LayerNormalization

Description

Standard tensor.

Input

Three inputs:

X: tensor of type float16.

scale: tensor of type float16, specifying the scale factor.

B: (optional) tensor of type float16, specifying the offset.

Output

Three outputs:

Y: normalized tensor of type float16 or float.

mean: average value, which is an optional output.

InvStdDev: (optional) reciprocal of the standard deviation.

Property

axis: int. First dimension to be normalized. Defaults to -1.

epsilon: float. The default value is 1e-05. Adds a small value to var to avoid division by 0.

stash_type: int, specifying the types of Mean and InvStdDev. Defaults to 1.

ONNX Opset Support

Opset v17