Diffusers documentation
AllegroTransformer3DModel
AllegroTransformer3DModel
A Diffusion Transformer model for 3D data from 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 AllegroTransformer3DModel
transformer = AllegroTransformer3DModel.from_pretrained("rhymes-ai/Allegro", subfolder="transformer", dtype=torch.bfloat16).to("cuda")AllegroTransformer3DModel
class diffusers.AllegroTransformer3DModel
< source >( patch_size: int = 2patch_size_t: int = 1num_attention_heads: int = 24attention_head_dim: int = 96in_channels: int = 4out_channels: int = 4num_layers: int = 32dropout: float = 0.0cross_attention_dim: int = 2304attention_bias: bool = Truesample_height: int = 90sample_width: int = 160sample_frames: int = 22activation_fn: str = 'gelu-approximate'norm_elementwise_affine: bool = Falsenorm_eps: float = 1e-06caption_channels: int = 4096interpolation_scale_h: float = 2.0interpolation_scale_w: float = 2.0interpolation_scale_t: float = 2.2 )
forward
< source >( hidden_states: Tensorencoder_hidden_states: Tensortimestep: LongTensorattention_mask: typing.Optional[torch.Tensor] = Noneencoder_attention_mask: typing.Optional[torch.Tensor] = Noneimage_rotary_emb: tuple[torch.Tensor, torch.Tensor] | None = Nonereturn_dict: bool = True )
Parameters
- hidden_states (
torch.Tensorof shape(batch_size, num_channels, num_frames, height, width)) — Inputhidden_states. - encoder_hidden_states (
torch.Tensorof shape(batch_size, sequence_len, embed_dims)) — Conditional embeddings (embeddings computed from the input conditions such as prompts) to use. - timestep (
torch.LongTensor) — Used to indicate denoising step. - attention_mask (
torch.Tensor, optional) — Self-attention mask applied tohidden_states. - encoder_attention_mask (
torch.Tensor, optional) — Cross-attention mask applied toencoder_hidden_states. - image_rotary_emb (
tupleoftorch.Tensor, optional) — Pre-computed rotary positional embeddings. - return_dict (
bool, optional, defaults toTrue) — Whether or not to return a~models.transformer_2d.Transformer2DModelOutputinstead of a plain tuple.
The AllegroTransformer3DModel forward method.
Transformer2DModelOutput
class diffusers.models.modeling_outputs.Transformer2DModelOutput
< source >( sample: torch.Tensor )
Parameters
- sample (
torch.Tensorof shape(batch_size, num_channels, height, width)or(batch size, num_vector_embeds - 1, num_latent_pixels)if Transformer2DModel is discrete) — The hidden states output conditioned on theencoder_hidden_statesinput. If discrete, returns probability distributions for the unnoised latent pixels.
The output of Transformer2DModel.