Evaluation of fine-tuned models in Prompt Engineering / GenAI - Cheat Sheet & Quick Revision

Recall & Review

beginner

What is the main purpose of evaluating a fine-tuned model?

To check how well the model performs on new, unseen data after training, ensuring it learned useful patterns and can make accurate predictions.

beginner

Name two common metrics used to evaluate classification models.

Accuracy and F1-score are common metrics. Accuracy measures the percentage of correct predictions, while F1-score balances precision and recall.

intermediate

Why is it important to use a separate test set when evaluating a fine-tuned model?

Using a separate test set helps measure how the model performs on data it has never seen before, preventing overly optimistic results from training data.

intermediate

What does overfitting mean in the context of fine-tuned models?

Overfitting happens when a model learns the training data too well, including noise, and performs poorly on new data.

intermediate

How can you visually inspect a fine-tuned model's performance on classification tasks?

By using a confusion matrix, which shows correct and incorrect predictions for each class, helping identify where the model makes mistakes.

Which metric is best to use when classes are imbalanced?

A Accuracy

B F1-score

C Training loss

D Epoch count

F1-score balances precision and recall, making it better for imbalanced classes than accuracy.

What does a high training accuracy but low test accuracy usually indicate?

A Good generalization

B Underfitting

C Data leakage

D Overfitting

High training accuracy but low test accuracy means the model learned training data too well but fails on new data, which is overfitting.

Which dataset is used to tune model parameters during fine-tuning?

A Training set

B Validation set

C Test set

D Unlabeled data

The training set is used to adjust model parameters during fine-tuning.

What is the role of the test set in model evaluation?

A To assess final model performance

B To tune hyperparameters

C To train the model

D To generate synthetic data

The test set is used only to assess the final performance of the model on unseen data.

Which visualization helps understand classification errors?

A Scatter plot

B Loss curve

C Confusion matrix

D Histogram

A confusion matrix shows how many predictions were correct or wrong for each class.

Explain why evaluating a fine-tuned model on unseen data is crucial and describe common metrics used.

Think about how to know if the model learned well beyond just memorizing.

Describe overfitting in fine-tuned models and how you can detect it using evaluation results.

Consider what happens when a model performs great on training but poorly on new data.