Azure AI Foundry portal. The fine-tuned model is overfitting. You plan to correct overfitting by fine-tuning the model again. You need…

The correct answer is D. Decrease the learning_rate_multiplier. Overfitting means the model has learned the training data too aggressively and lacks generalization. A high learning_rate_multiplier causes the model to make large weight updates, which accelerates fitting to training examples and promotes overfitting. Decreasing the…

Optimize language models for AI applications

Question

You have fine-tuned an Azure OpenAI Service model by using the Azure AI Foundry portal. The fine-tuned model is overfitting. You plan to correct overfitting by fine-tuning the model again. You need to modify the default value of a fine-tuning task parameter to minimize the possibility of overfitting. Which modification should you apply?

Options

AIncrease the batch_size.
BIncrease the learning_rate_multiplier.
CDecrease the batch_size.
DDecrease the learning_rate_multiplier.

How the community answered

(22 responses)

A
18% (4)
B
5% (1)
C
5% (1)
D
73% (16)

Explanation

Overfitting means the model has learned the training data too aggressively and lacks generalization. A high learning_rate_multiplier causes the model to make large weight updates, which accelerates fitting to training examples and promotes overfitting. Decreasing the learning_rate_multiplier slows down the learning process, resulting in smaller, more conservative weight updates that reduce the risk of overfitting. Increasing batch_size (Option A) can also help, but the most direct lever for overfitting caused by aggressive learning is reducing the learning rate.

Topics

#Azure OpenAI Fine-tuning#Overfitting#Hyperparameter Tuning#Learning Rate

Community Discussion

No community discussion yet for this question.

Full DP-100 Practice