( in_channels: int = 16 out_channels: int = 16 timestep_ratio_embedding_dim: int = 64 patch_size: int = 1 conditioning_dim: int = 2048 block_out_channels: typing.Tuple[int] = (2048, 2048) num_attention_heads: typing.Tuple[int] = (32, 32) down_num_layers_per_block: typing.Tuple[int] = (8, 24) up_num_layers_per_block: typing.Tuple[int] = (24, 8) down_blocks_repeat_mappers: typing.Optional[typing.Tuple[int]] = (1, 1) up_blocks_repeat_mappers: typing.Optional[typing.Tuple[int]] = (1, 1) block_types_per_layer: typing.Tuple[typing.Tuple[str]] = (('SDCascadeResBlock', 'SDCascadeTimestepBlock', 'SDCascadeAttnBlock'), ('SDCascadeResBlock', 'SDCascadeTimestepBlock', 'SDCascadeAttnBlock')) clip_text_in_channels: typing.Optional[int] = None clip_text_pooled_in_channels = 1280 clip_image_in_channels: typing.Optional[int] = None clip_seq = 4 effnet_in_channels: typing.Optional[int] = None pixel_mapper_in_channels: typing.Optional[int] = None kernel_size = 3 dropout: typing.Union[float, typing.Tuple[float]] = (0.1, 0.1) self_attn: typing.Union[bool, typing.Tuple[bool]] = True timestep_conditioning_type: typing.Tuple[str] = ('sca', 'crp') switch_level: typing.Optional[typing.Tuple[bool]] = None )