Customized normalization layers for supporting various models in 🤗 Diffusers.
( embedding_dim: int num_embeddings: int )
Norm layer modified to incorporate timestep embeddings.
( embedding_dim: int num_embeddings: int )
Norm layer adaptive layer norm zero (adaLN-Zero).
( embedding_dim: int use_additional_conditions: bool = False )
Norm layer adaptive layer norm single (adaLN-single).
As proposed in PixArt-Alpha (see: https://arxiv.org/abs/2310.00426; Section 2.3).
( embedding_dim: int out_dim: int num_groups: int act_fn: typing.Optional[str] = None eps: float = 1e-05 )
Parameters
int
) — The size of each embedding vector. int
) — The size of the embeddings dictionary. int
) — The number of groups to separate the channels into. str
, optional, defaults to None
) — The activation function to use. float
, optional, defaults to 1e-5
) — The epsilon value to use for numerical stability. GroupNorm layer modified to incorporate timestep embeddings.