Here are the top PyTorch interview questions to prepare for your next role.
1️⃣ What is CUDA?
- A) A framework for deep learning libraries
- B) An API for executing mathematical calculations on the GPU
- C) A type of neural network architecture
- D) A library for distributed computing
2️⃣ What does the requires_grad=True attribute on a Tensor signify?
- A) It allows the Tensor to track operations for computing gradients.
- B) It enables the Tensor to be saved to disk.
- C) It ensures that the Tensor will never be zero.
- D) It makes the Tensor immutable.
3️⃣ What is the difference between model.train() and model.eval() in PyTorch?
- A) It modifies the behavior of specific layers like Dropout and Batch Normalization to suit training or testing.
- B) It globally enables or disables gradient computation to optimize memory usage during inference.
- C) It freezes all model parameters to prevent the optimizer from updating weights during the backward pass.
- D) It automatically shuffles the input data loader and applies random augmentations to the features.
4️⃣ What is the role of loss.backward() in PyTorch?
- A) It updates the weights of the neural network.
- B) It computes the gradient of the loss with respect to the model parameters.
- C) It performs forward propagation.
- D) It initializes the model parameters.
5️⃣ What is the role of optimizer.step()?
- A) It zeroes the gradients of all parameters
- B) It updates the parameters based on the current gradients
- C) It computes the gradients based on the loss
- D) It resets the optimizer state