PyTorch vs TensorFlow comparison - Practice Questions

Challenge - 5 Problems

🧠 Conceptual

intermediate

Key difference in execution style between PyTorch and TensorFlow

Which statement best describes the main difference in how PyTorch and TensorFlow execute operations?

A Both PyTorch and TensorFlow exclusively use static computation graphs for faster execution.

B TensorFlow only supports eager execution, while PyTorch only supports static computation graphs.

C PyTorch uses eager execution by default, while TensorFlow originally used static graphs but now supports eager execution.

D PyTorch and TensorFlow both use eager execution only for training and static graphs only for inference.

💡 Hint

Think about how each framework builds and runs computations during model training.

PyTorch runs operations immediately (eager execution), making debugging easier. TensorFlow originally used static graphs but added eager mode later.

❓ Model Choice

intermediate

Choosing framework for dynamic neural networks

You want to build a neural network where the structure changes during training based on input data. Which framework is generally better suited for this?

A TensorFlow, because it has better support for static graphs.

B PyTorch, because it supports dynamic computation graphs naturally.

C TensorFlow, because it is faster for all types of models.

D Neither, both frameworks handle dynamic graphs equally well.

💡 Hint

Consider which framework builds the graph on the fly during execution.

PyTorch builds computation graphs dynamically during execution, making it easier to create models that change structure.

❓ Metrics

advanced

Comparing training speed metrics

You train the same model architecture on the same dataset using PyTorch and TensorFlow. Which metric would best compare their training speed fairly?

A Memory usage during inference.

B Final accuracy after training completes.

C Number of lines of code used to build the model.

D Total training time in seconds for a fixed number of epochs.

💡 Hint

Focus on measuring speed, not accuracy or code complexity.

Training time for a fixed number of epochs directly measures speed, while accuracy or code length do not.

🔧 Debug

advanced

Identifying cause of slower training in TensorFlow

You notice your TensorFlow model trains slower than the PyTorch version. Which of these is a likely cause?

A TensorFlow is running in graph mode without proper graph optimizations enabled.

B PyTorch does not support GPU acceleration, so it runs faster on CPU.

C TensorFlow models always use more memory, causing slowdowns.

D PyTorch uses static graphs which are faster by default.

💡 Hint

Think about how TensorFlow executes graphs and what might slow it down.

If TensorFlow runs in graph mode but graph optimizations are off, it can be slower. PyTorch uses eager mode by default.

❓ Predict Output

expert

Output difference in simple tensor operation

What is the output of this PyTorch code snippet? import torch x = torch.tensor([1.0, 2.0, 3.0], requires_grad=True) y = x * 2 y = y.mean() y.backward() print(x.grad)

PyTorch

import torch
x = torch.tensor([1.0, 2.0, 3.0], requires_grad=True)
y = x * 2
y = y.mean()
y.backward()
print(x.grad)

A tensor([0.6667, 0.6667, 0.6667])

B tensor([2.0, 2.0, 2.0])

C tensor([1.0, 1.0, 1.0])

D None

💡 Hint

Recall how gradients are computed for mean of scaled tensor.

The gradient of y = mean(x * 2) with respect to x is 2 divided by number of elements (3), so each element's gradient is 2/3 ≈ 0.6667.