GroupNorm

GroupNorm(num_groups, num_channels, eps=1e-05, affine=True, device=None, dtype=None, *, activation=None)

Each sample normalizes over each of ``num_groups`` groups of consecutive
channels and the spatial dimensions of channels-last ``[batch, *spatial,
num_channels]`` input, the same in training and evaluation.

Args:
    num_groups: the group count; it divides ``num_channels``.
    num_channels: the channel count of the input.
    eps: added to the variance for numerical stability.
    affine: declare the per-channel ``weight`` (scale) and ``bias``
        (shift) slots.
    device: where the layer's tensors live; ``None`` declares on the cpu.
    dtype: the dtype the layer declares its slots at; ``None`` declares
        the default dtype (:func:`~clika_runtime.get_default_dtype`).
    activation: a fused epilogue applied to every output; ``None`` for
        none.

__init__

__init__(self, num_groups: 'int', num_channels: 'int', eps: 'float' = 1e-05, affine: 'bool' = True, device: 'Device | str | None' = None, dtype: 'DtypeLike | None' = None, *, activation: 'ActivationLike' = None) -> 'None'

Initialize self. See help(type(self)) for accurate signature.

extra_repr

extra_repr(self) -> 'str'

extra_repr() -> str

One line of per-class detail for :meth:__repr__; a layer prints its geometry here (in_features=64, out_features=256).

forward

forward(self, input: 'Tensor') -> 'Tensor'

forward(input) -> Tensor

[batch, *spatial, num_channels] to the same shape; the output dtype follows the input.

set_weights

set_weights(self, weight: 'Tensor', bias: 'Tensor | None' = None) -> 'None'

set_weights(weight, bias=None) -> None

Bind the [num_channels] scale and shift; raises RuntimeError on a shape mismatch or on a layer made with affine=False.