Hugging Face's logo

Diffusers documentation

LatteTransformer3DModel

You are viewing main version, which requires installation from source. If you'd like regular pip install, checkout the latest stable version (v0.42.1).
Hugging Face's logo
Join the Hugging Face community

and get access to the augmented documentation experience

to get started

LatteTransformer3DModel

Model Diffusion Transformer untuk data 3D dari Latte.

LatteTransformer3DModel

kelas diffusers.LatteTransformer3DModel

< >

( num_attention_heads: int = 16attention_head_dim: int = 88in_channels: int | None = Noneout_channels: int | None = Nonenum_layers: int = 1dropout: float = 0.0cross_attention_dim: int | None = Noneattention_bias: bool = Falsesample_size: int = 64patch_size: int | None = Noneactivation_fn: str = 'geglu'num_embeds_ada_norm: int | None = Nonenorm_type: str = 'layer_norm'norm_elementwise_affine: bool = Truenorm_eps: float = 1e-05caption_channels: int = Nonevideo_length: int = 16 )

forward

< >

( hidden_states: Tensortimestep: typing.Optional[torch.LongTensor] = Noneencoder_hidden_states: typing.Optional[torch.Tensor] = Noneencoder_attention_mask: typing.Optional[torch.Tensor] = Noneenable_temporal_attentions: bool = Truereturn_dict: bool = True )

Parameter

  • hidden_states (torch.Tensor dengan bentuk (batch size, channel, num_frame, height, width)) — hidden_states masukan.
  • timestep ( torch.LongTensor, opsional) — Digunakan untuk menunjukkan langkah denoising. Timestep opsional yang diterapkan sebagai embedding di AdaLayerNorm.
  • encoder_hidden_states ( torch.FloatTensor dengan bentuk (batch size, sequence len, embed dims), opsional) — Embedding kondisional untuk lapisan cross attention. Jika tidak diberikan, cross-attention secara default menggunakan self-attention.
  • encoder_attention_mask ( torch.Tensor, opsional) — Mask cross-attention yang diterapkan pada encoder_hidden_states. Dua format didukung:
    • Mask (batcheight, sequence_length) True = simpan, False = buang.
    • Bias (batcheight, 1, sequence_length) 0 = simpan, -10000 = buang.

    Jika ndim == 2: akan diinterpretasikan sebagai mask, kemudian dikonversi menjadi bias yang konsisten dengan format di atas. Bias ini akan ditambahkan ke skor cross-attention.

  • enable_temporal_attentions — (bool, opsional, default True): Apakah akan mengaktifkan temporal attentions.
  • return_dict (bool, opsional, default True) — Apakah akan mengembalikan ~models.unet_2d_condition.UNet2DConditionOutput alih-alih tuple biasa.

Metode forward dari LatteTransformer3DModel.

Update on GitHub

←Krea2Transformer2DModel LongCatImageTransformer2DModel→