scale

paddle. scale ( x: Tensor, scale: float | Tensor = 1.0, bias: float = 0.0, bias_after_scale: bool = True, act: str | None = None, name: str | None = None ) → Tensor [source]

Scale operator.

Putting scale and bias to the input Tensor as following:

bias_after_scale is True:

\[Out=scale*X+bias\]

bias_after_scale is False:

\[Out=scale*(X+bias)\]
Parameters :

  • x (Tensor) – Input N-D Tensor of scale operator. Data type can be bfloat16, float16, float32, float64, int8, int16, int32, int64, uint8, complex64, complex128.

  • scale (float|Tensor) – The scale factor of the input, it should be a float number or a 0-D Tensor with shape [] and data type as float32.

  • bias (float) – The bias to be put on the input.

  • bias_after_scale (bool) – Apply bias addition after or before scaling. It is useful for numeric stability in some circumstances.

  • act (str|None, optional) – Activation applied to the output such as tanh, softmax, sigmoid, relu.

  • name (str|None, optional) – Name for the operation (optional, default is None). For more information, please refer to Name.

Returns :

Output Tensor of scale operator, with shape and data type same as input.

Return type :

Tensor

Examples

>>> # scale as a float32 number
>>> import paddle

>>> data = paddle.arange(6).astype("float32").reshape([2, 3])
>>> print(data)
Tensor(shape=[2, 3], dtype=float32, place=Place(cpu), stop_gradient=True,
[[0., 1., 2.],
 [3., 4., 5.]])
>>> res = paddle.scale(data, scale=2.0, bias=1.0)
>>> print(res)
Tensor(shape=[2, 3], dtype=float32, place=Place(cpu), stop_gradient=True,
[[1. , 3. , 5. ],
 [7. , 9. , 11.]])

>>> # scale with parameter scale as a Tensor
>>> import paddle

>>> data = paddle.arange(6).astype("float32").reshape([2, 3])
>>> print(data)
Tensor(shape=[2, 3], dtype=float32, place=Place(cpu), stop_gradient=True,
[[0., 1., 2.],
 [3., 4., 5.]])
>>> factor = paddle.to_tensor([2], dtype='float32')
>>> res = paddle.scale(data, scale=factor, bias=1.0)
>>> print(res)
Tensor(shape=[2, 3], dtype=float32, place=Place(cpu), stop_gradient=True,
[[1. , 3. , 5. ],
 [7. , 9. , 11.]])