Pre-training and fine-tuning concept in Prompt Engineering / GenAI - Practice Problems & Coding Challenges

Challenge - 5 Problems

🧠 Conceptual

intermediate

Understanding the role of pre-training

Which statement best describes the purpose of pre-training in machine learning models?

A Pre-training initializes a model by learning general patterns from a large dataset before adapting to a specific task.

B Pre-training involves manually labeling data to improve model performance.

C Pre-training is the final step where the model is tested on unseen data to check accuracy.

D Pre-training is used to reduce the size of the dataset by removing irrelevant samples.

💡 Hint

Think about how a model learns general knowledge before focusing on a specific problem.

Pre-training helps the model learn broad features from a large dataset, which can then be fine-tuned for a specific task.

❓ Model Choice

intermediate

Choosing a model for fine-tuning

You want to build a sentiment analysis tool using a large language model. Which model type is best suited for fine-tuning on your specific dataset?

A A simple linear regression model trained from scratch

B A pre-trained transformer model like BERT or GPT

C A random forest model without any pre-training

D A clustering algorithm like K-means

💡 Hint

Consider models that have already learned language patterns and can be adapted.

Pre-trained transformer models have learned language features and can be fine-tuned efficiently for tasks like sentiment analysis.

❓ Predict Output

advanced

Output of fine-tuning training metrics

What will be the output of the training accuracy after fine-tuning this simple model for 3 epochs?

Prompt Engineering / GenAI

import torch
import torch.nn as nn
import torch.optim as optim

class SimpleModel(nn.Module):
    def __init__(self):
        super().__init__()
        self.linear = nn.Linear(2, 2)

    def forward(self, x):
        return self.linear(x)

model = SimpleModel()
criterion = nn.CrossEntropyLoss()
optimizer = optim.SGD(model.parameters(), lr=0.1)

# Dummy data: inputs and labels
inputs = torch.tensor([[1.0, 2.0], [2.0, 1.0], [1.5, 1.5], [3.0, 3.0]])
labels = torch.tensor([0, 1, 0, 1])

for epoch in range(3):
    optimizer.zero_grad()
    outputs = model(inputs)
    loss = criterion(outputs, labels)
    loss.backward()
    optimizer.step()

_, predicted = torch.max(outputs, 1)
correct = (predicted == labels).sum().item()
accuracy = correct / labels.size(0)
print(f"Accuracy after 3 epochs: {accuracy:.2f}")

A Accuracy after 3 epochs: 0.50

B Accuracy after 3 epochs: 1.00

C Accuracy after 3 epochs: 0.25

D Accuracy after 3 epochs: 0.75

💡 Hint

Check the model output and compare predicted labels to true labels after training.

The simple linear model trained for 3 epochs on this small dataset achieves 50% accuracy due to limited training and model capacity.

❓ Hyperparameter

advanced

Effect of learning rate during fine-tuning

During fine-tuning a pre-trained model, what is the typical effect of using a learning rate that is too high?

A The model will ignore the pre-trained weights and start training from scratch.

B The model will always converge faster and achieve better accuracy.

C The model will reduce overfitting by stopping training early.

D The model may fail to converge and training loss can fluctuate or increase.

💡 Hint

Think about how big steps in learning affect the model's ability to settle on good solutions.

A too high learning rate can cause the model to jump around the solution space, preventing stable convergence.

🔧 Debug

expert

Identifying the cause of poor fine-tuning results

You fine-tuned a large pre-trained model on a small dataset but the validation accuracy is very low and training loss does not improve. Which issue is most likely causing this?

A The model architecture is too simple for the task.

B The dataset is too large, causing overfitting.

C The learning rate is too high, causing unstable training.

D The pre-training was done on a similar domain.

💡 Hint

Consider what happens when training a complex model on limited data with aggressive settings.

A high learning rate can prevent the model from properly adjusting weights during fine-tuning, leading to poor accuracy and no loss improvement.