AI/diffusers
1
0
Fork 0
mirror of https://github.com/huggingface/diffusers.git synced 2026-01-29 07:22:12 +03:00
Code Activity

update

Browse Source
This commit is contained in:
DN6
2025-11-14 17:55:54 +05:30
parent 2cf6dd1d88
commit 799cf8de89
26 changed files with 0 additions and 465 deletions
Download Patch File Download Diff File Expand all files Collapse all files

15
src/diffusers/modular_pipelines/stable_diffusion_xl/decoders.py
View File

@@ -21,7 +21,6 @@ import torch
from ...configuration_utils import FrozenDict
from ...image_processor import VaeImageProcessor
from ...models import AutoencoderKL
from ...models.attention_processor import AttnProcessor2_0, XFormersAttnProcessor
from ...utils import logging
from ..modular_pipeline import (
ModularPipelineBlocks,
@@ -77,21 +76,7 @@ class StableDiffusionXLDecodeStep(ModularPipelineBlocks):
@staticmethod
# Copied from diffusers.pipelines.stable_diffusion.pipeline_stable_diffusion_upscale.StableDiffusionUpscalePipeline.upcast_vae with self->components
def upcast_vae(components):
dtype = components.vae.dtype
components.vae.to(dtype=torch.float32)
use_torch_2_0_or_xformers = isinstance(
components.vae.decoder.mid_block.attentions[0].processor,
(
AttnProcessor2_0,
XFormersAttnProcessor,
),
)
# if xformers or torch_2_0 is used attention block does not need
# to be in float32 which can save lots of memory
if use_torch_2_0_or_xformers:
components.vae.post_quant_conv.to(dtype)
components.vae.decoder.conv_in.to(dtype)
components.vae.decoder.mid_block.to(dtype)
@torch.no_grad()
def __call__(self, components, state: PipelineState) -> PipelineState:

16
src/diffusers/pipelines/aura_flow/pipeline_aura_flow.py
View File

@@ -21,7 +21,6 @@ from ...callbacks import MultiPipelineCallbacks, PipelineCallback
from ...image_processor import VaeImageProcessor
from ...loaders import AuraFlowLoraLoaderMixin
from ...models import AuraFlowTransformer2DModel, AutoencoderKL
from ...models.attention_processor import AttnProcessor2_0, FusedAttnProcessor2_0, XFormersAttnProcessor
from ...schedulers import FlowMatchEulerDiscreteScheduler
from ...utils import (
USE_PEFT_BACKEND,
@@ -406,22 +405,7 @@ class AuraFlowPipeline(DiffusionPipeline, AuraFlowLoraLoaderMixin):
# Copied from diffusers.pipelines.stable_diffusion_xl.pipeline_stable_diffusion_xl.StableDiffusionXLPipeline.upcast_vae
def upcast_vae(self):
dtype = self.vae.dtype
self.vae.to(dtype=torch.float32)
use_torch_2_0_or_xformers = isinstance(
self.vae.decoder.mid_block.attentions[0].processor,
(
AttnProcessor2_0,
XFormersAttnProcessor,
FusedAttnProcessor2_0,
),
)
# if xformers or torch_2_0 is used attention block does not need
# to be in float32 which can save lots of memory
if use_torch_2_0_or_xformers:
self.vae.post_quant_conv.to(dtype)
self.vae.decoder.conv_in.to(dtype)
self.vae.decoder.mid_block.to(dtype)
@property
def guidance_scale(self):

18
src/diffusers/pipelines/controlnet/pipeline_controlnet_inpaint_sd_xl.py
View File

@@ -36,10 +36,6 @@ from ...loaders import (
TextualInversionLoaderMixin,
)
from ...models import AutoencoderKL, ControlNetModel, ImageProjection, MultiControlNetModel, UNet2DConditionModel
from ...models.attention_processor import (
AttnProcessor2_0,
XFormersAttnProcessor,
)
from ...models.lora import adjust_lora_scale_text_encoder
from ...schedulers import KarrasDiffusionSchedulers
from ...utils import (
@@ -1132,21 +1128,7 @@ class StableDiffusionXLControlNetInpaintPipeline(
# Copied from diffusers.pipelines.stable_diffusion.pipeline_stable_diffusion_upscale.StableDiffusionUpscalePipeline.upcast_vae
def upcast_vae(self):
dtype = self.vae.dtype
self.vae.to(dtype=torch.float32)
use_torch_2_0_or_xformers = isinstance(
self.vae.decoder.mid_block.attentions[0].processor,
(
AttnProcessor2_0,
XFormersAttnProcessor,
),
)
# if xformers or torch_2_0 is used attention block does not need
# to be in float32 which can save lots of memory
if use_torch_2_0_or_xformers:
self.vae.post_quant_conv.to(dtype)
self.vae.decoder.conv_in.to(dtype)
self.vae.decoder.mid_block.to(dtype)
@property
def guidance_scale(self):

