GELU

class paddle.nn. GELU ( approximate: Union[Literal['tanh', 'none'], bool] = False, name: Optional[str] = None ) [source]

GELU Activation.

approximate parameter must be True, False, “tanh”, “none”.

If approximate is True or “tanh”

\[GELU(x) = 0.5 * x * (1 + tanh(\sqrt{\frac{2}{\pi}} * (x + 0.044715x^{3})))\]

else

\[GELU(x) = 0.5 * x * (1 + erf(\frac{x}{\sqrt{2}}))\]

Parameters

• approximate (str|bool, optional) – Whether to enable approximation. Default is False.

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

Shape:

• input: Tensor with any shape.

• output: Tensor with the same shape as input.

Examples

>>> import paddle
>>> x = paddle.to_tensor([[-1, 0.5],[1, 1.5]])
>>> m = paddle.nn.GELU()
>>> out = m(x)
>>> print(out)
Tensor(shape=[2, 2], dtype=float32, place=Place(cpu), stop_gradient=True,
[[-0.15865529,  0.34573123],
 [ 0.84134471,  1.39978933]])
>>> m = paddle.nn.GELU(False)
>>> out = m(x)
>>> print(out)
Tensor(shape=[2, 2], dtype=float32, place=Place(cpu), stop_gradient=True,
[[-0.15865529,  0.34573123],
 [ 0.84134471,  1.39978933]])
>>> m = paddle.nn.GELU("none")
>>> out = m(x)
>>> print(out)
Tensor(shape=[2, 2], dtype=float32, place=Place(cpu), stop_gradient=True,
[[-0.15865529,  0.34573123],
 [ 0.84134471,  1.39978933]])
>>> m = paddle.nn.GELU("tanh")
>>> out = m(x)
>>> print(out)
Tensor(shape=[2, 2], dtype=float32, place=Place(cpu), stop_gradient=True,
[[-0.15880796,  0.34571400],
 [ 0.84119201,  1.39957154]])
>>> m = paddle.nn.GELU(True)
>>> out = m(x)
>>> print(out)
Tensor(shape=[2, 2], dtype=float32, place=Place(cpu), stop_gradient=True,
[[-0.15880796,  0.34571400],
 [ 0.84119201,  1.39957154]])

forward ( x: Tensor ) → Tensor

forward¶

Defines the computation performed at every call. Should be overridden by all subclasses.

Parameters

• *inputs (tuple) – unpacked tuple arguments

• **kwargs (dict) – unpacked dict arguments

extra_repr ( ) → str

extra_repr¶

Extra representation of this layer, you can have custom implementation of your own layer.