diffusers 0.32.2__py3-none-any.whl → 0.33.0__py3-none-any.whl

diffusers/pipelines/hunyuandit/pipeline_hunyuandit.py

@@ -207,8 +207,8 @@ class HunyuanDiTPipeline(DiffusionPipeline):
         safety_checker: StableDiffusionSafetyChecker,
         feature_extractor: CLIPImageProcessor,
         requires_safety_checker: bool = True,
-        text_encoder_2=T5EncoderModel,
-        tokenizer_2=MT5Tokenizer,
+        text_encoder_2: Optional[T5EncoderModel] = None,
+        tokenizer_2: Optional[MT5Tokenizer] = None,
     ):
         super().__init__()
 
@@ -240,9 +240,7 @@ class HunyuanDiTPipeline(DiffusionPipeline):
                 " checker. If you do not want to use the safety checker, you can pass `'safety_checker=None'` instead."
             )
 
-        self.vae_scale_factor = (
-            2 ** (len(self.vae.config.block_out_channels) - 1) if hasattr(self, "vae") and self.vae is not None else 8
-        )
+        self.vae_scale_factor = 2 ** (len(self.vae.config.block_out_channels) - 1) if getattr(self, "vae", None) else 8
         self.image_processor = VaeImageProcessor(vae_scale_factor=self.vae_scale_factor)
         self.register_to_config(requires_safety_checker=requires_safety_checker)
         self.default_sample_size = (

diffusers/pipelines/i2vgen_xl/pipeline_i2vgen_xl.py

@@ -27,6 +27,7 @@ from ...models.unets.unet_i2vgen_xl import I2VGenXLUNet
 from ...schedulers import DDIMScheduler
 from ...utils import (
     BaseOutput,
+    is_torch_xla_available,
     logging,
     replace_example_docstring,
 )
@@ -35,8 +36,16 @@ from ...video_processor import VideoProcessor
 from ..pipeline_utils import DiffusionPipeline, StableDiffusionMixin
 
 
+if is_torch_xla_available():
+    import torch_xla.core.xla_model as xm
+
+    XLA_AVAILABLE = True
+else:
+    XLA_AVAILABLE = False
+
 logger = logging.get_logger(__name__)  # pylint: disable=invalid-name
 
+
 EXAMPLE_DOC_STRING = """
     Examples:
         ```py
@@ -133,7 +142,7 @@ class I2VGenXLPipeline(
             unet=unet,
             scheduler=scheduler,
         )
-        self.vae_scale_factor = 2 ** (len(self.vae.config.block_out_channels) - 1)
+        self.vae_scale_factor = 2 ** (len(self.vae.config.block_out_channels) - 1) if getattr(self, "vae", None) else 8
         # `do_resize=False` as we do custom resizing.
         self.video_processor = VideoProcessor(vae_scale_factor=self.vae_scale_factor, do_resize=False)
 
@@ -711,6 +720,9 @@ class I2VGenXLPipeline(
                 if i == len(timesteps) - 1 or ((i + 1) > num_warmup_steps and (i + 1) % self.scheduler.order == 0):
                     progress_bar.update()
 
+                if XLA_AVAILABLE:
+                    xm.mark_step()
+
         # 8. Post processing
         if output_type == "latent":
             video = latents

diffusers/pipelines/kandinsky/pipeline_kandinsky.py

@@ -22,6 +22,7 @@ from transformers import (
 from ...models import UNet2DConditionModel, VQModel
 from ...schedulers import DDIMScheduler, DDPMScheduler
 from ...utils import (
+    is_torch_xla_available,
     logging,
     replace_example_docstring,
 )
@@ -30,8 +31,16 @@ from ..pipeline_utils import DiffusionPipeline, ImagePipelineOutput
 from .text_encoder import MultilingualCLIP
 
 
+if is_torch_xla_available():
+    import torch_xla.core.xla_model as xm
+
+    XLA_AVAILABLE = True
+else:
+    XLA_AVAILABLE = False
+
 logger = logging.get_logger(__name__)  # pylint: disable=invalid-name
 
+
 EXAMPLE_DOC_STRING = """
     Examples:
         ```py
@@ -385,6 +394,9 @@ class KandinskyPipeline(DiffusionPipeline):
                 step_idx = i // getattr(self.scheduler, "order", 1)
                 callback(step_idx, t, latents)
 
