diffusers 0.33.1__py3-none-any.whl → 0.35.0__py3-none-any.whl

diffusers/pipelines/pag/pipeline_pag_sd_xl_inpaint.py

@@ -1,4 +1,4 @@
-# Copyright 2024 The HuggingFace Team. All rights reserved.
+# Copyright 2025 The HuggingFace Team. All rights reserved.
 #
 # Licensed under the Apache License, Version 2.0 (the "License");
 # you may not use this file except in compliance with the License.
@@ -108,7 +108,7 @@ def rescale_noise_cfg(noise_cfg, noise_pred_text, guidance_rescale=0.0):
     r"""
     Rescales `noise_cfg` tensor based on `guidance_rescale` to improve image quality and fix overexposure. Based on
     Section 3.4 from [Common Diffusion Noise Schedules and Sample Steps are
-    Flawed](https://arxiv.org/pdf/2305.08891.pdf).
+    Flawed](https://huggingface.co/papers/2305.08891).
 
     Args:
         noise_cfg (`torch.Tensor`):
@@ -643,7 +643,7 @@ class StableDiffusionXLPAGInpaintPipeline(
     def prepare_extra_step_kwargs(self, generator, eta):
         # prepare extra kwargs for the scheduler step, since not all schedulers have the same signature
         # eta (η) is only used with the DDIMScheduler, it will be ignored for other schedulers.
-        # eta corresponds to η in DDIM paper: https://arxiv.org/abs/2010.02502
+        # eta corresponds to η in DDIM paper: https://huggingface.co/papers/2010.02502
         # and should be between [0, 1]
 
         accepts_eta = "eta" in set(inspect.signature(self.scheduler.step).parameters.keys())
@@ -1061,7 +1061,7 @@ class StableDiffusionXLPAGInpaintPipeline(
         return self._clip_skip
 
     # here `guidance_scale` is defined analog to the guidance weight `w` of equation (2)
-    # of the Imagen paper: https://arxiv.org/pdf/2205.11487.pdf . `guidance_scale = 1`
+    # of the Imagen paper: https://huggingface.co/papers/2205.11487 . `guidance_scale = 1`
     # corresponds to doing no classifier free guidance.
     @property
     def do_classifier_free_guidance(self):
@@ -1208,11 +1208,11 @@ class StableDiffusionXLPAGInpaintPipeline(
                 forms a part of a "Mixture of Denoisers" multi-pipeline setup, as elaborated in [**Refining the Image
                 Output**](https://huggingface.co/docs/diffusers/api/pipelines/stable_diffusion/stable_diffusion_xl#refining-the-image-output).
             guidance_scale (`float`, *optional*, defaults to 7.5):
-                Guidance scale as defined in [Classifier-Free Diffusion Guidance](https://arxiv.org/abs/2207.12598).
-                `guidance_scale` is defined as `w` of equation 2. of [Imagen
-                Paper](https://arxiv.org/pdf/2205.11487.pdf). Guidance scale is enabled by setting `guidance_scale >
-                1`. Higher guidance scale encourages to generate images that are closely linked to the text `prompt`,
-                usually at the expense of lower image quality.
+                Guidance scale as defined in [Classifier-Free Diffusion
+                Guidance](https://huggingface.co/papers/2207.12598). `guidance_scale` is defined as `w` of equation 2.
+                of [Imagen Paper](https://huggingface.co/papers/2205.11487). Guidance scale is enabled by setting
+                `guidance_scale > 1`. Higher guidance scale encourages to generate images that are closely linked to
+                the text `prompt`, usually at the expense of lower image quality.
             negative_prompt (`str` or `List[str]`, *optional*):
                 The prompt or prompts not to guide the image generation. If not defined, one has to pass
                 `negative_prompt_embeds` instead. Ignored when not using guidance (i.e., ignored if `guidance_scale` is
@@ -1243,8 +1243,8 @@ class StableDiffusionXLPAGInpaintPipeline(
             num_images_per_prompt (`int`, *optional*, defaults to 1):
                 The number of images to generate per prompt.
             eta (`float`, *optional*, defaults to 0.0):
-                Corresponds to parameter eta (η) in the DDIM paper: https://arxiv.org/abs/2010.02502. Only applies to
-                [`schedulers.DDIMScheduler`], will be ignored for others.
+                Corresponds to parameter eta (η) in the DDIM paper: https://huggingface.co/papers/2010.02502. Only
+                applies to [`schedulers.DDIMScheduler`], will be ignored for others.
             generator (`torch.Generator`, *optional*):
                 One or a list of [torch generator(s)](https://pytorch.org/docs/stable/generated/torch.Generator.html)
                 to make generation deterministic.
@@ -1673,7 +1673,7 @@ class StableDiffusionXLPAGInpaintPipeline(
                     noise_pred = noise_pred_uncond + self.guidance_scale * (noise_pred_text - noise_pred_uncond)
 
