Hugging Face's logo

Diffusers documentation

AutoencoderKLAllegro

Hugging Face's logo
Join the Hugging Face community

and get access to the augmented documentation experience

to get started

AutoencoderKLAllegro

The 3D variational autoencoder (VAE) model with KL loss used in Allegro was introduced in Allegro: Open the Black Box of Commercial-Level Video Generation Model by RhymesAI.

The model can be loaded with the following code snippet.

from diffusers import AutoencoderKLAllegro

vae = AutoencoderKLCogVideoX.from_pretrained("rhymes-ai/Allegro", subfolder="vae", torch_dtype=torch.float32).to("cuda")

AutoencoderKLAllegro

class diffusers.AutoencoderKLAllegro

< >

( in_channels: int = 3 out_channels: int = 3 down_block_types: typing.Tuple[str, ...] = ('AllegroDownBlock3D', 'AllegroDownBlock3D', 'AllegroDownBlock3D', 'AllegroDownBlock3D') up_block_types: typing.Tuple[str, ...] = ('AllegroUpBlock3D', 'AllegroUpBlock3D', 'AllegroUpBlock3D', 'AllegroUpBlock3D') block_out_channels: typing.Tuple[int, ...] = (128, 256, 512, 512) temporal_downsample_blocks: typing.Tuple[bool, ...] = (True, True, False, False) temporal_upsample_blocks: typing.Tuple[bool, ...] = (False, True, True, False) latent_channels: int = 4 layers_per_block: int = 2 act_fn: str = 'silu' norm_num_groups: int = 32 temporal_compression_ratio: float = 4 sample_size: int = 320 scaling_factor: float = 0.13 force_upcast: bool = True )

Parameters

  • in_channels (int, defaults to 3) — Number of channels in the input image.
  • out_channels (int, defaults to 3) — Number of channels in the output.
  • down_block_types (Tuple[str, ...], defaults to ("AllegroDownBlock3D", "AllegroDownBlock3D", "AllegroDownBlock3D", "AllegroDownBlock3D")) — Tuple of strings denoting which types of down blocks to use.
  • up_block_types (Tuple[str, ...], defaults to ("AllegroUpBlock3D", "AllegroUpBlock3D", "AllegroUpBlock3D", "AllegroUpBlock3D")) — Tuple of strings denoting which types of up blocks to use.
  • block_out_channels (Tuple[int, ...], defaults to (128, 256, 512, 512)) — Tuple of integers denoting number of output channels in each block.
  • temporal_downsample_blocks (Tuple[bool, ...], defaults to (True, True, False, False)) — Tuple of booleans denoting which blocks to enable temporal downsampling in.
  • latent_channels (int, defaults to 4) — Number of channels in latents.
  • layers_per_block (int, defaults to 2) — Number of resnet or attention or temporal convolution layers per down/up block.
  • act_fn (str, defaults to "silu") — The activation function to use.
  • norm_num_groups (int, defaults to 32) — Number of groups to use in normalization layers.
  • temporal_compression_ratio (int, defaults to 4) — Ratio by which temporal dimension of samples are compressed.
  • sample_size (int, defaults to 320) — Default latent size.
  • scaling_factor (float, defaults to 0.13235) — The component-wise standard deviation of the trained latent space computed using the first batch of the training set. This is used to scale the latent space to have unit variance when training the diffusion model. The latents are scaled with the formula z = z * scaling_factor before being passed to the diffusion model. When decoding, the latents are scaled back to the original scale with the formula: z = 1 / scaling_factor * z. For more details, refer to sections 4.3.2 and D.1 of the High-Resolution Image Synthesis with Latent Diffusion Models paper.
  • force_upcast (bool, default to True) — If enabled it will force the VAE to run in float32 for high image resolution pipelines, such as SD-XL. VAE can be fine-tuned / trained to a lower range without loosing too much precision in which case force_upcast can be set to False - see: https://huggingface.co/madebyollin/sdxl-vae-fp16-fix

A VAE model with KL loss for encoding videos into latents and decoding latent representations into videos. Used in Allegro.

This model inherits from ModelMixin. Check the superclass documentation for it’s generic methods implemented for all models (such as downloading or saving).

wrapper

< >

( *args **kwargs )

wrapper

< >

( *args **kwargs )

disable_slicing

< >

( )

Disable sliced VAE decoding. If enable_slicing was previously enabled, this method will go back to computing decoding in one step.

disable_tiling

< >

( )

Disable tiled VAE decoding. If enable_tiling was previously enabled, this method will go back to computing decoding in one step.

enable_slicing

< >

( )

Enable sliced VAE decoding. When this option is enabled, the VAE will split the input tensor in slices to compute decoding in several steps. This is useful to save some memory and allow larger batch sizes.

enable_tiling

< >

( )

Enable tiled VAE decoding. When this option is enabled, the VAE will split the input tensor into tiles to compute decoding and encoding in several steps. This is useful for saving a large amount of memory and to allow processing larger images.

forward

< >

( sample: Tensor sample_posterior: bool = False return_dict: bool = True generator: typing.Optional[torch._C.Generator] = None )

Parameters

  • sample (torch.Tensor) — Input sample.
  • sample_posterior (bool, optional, defaults to False) — Whether to sample from the posterior.
  • return_dict (bool, optional, defaults to True) — Whether or not to return a DecoderOutput instead of a plain tuple.
  • generator (torch.Generator, optional) — PyTorch random number generator.

AutoencoderKLOutput

class diffusers.models.modeling_outputs.AutoencoderKLOutput

< >

( latent_dist: DiagonalGaussianDistribution )

Parameters

  • latent_dist (DiagonalGaussianDistribution) — Encoded outputs of Encoder represented as the mean and logvar of DiagonalGaussianDistribution. DiagonalGaussianDistribution allows for sampling latents from the distribution.

Output of AutoencoderKL encoding method.

DecoderOutput

class diffusers.models.autoencoders.vae.DecoderOutput

< >

( sample: Tensor commit_loss: typing.Optional[torch.FloatTensor] = None )

Parameters

  • sample (torch.Tensor of shape (batch_size, num_channels, height, width)) — The decoded output sample from the last layer of the model.

Output of decoding method.

< > Update on GitHub

←AutoencoderKL AutoencoderKLCogVideoX→