+            if XLA_AVAILABLE:
+                xm.mark_step()
+
         # post-processing
         image = self.movq.decode(latents, force_not_quantize=True)["sample"]
 

diffusers/pipelines/kandinsky/pipeline_kandinsky_combined.py

@@ -360,7 +360,7 @@ class KandinskyImg2ImgCombinedPipeline(DiffusionPipeline):
     """
 
     _load_connected_pipes = True
-    model_cpu_offload_seq = "prior_text_encoder->prior_image_encoder->prior_prior->" "text_encoder->unet->movq"
+    model_cpu_offload_seq = "prior_text_encoder->prior_image_encoder->prior_prior->text_encoder->unet->movq"
     _exclude_from_cpu_offload = ["prior_prior"]
 
     def __init__(

diffusers/pipelines/kandinsky/pipeline_kandinsky_img2img.py

@@ -25,6 +25,7 @@ from transformers import (
 from ...models import UNet2DConditionModel, VQModel
 from ...schedulers import DDIMScheduler
 from ...utils import (
+    is_torch_xla_available,
     logging,
     replace_example_docstring,
 )
@@ -33,8 +34,16 @@ from ..pipeline_utils import DiffusionPipeline, ImagePipelineOutput
 from .text_encoder import MultilingualCLIP
 
 
+if is_torch_xla_available():
+    import torch_xla.core.xla_model as xm
+
+    XLA_AVAILABLE = True
+else:
+    XLA_AVAILABLE = False
+
 logger = logging.get_logger(__name__)  # pylint: disable=invalid-name
 
+
 EXAMPLE_DOC_STRING = """
     Examples:
         ```py
@@ -478,6 +487,9 @@ class KandinskyImg2ImgPipeline(DiffusionPipeline):
                 step_idx = i // getattr(self.scheduler, "order", 1)
                 callback(step_idx, t, latents)
 
+            if XLA_AVAILABLE:
+                xm.mark_step()
+
         # 7. post-processing
         image = self.movq.decode(latents, force_not_quantize=True)["sample"]
 

diffusers/pipelines/kandinsky/pipeline_kandinsky_inpaint.py

@@ -29,6 +29,7 @@ from ... import __version__
 from ...models import UNet2DConditionModel, VQModel
 from ...schedulers import DDIMScheduler
 from ...utils import (
+    is_torch_xla_available,
     logging,
     replace_example_docstring,
 )
@@ -37,8 +38,16 @@ from ..pipeline_utils import DiffusionPipeline, ImagePipelineOutput
 from .text_encoder import MultilingualCLIP
 
 
+if is_torch_xla_available():
+    import torch_xla.core.xla_model as xm
+
+    XLA_AVAILABLE = True
+else:
+    XLA_AVAILABLE = False
+
 logger = logging.get_logger(__name__)  # pylint: disable=invalid-name
 
+
 EXAMPLE_DOC_STRING = """
     Examples:
         ```py
@@ -570,7 +579,7 @@ class KandinskyInpaintPipeline(DiffusionPipeline):
                 f"Incorrect configuration settings! The config of `pipeline.unet`: {self.unet.config} expects"
                 f" {self.unet.config.in_channels} but received `num_channels_latents`: {num_channels_latents} +"
                 f" `num_channels_mask`: {num_channels_mask} + `num_channels_masked_image`: {num_channels_masked_image}"
-                f" = {num_channels_latents+num_channels_masked_image+num_channels_mask}. Please verify the config of"
+                f" = {num_channels_latents + num_channels_masked_image + num_channels_mask}. Please verify the config of"
                 " `pipeline.unet` or your `mask_image` or `image` input."
             )
 
@@ -613,6 +622,9 @@ class KandinskyInpaintPipeline(DiffusionPipeline):
                 step_idx = i // getattr(self.scheduler, "order", 1)
                 callback(step_idx, t, latents)
 
