Evaluation of fine-tuned models in Prompt Engineering / GenAI - Model Metrics & Evaluation

Metrics & Evaluation - Evaluation of fine-tuned models

Which metric matters for Evaluation of fine-tuned models and WHY

When we fine-tune a model, we want to see if it learned better than before. Common metrics include accuracy for simple tasks, but often precision, recall, and F1 score matter more. These metrics tell us how well the model predicts the right answers and avoids mistakes. For tasks like text generation or classification, we also check loss to see if the model is improving during training. Choosing the right metric depends on the task and what mistakes cost more.

Confusion matrix example for fine-tuned classification model

      | Predicted Positive | Predicted Negative |
      |--------------------|--------------------|
      | True Positive (TP): 85 | False Negative (FN): 15 |
      | False Positive (FP): 10 | True Negative (TN): 90 |

      Total samples = 85 + 15 + 10 + 90 = 200

      Precision = TP / (TP + FP) = 85 / (85 + 10) = 0.8947
      Recall = TP / (TP + FN) = 85 / (85 + 15) = 0.85
      F1 Score = 2 * (Precision * Recall) / (Precision + Recall) = 0.871
    

Precision vs Recall tradeoff with examples

Precision means when the model says "yes," it is usually right. This is important when false alarms are costly. For example, in spam detection, high precision means fewer good emails marked as spam.

Recall means the model finds most of the true positives. This is important when missing a positive is bad. For example, in medical diagnosis, high recall means fewer sick patients are missed.

Fine-tuning can improve one metric but may reduce the other. We must balance them based on the task.

What "good" vs "bad" metric values look like for fine-tuned models

Good: Precision and recall above 0.85, F1 score close to 0.9 or higher, and steadily decreasing loss during training. This means the model predicts well and learns from data.

Bad: High accuracy but very low recall (e.g., recall 0.2) means the model misses many true cases. Or if loss stops improving or increases, the model may not be learning well.

Common pitfalls in evaluating fine-tuned models

• Accuracy paradox: High accuracy can be misleading if data is unbalanced.
• Data leakage: Using test data during training inflates metrics falsely.
• Overfitting: Model performs well on training but poorly on new data.
• Ignoring task needs: Using wrong metrics for the problem can hide issues.

Self-check question

Your fine-tuned model has 98% accuracy but only 12% recall on fraud detection. Is it good for production? Why or why not?

Answer: No, it is not good. Even though accuracy is high, the model misses 88% of fraud cases (low recall). This means many frauds go undetected, which is risky. For fraud detection, high recall is critical to catch as many frauds as possible.

Key Result

For fine-tuned models, balance precision and recall to ensure meaningful improvements beyond accuracy.

Practice

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

easy

A. To reduce the number of model layers

B. To check how well the model performs on new, unseen data

C. To speed up the training process

D. To increase the size of the training dataset

Solution

1. Step 1: Understand model evaluation

   Evaluation measures how well the model predicts on data it has not seen before.

2. Step 2: Identify the purpose of evaluation

   It helps us know if the model learned useful patterns or just memorized training data.

3. Final Answer:

   To check how well the model performs on new, unseen data -> Option B

4. Quick Check:

   Evaluation = performance on new data [OK]

Hint: Evaluation checks model on new data, not training data [OK]

Common Mistakes:

• Confusing evaluation with training
• Thinking evaluation changes model structure
• Believing evaluation increases data size

2. Which of the following is the correct way to evaluate a fine-tuned model in Python using TensorFlow?

easy

A. model.compile(optimizer='adam')

B. model.train(test_data, test_labels)

C. model.predict(train_data)

D. model.evaluate(test_data, test_labels)

Solution

1. Step 1: Recall TensorFlow evaluation method

   TensorFlow models use model.evaluate() to measure performance on test data.

2. Step 2: Identify correct usage

   model.evaluate(test_data, test_labels) returns loss and metrics on unseen data.

3. Final Answer:

   model.evaluate(test_data, test_labels) -> Option D

4. Quick Check:

   Use model.evaluate() for testing [OK]

Hint: Use model.evaluate() with test data for evaluation [OK]

Common Mistakes:

• Using model.train() instead of evaluate
• Calling predict() without labels for evaluation
• Confusing compile() with evaluation

3. Given the code below, what will be the output of print(loss, accuracy)?

loss, accuracy = model.evaluate(x_test, y_test)
print(loss, accuracy)

medium

A. The loss value and accuracy metric on the test set

B. The training loss and accuracy values

C. A syntax error because evaluate returns only one value

D. The predicted labels for x_test

Solution

1. Step 1: Understand model.evaluate() output

   It returns loss and metrics (like accuracy) on the test data.

2. Step 2: Analyze the print statement

   Printing loss, accuracy shows these two values from evaluation.

3. Final Answer:

   The loss value and accuracy metric on the test set -> Option A

4. Quick Check:

   evaluate() returns loss and accuracy [OK]

Hint: model.evaluate() returns loss and metrics tuple [OK]

Common Mistakes:

• Thinking evaluate returns training metrics
• Assuming evaluate returns predictions
• Believing evaluate returns only one value

4. You ran model.evaluate(x_test) but got an error. What is the likely cause?

medium

A. The model is not compiled

B. The test data x_test is empty

C. Missing the true labels y_test in the evaluate call

D. The model has too many layers

Solution

1. Step 1: Check evaluate method requirements

   model.evaluate() needs both input data and true labels to compute metrics.

2. Step 2: Identify missing argument

   Calling model.evaluate(x_test) misses y_test, causing an error.

3. Final Answer:

   Missing the true labels y_test in the evaluate call -> Option C

4. Quick Check:

   evaluate() needs inputs and labels [OK]

Hint: Always pass both data and labels to evaluate() [OK]

Common Mistakes:

• Forgetting to pass labels to evaluate()
• Assuming evaluate works with inputs only
• Ignoring model compilation status

5. You fine-tuned two models and got these evaluation results on the same test set:

• Model A: loss=0.25, accuracy=0.90
• Model B: loss=0.20, accuracy=0.85

Which model should you choose and why?

hard

A. Model A, because it has higher accuracy which is more important than loss

B. Model B, because it has lower loss indicating better overall fit

C. Model A, because loss and accuracy must both be higher

D. Model B, because accuracy is less important than loss

Solution

1. Step 1: Understand evaluation metrics

   Accuracy shows correct predictions percentage; loss shows error magnitude.

2. Step 2: Compare models on accuracy and loss

   Model A has higher accuracy (0.90) but slightly higher loss (0.25) than Model B.

3. Step 3: Decide based on goal

   For classification, accuracy is usually more important to pick the better model.

4. Final Answer:

   Model A, because it has higher accuracy which is more important than loss -> Option A

5. Quick Check:

   Higher accuracy preferred for classification [OK]

Hint: Pick model with higher accuracy for classification tasks [OK]

Common Mistakes:

• Choosing model with lower loss but worse accuracy
• Ignoring accuracy when loss differs
• Assuming loss always trumps accuracy