gelu¶
逐元素计算 Gelu激活函数。更多细节请参考 Gaussian Error Linear Units 。
如果使用近似计算:
\[out = 0.5 * x * (1 + tanh(\sqrt{\frac{2}{\pi}} * (x + 0.044715x^{3})))\]
如果不使用近似计算:
\[out = 0.5 * x * (1 + erf(\frac{x}{\sqrt{2}}))\]
- 参数:
-
x (Variable) - Gelu Op 的输入,多维 Tensor 或 LoDTensor,数据类型为 float32 或 float64。
approximate (bool, 可选) - 是否使用近似计算,默认值为 False。
- 返回:
-
多维 Tensor 或 LoDTensor, 数据类型为 float32 或 float64, 和输入 x 的数据类型相同,形状和输入 x 相同。
- 返回类型:
-
Variable
代码示例:
# declarative mode
import numpy as np
from paddle import fluid
x = fluid.data(name="x", shape=(-1, 3), dtype="float32")
y = fluid.layers.gelu(x)
place = fluid.CPUPlace()
exe = fluid.Executor(place)
start = fluid.default_startup_program()
main = fluid.default_main_program()
data = np.random.randn(2, 3).astype("float32")
exe.run(start)
y_np, = exe.run(main, feed={"x": data}, fetch_list=[y])
data
# array([[ 0.87165993, -1.0541513 , -0.37214822],
# [ 0.15647964, 0.32496083, 0.33045998]], dtype=float32)
y_np
# array([[ 0.70456535, -0.15380788, -0.13207214],
# [ 0.08796856, 0.20387867, 0.2080159 ]], dtype=float32)
# imperative mode
import numpy as np
from paddle import fluid
import paddle.fluid.dygraph as dg
data = np.random.randn(2, 3).astype("float32")
place = fluid.CPUPlace()
with dg.guard(place) as g:
x = dg.to_variable(data)
y = fluid.layers.gelu(x)
y_np = y.numpy()
data
# array([[ 0.87165993, -1.0541513 , -0.37214822],
# [ 0.15647964, 0.32496083, 0.33045998]], dtype=float32)
y_np
# array([[ 0.70456535, -0.15380788, -0.13207214],
# [ 0.08796856, 0.20387867, 0.2080159 ]], dtype=float32)