+            if XLA_AVAILABLE:
+                xm.mark_step()
+
         # post-processing
         image = self.movq.decode(latents, force_not_quantize=True)["sample"]
 

diffusers/pipelines/kandinsky/pipeline_kandinsky_prior.py

@@ -24,6 +24,7 @@ from ...models import PriorTransformer
 from ...schedulers import UnCLIPScheduler
 from ...utils import (
     BaseOutput,
+    is_torch_xla_available,
     logging,
     replace_example_docstring,
 )
@@ -31,8 +32,16 @@ from ...utils.torch_utils import randn_tensor
 from ..pipeline_utils import DiffusionPipeline
 
 
+if is_torch_xla_available():
+    import torch_xla.core.xla_model as xm
+
+    XLA_AVAILABLE = True
+else:
+    XLA_AVAILABLE = False
+
 logger = logging.get_logger(__name__)  # pylint: disable=invalid-name
 
+
 EXAMPLE_DOC_STRING = """
     Examples:
         ```py
@@ -519,6 +528,9 @@ class KandinskyPriorPipeline(DiffusionPipeline):
                 prev_timestep=prev_timestep,
             ).prev_sample
 
+            if XLA_AVAILABLE:
+                xm.mark_step()
+
         latents = self.prior.post_process_latents(latents)
 
         image_embeddings = latents

diffusers/pipelines/kandinsky2_2/pipeline_kandinsky2_2.py

@@ -18,13 +18,21 @@ import torch
 
 from ...models import UNet2DConditionModel, VQModel
 from ...schedulers import DDPMScheduler
-from ...utils import deprecate, logging, replace_example_docstring
+from ...utils import deprecate, is_torch_xla_available, logging, replace_example_docstring
 from ...utils.torch_utils import randn_tensor
 from ..pipeline_utils import DiffusionPipeline, ImagePipelineOutput
 
 
+if is_torch_xla_available():
+    import torch_xla.core.xla_model as xm
+
+    XLA_AVAILABLE = True
+else:
+    XLA_AVAILABLE = False
+
 logger = logging.get_logger(__name__)  # pylint: disable=invalid-name
 
+
 EXAMPLE_DOC_STRING = """
     Examples:
         ```py
@@ -296,6 +304,9 @@ class KandinskyV22Pipeline(DiffusionPipeline):
                 step_idx = i // getattr(self.scheduler, "order", 1)
                 callback(step_idx, t, latents)
 
+            if XLA_AVAILABLE:
+                xm.mark_step()
+
         if output_type not in ["pt", "np", "pil", "latent"]:
             raise ValueError(f"Only the output types `pt`, `pil` and `np` are supported not output_type={output_type}")
 

diffusers/pipelines/kandinsky2_2/pipeline_kandinsky2_2_controlnet.py

@@ -19,14 +19,23 @@ import torch
 from ...models import UNet2DConditionModel, VQModel
 from ...schedulers import DDPMScheduler
 from ...utils import (
+    is_torch_xla_available,
     logging,
 )
 from ...utils.torch_utils import randn_tensor
 from ..pipeline_utils import DiffusionPipeline, ImagePipelineOutput
 
 
+if is_torch_xla_available():
+    import torch_xla.core.xla_model as xm
+
+    XLA_AVAILABLE = True
+else:
+    XLA_AVAILABLE = False
+
 logger = logging.get_logger(__name__)  # pylint: disable=invalid-name
 
+
 EXAMPLE_DOC_STRING = """
     Examples:
         ```py
@@ -297,6 +306,10 @@ class KandinskyV22ControlnetPipeline(DiffusionPipeline):
             if callback is not None and i % callback_steps == 0:
                 step_idx = i // getattr(self.scheduler, "order", 1)
                 callback(step_idx, t, latents)
+
+            if XLA_AVAILABLE:
+                xm.mark_step()
+
         # post-processing
         image = self.movq.decode(latents, force_not_quantize=True)["sample"]
 

diffusers/pipelines/kandinsky2_2/pipeline_kandinsky2_2_controlnet_img2img.py

@@ -22,14 +22,23 @@ from PIL import Image
 from ...models import UNet2DConditionModel, VQModel
 from ...schedulers import DDPMScheduler
 from ...utils import (
+    is_torch_xla_available,
     logging,
 )
 from ...utils.torch_utils import randn_tensor
 from ..pipeline_utils import DiffusionPipeline, ImagePipelineOutput
 