                 if self.do_classifier_free_guidance and self.guidance_rescale > 0.0:
-                    # Based on 3.4. in https://arxiv.org/pdf/2305.08891.pdf
+                    # Based on 3.4. in https://huggingface.co/papers/2305.08891
                     noise_pred = rescale_noise_cfg(noise_pred, noise_pred_text, guidance_rescale=self.guidance_rescale)
 
                 # compute the previous noisy sample x_t -> x_t-1

diffusers/pipelines/paint_by_example/image_encoder.py

@@ -1,4 +1,4 @@
-# Copyright 2024 The HuggingFace Team. All rights reserved.
+# Copyright 2025 The HuggingFace Team. All rights reserved.
 #
 # Licensed under the Apache License, Version 2.0 (the "License");
 # you may not use this file except in compliance with the License.

diffusers/pipelines/paint_by_example/pipeline_paint_by_example.py

@@ -1,4 +1,4 @@
-# Copyright 2024 The HuggingFace Team. All rights reserved.
+# Copyright 2025 The HuggingFace Team. All rights reserved.
 #
 # Licensed under the Apache License, Version 2.0 (the "License");
 # you may not use this file except in compliance with the License.
@@ -25,7 +25,7 @@ from ...models import AutoencoderKL, UNet2DConditionModel
 from ...schedulers import DDIMScheduler, LMSDiscreteScheduler, PNDMScheduler
 from ...utils import deprecate, is_torch_xla_available, logging
 from ...utils.torch_utils import randn_tensor
-from ..pipeline_utils import DiffusionPipeline, StableDiffusionMixin
+from ..pipeline_utils import DeprecatedPipelineMixin, DiffusionPipeline, StableDiffusionMixin
 from ..stable_diffusion import StableDiffusionPipelineOutput
 from ..stable_diffusion.safety_checker import StableDiffusionSafetyChecker
 from .image_encoder import PaintByExampleImageEncoder
@@ -155,7 +155,8 @@ def prepare_mask_and_masked_image(image, mask):
     return mask, masked_image
 
 
-class PaintByExamplePipeline(DiffusionPipeline, StableDiffusionMixin):
+class PaintByExamplePipeline(DeprecatedPipelineMixin, DiffusionPipeline, StableDiffusionMixin):
+    _last_supported_version = "0.33.1"
     r"""
     <Tip warning={true}>
 
@@ -239,7 +240,7 @@ class PaintByExamplePipeline(DiffusionPipeline, StableDiffusionMixin):
     def prepare_extra_step_kwargs(self, generator, eta):
         # prepare extra kwargs for the scheduler step, since not all schedulers have the same signature
         # eta (η) is only used with the DDIMScheduler, it will be ignored for other schedulers.
-        # eta corresponds to η in DDIM paper: https://arxiv.org/abs/2010.02502
+        # eta corresponds to η in DDIM paper: https://huggingface.co/papers/2010.02502
         # and should be between [0, 1]
 