18
src/diffusers/pipelines/controlnet/pipeline_controlnet_sd_xl.py
View File

@@ -39,10 +39,6 @@ from ...loaders import (
TextualInversionLoaderMixin,
)
from ...models import AutoencoderKL, ControlNetModel, ImageProjection, MultiControlNetModel, UNet2DConditionModel
from ...models.attention_processor import (
AttnProcessor2_0,
XFormersAttnProcessor,
)
from ...models.lora import adjust_lora_scale_text_encoder
from ...schedulers import KarrasDiffusionSchedulers
from ...utils import (
@@ -929,21 +925,7 @@ class StableDiffusionXLControlNetPipeline(
# Copied from diffusers.pipelines.stable_diffusion.pipeline_stable_diffusion_upscale.StableDiffusionUpscalePipeline.upcast_vae
def upcast_vae(self):
dtype = self.vae.dtype
self.vae.to(dtype=torch.float32)
use_torch_2_0_or_xformers = isinstance(
self.vae.decoder.mid_block.attentions[0].processor,
(
AttnProcessor2_0,
XFormersAttnProcessor,
),
)
# if xformers or torch_2_0 is used attention block does not need
# to be in float32 which can save lots of memory
if use_torch_2_0_or_xformers:
self.vae.post_quant_conv.to(dtype)
self.vae.decoder.conv_in.to(dtype)
self.vae.decoder.mid_block.to(dtype)
# Copied from diffusers.pipelines.latent_consistency_models.pipeline_latent_consistency_text2img.LatentConsistencyModelPipeline.get_guidance_scale_embedding
def get_guidance_scale_embedding(

18
src/diffusers/pipelines/controlnet/pipeline_controlnet_sd_xl_img2img.py
View File

@@ -39,10 +39,6 @@ from ...loaders import (
TextualInversionLoaderMixin,
)
from ...models import AutoencoderKL, ControlNetModel, ImageProjection, MultiControlNetModel, UNet2DConditionModel
from ...models.attention_processor import (
AttnProcessor2_0,
XFormersAttnProcessor,
)
from ...models.lora import adjust_lora_scale_text_encoder
from ...schedulers import KarrasDiffusionSchedulers
from ...utils import (
@@ -1044,21 +1040,7 @@ class StableDiffusionXLControlNetImg2ImgPipeline(
# Copied from diffusers.pipelines.stable_diffusion.pipeline_stable_diffusion_upscale.StableDiffusionUpscalePipeline.upcast_vae
def upcast_vae(self):
dtype = self.vae.dtype
self.vae.to(dtype=torch.float32)
use_torch_2_0_or_xformers = isinstance(
self.vae.decoder.mid_block.attentions[0].processor,
(
AttnProcessor2_0,
XFormersAttnProcessor,
),
)
# if xformers or torch_2_0 is used attention block does not need
# to be in float32 which can save lots of memory
if use_torch_2_0_or_xformers:
self.vae.post_quant_conv.to(dtype)
self.vae.decoder.conv_in.to(dtype)
self.vae.decoder.mid_block.to(dtype)
@property
def guidance_scale(self):

18
src/diffusers/pipelines/controlnet/pipeline_controlnet_union_inpaint_sd_xl.py
View File

@@ -41,10 +41,6 @@ from ...models import (
MultiControlNetUnionModel,
UNet2DConditionModel,
)
from ...models.attention_processor import (
AttnProcessor2_0,
XFormersAttnProcessor,
)
from ...models.lora import adjust_lora_scale_text_encoder
from ...schedulers import KarrasDiffusionSchedulers
from ...utils import (
@@ -1111,21 +1107,7 @@ class StableDiffusionXLControlNetUnionInpaintPipeline(
# Copied from diffusers.pipelines.stable_diffusion.pipeline_stable_diffusion_upscale.StableDiffusionUpscalePipeline.upcast_vae
def upcast_vae(self):
dtype = self.vae.dtype
self.vae.to(dtype=torch.float32)
use_torch_2_0_or_xformers = isinstance(
self.vae.decoder.mid_block.attentions[0].processor,
(
AttnProcessor2_0,
XFormersAttnProcessor,
),
)
# if xformers or torch_2_0 is used attention block does not need
# to be in float32 which can save lots of memory
if use_torch_2_0_or_xformers:
self.vae.post_quant_conv.to(dtype)
self.vae.decoder.conv_in.to(dtype)
self.vae.decoder.mid_block.to(dtype)
@property
def guidance_scale(self):

18
src/diffusers/pipelines/controlnet/pipeline_controlnet_union_sd_xl.py
View File

@@ -44,10 +44,6 @@ from ...models import (
MultiControlNetUnionModel,
UNet2DConditionModel,
)
from ...models.attention_processor import (
AttnProcessor2_0,
XFormersAttnProcessor,
)
from ...models.lora import adjust_lora_scale_text_encoder
from ...schedulers import KarrasDiffusionSchedulers
from ...utils import (
@@ -904,21 +900,7 @@ class StableDiffusionXLControlNetUnionPipeline(
# Copied from diffusers.pipelines.stable_diffusion.pipeline_stable_diffusion_upscale.StableDiffusionUpscalePipeline.upcast_vae
def upcast_vae(self):
dtype = self.vae.dtype
self.vae.to(dtype=torch.float32)
use_torch_2_0_or_xformers = isinstance(
self.vae.decoder.mid_block.attentions[0].processor,
(
AttnProcessor2_0,
XFormersAttnProcessor,
),
)
# if xformers or torch_2_0 is used attention block does not need
# to be in float32 which can save lots of memory
if use_torch_2_0_or_xformers:
self.vae.post_quant_conv.to(dtype)
self.vae.decoder.conv_in.to(dtype)
self.vae.decoder.mid_block.to(dtype)
# Copied from diffusers.pipelines.latent_consistency_models.pipeline_latent_consistency_text2img.LatentConsistencyModelPipeline.get_guidance_scale_embedding
def get_guidance_scale_embedding(

