Fine-tuning strategy in PyTorch - Practice Problems & Coding Challenges

Challenge - 5 Problems

🧠 Conceptual

intermediate

Understanding Fine-tuning Layers

When fine-tuning a pretrained neural network, which layers are typically updated to adapt the model to a new task?

A Only the first few layers are updated, and the rest remain frozen.

B All layers are frozen and no weights are updated during fine-tuning.

C Only the final classification layers are updated, while earlier layers remain frozen.

D Random layers are updated without any specific strategy.

💡 Hint

Think about which parts of the model capture general features versus task-specific features.

Typically, the early layers capture general features like edges or textures, so they are frozen. The final layers are updated to learn task-specific patterns.

❓ Hyperparameter

intermediate

Choosing Learning Rate for Fine-tuning

When fine-tuning a pretrained model, which learning rate strategy is generally recommended?

A Use a much smaller learning rate than training from scratch to avoid destroying pretrained weights.

B Use a very large learning rate to quickly adapt the model.

C Use the same learning rate as training from scratch.

D Use a learning rate that increases over time.

💡 Hint

Consider how sensitive pretrained weights are to big changes.

A smaller learning rate helps preserve useful pretrained features while allowing gradual adaptation to the new task.

❓ Predict Output

advanced

Output of Fine-tuning Code Snippet

What will be the output of the following PyTorch code snippet that freezes all layers except the last linear layer?

PyTorch

import torch
import torch.nn as nn

class SimpleModel(nn.Module):
    def __init__(self):
        super().__init__()
        self.features = nn.Sequential(
            nn.Linear(10, 20),
            nn.ReLU(),
            nn.Linear(20, 10)
        )
        self.classifier = nn.Linear(10, 2)

    def forward(self, x):
        x = self.features(x)
        return self.classifier(x)

model = SimpleModel()

# Freeze all layers except classifier
for param in model.features.parameters():
    param.requires_grad = False

trainable_params = [p for p in model.parameters() if p.requires_grad]
print(len(trainable_params))

A 6

B 0

C 4

D 2

💡 Hint

Count the number of parameters in the classifier layer that require gradients.

The classifier layer has weights and biases, so 2 parameter tensors require gradients. All feature layers are frozen.

❓ Metrics

advanced

Interpreting Fine-tuning Training Metrics

During fine-tuning, you observe the training loss decreases steadily but validation accuracy plateaus early. What is the most likely explanation?

A The model is overfitting the training data and not generalizing well.

B The learning rate is too low causing slow learning.

C The model is underfitting and needs more layers unfrozen.

D The dataset is too small to train any model.

💡 Hint

Think about what it means when training improves but validation does not.

When training loss improves but validation accuracy stops improving, the model is likely memorizing training data and not generalizing.

🔧 Debug

expert

Debugging Fine-tuning Freezing Issue

You try to freeze layers in a pretrained model but notice all parameters are still updating during training. Which code snippet correctly freezes the feature extractor layers in PyTorch?

A

for param in model.features:
    param.requires_grad = False

B

for param in model.features.parameters():
    param.requires_grad = False

C model.features.requires_grad = False

D model.features.freeze()

💡 Hint

Remember how to access parameters in a PyTorch module.

Only iterating over model.features.parameters() correctly sets requires_grad to False for all parameters. Other options are invalid or do not affect parameters.