diffusers 0.34.0__py3-none-any.whl → 0.35.1__py3-none-any.whl

diffusers/models/auto_model.py

@@ -117,8 +117,8 @@ class AutoModel(ConfigMixin):
 
         <Tip>
 
-        To use private or [gated models](https://huggingface.co/docs/hub/models-gated#gated-models), log-in with
-        `huggingface-cli login`. You can also activate the special
+        To use private or [gated models](https://huggingface.co/docs/hub/models-gated#gated-models), log-in with `hf
+        auth login`. You can also activate the special
         ["offline-mode"](https://huggingface.co/diffusers/installation.html#offline-mode) to use this method in a
         firewalled environment.
 

diffusers/models/autoencoders/__init__.py

@@ -8,6 +8,7 @@ from .autoencoder_kl_hunyuan_video import AutoencoderKLHunyuanVideo
 from .autoencoder_kl_ltx import AutoencoderKLLTXVideo
 from .autoencoder_kl_magvit import AutoencoderKLMagvit
 from .autoencoder_kl_mochi import AutoencoderKLMochi
+from .autoencoder_kl_qwenimage import AutoencoderKLQwenImage
 from .autoencoder_kl_temporal_decoder import AutoencoderKLTemporalDecoder
 from .autoencoder_kl_wan import AutoencoderKLWan
 from .autoencoder_oobleck import AutoencoderOobleck

diffusers/models/autoencoders/autoencoder_dc.py

@@ -299,6 +299,7 @@ class Decoder(nn.Module):
         act_fn: Union[str, Tuple[str]] = "silu",
         upsample_block_type: str = "pixel_shuffle",
         in_shortcut: bool = True,
+        conv_act_fn: str = "relu",
     ):
         super().__init__()
 
@@ -349,7 +350,7 @@ class Decoder(nn.Module):
         channels = block_out_channels[0] if layers_per_block[0] > 0 else block_out_channels[1]
 
         self.norm_out = RMSNorm(channels, 1e-5, elementwise_affine=True, bias=True)
-        self.conv_act = nn.ReLU()
+        self.conv_act = get_activation(conv_act_fn)
         self.conv_out = None
 
         if layers_per_block[0] > 0:
@@ -414,6 +415,12 @@ class AutoencoderDC(ModelMixin, ConfigMixin, FromOriginalModelMixin):
             The normalization type(s) to use in the decoder.
         decoder_act_fns (`Union[str, Tuple[str]]`, defaults to `"silu"`):
             The activation function(s) to use in the decoder.
+        encoder_out_shortcut  (`bool`, defaults to `True`):
+            Whether to use shortcut at the end of the encoder.
+        decoder_in_shortcut (`bool`, defaults to `True`):
+            Whether to use shortcut at the beginning of the decoder.
+        decoder_conv_act_fn (`str`, defaults to `"relu"`):
+            The activation function to use at the end of the decoder.
         scaling_factor (`float`, defaults to `1.0`):
             The multiplicative inverse of the root mean square of the latent features. This is used to scale the latent
             space to have unit variance when training the diffusion model. The latents are scaled with the formula `z =
@@ -441,6 +448,9 @@ class AutoencoderDC(ModelMixin, ConfigMixin, FromOriginalModelMixin):
         downsample_block_type: str = "pixel_unshuffle",
         decoder_norm_types: Union[str, Tuple[str]] = "rms_norm",
         decoder_act_fns: Union[str, Tuple[str]] = "silu",
+        encoder_out_shortcut: bool = True,
+        decoder_in_shortcut: bool = True,
+        decoder_conv_act_fn: str = "relu",
         scaling_factor: float = 1.0,
     ) -> None:
         super().__init__()
@@ -454,6 +464,7 @@ class AutoencoderDC(ModelMixin, ConfigMixin, FromOriginalModelMixin):
             layers_per_block=encoder_layers_per_block,
             qkv_multiscales=encoder_qkv_multiscales,
             downsample_block_type=downsample_block_type,
+            out_shortcut=encoder_out_shortcut,
         )
         self.decoder = Decoder(
             in_channels=in_channels,
@@ -466,6 +477,8 @@ class AutoencoderDC(ModelMixin, ConfigMixin, FromOriginalModelMixin):
             norm_type=decoder_norm_types,
             act_fn=decoder_act_fns,
             upsample_block_type=upsample_block_type,
+            in_shortcut=decoder_in_shortcut,
+            conv_act_fn=decoder_conv_act_fn,
         )
 
         self.spatial_compression_ratio = 2 ** (len(encoder_block_out_channels) - 1)

diffusers/models/autoencoders/autoencoder_kl.py

@@ -90,7 +90,7 @@ class AutoencoderKL(ModelMixin, ConfigMixin, FromOriginalModelMixin, PeftAdapter
         shift_factor: Optional[float] = None,
         latents_mean: Optional[Tuple[float]] = None,
         latents_std: Optional[Tuple[float]] = None,
-        force_upcast: float = True,
+        force_upcast: bool = True,
         use_quant_conv: bool = True,
         use_post_quant_conv: bool = True,
         mid_block_add_attention: bool = True,

diffusers/models/autoencoders/autoencoder_kl_cosmos.py

@@ -168,7 +168,9 @@ class CosmosPatchEmbed3d(nn.Module):
         batch_size, num_channels, num_frames, height, width = hidden_states.shape
         p = self.patch_size
 
-        hidden_states = torch.reshape(batch_size, num_channels, num_frames // p, p, height // p, p, width // p, p)
+        hidden_states = hidden_states.reshape(
+            batch_size, num_channels, num_frames // p, p, height // p, p, width // p, p
+        )
         hidden_states = hidden_states.permute(0, 1, 3, 5, 7, 2, 4, 6).flatten(1, 4).contiguous()
         return hidden_states