--- title: lingbot_video ---

Architecture configuration for LingBot-Video Dense and MoE DiTs.

fastvideo.configs.models.dits.lingbot_video.DiTArchConfig

DiTArchConfig(stacked_params_mapping: list[tuple[str, str, str]] = list(), _fsdp_shard_conditions: list = list(), _compile_conditions: list = list(), param_names_mapping: dict = dict(), reverse_param_names_mapping: dict = dict(), lora_param_names_mapping: dict = dict(), cast_prompt_embeds_to_dit_dtype: bool = False, _supported_attention_backends: tuple[AttentionBackendEnum, ...] = (AttentionBackendEnum.SAGE_ATTN, AttentionBackendEnum.FLASH_ATTN, AttentionBackendEnum.TORCH_SDPA, AttentionBackendEnum.VIDEO_SPARSE_ATTN, AttentionBackendEnum.VMOBA_ATTN, AttentionBackendEnum.SAGE_ATTN_THREE, AttentionBackendEnum.ATTN_QAT_INFER, AttentionBackendEnum.ATTN_QAT_TRAIN, AttentionBackendEnum.SLA_ATTN, AttentionBackendEnum.SAGE_SLA_ATTN), hidden_size: int = 0, num_attention_heads: int = 0, num_channels_latents: int = 0, in_channels: int = 0, out_channels: int = 0, exclude_lora_layers: list[str] = list(), boundary_ratio: float | None = None)

fastvideo.configs.models.dits.lingbot_video.DiTConfig

DiTConfig(arch_config: DiTArchConfig = DiTArchConfig(), prefix: str = '', quant_config: QuantizationConfig | None = None, *, _resolved_attention_backend: AttentionBackendEnum | None = None)

fastvideo.configs.models.dits.lingbot_video.DiTConfig.add_cli_args

add_cli_args(parser: Any, prefix: str = 'dit-config') -> Any

Add CLI arguments for DiTConfig fields

fastvideo.configs.models.dits.lingbot_video.AttentionBackendEnum

AttentionBackendEnum()

fastvideo.configs.models.dits.lingbot_video.LingBotVideoArchConfig

LingBotVideoArchConfig(stacked_params_mapping: list[tuple[str, str, str]] = list(), _fsdp_shard_conditions: list = (lambda: [_is_lingbot_video_block])(), _compile_conditions: list = list(), param_names_mapping: dict = (lambda: {'^(.*)$': '\\1'})(), reverse_param_names_mapping: dict = dict(), lora_param_names_mapping: dict = dict(), cast_prompt_embeds_to_dit_dtype: bool = False, _supported_attention_backends: tuple[AttentionBackendEnum, ...] = (AttentionBackendEnum.TORCH_SDPA, AttentionBackendEnum.FLASH_ATTN), hidden_size: int = 2048, num_attention_heads: int = 16, num_channels_latents: int = 0, in_channels: int = 16, out_channels: int = 16, exclude_lora_layers: list[str] = list(), boundary_ratio: float | None = None, patch_size: tuple[int, int, int] = (1, 2, 2), depth: int = 24, intermediate_size: int = 6144, text_dim: int = 2560, freq_dim: int = 256, norm_eps: float = 1e-06, rope_theta: float = 256.0, axes_dims: tuple[int, int, int] = (32, 48, 48), axes_lens: tuple[int, int, int] = (8192, 1024, 1024), qkv_bias: bool = False, out_bias: bool = True, patch_embed_bias: bool = True, timestep_mlp_bias: bool = True, num_experts: int = 0, num_experts_per_tok: int = 8, moe_intermediate_size: int = 512, decoder_sparse_step: int = 1, mlp_only_layers: tuple[int, ...] = (), n_shared_experts: int | None = None, score_func: str = 'sigmoid', norm_topk_prob: bool = True, n_group: int | None = None, topk_group: int | None = None, routed_scaling_factor: float = 1.0)

One-to-one representation of the released transformer config JSON.

Source
@dataclass
class LingBotVideoArchConfig(DiTArchConfig):
    """One-to-one representation of the released transformer config JSON."""

    _fsdp_shard_conditions: list = field(default_factory=lambda: [_is_lingbot_video_block])
    _supported_attention_backends: tuple[AttentionBackendEnum, ...] = (
        AttentionBackendEnum.TORCH_SDPA,
        AttentionBackendEnum.FLASH_ATTN,
    )
    param_names_mapping: dict = field(default_factory=lambda: {r"^(.*)$": r"\1"})

    patch_size: tuple[int, int, int] = (1, 2, 2)
    in_channels: int = 16
    out_channels: int = 16
    hidden_size: int = 2048
    num_attention_heads: int = 16
    depth: int = 24
    intermediate_size: int = 6144
    text_dim: int = 2560
    freq_dim: int = 256
    norm_eps: float = 1e-6
    rope_theta: float = 256.0
    axes_dims: tuple[int, int, int] = (32, 48, 48)
    axes_lens: tuple[int, int, int] = (8192, 1024, 1024)
    qkv_bias: bool = False
    out_bias: bool = True
    patch_embed_bias: bool = True
    timestep_mlp_bias: bool = True
    num_experts: int = 0
    num_experts_per_tok: int = 8
    moe_intermediate_size: int = 512
    decoder_sparse_step: int = 1
    mlp_only_layers: tuple[int, ...] = ()
    n_shared_experts: int | None = None
    score_func: str = "sigmoid"
    norm_topk_prob: bool = True
    n_group: int | None = None
    topk_group: int | None = None
    routed_scaling_factor: float = 1.0

    def __post_init__(self) -> None:
        """Populate FastVideo loader fields from the released architecture."""
        super().__post_init__()
        self.num_channels_latents = self.in_channels
        self.attention_head_dim = self.hidden_size // self.num_attention_heads

fastvideo.configs.models.dits.lingbot_video.LingBotVideoConfig

LingBotVideoConfig(arch_config: DiTArchConfig = LingBotVideoArchConfig(), prefix: str = '', quant_config: QuantizationConfig | None = None, *, _resolved_attention_backend: AttentionBackendEnum | None = None)

FastVideo component configuration for LingBot-Video transformers.

Source
@dataclass
class LingBotVideoConfig(DiTConfig):
    """FastVideo component configuration for LingBot-Video transformers."""

    arch_config: DiTArchConfig = field(default_factory=LingBotVideoArchConfig)