18
src/diffusers/pipelines/controlnet/pipeline_controlnet_union_sd_xl_img2img.py
View File

@@ -44,10 +44,6 @@ from ...models import (
MultiControlNetUnionModel,
UNet2DConditionModel,
)
from ...models.attention_processor import (
AttnProcessor2_0,
XFormersAttnProcessor,
)
from ...models.lora import adjust_lora_scale_text_encoder
from ...schedulers import KarrasDiffusionSchedulers
from ...utils import (
@@ -1032,21 +1028,7 @@ class StableDiffusionXLControlNetUnionImg2ImgPipeline(
# Copied from diffusers.pipelines.stable_diffusion.pipeline_stable_diffusion_upscale.StableDiffusionUpscalePipeline.upcast_vae
def upcast_vae(self):
dtype = self.vae.dtype
self.vae.to(dtype=torch.float32)
use_torch_2_0_or_xformers = isinstance(
self.vae.decoder.mid_block.attentions[0].processor,
(
AttnProcessor2_0,
XFormersAttnProcessor,
),
)
# if xformers or torch_2_0 is used attention block does not need
# to be in float32 which can save lots of memory
if use_torch_2_0_or_xformers:
self.vae.post_quant_conv.to(dtype)
self.vae.decoder.conv_in.to(dtype)
self.vae.decoder.mid_block.to(dtype)
@property
def guidance_scale(self):

18
src/diffusers/pipelines/controlnet_xs/pipeline_controlnet_xs_sd_xl.py
View File

@@ -32,10 +32,6 @@ from ...callbacks import MultiPipelineCallbacks, PipelineCallback
from ...image_processor import PipelineImageInput, VaeImageProcessor
from ...loaders import FromSingleFileMixin, StableDiffusionXLLoraLoaderMixin, TextualInversionLoaderMixin
from ...models import AutoencoderKL, ControlNetXSAdapter, UNet2DConditionModel, UNetControlNetXSModel
from ...models.attention_processor import (
AttnProcessor2_0,
XFormersAttnProcessor,
)
from ...models.lora import adjust_lora_scale_text_encoder
from ...schedulers import KarrasDiffusionSchedulers
from ...utils import (
@@ -685,21 +681,7 @@ class StableDiffusionXLControlNetXSPipeline(
# Copied from diffusers.pipelines.stable_diffusion.pipeline_stable_diffusion_upscale.StableDiffusionUpscalePipeline.upcast_vae
def upcast_vae(self):
dtype = self.vae.dtype
self.vae.to(dtype=torch.float32)
use_torch_2_0_or_xformers = isinstance(
self.vae.decoder.mid_block.attentions[0].processor,
(
AttnProcessor2_0,
XFormersAttnProcessor,
),
)
# if xformers or torch_2_0 is used attention block does not need
# to be in float32 which can save lots of memory
if use_torch_2_0_or_xformers:
self.vae.post_quant_conv.to(dtype)
self.vae.decoder.conv_in.to(dtype)
self.vae.decoder.mid_block.to(dtype)
@property
# Copied from diffusers.pipelines.stable_diffusion.pipeline_stable_diffusion.StableDiffusionPipeline.guidance_scale

16
src/diffusers/pipelines/kolors/pipeline_kolors.py
View File

@@ -21,7 +21,6 @@ from ...callbacks import MultiPipelineCallbacks, PipelineCallback
from ...image_processor import PipelineImageInput, VaeImageProcessor
from ...loaders import IPAdapterMixin, StableDiffusionLoraLoaderMixin
from ...models import AutoencoderKL, ImageProjection, UNet2DConditionModel
from ...models.attention_processor import AttnProcessor2_0, FusedAttnProcessor2_0, XFormersAttnProcessor
from ...schedulers import KarrasDiffusionSchedulers
from ...utils import is_torch_xla_available, logging, replace_example_docstring
from ...utils.torch_utils import randn_tensor
@@ -580,22 +579,7 @@ class KolorsPipeline(DiffusionPipeline, StableDiffusionMixin, StableDiffusionLor
# Copied from diffusers.pipelines.stable_diffusion_xl.pipeline_stable_diffusion_xl.StableDiffusionXLPipeline.upcast_vae
def upcast_vae(self):
dtype = self.vae.dtype
self.vae.to(dtype=torch.float32)
use_torch_2_0_or_xformers = isinstance(
self.vae.decoder.mid_block.attentions[0].processor,
(
AttnProcessor2_0,
XFormersAttnProcessor,
FusedAttnProcessor2_0,
),
)
# if xformers or torch_2_0 is used attention block does not need
# to be in float32 which can save lots of memory
if use_torch_2_0_or_xformers:
self.vae.post_quant_conv.to(dtype)
self.vae.decoder.conv_in.to(dtype)
self.vae.decoder.mid_block.to(dtype)
# Copied from diffusers.pipelines.latent_consistency_models.pipeline_latent_consistency_text2img.LatentConsistencyModelPipeline.get_guidance_scale_embedding
def get_guidance_scale_embedding(