 
+if is_torch_xla_available():
+    import torch_xla.core.xla_model as xm
+
+    XLA_AVAILABLE = True
+else:
+    XLA_AVAILABLE = False
+
 logger = logging.get_logger(__name__)  # pylint: disable=invalid-name
 
+
 EXAMPLE_DOC_STRING = """
     Examples:
         ```py
@@ -358,6 +367,9 @@ class KandinskyV22ControlnetImg2ImgPipeline(DiffusionPipeline):
                 step_idx = i // getattr(self.scheduler, "order", 1)
                 callback(step_idx, t, latents)
 
+            if XLA_AVAILABLE:
+                xm.mark_step()
+
         # post-processing
         image = self.movq.decode(latents, force_not_quantize=True)["sample"]
 

diffusers/pipelines/kandinsky2_2/pipeline_kandinsky2_2_img2img.py

@@ -21,13 +21,21 @@ from PIL import Image
 
 from ...models import UNet2DConditionModel, VQModel
 from ...schedulers import DDPMScheduler
-from ...utils import deprecate, logging
+from ...utils import deprecate, is_torch_xla_available, logging
 from ...utils.torch_utils import randn_tensor
 from ..pipeline_utils import DiffusionPipeline, ImagePipelineOutput
 
 
+if is_torch_xla_available():
+    import torch_xla.core.xla_model as xm
+
+    XLA_AVAILABLE = True
+else:
+    XLA_AVAILABLE = False
+
 logger = logging.get_logger(__name__)  # pylint: disable=invalid-name
 
+
 EXAMPLE_DOC_STRING = """
     Examples:
         ```py
@@ -372,6 +380,9 @@ class KandinskyV22Img2ImgPipeline(DiffusionPipeline):
                 step_idx = i // getattr(self.scheduler, "order", 1)
                 callback(step_idx, t, latents)
 
+            if XLA_AVAILABLE:
+                xm.mark_step()
+
         if output_type not in ["pt", "np", "pil", "latent"]:
             raise ValueError(
                 f"Only the output types `pt`, `pil` ,`np` and `latent` are supported not output_type={output_type}"

diffusers/pipelines/kandinsky2_2/pipeline_kandinsky2_2_inpainting.py

@@ -25,13 +25,21 @@ from PIL import Image
 from ... import __version__
 from ...models import UNet2DConditionModel, VQModel
 from ...schedulers import DDPMScheduler
-from ...utils import deprecate, logging
+from ...utils import deprecate, is_torch_xla_available, logging
 from ...utils.torch_utils import randn_tensor
 from ..pipeline_utils import DiffusionPipeline, ImagePipelineOutput
 
 
+if is_torch_xla_available():
+    import torch_xla.core.xla_model as xm
+
+    XLA_AVAILABLE = True
+else:
+    XLA_AVAILABLE = False
+
 logger = logging.get_logger(__name__)  # pylint: disable=invalid-name
 
+
 EXAMPLE_DOC_STRING = """
     Examples:
         ```py
@@ -526,6 +534,9 @@ class KandinskyV22InpaintPipeline(DiffusionPipeline):
                 step_idx = i // getattr(self.scheduler, "order", 1)
                 callback(step_idx, t, latents)
 
+            if XLA_AVAILABLE:
+                xm.mark_step()
+
         # post-processing
         latents = mask_image[:1] * image[:1] + (1 - mask_image[:1]) * latents
 

diffusers/pipelines/kandinsky2_2/pipeline_kandinsky2_2_prior.py

@@ -7,6 +7,7 @@ from transformers import CLIPImageProcessor, CLIPTextModelWithProjection, CLIPTo
 from ...models import PriorTransformer
 from ...schedulers import UnCLIPScheduler
 from ...utils import (
+    is_torch_xla_available,
     logging,
     replace_example_docstring,
 )
@@ -15,8 +16,16 @@ from ..kandinsky import KandinskyPriorPipelineOutput
 from ..pipeline_utils import DiffusionPipeline
 
 
+if is_torch_xla_available():
+    import torch_xla.core.xla_model as xm
+
+    XLA_AVAILABLE = True
+else:
+    XLA_AVAILABLE = False
+
 logger = logging.get_logger(__name__)  # pylint: disable=invalid-name
 
