## LatteTransformer3DModel

来自 [Latte](https://github.com/Vchitect/Latte) 的用于 3D 数据的 Diffusion Transformer 模型。

## LatteTransformer3DModel[[diffusers.LatteTransformer3DModel]]

#### diffusers.LatteTransformer3DModel[[diffusers.LatteTransformer3DModel]]

```python
diffusers.LatteTransformer3DModel(num_attention_heads: int = 16, attention_head_dim: int = 88, in_channels: int | None = None, out_channels: int | None = None, num_layers: int = 1, dropout: float = 0.0, cross_attention_dim: int | None = None, attention_bias: bool = False, sample_size: int = 64, patch_size: int | None = None, activation_fn: str = 'geglu', num_embeds_ada_norm: int | None = None, norm_type: str = 'layer_norm', norm_elementwise_affine: bool = True, norm_eps: float = 1e-05, caption_channels: int = None, video_length: int = 16)
```

[源代码](https://github.com/huggingface/diffusers/blob/v0.43.0/src/diffusers/models/transformers/latte_transformer_3d.py#L27)

#### forward[[diffusers.LatteTransformer3DModel.forward]]

```python
forward(hidden_states: Tensor, timestep: typing.Optional[torch.LongTensor] = None, encoder_hidden_states: typing.Optional[torch.Tensor] = None, encoder_attention_mask: typing.Optional[torch.Tensor] = None, enable_temporal_attentions: bool = True, return_dict: bool = True)
```

[源代码](https://github.com/huggingface/diffusers/blob/v0.43.0/src/diffusers/models/transformers/latte_transformer_3d.py#L166)

**参数：**

hidden_states (`torch.Tensor`，形状为 `(batch size, channel, num_frame, height, width)`) : 输入 `hidden_states`。

timestep ( `torch.LongTensor`, *可选*) : 用于指示去噪步骤。作为嵌入应用于 `AdaLayerNorm` 的可选时间步。

encoder_hidden_states ( `torch.FloatTensor`，形状为 `(batch size, sequence len, embed dims)`，*可选*)：用于交叉注意力层的条件嵌入。如果未提供，交叉注意力将默认为自注意力。

encoder_attention_mask ( `torch.Tensor`, *可选*) : 应用于 `encoder_hidden_states` 的交叉注意力掩码。支持两种格式：  * 掩码 `(batcheight, sequence_length)` True = 保留，False = 丢弃。 * 偏置 `(batcheight, 1, sequence_length)` 0 = 保留，-10000 = 丢弃。  如果 `ndim == 2`：将被解释为掩码，然后转换为与上述格式一致的偏置。该偏置将被加到交叉注意力分数上。

enable_temporal_attentions : (`bool`，*可选*，默认为 `True`)：是否启用时间注意力。

return_dict (`bool`，*可选*，默认为 `True`)：是否返回 `~models.unet_2d_condition.UNet2DConditionOutput` 而不是普通元组。

**返回值：**

如果 `return_dict` 为 True，则返回 `~models.transformer_2d.Transformer2DModelOutput`，否则返回一个 `tuple`，其中第一个元素是样本张量。

[LatteTransformer3DModel](/docs/diffusers/v0.43.0/zh/api/models/latte_transformer3d#diffusers.LatteTransformer3DModel) 的前向方法。