16
src/diffusers/pipelines/kolors/pipeline_kolors_img2img.py
View File

@@ -22,7 +22,6 @@ from ...callbacks import MultiPipelineCallbacks, PipelineCallback
from ...image_processor import PipelineImageInput, VaeImageProcessor
from ...loaders import IPAdapterMixin, StableDiffusionXLLoraLoaderMixin
from ...models import AutoencoderKL, ImageProjection, UNet2DConditionModel
from ...models.attention_processor import AttnProcessor2_0, FusedAttnProcessor2_0, XFormersAttnProcessor
from ...schedulers import KarrasDiffusionSchedulers
from ...utils import is_torch_xla_available, logging, replace_example_docstring
from ...utils.torch_utils import empty_device_cache, randn_tensor
@@ -708,22 +707,7 @@ class KolorsImg2ImgPipeline(DiffusionPipeline, StableDiffusionMixin, StableDiffu
# Copied from diffusers.pipelines.stable_diffusion_xl.pipeline_stable_diffusion_xl.StableDiffusionXLPipeline.upcast_vae
def upcast_vae(self):
dtype = self.vae.dtype
self.vae.to(dtype=torch.float32)
use_torch_2_0_or_xformers = isinstance(
self.vae.decoder.mid_block.attentions[0].processor,
(
AttnProcessor2_0,
XFormersAttnProcessor,
FusedAttnProcessor2_0,
),
)
# if xformers or torch_2_0 is used attention block does not need
# to be in float32 which can save lots of memory
if use_torch_2_0_or_xformers:
self.vae.post_quant_conv.to(dtype)
self.vae.decoder.conv_in.to(dtype)
self.vae.decoder.mid_block.to(dtype)
# Copied from diffusers.pipelines.latent_consistency_models.pipeline_latent_consistency_text2img.LatentConsistencyModelPipeline.get_guidance_scale_embedding
def get_guidance_scale_embedding(

16
src/diffusers/pipelines/ledits_pp/pipeline_leditspp_stable_diffusion_xl.py
View File

@@ -37,8 +37,6 @@ from ...models import AutoencoderKL, UNet2DConditionModel
from ...models.attention_processor import (
Attention,
AttnProcessor,
AttnProcessor2_0,
XFormersAttnProcessor,
)
from ...models.lora import adjust_lora_scale_text_encoder
from ...schedulers import DDIMScheduler, DPMSolverMultistepScheduler
@@ -688,21 +686,7 @@ class LEditsPPPipelineStableDiffusionXL(
# Copied from diffusers.pipelines.stable_diffusion.pipeline_stable_diffusion_upscale.StableDiffusionUpscalePipeline.upcast_vae
def upcast_vae(self):
dtype = self.vae.dtype
self.vae.to(dtype=torch.float32)
use_torch_2_0_or_xformers = isinstance(
self.vae.decoder.mid_block.attentions[0].processor,
(
AttnProcessor2_0,
XFormersAttnProcessor,
),
)
# if xformers or torch_2_0 is used attention block does not need
# to be in float32 which can save lots of memory
if use_torch_2_0_or_xformers:
self.vae.post_quant_conv.to(dtype)
self.vae.decoder.conv_in.to(dtype)
self.vae.decoder.mid_block.to(dtype)
# Copied from diffusers.pipelines.latent_consistency_models.pipeline_latent_consistency_text2img.LatentConsistencyModelPipeline.get_guidance_scale_embedding
def get_guidance_scale_embedding(

18
src/diffusers/pipelines/pag/pipeline_pag_controlnet_sd_xl.py
View File

@@ -39,10 +39,6 @@ from ...loaders import (
TextualInversionLoaderMixin,
)
from ...models import AutoencoderKL, ControlNetModel, ImageProjection, MultiControlNetModel, UNet2DConditionModel
from ...models.attention_processor import (
AttnProcessor2_0,
XFormersAttnProcessor,
)
from ...models.lora import adjust_lora_scale_text_encoder
from ...schedulers import KarrasDiffusionSchedulers
from ...utils import (
@@ -936,21 +932,7 @@ class StableDiffusionXLControlNetPAGPipeline(
# Copied from diffusers.pipelines.stable_diffusion.pipeline_stable_diffusion_upscale.StableDiffusionUpscalePipeline.upcast_vae
def upcast_vae(self):
dtype = self.vae.dtype
self.vae.to(dtype=torch.float32)
use_torch_2_0_or_xformers = isinstance(
self.vae.decoder.mid_block.attentions[0].processor,
(
AttnProcessor2_0,
XFormersAttnProcessor,
),
)
# if xformers or torch_2_0 is used attention block does not need
# to be in float32 which can save lots of memory
if use_torch_2_0_or_xformers:
self.vae.post_quant_conv.to(dtype)
self.vae.decoder.conv_in.to(dtype)
self.vae.decoder.mid_block.to(dtype)
# Copied from diffusers.pipelines.latent_consistency_models.pipeline_latent_consistency_text2img.LatentConsistencyModelPipeline.get_guidance_scale_embedding
def get_guidance_scale_embedding(

