from_dlpack¶

paddle. from_dlpack ( dlpack: SupportDLPack | CapsuleType ) → Tensor [source]

Decodes a DLPack to a tensor. The returned Paddle tensor will share the memory with the tensor from given dlpack.

Parameters

dlpack (SupportDLPack | CapsuleType) –

A PyCapsule object with the dltensor, or that implements ‘__dlpack__’ and ‘__dlpack_device__’ methods.

If dlpack is a tensor (or ndarray) object, it must support the __dlpack__ protocol (i.e., have a dlpack.__dlpack__ method). Otherwise dlpack may be a DLPack capsule, which is an opaque PyCapsule instance, typically produced by a to_dlpack function or method.

Returns

A tensor decoded from DLPack. The data type of returned tensor

can be one of: int32, int64, float16, float32 and float64. The device of returned tensor can be one of: CPU, CUDAPlace, CUDAPinnedPlace.

Return type

out (Tensor)

Examples

>>> import paddle
>>> # From DLPack capsule
>>> x = paddle.to_tensor([[0.2, 0.3, 0.5, 0.9],
...                       [0.1, 0.2, 0.6, 0.7]], place="cpu")
>>> dlpack = paddle.utils.dlpack.to_dlpack(x)

>>> y = paddle.utils.dlpack.from_dlpack(dlpack)
>>> # dlpack capsule will be renamed to 'used_dltensor' after decoded
>>> print(dlpack)
>>> 
<capsule object "used_dltensor" at 0x7f6103c681b0>

>>> print(y)
Tensor(shape=[2, 4], dtype=float32, place=Place(cpu), stop_gradient=True,
       [[0.20000000, 0.30000001, 0.50000000, 0.89999998],
        [0.10000000, 0.20000000, 0.60000002, 0.69999999]])
>>> # data of tensor x is shared with tensor y
>>> y[0, 0] = 10.0
>>> print(x)
Tensor(shape=[2, 4], dtype=float32, place=Place(gpu:0), stop_gradient=True,
       [[10.       , 0.30000001, 0.50000000, 0.89999998],
        [0.10000000, 0.20000000, 0.60000002, 0.69999999]])

>>> # Directly from external tensor that implements '__dlpack__' and '__dlpack_device__' methods
>>> import paddle
>>> import numpy as np
>>> x = np.array([[0.2, 0.3, 0.5, 0.9],
...              [0.1, 0.2, 0.6, 0.7]])
>>> y = paddle.utils.dlpack.from_dlpack(x)
>>> y[0, 0] = 10.0
>>> # data of tensor x is shared with tensor y
>>> print(x)
[[10.   0.3  0.5  0.9]
[ 0.1  0.2  0.6  0.7]]