         accepts_eta = "eta" in set(inspect.signature(self.scheduler.step).parameters.keys())
@@ -447,8 +448,8 @@ class PaintByExamplePipeline(DiffusionPipeline, StableDiffusionMixin):
             num_images_per_prompt (`int`, *optional*, defaults to 1):
                 The number of images to generate per prompt.
             eta (`float`, *optional*, defaults to 0.0):
-                Corresponds to parameter eta (η) from the [DDIM](https://arxiv.org/abs/2010.02502) paper. Only applies
-                to the [`~schedulers.DDIMScheduler`], and is ignored in other schedulers.
+                Corresponds to parameter eta (η) from the [DDIM](https://huggingface.co/papers/2010.02502) paper. Only
+                applies to the [`~schedulers.DDIMScheduler`], and is ignored in other schedulers.
             generator (`torch.Generator` or `List[torch.Generator]`, *optional*):
                 A [`torch.Generator`](https://pytorch.org/docs/stable/generated/torch.Generator.html) to make
                 generation deterministic.
@@ -521,7 +522,7 @@ class PaintByExamplePipeline(DiffusionPipeline, StableDiffusionMixin):
             batch_size = image.shape[0]
         device = self._execution_device
         # here `guidance_scale` is defined analog to the guidance weight `w` of equation (2)
-        # of the Imagen paper: https://arxiv.org/pdf/2205.11487.pdf . `guidance_scale = 1`
+        # of the Imagen paper: https://huggingface.co/papers/2205.11487 . `guidance_scale = 1`
         # corresponds to doing no classifier free guidance.
         do_classifier_free_guidance = guidance_scale > 1.0
 

diffusers/pipelines/pia/pipeline_pia.py

@@ -1,4 +1,4 @@
-# Copyright 2024 The HuggingFace Team. All rights reserved.
+# Copyright 2025 The HuggingFace Team. All rights reserved.
 #
 # Licensed under the Apache License, Version 2.0 (the "License");
 # you may not use this file except in compliance with the License.
@@ -46,7 +46,7 @@ from ...utils import (
 from ...utils.torch_utils import randn_tensor
 from ...video_processor import VideoProcessor
 from ..free_init_utils import FreeInitMixin
-from ..pipeline_utils import DiffusionPipeline, StableDiffusionMixin
+from ..pipeline_utils import DeprecatedPipelineMixin, DiffusionPipeline, StableDiffusionMixin
 
 
 if is_torch_xla_available():
@@ -132,6 +132,7 @@ class PIAPipelineOutput(BaseOutput):
 
 
 class PIAPipeline(
+    DeprecatedPipelineMixin,
     DiffusionPipeline,
     StableDiffusionMixin,
     TextualInversionLoaderMixin,
@@ -140,6 +141,7 @@ class PIAPipeline(
     FromSingleFileMixin,
     FreeInitMixin,
 ):
+    _last_supported_version = "0.33.1"
     r"""
     Pipeline for text-to-video generation.
 
@@ -432,7 +434,7 @@ class PIAPipeline(
     def prepare_extra_step_kwargs(self, generator, eta):
         # prepare extra kwargs for the scheduler step, since not all schedulers have the same signature
         # eta (η) is only used with the DDIMScheduler, it will be ignored for other schedulers.
-        # eta corresponds to η in DDIM paper: https://arxiv.org/abs/2010.02502
+        # eta corresponds to η in DDIM paper: https://huggingface.co/papers/2010.02502
         # and should be between [0, 1]
 
         accepts_eta = "eta" in set(inspect.signature(self.scheduler.step).parameters.keys())
@@ -653,7 +655,7 @@ class PIAPipeline(
         return self._clip_skip
 
     # here `guidance_scale` is defined analog to the guidance weight `w` of equation (2)
-    # of the Imagen paper: https://arxiv.org/pdf/2205.11487.pdf . `guidance_scale = 1`
+    # of the Imagen paper: https://huggingface.co/papers/2205.11487 . `guidance_scale = 1`
     # corresponds to doing no classifier free guidance.
     @property
     def do_classifier_free_guidance(self):
@@ -723,8 +725,8 @@ class PIAPipeline(
                 The prompt or prompts to guide what to not include in image generation. If not defined, you need to
                 pass `negative_prompt_embeds` instead. Ignored when not using guidance (`guidance_scale < 1`).
             eta (`float`, *optional*, defaults to 0.0):
-                Corresponds to parameter eta (η) from the [DDIM](https://arxiv.org/abs/2010.02502) paper. Only applies
-                to the [`~schedulers.DDIMScheduler`], and is ignored in other schedulers.
+                Corresponds to parameter eta (η) from the [DDIM](https://huggingface.co/papers/2010.02502) paper. Only
+                applies to the [`~schedulers.DDIMScheduler`], and is ignored in other schedulers.
             generator (`torch.Generator` or `List[torch.Generator]`, *optional*):
                 A [`torch.Generator`](https://pytorch.org/docs/stable/generated/torch.Generator.html) to make
                 generation deterministic.