18
src/diffusers/pipelines/pag/pipeline_pag_controlnet_sd_xl_img2img.py
View File

@@ -39,10 +39,6 @@ from ...loaders import (
TextualInversionLoaderMixin,
)
from ...models import AutoencoderKL, ControlNetModel, ImageProjection, MultiControlNetModel, UNet2DConditionModel
from ...models.attention_processor import (
AttnProcessor2_0,
XFormersAttnProcessor,
)
from ...models.lora import adjust_lora_scale_text_encoder
from ...schedulers import KarrasDiffusionSchedulers
from ...utils import (
@@ -1049,21 +1045,7 @@ class StableDiffusionXLControlNetPAGImg2ImgPipeline(
# Copied from diffusers.pipelines.stable_diffusion.pipeline_stable_diffusion_upscale.StableDiffusionUpscalePipeline.upcast_vae
def upcast_vae(self):
dtype = self.vae.dtype
self.vae.to(dtype=torch.float32)
use_torch_2_0_or_xformers = isinstance(
self.vae.decoder.mid_block.attentions[0].processor,
(
AttnProcessor2_0,
XFormersAttnProcessor,
),
)
# if xformers or torch_2_0 is used attention block does not need
# to be in float32 which can save lots of memory
if use_torch_2_0_or_xformers:
self.vae.post_quant_conv.to(dtype)
self.vae.decoder.conv_in.to(dtype)
self.vae.decoder.mid_block.to(dtype)
@property
def guidance_scale(self):

16
src/diffusers/pipelines/pag/pipeline_pag_kolors.py
View File

@@ -21,7 +21,6 @@ from ...callbacks import MultiPipelineCallbacks, PipelineCallback
from ...image_processor import PipelineImageInput, VaeImageProcessor
from ...loaders import IPAdapterMixin, StableDiffusionXLLoraLoaderMixin
from ...models import AutoencoderKL, ImageProjection, UNet2DConditionModel
from ...models.attention_processor import AttnProcessor2_0, FusedAttnProcessor2_0, XFormersAttnProcessor
from ...schedulers import KarrasDiffusionSchedulers
from ...utils import is_torch_xla_available, logging, replace_example_docstring
from ...utils.torch_utils import randn_tensor
@@ -598,22 +597,7 @@ class KolorsPAGPipeline(
# Copied from diffusers.pipelines.stable_diffusion_xl.pipeline_stable_diffusion_xl.StableDiffusionXLPipeline.upcast_vae
def upcast_vae(self):
dtype = self.vae.dtype
self.vae.to(dtype=torch.float32)
use_torch_2_0_or_xformers = isinstance(
self.vae.decoder.mid_block.attentions[0].processor,
(
AttnProcessor2_0,
XFormersAttnProcessor,
FusedAttnProcessor2_0,
),
)
# if xformers or torch_2_0 is used attention block does not need
# to be in float32 which can save lots of memory
if use_torch_2_0_or_xformers:
self.vae.post_quant_conv.to(dtype)
self.vae.decoder.conv_in.to(dtype)
self.vae.decoder.mid_block.to(dtype)
# Copied from diffusers.pipelines.latent_consistency_models.pipeline_latent_consistency_text2img.LatentConsistencyModelPipeline.get_guidance_scale_embedding
def get_guidance_scale_embedding(

20
src/diffusers/pipelines/pag/pipeline_pag_sd_xl.py
View File

@@ -32,11 +32,6 @@ from ...loaders import (
TextualInversionLoaderMixin,
)
from ...models import AutoencoderKL, ImageProjection, UNet2DConditionModel
from ...models.attention_processor import (
AttnProcessor2_0,
FusedAttnProcessor2_0,
XFormersAttnProcessor,
)
from ...models.lora import adjust_lora_scale_text_encoder
from ...schedulers import KarrasDiffusionSchedulers
from ...utils import (
@@ -762,22 +757,7 @@ class StableDiffusionXLPAGPipeline(
# Copied from diffusers.pipelines.stable_diffusion_xl.pipeline_stable_diffusion_xl.StableDiffusionXLPipeline.upcast_vae
def upcast_vae(self):
dtype = self.vae.dtype
self.vae.to(dtype=torch.float32)
use_torch_2_0_or_xformers = isinstance(
self.vae.decoder.mid_block.attentions[0].processor,
(
AttnProcessor2_0,
XFormersAttnProcessor,
FusedAttnProcessor2_0,
),
)
# if xformers or torch_2_0 is used attention block does not need
# to be in float32 which can save lots of memory
if use_torch_2_0_or_xformers:
self.vae.post_quant_conv.to(dtype)
self.vae.decoder.conv_in.to(dtype)
self.vae.decoder.mid_block.to(dtype)
# Copied from diffusers.pipelines.latent_consistency_models.pipeline_latent_consistency_text2img.LatentConsistencyModelPipeline.get_guidance_scale_embedding
def get_guidance_scale_embedding(

