Why is `id(tensors.storage())` different every time?

1

Demo:

a = torch.randn(10, 10)
b = a[1:, :]
a.shape
Out[45]: torch.Size([10, 10])
b.shape
Out[46]: torch.Size([9, 10])
id(a)
Out[47]: 139889952764416
id(a)
Out[48]: 139889952764416
id(b)
Out[49]: 139889948003072
id(b)
Out[50]: 139889948003072
id(a.storage())
Out[51]: 139889948211136
id(a.storage())
Out[52]: 139889948120832
id(b.storage())
Out[53]: 139889952743616
id(b.storage())
Out[54]: 139892027430848

Why is id(a.storage()) or id(b.storage()) different every time?

AND How can I tell if a and b here take up the same memory space?

2

the python storage object is construct when calling .storage(), so id is different.

test x.storage().data_ptr()

1 Like