diffusers/pipelines/pipeline_flax_utils.py

@@ -248,9 +248,8 @@ class FlaxDiffusionPipeline(ConfigMixin, PushToHubMixin):
                       pretrained pipeline hosted on the Hub.
                     - A path to a *directory* (for example `./my_model_directory`) containing the model weights saved
                       using [`~FlaxDiffusionPipeline.save_pretrained`].
-            dtype (`str` or `jnp.dtype`, *optional*):
-                Override the default `jnp.dtype` and load the model under this dtype. If `"auto"`, the dtype is
-                automatically derived from the model's weights.
+            dtype (`jnp.dtype`, *optional*):
+                Override the default `jnp.dtype` and load the model under this dtype.
             force_download (`bool`, *optional*, defaults to `False`):
                 Whether or not to force the (re-)download of the model weights and configuration files, overriding the
                 cached versions if they exist.
@@ -279,8 +278,8 @@ class FlaxDiffusionPipeline(ConfigMixin, PushToHubMixin):
 
         <Tip>
 
-        To use private or [gated models](https://huggingface.co/docs/hub/models-gated#gated-models), log-in with
-        `huggingface-cli login`.
+        To use private or [gated models](https://huggingface.co/docs/hub/models-gated#gated-models), log-in with `hf
+        auth login`.
 
         </Tip>
 
@@ -469,7 +468,7 @@ class FlaxDiffusionPipeline(ConfigMixin, PushToHubMixin):
                 class_obj = import_flax_or_no_model(pipeline_module, class_name)
 
                 importable_classes = ALL_IMPORTABLE_CLASSES
-                class_candidates = {c: class_obj for c in importable_classes.keys()}
+                class_candidates = dict.fromkeys(importable_classes.keys(), class_obj)
             else:
                 # else we just import it from the library.
                 library = importlib.import_module(library_name)

diffusers/pipelines/pipeline_loading_utils.py

@@ -92,7 +92,7 @@ for library in LOADABLE_CLASSES:
     ALL_IMPORTABLE_CLASSES.update(LOADABLE_CLASSES[library])
 
 
-def is_safetensors_compatible(filenames, passed_components=None, folder_names=None) -> bool:
+def is_safetensors_compatible(filenames, passed_components=None, folder_names=None, variant=None) -> bool:
     """
     Checking for safetensors compatibility:
     - The model is safetensors compatible only if there is a safetensors file for each model component present in
@@ -103,6 +103,31 @@ def is_safetensors_compatible(filenames, passed_components=None, folder_names=No
     - For models from the transformers library, the filename changes from "pytorch_model" to "model", and the ".bin"
       extension is replaced with ".safetensors"
     """
+    weight_names = [
+        WEIGHTS_NAME,
+        SAFETENSORS_WEIGHTS_NAME,
+        FLAX_WEIGHTS_NAME,
+        ONNX_WEIGHTS_NAME,
+        ONNX_EXTERNAL_WEIGHTS_NAME,
+    ]
+
+    if is_transformers_available():
+        weight_names += [TRANSFORMERS_WEIGHTS_NAME, TRANSFORMERS_SAFE_WEIGHTS_NAME, TRANSFORMERS_FLAX_WEIGHTS_NAME]
+
+    # model_pytorch, diffusion_model_pytorch, ...
+    weight_prefixes = [w.split(".")[0] for w in weight_names]
+    # .bin, .safetensors, ...
+    weight_suffixs = [w.split(".")[-1] for w in weight_names]
+    # -00001-of-00002
+    transformers_index_format = r"\d{5}-of-\d{5}"
+    # `diffusion_pytorch_model.bin` as well as `model-00001-of-00002.safetensors`
+    variant_file_re = re.compile(
+        rf"({'|'.join(weight_prefixes)})\.({variant}|{variant}-{transformers_index_format})\.({'|'.join(weight_suffixs)})$"
+    )
+    non_variant_file_re = re.compile(
+        rf"({'|'.join(weight_prefixes)})(-{transformers_index_format})?\.({'|'.join(weight_suffixs)})$"
+    )
+
     passed_components = passed_components or []
     if folder_names:
         filenames = {f for f in filenames if os.path.split(f)[0] in folder_names}
@@ -121,15 +146,29 @@ def is_safetensors_compatible(filenames, passed_components=None, folder_names=No
         components[component].append(component_filename)
 
