clone

→Tensor

clone(input: Tensor)

source

Return a deep copy of input with its own storage.

The clone shares no memory with the source. Autograd state is preserved: gradients flow back through clone as the identity. To detach from the graph as well, compose with detach.

Parameters

inputTensor

Source tensor.

Returns

Tensor

Newly-allocated tensor with the same shape, dtype, device and contents as input.

Notes

Use clone (not contiguous) when you specifically need a fresh buffer regardless of the input's current layout — for example before in-place mutation that must not alias the source.

Examples

>>> import lucid
>>> x = lucid.tensor([1.0, 2.0, 3.0])
>>> y = lucid.clone(x)
>>> y.data_ptr() == x.data_ptr()
False