GPU tensors (to, cuda) in PyTorch - Cheat Sheet & Quick Revision

Recall & Review

beginner

What is a GPU tensor in PyTorch?

A GPU tensor is a tensor stored in the GPU's memory, allowing faster computation by using the GPU's parallel processing power.

beginner

How do you move a tensor to the GPU using the to() method?

You call tensor.to('cuda') to move the tensor to the GPU if available.

beginner

What does tensor.cuda() do in PyTorch?

It moves the tensor to the default GPU device, similar to tensor.to('cuda').

intermediate

Why should you check if a GPU is available before moving tensors to it?

Because if no GPU is available, trying to move tensors to GPU will cause errors. Use torch.cuda.is_available() to check.

intermediate

What is the difference between tensor.to('cuda') and tensor.cuda()?

tensor.to('cuda') can specify any device and is more flexible. tensor.cuda() moves tensor to the default GPU device. Both move tensors to GPU memory.

Which PyTorch method moves a tensor to the GPU?

A tensor.numpy()

B tensor.to('cpu')

C tensor.reshape()

D tensor.to('cuda')

The to('cuda') method moves a tensor to the GPU device.

What does torch.cuda.is_available() check?

A If GPU is available

B If CPU is available

C If tensor is on GPU

D If CUDA is installed

It checks if a GPU device is available for PyTorch to use.

What happens if you try to move a tensor to GPU when no GPU is available?

A Tensor moves to GPU anyway

B Tensor stays on CPU

C Error occurs

D Tensor is deleted

Trying to move a tensor to GPU without a GPU causes an error.

Which method is more flexible for moving tensors between devices?

A tensor.cuda()

B tensor.to()

C tensor.cpu()

D tensor.numpy()

tensor.to() can move tensors to any device, not just GPU.

What device does tensor.cuda() move the tensor to?

A Default GPU device

B CPU

C Random device

D Disk storage

tensor.cuda() moves the tensor to the default GPU device.

Explain how to move a tensor to GPU in PyTorch and why it is useful.

Think about device names and checking if GPU exists.

Describe the difference between tensor.to('cuda') and tensor.cuda().

Consider device flexibility and default behavior.