Class SkipDownBlock2D

src/diffusers/models/unets/unet_2d_blocks.py:1666–1753

class SkipDownBlock2D(nn.Module):
    def __init__(
        self,
        in_channels: int,
        out_channels: int,
        temb_channels: int,
        dropout: float = 0.0,
        num_layers: int = 1,
        resnet_eps: float = 1e-6,
        resnet_time_scale_shift: str = "default",
        resnet_act_fn: str = "swish",
        resnet_pre_norm: bool = True,
        output_scale_factor: float = np.sqrt(2.0),
        add_downsample: bool = True,
        downsample_padding: int = 1,
    ):
        super().__init__()
        self.resnets = nn.ModuleList([])

        for i in range(num_layers):
            in_channels = in_channels if i == 0 else out_channels
            self.resnets.append(
                ResnetBlock2D(
                    in_channels=in_channels,
                    out_channels=out_channels,
                    temb_channels=temb_channels,
                    eps=resnet_eps,
                    groups=min(in_channels // 4, 32),
                    groups_out=min(out_channels // 4, 32),
                    dropout=dropout,
                    time_embedding_norm=resnet_time_scale_shift,
                    non_linearity=resnet_act_fn,
                    output_scale_factor=output_scale_factor,
                    pre_norm=resnet_pre_norm,
                )
            )

        if add_downsample:
            self.resnet_down = ResnetBlock2D(
                in_channels=out_channels,
                out_channels=out_channels,
                temb_channels=temb_channels,
                eps=resnet_eps,
                groups=min(out_channels // 4, 32),
                dropout=dropout,
                time_embedding_norm=resnet_time_scale_shift,
                non_linearity=resnet_act_fn,
                output_scale_factor=output_scale_factor,
                pre_norm=resnet_pre_norm,
                use_in_shortcut=True,
                down=True,
                kernel="fir",
            )
            self.downsamplers = nn.ModuleList([FirDownsample2D(out_channels, out_channels=out_channels)])
            self.skip_conv = nn.Conv2d(3, out_channels, kernel_size=(1, 1), stride=(1, 1))
        else:
            self.resnet_down = None
            self.downsamplers = None