fastvideo.v1.configs.models.dits.hunyuanvideo

Contents

`fastvideo.v1.configs.models.dits.hunyuanvideo`#

Module Contents#

Classes#

`HunyuanVideoArchConfig`
`HunyuanVideoConfig`

Functions#

`is_double_block`
`is_refiner_block`
`is_single_block`
`is_txt_in`

API#

class fastvideo.v1.configs.models.dits.hunyuanvideo.HunyuanVideoArchConfig[source]#

Bases: fastvideo.v1.configs.models.dits.base.DiTArchConfig

attention_head_dim: int[source]#: 128

dtype: Optional[torch.dtype][source]#: None

guidance_embeds: bool[source]#: False

in_channels: int[source]#: 16

mlp_ratio: float[source]#: 4.0

num_attention_heads: int[source]#: 24

num_layers: int[source]#: 20

num_refiner_layers: int[source]#: 2

num_single_layers: int[source]#: 40

out_channels: int[source]#: 16

patch_size: int[source]#: 2

patch_size_t: int[source]#: 1

pooled_projection_dim: int[source]#: 768

qk_norm: str[source]#: ‘rms_norm’

rope_axes_dim: Tuple[int, int, int][source]#: (16, 56, 56)

rope_theta: int[source]#: 256

text_embed_dim: int[source]#: 4096

class fastvideo.v1.configs.models.dits.hunyuanvideo.HunyuanVideoConfig[source]#

Bases: fastvideo.v1.configs.models.dits.base.DiTConfig

arch_config: fastvideo.v1.configs.models.dits.base.DiTArchConfig[source]#: ‘field(…)’

prefix: str#: ‘Hunyuan’

fastvideo.v1.configs.models.dits.hunyuanvideo.is_double_block(n: str, m) → bool[source]#

fastvideo.v1.configs.models.dits.hunyuanvideo.is_refiner_block(n: str, m) → bool[source]#

fastvideo.v1.configs.models.dits.hunyuanvideo.is_single_block(n: str, m) → bool[source]#

fastvideo.v1.configs.models.dits.hunyuanvideo.is_txt_in(n: str, m) → bool[source]#