+
 EXAMPLE_DOC_STRING = """
     Examples:
         ```py
@@ -524,6 +533,9 @@ class KandinskyV22PriorPipeline(DiffusionPipeline):
                 )
                 text_mask = callback_outputs.pop("text_mask", text_mask)
 
+            if XLA_AVAILABLE:
+                xm.mark_step()
+
         latents = self.prior.post_process_latents(latents)
 
         image_embeddings = latents

diffusers/pipelines/kandinsky2_2/pipeline_kandinsky2_2_prior_emb2emb.py

@@ -7,6 +7,7 @@ from transformers import CLIPImageProcessor, CLIPTextModelWithProjection, CLIPTo
 from ...models import PriorTransformer
 from ...schedulers import UnCLIPScheduler
 from ...utils import (
+    is_torch_xla_available,
     logging,
     replace_example_docstring,
 )
@@ -15,8 +16,16 @@ from ..kandinsky import KandinskyPriorPipelineOutput
 from ..pipeline_utils import DiffusionPipeline
 
 
+if is_torch_xla_available():
+    import torch_xla.core.xla_model as xm
+
+    XLA_AVAILABLE = True
+else:
+    XLA_AVAILABLE = False
+
 logger = logging.get_logger(__name__)  # pylint: disable=invalid-name
 
+
 EXAMPLE_DOC_STRING = """
     Examples:
         ```py
@@ -538,6 +547,9 @@ class KandinskyV22PriorEmb2EmbPipeline(DiffusionPipeline):
                 prev_timestep=prev_timestep,
             ).prev_sample
 
+            if XLA_AVAILABLE:
+                xm.mark_step()
+
         latents = self.prior.post_process_latents(latents)
 
         image_embeddings = latents

diffusers/pipelines/kandinsky3/pipeline_kandinsky3.py

@@ -8,6 +8,7 @@ from ...models import Kandinsky3UNet, VQModel
 from ...schedulers import DDPMScheduler
 from ...utils import (
     deprecate,
+    is_torch_xla_available,
     logging,
     replace_example_docstring,
 )
@@ -15,8 +16,16 @@ from ...utils.torch_utils import randn_tensor
 from ..pipeline_utils import DiffusionPipeline, ImagePipelineOutput
 
 
+if is_torch_xla_available():
+    import torch_xla.core.xla_model as xm
+
+    XLA_AVAILABLE = True
+else:
+    XLA_AVAILABLE = False
+
 logger = logging.get_logger(__name__)  # pylint: disable=invalid-name
 
+
 EXAMPLE_DOC_STRING = """
     Examples:
         ```py
@@ -549,6 +558,9 @@ class Kandinsky3Pipeline(DiffusionPipeline, StableDiffusionLoraLoaderMixin):
                         step_idx = i // getattr(self.scheduler, "order", 1)
                         callback(step_idx, t, latents)
 
+                if XLA_AVAILABLE:
+                    xm.mark_step()
+
             # post-processing
             if output_type not in ["pt", "np", "pil", "latent"]:
                 raise ValueError(

diffusers/pipelines/kandinsky3/pipeline_kandinsky3_img2img.py

@@ -12,6 +12,7 @@ from ...models import Kandinsky3UNet, VQModel
 from ...schedulers import DDPMScheduler
 from ...utils import (
     deprecate,
+    is_torch_xla_available,
     logging,
     replace_example_docstring,
 )
@@ -19,8 +20,16 @@ from ...utils.torch_utils import randn_tensor
 from ..pipeline_utils import DiffusionPipeline, ImagePipelineOutput
 
 
+if is_torch_xla_available():
+    import torch_xla.core.xla_model as xm
+
+    XLA_AVAILABLE = True
+else:
+    XLA_AVAILABLE = False
+
 logger = logging.get_logger(__name__)  # pylint: disable=invalid-name
 
+
 EXAMPLE_DOC_STRING = """
     Examples:
         ```py
@@ -617,6 +626,9 @@ class Kandinsky3Img2ImgPipeline(DiffusionPipeline, StableDiffusionLoraLoaderMixi
                         step_idx = i // getattr(self.scheduler, "order", 1)
                         callback(step_idx, t, latents)
 