18
src/diffusers/pipelines/pag/pipeline_pag_sd_xl_img2img.py
View File

@@ -34,10 +34,6 @@ from ...loaders import (
TextualInversionLoaderMixin,
)
from ...models import AutoencoderKL, ImageProjection, UNet2DConditionModel
from ...models.attention_processor import (
AttnProcessor2_0,
XFormersAttnProcessor,
)
from ...models.lora import adjust_lora_scale_text_encoder
from ...schedulers import KarrasDiffusionSchedulers
from ...utils import (
@@ -910,21 +906,7 @@ class StableDiffusionXLPAGImg2ImgPipeline(
# Copied from diffusers.pipelines.stable_diffusion.pipeline_stable_diffusion_upscale.StableDiffusionUpscalePipeline.upcast_vae
def upcast_vae(self):
dtype = self.vae.dtype
self.vae.to(dtype=torch.float32)
use_torch_2_0_or_xformers = isinstance(
self.vae.decoder.mid_block.attentions[0].processor,
(
AttnProcessor2_0,
XFormersAttnProcessor,
),
)
# if xformers or torch_2_0 is used attention block does not need
# to be in float32 which can save lots of memory
if use_torch_2_0_or_xformers:
self.vae.post_quant_conv.to(dtype)
self.vae.decoder.conv_in.to(dtype)
self.vae.decoder.mid_block.to(dtype)
# Copied from diffusers.pipelines.latent_consistency_models.pipeline_latent_consistency_text2img.LatentConsistencyModelPipeline.get_guidance_scale_embedding
def get_guidance_scale_embedding(

18
src/diffusers/pipelines/pag/pipeline_pag_sd_xl_inpaint.py
View File

@@ -34,10 +34,6 @@ from ...loaders import (
TextualInversionLoaderMixin,
)
from ...models import AutoencoderKL, ImageProjection, UNet2DConditionModel
from ...models.attention_processor import (
AttnProcessor2_0,
XFormersAttnProcessor,
)
from ...models.lora import adjust_lora_scale_text_encoder
from ...schedulers import KarrasDiffusionSchedulers
from ...utils import (
@@ -1001,21 +997,7 @@ class StableDiffusionXLPAGInpaintPipeline(
# Copied from diffusers.pipelines.stable_diffusion.pipeline_stable_diffusion_upscale.StableDiffusionUpscalePipeline.upcast_vae
def upcast_vae(self):
dtype = self.vae.dtype
self.vae.to(dtype=torch.float32)
use_torch_2_0_or_xformers = isinstance(
self.vae.decoder.mid_block.attentions[0].processor,
(
AttnProcessor2_0,
XFormersAttnProcessor,
),
)
# if xformers or torch_2_0 is used attention block does not need
# to be in float32 which can save lots of memory
if use_torch_2_0_or_xformers:
self.vae.post_quant_conv.to(dtype)
self.vae.decoder.conv_in.to(dtype)
self.vae.decoder.mid_block.to(dtype)
# Copied from diffusers.pipelines.latent_consistency_models.pipeline_latent_consistency_text2img.LatentConsistencyModelPipeline.get_guidance_scale_embedding
def get_guidance_scale_embedding(

18
src/diffusers/pipelines/stable_diffusion/pipeline_stable_diffusion_upscale.py
View File

@@ -24,10 +24,6 @@ from transformers import CLIPImageProcessor, CLIPTextModel, CLIPTokenizer
from ...image_processor import PipelineImageInput, VaeImageProcessor
from ...loaders import FromSingleFileMixin, StableDiffusionLoraLoaderMixin, TextualInversionLoaderMixin
from ...models import AutoencoderKL, UNet2DConditionModel
from ...models.attention_processor import (
AttnProcessor2_0,
XFormersAttnProcessor,
)
from ...models.lora import adjust_lora_scale_text_encoder
from ...schedulers import DDPMScheduler, KarrasDiffusionSchedulers
from ...utils import (
@@ -529,21 +525,7 @@ class StableDiffusionUpscalePipeline(
return latents
def upcast_vae(self):
dtype = self.vae.dtype
self.vae.to(dtype=torch.float32)
use_torch_2_0_or_xformers = isinstance(
self.vae.decoder.mid_block.attentions[0].processor,
(
AttnProcessor2_0,
XFormersAttnProcessor,
),
)
# if xformers or torch_2_0 is used attention block does not need
# to be in float32 which can save lots of memory
if use_torch_2_0_or_xformers:
self.vae.post_quant_conv.to(dtype)
self.vae.decoder.conv_in.to(dtype)
self.vae.decoder.mid_block.to(dtype)
@torch.no_grad()
def __call__(

20
src/diffusers/pipelines/stable_diffusion_k_diffusion/pipeline_stable_diffusion_xl_k_diffusion.py
View File

@@ -33,11 +33,6 @@ from ...loaders import (
TextualInversionLoaderMixin,
)
from ...models import AutoencoderKL, UNet2DConditionModel
from ...models.attention_processor import (
AttnProcessor2_0,
FusedAttnProcessor2_0,
XFormersAttnProcessor,
)
from ...models.lora import adjust_lora_scale_text_encoder
from ...schedulers import KarrasDiffusionSchedulers, LMSDiscreteScheduler
from ...utils import (
@@ -545,22 +540,7 @@ class StableDiffusionXLKDiffusionPipeline(
# Copied from diffusers.pipelines.stable_diffusion_xl.pipeline_stable_diffusion_xl.StableDiffusionXLPipeline.upcast_vae
def upcast_vae(self):
dtype = self.vae.dtype
self.vae.to(dtype=torch.float32)
use_torch_2_0_or_xformers = isinstance(
self.vae.decoder.mid_block.attentions[0].processor,
(
AttnProcessor2_0,
XFormersAttnProcessor,
FusedAttnProcessor2_0,
),
)
# if xformers or torch_2_0 is used attention block does not need
# to be in float32 which can save lots of memory
if use_torch_2_0_or_xformers:
self.vae.post_quant_conv.to(dtype)
self.vae.decoder.conv_in.to(dtype)
self.vae.decoder.mid_block.to(dtype)
@property
def guidance_scale(self):

