Method forward

src/diffusers/models/resnet.py:152–192

(
        self,
        input_tensor: torch.FloatTensor,
        temb: torch.FloatTensor,
        scale: float = 1.0,
    )

            )

    def forward(
        self,
        input_tensor: torch.FloatTensor,
        temb: torch.FloatTensor,
        scale: float = 1.0,
    ) -> torch.FloatTensor:
        hidden_states = input_tensor

        hidden_states = self.norm1(hidden_states, temb)

        hidden_states = self.nonlinearity(hidden_states)

        if self.upsample is not None:
            # upsample_nearest_nhwc fails with large batch sizes. see https://github.com/huggingface/diffusers/issues/984
            if hidden_states.shape[0] >= 64:
                input_tensor = input_tensor.contiguous()
                hidden_states = hidden_states.contiguous()
            input_tensor = self.upsample(input_tensor, scale=scale)
            hidden_states = self.upsample(hidden_states, scale=scale)

        elif self.downsample is not None:
            input_tensor = self.downsample(input_tensor, scale=scale)
            hidden_states = self.downsample(hidden_states, scale=scale)

        hidden_states = self.conv1(hidden_states, scale) if not USE_PEFT_BACKEND else self.conv1(hidden_states)

        hidden_states = self.norm2(hidden_states, temb)

        hidden_states = self.nonlinearity(hidden_states)

        hidden_states = self.dropout(hidden_states)
        hidden_states = self.conv2(hidden_states, scale) if not USE_PEFT_BACKEND else self.conv2(hidden_states)

        if self.conv_shortcut is not None:
            input_tensor = (
                self.conv_shortcut(input_tensor, scale) if not USE_PEFT_BACKEND else self.conv_shortcut(input_tensor)
            )

        output_tensor = (input_tensor + hidden_states) / self.output_scale_factor

        return output_tensor


class ResnetBlock2D(nn.Module):