group_norm

Normalize channel groups per example, with learned per-channel affine.

Category: normalization · Identity: group_norm@1

Shape

x[B, C, ...] -> out[B, C, ...]

Port	Direction	Pattern	dtype
x	input	x[B, C, ...]	compute
out	output	out[B, C, ...]	compute

Arguments

Name	Type	Default	Constraints	Description
num_groups (positional)	int	required	>= 1	Number of channel groups; must divide the channel count.
num_channels	int	inferred	>= 1; <= 2147483647; binds C	Channels at axis 1. Inferred from the incoming tensor.
eps	float	1e-05	> 0	Added to the variance for stability.
affine	bool	True	—	Learn per-channel scale and bias.

Description

Splits the channel axis into num_groups groups and normalizes each group over its channels and spatial positions using population statistics. Behavior is identical in train and eval mode. Parameters are weight and bias when affine is true.

Examples

Example 1

conv(16, kernel_size=3, padding=1)
group_norm(4)
relu()

Input ['B', 3, 8, 8] → output ['B', 16, 8, 8].

Network: [B, 3, 8, 8] -> [B, 16, 8, 8]  dtype=float32
index  name  operation   input shapes     output shapes
0      n0    conv        x=[B, 3, 8, 8]   out=[B, 16, 8, 8]
1      n1    group_norm  x=[B, 16, 8, 8]  out=[B, 16, 8, 8]
2      n2    relu        x=[B, 16, 8, 8]  out=[B, 16, 8, 8]

Parameters: 480