     # If there are no component folders check the main directory for safetensors files
+    filtered_filenames = set()
     if not components:
-        return any(".safetensors" in filename for filename in filenames)
+        if variant is not None:
+            filtered_filenames = filter_with_regex(filenames, variant_file_re)
+
+        # If no variant filenames exist check if non-variant files are available
+        if not filtered_filenames:
+            filtered_filenames = filter_with_regex(filenames, non_variant_file_re)
+        return any(".safetensors" in filename for filename in filtered_filenames)
 
     # iterate over all files of a component
     # check if safetensor files exist for that component
-    # if variant is provided check if the variant of the safetensors exists
     for component, component_filenames in components.items():
         matches = []
-        for component_filename in component_filenames:
+        filtered_component_filenames = set()
+        # if variant is provided check if the variant of the safetensors exists
+        if variant is not None:
+            filtered_component_filenames = filter_with_regex(component_filenames, variant_file_re)
+
+        # if variant safetensor files do not exist check for non-variants
+        if not filtered_component_filenames:
+            filtered_component_filenames = filter_with_regex(component_filenames, non_variant_file_re)
+        for component_filename in filtered_component_filenames:
             filename, extension = os.path.splitext(component_filename)
 
             match_exists = extension == ".safetensors"
@@ -159,6 +198,10 @@ def filter_model_files(filenames):
     return [f for f in filenames if any(f.endswith(extension) for extension in allowed_extensions)]
 
 
+def filter_with_regex(filenames, pattern_re):
+    return {f for f in filenames if pattern_re.match(f.split("/")[-1]) is not None}
+
+
 def variant_compatible_siblings(filenames, variant=None, ignore_patterns=None) -> Union[List[os.PathLike], str]:
     weight_names = [
         WEIGHTS_NAME,
@@ -207,9 +250,6 @@ def variant_compatible_siblings(filenames, variant=None, ignore_patterns=None) -
         # interested in the extension name
         return {f for f in filenames if not any(f.endswith(pat.lstrip("*.")) for pat in ignore_patterns)}
 
-    def filter_with_regex(filenames, pattern_re):
-        return {f for f in filenames if pattern_re.match(f.split("/")[-1]) is not None}
-
     # Group files by component
     components = {}
     for filename in filenames:
@@ -331,23 +371,39 @@ def maybe_raise_or_warn(
         )
 
 
+# a simpler version of get_class_obj_and_candidates, it won't work with custom code
+def simple_get_class_obj(library_name, class_name):
+    from diffusers import pipelines
+
+    is_pipeline_module = hasattr(pipelines, library_name)
+
+    if is_pipeline_module:
+        pipeline_module = getattr(pipelines, library_name)
+        class_obj = getattr(pipeline_module, class_name)
+    else:
+        library = importlib.import_module(library_name)
+        class_obj = getattr(library, class_name)
+
+    return class_obj
+
+
 def get_class_obj_and_candidates(
     library_name, class_name, importable_classes, pipelines, is_pipeline_module, component_name=None, cache_dir=None
 ):
     """Simple helper method to retrieve class object of module as well as potential parent class objects"""
-    component_folder = os.path.join(cache_dir, component_name)
+    component_folder = os.path.join(cache_dir, component_name) if component_name and cache_dir else None
 
     if is_pipeline_module:
         pipeline_module = getattr(pipelines, library_name)
 
         class_obj = getattr(pipeline_module, class_name)
-        class_candidates = {c: class_obj for c in importable_classes.keys()}
-    elif os.path.isfile(os.path.join(component_folder, library_name + ".py")):
+        class_candidates = dict.fromkeys(importable_classes.keys(), class_obj)
+    elif component_folder and os.path.isfile(os.path.join(component_folder, library_name + ".py")):
         # load custom component
         class_obj = get_class_from_dynamic_module(
             component_folder, module_file=library_name + ".py", class_name=class_name
         )
-        class_candidates = {c: class_obj for c in importable_classes.keys()}
+        class_candidates = dict.fromkeys(importable_classes.keys(), class_obj)
     else:
         # else we just import it from the library.
         library = importlib.import_module(library_name)
@@ -412,7 +468,7 @@ def _get_pipeline_class(
             revision=revision,
         )
 