+                if XLA_AVAILABLE:
+                    xm.mark_step()
+
             # post-processing
             if output_type not in ["pt", "np", "pil", "latent"]:
                 raise ValueError(

diffusers/pipelines/kolors/pipeline_kolors.py

@@ -19,7 +19,7 @@ from transformers import CLIPImageProcessor, CLIPVisionModelWithProjection
 
 from ...callbacks import MultiPipelineCallbacks, PipelineCallback
 from ...image_processor import PipelineImageInput, VaeImageProcessor
-from ...loaders import IPAdapterMixin, StableDiffusionXLLoraLoaderMixin
+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
@@ -121,7 +121,7 @@ def retrieve_timesteps(
     return timesteps, num_inference_steps
 
 
-class KolorsPipeline(DiffusionPipeline, StableDiffusionMixin, StableDiffusionXLLoraLoaderMixin, IPAdapterMixin):
+class KolorsPipeline(DiffusionPipeline, StableDiffusionMixin, StableDiffusionLoraLoaderMixin, IPAdapterMixin):
     r"""
     Pipeline for text-to-image generation using Kolors.
 
@@ -129,8 +129,8 @@ class KolorsPipeline(DiffusionPipeline, StableDiffusionMixin, StableDiffusionXLL
     library implements for all the pipelines (such as downloading or saving, running on a particular device, etc.)
 
     The pipeline also inherits the following loading methods:
-        - [`~loaders.StableDiffusionXLLoraLoaderMixin.load_lora_weights`] for loading LoRA weights
-        - [`~loaders.StableDiffusionXLLoraLoaderMixin.save_lora_weights`] for saving LoRA weights
+        - [`~loaders.StableDiffusionLoraLoaderMixin.load_lora_weights`] for loading LoRA weights
+        - [`~loaders.StableDiffusionLoraLoaderMixin.save_lora_weights`] for saving LoRA weights
         - [`~loaders.IPAdapterMixin.load_ip_adapter`] for loading IP Adapters
 
     Args:
@@ -188,12 +188,14 @@ class KolorsPipeline(DiffusionPipeline, StableDiffusionMixin, StableDiffusionXLL
             feature_extractor=feature_extractor,
         )
         self.register_to_config(force_zeros_for_empty_prompt=force_zeros_for_empty_prompt)
-        self.vae_scale_factor = (
-            2 ** (len(self.vae.config.block_out_channels) - 1) if hasattr(self, "vae") and self.vae is not None else 8
-        )
+        self.vae_scale_factor = 2 ** (len(self.vae.config.block_out_channels) - 1) if getattr(self, "vae", None) else 8
         self.image_processor = VaeImageProcessor(vae_scale_factor=self.vae_scale_factor)
 
-        self.default_sample_size = self.unet.config.sample_size
+        self.default_sample_size = (
+            self.unet.config.sample_size
+            if hasattr(self, "unet") and self.unet is not None and hasattr(self.unet.config, "sample_size")
+            else 128
+        )
 
     def encode_prompt(
         self,

diffusers/pipelines/kolors/pipeline_kolors_img2img.py

@@ -207,12 +207,14 @@ class KolorsImg2ImgPipeline(DiffusionPipeline, StableDiffusionMixin, StableDiffu
             feature_extractor=feature_extractor,
         )
         self.register_to_config(force_zeros_for_empty_prompt=force_zeros_for_empty_prompt)
-        self.vae_scale_factor = (
-            2 ** (len(self.vae.config.block_out_channels) - 1) if hasattr(self, "vae") and self.vae is not None else 8
-        )
+        self.vae_scale_factor = 2 ** (len(self.vae.config.block_out_channels) - 1) if getattr(self, "vae", None) else 8
         self.image_processor = VaeImageProcessor(vae_scale_factor=self.vae_scale_factor)
 
-        self.default_sample_size = self.unet.config.sample_size
+        self.default_sample_size = (
+            self.unet.config.sample_size
+            if hasattr(self, "unet") and self.unet is not None and hasattr(self.unet.config, "sample_size")
+            else 128
+        )
 
     # Copied from diffusers.pipelines.kolors.pipeline_kolors.KolorsPipeline.encode_prompt
     def encode_prompt(