(self, x, temb, zq)
| 139 | self.nin_shortcut = torch.nn.Conv2d(in_channels, out_channels, kernel_size=1, stride=1, padding=0) |
| 140 | |
| 141 | def forward(self, x, temb, zq): |
| 142 | h = x |
| 143 | h = self.norm1(h, zq) |
| 144 | h = nonlinearity(h) |
| 145 | h = self.conv1(h) |
| 146 | |
| 147 | if temb is not None: |
| 148 | h = h + self.temb_proj(nonlinearity(temb))[:, :, None, None] |
| 149 | |
| 150 | h = self.norm2(h, zq) |
| 151 | h = nonlinearity(h) |
| 152 | h = self.dropout(h) |
| 153 | h = self.conv2(h) |
| 154 | |
| 155 | if self.in_channels != self.out_channels: |
| 156 | if self.use_conv_shortcut: |
| 157 | x = self.conv_shortcut(x) |
| 158 | else: |
| 159 | x = self.nin_shortcut(x) |
| 160 | |
| 161 | return x + h |
| 162 | |
| 163 | |
| 164 | class AttnBlock(nn.Module): |
nothing calls this directly
no test coverage detected