Diffusers documentation

Normalization layers

Diffusers

You are viewing main version, which requires installation from source. If you'd like regular pip install, checkout the latest stable version (v0.32.2).

Join the Hugging Face community

and get access to the augmented documentation experience

Collaborate on models, datasets and Spaces

Faster examples with accelerated inference

Switch between documentation themes

to get started

Normalization layers

Customized normalization layers for supporting various models in 🤗 Diffusers.

AdaLayerNorm

class diffusers.models.normalization.AdaLayerNorm

< source >

( embedding_dim: int num_embeddings: typing.Optional[int] = None output_dim: typing.Optional[int] = None norm_elementwise_affine: bool = False norm_eps: float = 1e-05 chunk_dim: int = 0 )

Parameters

embedding_dim (int) — The size of each embedding vector.
num_embeddings (int, optional) — The size of the embeddings dictionary.
output_dim (int, optional) —
norm_elementwise_affine (bool, defaults to `False) —
norm_eps (bool, defaults to False) —
chunk_dim (int, defaults to 0) —

Norm layer modified to incorporate timestep embeddings.

AdaLayerNormZero

class diffusers.models.normalization.AdaLayerNormZero

< source >

( embedding_dim: int num_embeddings: typing.Optional[int] = None norm_type = 'layer_norm' bias = True )

Parameters

embedding_dim (int) — The size of each embedding vector.
num_embeddings (int) — The size of the embeddings dictionary.

Norm layer adaptive layer norm zero (adaLN-Zero).

AdaLayerNormSingle

class diffusers.models.normalization.AdaLayerNormSingle

< source >

( embedding_dim: int use_additional_conditions: bool = False )

Parameters

embedding_dim (int) — The size of each embedding vector.
use_additional_conditions (bool) — To use additional conditions for normalization or not.

Norm layer adaptive layer norm single (adaLN-single).

As proposed in PixArt-Alpha (see: https://arxiv.org/abs/2310.00426; Section 2.3).

AdaGroupNorm

class diffusers.models.normalization.AdaGroupNorm

< source >

( embedding_dim: int out_dim: int num_groups: int act_fn: typing.Optional[str] = None eps: float = 1e-05 )

Parameters

embedding_dim (int) — The size of each embedding vector.
num_embeddings (int) — The size of the embeddings dictionary.
num_groups (int) — The number of groups to separate the channels into.
act_fn (str, optional, defaults to None) — The activation function to use.
eps (float, optional, defaults to 1e-5) — The epsilon value to use for numerical stability.

GroupNorm layer modified to incorporate timestep embeddings.

AdaLayerNormContinuous

class diffusers.models.normalization.AdaLayerNormContinuous

< source >

( embedding_dim: int conditioning_embedding_dim: int elementwise_affine = True eps = 1e-05 bias = True norm_type = 'layer_norm' )

Parameters

embedding_dim (int) — Embedding dimension to use during projection.
conditioning_embedding_dim (int) — Dimension of the input condition.
elementwise_affine (bool, defaults to True) — Boolean flag to denote if affine transformation should be applied.
eps (float, defaults to 1e-5) — Epsilon factor.
bias (bias, defaults to True) — Boolean flag to denote if bias should be use.
norm_type (str, defaults to "layer_norm") — Normalization layer to use. Values supported: “layer_norm”, “rms_norm”.

Adaptive normalization layer with a norm layer (layer_norm or rms_norm).

RMSNorm

class diffusers.models.normalization.RMSNorm

< source >

( dim eps: float elementwise_affine: bool = True bias: bool = False )

Parameters

dim (int) — Number of dimensions to use for weights. Only effective when elementwise_affine is True.
eps (float) — Small value to use when calculating the reciprocal of the square-root.
elementwise_affine (bool, defaults to True) — Boolean flag to denote if affine transformation should be applied.
bias (bool, defaults to False) — If also training the bias param.

RMS Norm as introduced in https://arxiv.org/abs/1910.07467 by Zhang et al.

GlobalResponseNorm

class diffusers.models.normalization.GlobalResponseNorm

< source >

( dim )

Parameters

dim (int) — Number of dimensions to use for the gamma and beta.

Global response normalization as introduced in ConvNeXt-v2 (https://arxiv.org/abs/2301.00808).

LuminaLayerNormContinuous

class diffusers.models.normalization.LuminaLayerNormContinuous

< source >

( embedding_dim: int conditioning_embedding_dim: int elementwise_affine = True eps = 1e-05 bias = True norm_type = 'layer_norm' out_dim: typing.Optional[int] = None )