20
src/diffusers/pipelines/stable_diffusion_xl/pipeline_stable_diffusion_xl.py
View File

@@ -33,11 +33,6 @@ from ...loaders import (
TextualInversionLoaderMixin,
)
from ...models import AutoencoderKL, ImageProjection, UNet2DConditionModel
from ...models.attention_processor import (
AttnProcessor2_0,
FusedAttnProcessor2_0,
XFormersAttnProcessor,
)
from ...models.lora import adjust_lora_scale_text_encoder
from ...schedulers import KarrasDiffusionSchedulers
from ...utils import (
@@ -750,22 +745,7 @@ class StableDiffusionXLPipeline(
return add_time_ids
def upcast_vae(self):
dtype = self.vae.dtype
self.vae.to(dtype=torch.float32)
use_torch_2_0_or_xformers = isinstance(
self.vae.decoder.mid_block.attentions[0].processor,
(
AttnProcessor2_0,
XFormersAttnProcessor,
FusedAttnProcessor2_0,
),
)
# if xformers or torch_2_0 is used attention block does not need
# to be in float32 which can save lots of memory
if use_torch_2_0_or_xformers:
self.vae.post_quant_conv.to(dtype)
self.vae.decoder.conv_in.to(dtype)
self.vae.decoder.mid_block.to(dtype)
# Copied from diffusers.pipelines.latent_consistency_models.pipeline_latent_consistency_text2img.LatentConsistencyModelPipeline.get_guidance_scale_embedding
def get_guidance_scale_embedding(

18
src/diffusers/pipelines/stable_diffusion_xl/pipeline_stable_diffusion_xl_img2img.py
View File

@@ -34,10 +34,6 @@ from ...loaders import (
TextualInversionLoaderMixin,
)
from ...models import AutoencoderKL, ImageProjection, UNet2DConditionModel
from ...models.attention_processor import (
AttnProcessor2_0,
XFormersAttnProcessor,
)
from ...models.lora import adjust_lora_scale_text_encoder
from ...schedulers import KarrasDiffusionSchedulers
from ...utils import (
@@ -897,21 +893,7 @@ class StableDiffusionXLImg2ImgPipeline(
# Copied from diffusers.pipelines.stable_diffusion.pipeline_stable_diffusion_upscale.StableDiffusionUpscalePipeline.upcast_vae
def upcast_vae(self):
dtype = self.vae.dtype
self.vae.to(dtype=torch.float32)
use_torch_2_0_or_xformers = isinstance(
self.vae.decoder.mid_block.attentions[0].processor,
(
AttnProcessor2_0,
XFormersAttnProcessor,
),
)
# if xformers or torch_2_0 is used attention block does not need
# to be in float32 which can save lots of memory
if use_torch_2_0_or_xformers:
self.vae.post_quant_conv.to(dtype)
self.vae.decoder.conv_in.to(dtype)
self.vae.decoder.mid_block.to(dtype)
# Copied from diffusers.pipelines.latent_consistency_models.pipeline_latent_consistency_text2img.LatentConsistencyModelPipeline.get_guidance_scale_embedding
def get_guidance_scale_embedding(

18
src/diffusers/pipelines/stable_diffusion_xl/pipeline_stable_diffusion_xl_inpaint.py
View File

@@ -35,10 +35,6 @@ from ...loaders import (
TextualInversionLoaderMixin,
)
from ...models import AutoencoderKL, ImageProjection, UNet2DConditionModel
from ...models.attention_processor import (
AttnProcessor2_0,
XFormersAttnProcessor,
)
from ...models.lora import adjust_lora_scale_text_encoder
from ...schedulers import KarrasDiffusionSchedulers
from ...utils import (
@@ -1002,21 +998,7 @@ class StableDiffusionXLInpaintPipeline(
# Copied from diffusers.pipelines.stable_diffusion.pipeline_stable_diffusion_upscale.StableDiffusionUpscalePipeline.upcast_vae
def upcast_vae(self):
dtype = self.vae.dtype
self.vae.to(dtype=torch.float32)
use_torch_2_0_or_xformers = isinstance(
self.vae.decoder.mid_block.attentions[0].processor,
(
AttnProcessor2_0,
XFormersAttnProcessor,
),
)
# if xformers or torch_2_0 is used attention block does not need
# to be in float32 which can save lots of memory
if use_torch_2_0_or_xformers:
self.vae.post_quant_conv.to(dtype)
self.vae.decoder.conv_in.to(dtype)
self.vae.decoder.mid_block.to(dtype)
# Copied from diffusers.pipelines.latent_consistency_models.pipeline_latent_consistency_text2img.LatentConsistencyModelPipeline.get_guidance_scale_embedding
def get_guidance_scale_embedding(