-    if class_obj.__name__ != "DiffusionPipeline":
+    if class_obj.__name__ != "DiffusionPipeline" and class_obj.__name__ != "ModularPipeline":
         return class_obj
 
     diffusers_module = importlib.import_module(class_obj.__module__.split(".")[0])
@@ -557,6 +613,9 @@ def _assign_components_to_devices(
 
 
 def _get_final_device_map(device_map, pipeline_class, passed_class_obj, init_dict, library, max_memory, **kwargs):
+    # TODO: seperate out different device_map methods when it gets to it.
+    if device_map != "balanced":
+        return device_map
     # To avoid circular import problem.
     from diffusers import pipelines
 
@@ -675,8 +734,10 @@ def load_sub_model(
     use_safetensors: bool,
     dduf_entries: Optional[Dict[str, DDUFEntry]],
     provider_options: Any,
+    quantization_config: Optional[Any] = None,
 ):
     """Helper method to load the module `name` from `library_name` and `class_name`"""
+    from ..quantizers import PipelineQuantizationConfig
 
     # retrieve class candidates
 
@@ -769,6 +830,17 @@ def load_sub_model(
         else:
             loading_kwargs["low_cpu_mem_usage"] = False
 
+    if (
+        quantization_config is not None
+        and isinstance(quantization_config, PipelineQuantizationConfig)
+        and issubclass(class_obj, torch.nn.Module)
+    ):
+        model_quant_config = quantization_config._resolve_quant_config(
+            is_diffusers=is_diffusers_model, module_name=name
+        )
+        if model_quant_config is not None:
+            loading_kwargs["quantization_config"] = model_quant_config
+
     # check if the module is in a subdirectory
     if dduf_entries:
         loading_kwargs["dduf_entries"] = dduf_entries
@@ -839,7 +911,10 @@ def _fetch_class_library_tuple(module):
         library = not_compiled_module.__module__
 
     # retrieve class_name
-    class_name = not_compiled_module.__class__.__name__
+    if isinstance(not_compiled_module, type):
+        class_name = not_compiled_module.__name__
+    else:
+        class_name = not_compiled_module.__class__.__name__
 
     return (library, class_name)
 
@@ -984,7 +1059,7 @@ def _get_ignore_patterns(
         use_safetensors
         and not allow_pickle
         and not is_safetensors_compatible(
-            model_filenames, passed_components=passed_components, folder_names=model_folder_names
+            model_filenames, passed_components=passed_components, folder_names=model_folder_names, variant=variant
         )
     ):
         raise EnvironmentError(
@@ -995,7 +1070,7 @@ def _get_ignore_patterns(
         ignore_patterns = ["*.bin", "*.safetensors", "*.onnx", "*.pb"]
 
     elif use_safetensors and is_safetensors_compatible(
-        model_filenames, passed_components=passed_components, folder_names=model_folder_names
+        model_filenames, passed_components=passed_components, folder_names=model_folder_names, variant=variant
     ):
         ignore_patterns = ["*.bin", "*.msgpack"]
 
@@ -1078,3 +1153,26 @@ def _maybe_raise_error_for_incorrect_transformers(config_dict):
                 break
     if has_transformers_component and not is_transformers_version(">", "4.47.1"):
         raise ValueError("Please upgrade your `transformers` installation to the latest version to use DDUF.")
+
+
+def _maybe_warn_for_wrong_component_in_quant_config(pipe_init_dict, quant_config):
+    if quant_config is None:
+        return
+
+    actual_pipe_components = set(pipe_init_dict.keys())
+    missing = ""
+    quant_components = None
+    if getattr(quant_config, "components_to_quantize", None) is not None:
+        quant_components = set(quant_config.components_to_quantize)
+    elif getattr(quant_config, "quant_mapping", None) is not None and isinstance(quant_config.quant_mapping, dict):
+        quant_components = set(quant_config.quant_mapping.keys())
+
+    if quant_components and not quant_components.issubset(actual_pipe_components):
+        missing = quant_components - actual_pipe_components
+
+    if missing:
+        logger.warning(
+            f"The following components in the quantization config {missing} will be ignored "
+            "as they do not belong to the underlying pipeline. Acceptable values for the pipeline "
+            f"components are: {', '.join(actual_pipe_components)}."
+        )