Parameter

class paddle.nn. Parameter ( data: Tensor | None = None, requires_grad: bool = True ) [source]

Parameter is a subclass of Tensor, which is a persistable Tensor that can be updated by optimizers during training.

Parameters

• data (Tensor, optional) – The initial data for the Parameter. If None, an empty Tensor will be created. Default: None.

• requires_grad (bool, optional) – Whether this Parameter requires gradient computation. If True, the Parameter will accumulate gradients during backward pass. Default: True.

Examples

>>> import paddle
>>> # Create a Parameter from existing Tensor
>>> weight = paddle.to_tensor([1.0, 2.0, 3.0])
>>> param = paddle.nn.Parameter(weight)
>>> print(param)

>>> # Create a Parameter without initial data
>>> param = paddle.nn.Parameter()
>>> print(param)