20
src/diffusers/pipelines/stable_diffusion_xl/pipeline_stable_diffusion_xl_instruct_pix2pix.py
View File

@@ -22,11 +22,6 @@ from transformers import CLIPTextModel, CLIPTextModelWithProjection, CLIPTokeniz
from ...image_processor import PipelineImageInput, VaeImageProcessor
from ...loaders import FromSingleFileMixin, StableDiffusionXLLoraLoaderMixin, TextualInversionLoaderMixin
from ...models import AutoencoderKL, UNet2DConditionModel
from ...models.attention_processor import (
AttnProcessor2_0,
FusedAttnProcessor2_0,
XFormersAttnProcessor,
)
from ...models.lora import adjust_lora_scale_text_encoder
from ...schedulers import KarrasDiffusionSchedulers
from ...utils import (
@@ -590,22 +585,7 @@ class StableDiffusionXLInstructPix2PixPipeline(
# Copied from diffusers.pipelines.stable_diffusion_xl.pipeline_stable_diffusion_xl.StableDiffusionXLPipeline.upcast_vae
def upcast_vae(self):
dtype = self.vae.dtype
self.vae.to(dtype=torch.float32)
use_torch_2_0_or_xformers = isinstance(
self.vae.decoder.mid_block.attentions[0].processor,
(
AttnProcessor2_0,
XFormersAttnProcessor,
FusedAttnProcessor2_0,
),
)
# if xformers or torch_2_0 is used attention block does not need
# to be in float32 which can save lots of memory
if use_torch_2_0_or_xformers:
self.vae.post_quant_conv.to(dtype)
self.vae.decoder.conv_in.to(dtype)
self.vae.decoder.mid_block.to(dtype)
@torch.no_grad()
@replace_example_docstring(EXAMPLE_DOC_STRING)

18
src/diffusers/pipelines/t2i_adapter/pipeline_stable_diffusion_xl_adapter.py
View File

@@ -34,10 +34,6 @@ from ...loaders import (
TextualInversionLoaderMixin,
)
from ...models import AutoencoderKL, ImageProjection, MultiAdapter, T2IAdapter, UNet2DConditionModel
from ...models.attention_processor import (
AttnProcessor2_0,
XFormersAttnProcessor,
)
from ...models.lora import adjust_lora_scale_text_encoder
from ...schedulers import KarrasDiffusionSchedulers
from ...utils import (
@@ -779,21 +775,7 @@ class StableDiffusionXLAdapterPipeline(
# Copied from diffusers.pipelines.stable_diffusion.pipeline_stable_diffusion_upscale.StableDiffusionUpscalePipeline.upcast_vae
def upcast_vae(self):
dtype = self.vae.dtype
self.vae.to(dtype=torch.float32)
use_torch_2_0_or_xformers = isinstance(
self.vae.decoder.mid_block.attentions[0].processor,
(
AttnProcessor2_0,
XFormersAttnProcessor,
),
)
# if xformers or torch_2_0 is used attention block does not need
# to be in float32 which can save lots of memory
if use_torch_2_0_or_xformers:
self.vae.post_quant_conv.to(dtype)
self.vae.decoder.conv_in.to(dtype)
self.vae.decoder.mid_block.to(dtype)
# Copied from diffusers.pipelines.t2i_adapter.pipeline_stable_diffusion_adapter.StableDiffusionAdapterPipeline._default_height_width
def _default_height_width(self, height, width, image):

20
src/diffusers/pipelines/text_to_video_synthesis/pipeline_text_to_video_zero_sdxl.py
View File

@@ -19,11 +19,6 @@ from transformers import (
from ...image_processor import VaeImageProcessor
from ...loaders import StableDiffusionXLLoraLoaderMixin, TextualInversionLoaderMixin
from ...models import AutoencoderKL, UNet2DConditionModel
from ...models.attention_processor import (
AttnProcessor2_0,
FusedAttnProcessor2_0,
XFormersAttnProcessor,
)
from ...models.lora import adjust_lora_scale_text_encoder
from ...schedulers import KarrasDiffusionSchedulers
from ...utils import (
@@ -457,22 +452,7 @@ class TextToVideoZeroSDXLPipeline(
# Copied from diffusers.pipelines.stable_diffusion_xl.pipeline_stable_diffusion_xl.StableDiffusionXLPipeline.upcast_vae
def upcast_vae(self):
dtype = self.vae.dtype
self.vae.to(dtype=torch.float32)
use_torch_2_0_or_xformers = isinstance(
self.vae.decoder.mid_block.attentions[0].processor,
(
AttnProcessor2_0,
XFormersAttnProcessor,
FusedAttnProcessor2_0,
),
)
# if xformers or torch_2_0 is used attention block does not need
# to be in float32 which can save lots of memory
if use_torch_2_0_or_xformers:
self.vae.post_quant_conv.to(dtype)
self.vae.decoder.conv_in.to(dtype)
self.vae.decoder.mid_block.to(dtype)
# Copied from diffusers.pipelines.stable_diffusion_xl.pipeline_stable_diffusion_xl.StableDiffusionXLPipeline._get_add_time_ids
def _get_add_time_ids(