AI/diffusers
1
0
Fork 0
mirror of https://github.com/huggingface/diffusers.git synced 2026-01-27 17:22:53 +03:00
Code Activity

[core] remove unneeded autoencoder methods when subclassing from AutoencoderMixin (#12873)

Browse Source 
up
This commit is contained in:
Sayak Paul
2026-01-06 11:13:54 +05:30
committed by GitHub
parent 7c6d314549
commit b6098ca006
3 changed files with 6 additions and 69 deletions
Download Patch File Download Diff File Expand all files Collapse all files

25
src/diffusers/models/autoencoders/autoencoder_kl_hunyuanimage.py
View File

@@ -27,7 +27,7 @@ from ...utils.accelerate_utils import apply_forward_hook
from ..activations import get_activation
from ..modeling_outputs import AutoencoderKLOutput
from ..modeling_utils import ModelMixin
from .vae import DecoderOutput, DiagonalGaussianDistribution
from .vae import AutoencoderMixin, DecoderOutput, DiagonalGaussianDistribution
logger = logging.get_logger(__name__) # pylint: disable=invalid-name
@@ -410,7 +410,7 @@ class HunyuanImageDecoder2D(nn.Module):
return h
class AutoencoderKLHunyuanImage(ModelMixin, ConfigMixin, FromOriginalModelMixin):
class AutoencoderKLHunyuanImage(ModelMixin, AutoencoderMixin, ConfigMixin, FromOriginalModelMixin):
r"""
A VAE model for 2D images with spatial tiling support.
@@ -486,27 +486,6 @@ class AutoencoderKLHunyuanImage(ModelMixin, ConfigMixin, FromOriginalModelMixin)
self.tile_overlap_factor = tile_overlap_factor or self.tile_overlap_factor
self.tile_latent_min_size = self.tile_sample_min_size // self.config.spatial_compression_ratio
def disable_tiling(self) -> None:
r"""
Disable tiled VAE decoding. If `enable_tiling` was previously enabled, this method will go back to computing
decoding in one step.
"""
self.use_tiling = False
def enable_slicing(self) -> None:
r"""
Enable sliced VAE decoding. When this option is enabled, the VAE will split the input tensor in slices to
compute decoding in several steps. This is useful to save some memory and allow larger batch sizes.
"""
self.use_slicing = True
def disable_slicing(self) -> None:
r"""
Disable sliced VAE decoding. If `enable_slicing` was previously enabled, this method will go back to computing
decoding in one step.
"""
self.use_slicing = False
def _encode(self, x: torch.Tensor):
batch_size, num_channels, height, width = x.shape

25
src/diffusers/models/autoencoders/autoencoder_kl_hunyuanimage_refiner.py
View File

@@ -26,7 +26,7 @@ from ...utils.accelerate_utils import apply_forward_hook
from ..activations import get_activation
from ..modeling_outputs import AutoencoderKLOutput
from ..modeling_utils import ModelMixin
from .vae import DecoderOutput, DiagonalGaussianDistribution
from .vae import AutoencoderMixin, DecoderOutput, DiagonalGaussianDistribution
logger = logging.get_logger(__name__) # pylint: disable=invalid-name
@@ -584,7 +584,7 @@ class HunyuanImageRefinerDecoder3D(nn.Module):
return hidden_states
class AutoencoderKLHunyuanImageRefiner(ModelMixin, ConfigMixin):
class AutoencoderKLHunyuanImageRefiner(ModelMixin, AutoencoderMixin, ConfigMixin):
r"""
A VAE model with KL loss for encoding videos into latents and decoding latent representations into videos. Used for
HunyuanImage-2.1 Refiner.
@@ -685,27 +685,6 @@ class AutoencoderKLHunyuanImageRefiner(ModelMixin, ConfigMixin):
self.tile_sample_stride_width = tile_sample_stride_width or self.tile_sample_stride_width
self.tile_overlap_factor = tile_overlap_factor or self.tile_overlap_factor
def disable_tiling(self) -> None:
r"""
Disable tiled VAE decoding. If `enable_tiling` was previously enabled, this method will go back to computing
decoding in one step.
"""
self.use_tiling = False
def enable_slicing(self) -> None:
r"""
Enable sliced VAE decoding. When this option is enabled, the VAE will split the input tensor in slices to
compute decoding in several steps. This is useful to save some memory and allow larger batch sizes.
"""
self.use_slicing = True
def disable_slicing(self) -> None:
r"""
Disable sliced VAE decoding. If `enable_slicing` was previously enabled, this method will go back to computing
decoding in one step.
"""
self.use_slicing = False
def _encode(self, x: torch.Tensor) -> torch.Tensor:
_, _, _, height, width = x.shape

25
src/diffusers/models/autoencoders/autoencoder_kl_hunyuanvideo15.py
View File

@@ -26,7 +26,7 @@ from ...utils.accelerate_utils import apply_forward_hook
from ..activations import get_activation
from ..modeling_outputs import AutoencoderKLOutput
from ..modeling_utils import ModelMixin
from .vae import DecoderOutput, DiagonalGaussianDistribution
from .vae import AutoencoderMixin, DecoderOutput, DiagonalGaussianDistribution
logger = logging.get_logger(__name__) # pylint: disable=invalid-name
@@ -625,7 +625,7 @@ class HunyuanVideo15Decoder3D(nn.Module):
return hidden_states
class AutoencoderKLHunyuanVideo15(ModelMixin, ConfigMixin):
class AutoencoderKLHunyuanVideo15(ModelMixin, AutoencoderMixin, ConfigMixin):
r"""
A VAE model with KL loss for encoding videos into latents and decoding latent representations into videos. Used for
HunyuanVideo-1.5.
@@ -723,27 +723,6 @@ class AutoencoderKLHunyuanVideo15(ModelMixin, ConfigMixin):
self.tile_latent_min_width = tile_latent_min_width or self.tile_latent_min_width
self.tile_overlap_factor = tile_overlap_factor or self.tile_overlap_factor
def disable_tiling(self) -> None:
r"""
Disable tiled VAE decoding. If `enable_tiling` was previously enabled, this method will go back to computing
decoding in one step.
"""
self.use_tiling = False
def enable_slicing(self) -> None:
r"""
Enable sliced VAE decoding. When this option is enabled, the VAE will split the input tensor in slices to
compute decoding in several steps. This is useful to save some memory and allow larger batch sizes.
"""
self.use_slicing = True
def disable_slicing(self) -> None:
r"""
Disable sliced VAE decoding. If `enable_slicing` was previously enabled, this method will go back to computing
decoding in one step.
"""
self.use_slicing = False
def _encode(self, x: torch.Tensor) -> torch.Tensor:
_, _, _, height, width = x.shape