When to fine-tune vs prompt engineer in Prompt Engineering / GenAI - Practice Questions

Challenge - 5 Problems

🧠 Conceptual

intermediate

Choosing between fine-tuning and prompt engineering

You have a general language model and want it to perform well on a specific task with limited data. Which approach is best to start with?

A Fine-tune the model immediately using the limited data to specialize it.

B Ignore the task and use the model as is without any changes.

C Train a new model from scratch on the limited data.

D Use prompt engineering to guide the model without changing its weights.

💡 Hint

Think about the cost and data needed for fine-tuning versus prompt engineering.

Prompt engineering is usually the first step when data is limited because it does not require changing the model weights and is faster and cheaper. Fine-tuning is better when you have enough data and need more specialized behavior.

🧠 Conceptual

intermediate

When is fine-tuning preferred over prompt engineering?

Which situation best justifies fine-tuning a large language model instead of relying on prompt engineering?

A You want to use the model for many different unrelated tasks.

B You have a large, high-quality dataset specific to your task and need consistent output.

C You want to save computational resources and avoid retraining.

D You want to quickly test different instructions without changing the model.

💡 Hint

Consider when changing the model weights is beneficial.

Fine-tuning is preferred when you have enough task-specific data and want the model to consistently perform well on that task, as it adjusts the model weights to specialize it.

❓ Metrics

advanced

Evaluating fine-tuning vs prompt engineering performance

You fine-tune a model and also try prompt engineering on the base model. You measure accuracy on a test set. Which metric result indicates fine-tuning improved performance?

Prompt Engineering / GenAI

base_accuracy = 0.75
fine_tuned_accuracy = 0.82
prompt_engineered_accuracy = 0.78

A Fine-tuning improved accuracy by 7% over base and 4% over prompt engineering.

B Prompt engineering improved accuracy more than fine-tuning.

C Base model accuracy is highest, so no improvement.

D Fine-tuning decreased accuracy compared to prompt engineering.

💡 Hint

Compare the accuracy numbers carefully.

Fine-tuning accuracy (0.82) is higher than both base (0.75) and prompt engineered (0.78), showing it improved performance the most.

🔧 Debug

advanced

Why did fine-tuning not improve model performance?

You fine-tuned a model on a small dataset but test accuracy dropped. What is the most likely cause?

A Prompt engineering was not used before fine-tuning.

B The model architecture was changed accidentally.

C The dataset was too small causing overfitting during fine-tuning.

D The test set was too large compared to training.

💡 Hint

Think about what happens when fine-tuning with little data.

Fine-tuning on a small dataset can cause the model to memorize training data and perform worse on new data, known as overfitting.

❓ Model Choice

expert

Selecting approach for a multi-domain chatbot

You want to build a chatbot that handles many topics without retraining often. Which approach is best?

A Use prompt engineering with a single large base model to adapt responses dynamically.

B Fine-tune separate models for each domain and switch between them.

C Train a new model from scratch on combined domain data.

D Use a rule-based system instead of a language model.

💡 Hint

Consider flexibility and maintenance effort for many topics.

Prompt engineering with a single large model allows dynamic adaptation to many domains without retraining multiple models, making it more scalable and efficient.