| from typing import Optional |
|
|
| import torch |
| import torch.nn as nn |
| from distvae.models.upsampling import PatchUpsample2D |
| from distvae.modules.adapters.layers.conv_adapters import Conv2dAdapter |
|
|
| from diffusers.models.upsampling import Upsample2D |
|
|
| class Upsample2DAdapter(nn.Module): |
| def __init__( |
| self, |
| upsample2d: Upsample2D, |
| *, |
| conv_block_size = 0, |
| ): |
| super().__init__() |
| assert upsample2d.norm is None, "upsample2dBlock2DAdapter dose not support normalization" |
| if upsample2d.name == "conv": |
| assert not isinstance(upsample2d.conv, nn.ConvTranspose2d), "upsample2dBlock2DAdapter dose not support transpose conv" |
| else: |
| assert not isinstance(upsample2d.Conv2d_0, nn.ConvTranspose2d), "upsample2dBlock2DAdapter dose not support transpose conv" |
| self.upsample2d = PatchUpsample2D( |
| channels=upsample2d.channels, |
| use_conv=upsample2d.use_conv, |
| use_conv_transpose=upsample2d.use_conv_transpose, |
| out_channels=upsample2d.out_channels, |
| name=upsample2d.name, |
| kernel_size=None, |
| padding=1, |
| interpolate=upsample2d.interpolate |
| ) |
| if upsample2d.name == "conv": |
| self.upsample2d.conv = Conv2dAdapter(upsample2d.conv, block_size=conv_block_size) |
| else: |
| self.upsample2d.Conv2d_0 = Conv2dAdapter(upsample2d.Conv2d_0, block_size=conv_block_size) |
| |
|
|
| def forward( |
| self, hidden_states: torch.FloatTensor, output_size: Optional[int] = None, *args, **kwargs |
| ): |
| return self.upsample2d(hidden_states, output_size, *args, **kwargs) |
