Module norm 

Normalisation requests for the cuDNN actor: BatchNorm (training + inference + persistent), LayerNorm, InstanceNorm, GroupNorm, RMSNorm.

Structs

BatchNormRequest

BatchNorm — training-mode running-mean/var update + per-channel scale + bias.

GroupNormRequest

GroupNorm — generalisation of BatchNorm/LayerNorm/InstanceNorm with groups channel partitions.

InstanceNormRequest

InstanceNorm — normalises per-sample, per-channel.

LayerNormRequest

LayerNorm — normalises across the trailing axes, scale + bias applied per-feature.

NormBwdRequest

Norm backward request, applies to BatchNorm / LayerNorm / InstanceNorm / GroupNorm uniformly via mode.

Functions

build_norm_fwd_graph

Build the spec-side norm-fwd op graph.