Mapping of PyTorch tensors in device memory

I was wondering how are pytorch tensors mapped in memory.
For eg.

b = torch.rand(10)

Now if I do

>>> id(b[0])
>>> id(b[1])

Both of them are exactly same memory addresses?
I was hoping given that the values inside b are float32, the id's would be separated by 4 bytes.
Am I wrong somewhere?
I wanted to know how is a torch tensor mapped inside the memory?

Thank you!