diffusers 0.31.0__py3-none-any.whl → 0.32.1__py3-none-any.whl

diffusers/pipelines/pipeline_flax_utils.py

@@ -180,7 +180,7 @@ class FlaxDiffusionPipeline(ConfigMixin, PushToHubMixin):
 
         if push_to_hub:
             commit_message = kwargs.pop("commit_message", None)
-            private = kwargs.pop("private", False)
+            private = kwargs.pop("private", None)
             create_pr = kwargs.pop("create_pr", False)
             token = kwargs.pop("token", None)
             repo_id = kwargs.pop("repo_id", save_directory.split(os.path.sep)[-1])

diffusers/pipelines/pipeline_loading_utils.py

@@ -198,10 +198,31 @@ def variant_compatible_siblings(filenames, variant=None) -> Union[List[os.PathLi
             variant_filename = f"{filename.split('.')[0]}.{variant}.{filename.split('.')[1]}"
         return variant_filename
 
-    for f in non_variant_filenames:
-        variant_filename = convert_to_variant(f)
-        if variant_filename not in usable_filenames:
-            usable_filenames.add(f)
+    def find_component(filename):
+        if not len(filename.split("/")) == 2:
+            return
+        component = filename.split("/")[0]
+        return component
+
+    def has_sharded_variant(component, variant, variant_filenames):
+        # If component exists check for sharded variant index filename
+        # If component doesn't exist check main dir for sharded variant index filename
+        component = component + "/" if component else ""
+        variant_index_re = re.compile(
+            rf"{component}({'|'.join(weight_prefixes)})\.({'|'.join(weight_suffixs)})\.index\.{variant}\.json$"
+        )
+        return any(f for f in variant_filenames if variant_index_re.match(f) is not None)
+
+    for filename in non_variant_filenames:
+        if convert_to_variant(filename) in variant_filenames:
+            continue
+
+        component = find_component(filename)
+        # If a sharded variant exists skip adding to allowed patterns
+        if has_sharded_variant(component, variant, variant_filenames):
+            continue
+
+        usable_filenames.add(filename)
 
     return usable_filenames, variant_filenames
 

diffusers/pipelines/pipeline_utils.py

@@ -13,6 +13,7 @@
 # WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
 # See the License for the specific language governing permissions and
 # limitations under the License.
+import enum
 import fnmatch
 import importlib
 import inspect
@@ -66,7 +67,6 @@ from ..utils.torch_utils import is_compiled_module
 if is_torch_npu_available():
     import torch_npu  # noqa: F401
 
-
 from .pipeline_loading_utils import (
     ALL_IMPORTABLE_CLASSES,
     CONNECTED_PIPES_KEYS,
@@ -229,7 +229,7 @@ class DiffusionPipeline(ConfigMixin, PushToHubMixin):
 
         if push_to_hub:
             commit_message = kwargs.pop("commit_message", None)
-            private = kwargs.pop("private", False)
+            private = kwargs.pop("private", None)
             create_pr = kwargs.pop("create_pr", False)
             token = kwargs.pop("token", None)
             repo_id = kwargs.pop("repo_id", save_directory.split(os.path.sep)[-1])
@@ -388,6 +388,7 @@ class DiffusionPipeline(ConfigMixin, PushToHubMixin):
             )
 
         device = device or device_arg
+        pipeline_has_bnb = any(any((_check_bnb_status(module))) for _, module in self.components.items())
 
         # throw warning if pipeline is in "offloaded"-mode but user tries to manually set to GPU.
         def module_is_sequentially_offloaded(module):
@@ -410,10 +411,16 @@ class DiffusionPipeline(ConfigMixin, PushToHubMixin):
         pipeline_is_sequentially_offloaded = any(
             module_is_sequentially_offloaded(module) for _, module in self.components.items()
         )
-        if pipeline_is_sequentially_offloaded and device and torch.device(device).type == "cuda":
-            raise ValueError(
-                "It seems like you have activated sequential model offloading by calling `enable_sequential_cpu_offload`, but are now attempting to move the pipeline to GPU. This is not compatible with offloading. Please, move your pipeline `.to('cpu')` or consider removing the move altogether if you use sequential offloading."
-            )
+        if device and torch.device(device).type == "cuda":
+            if pipeline_is_sequentially_offloaded and not pipeline_has_bnb:
+                raise ValueError(
+                    "It seems like you have activated sequential model offloading by calling `enable_sequential_cpu_offload`, but are now attempting to move the pipeline to GPU. This is not compatible with offloading. Please, move your pipeline `.to('cpu')` or consider removing the move altogether if you use sequential offloading."
+                )
+            # PR: https://github.com/huggingface/accelerate/pull/3223/
+            elif pipeline_has_bnb and is_accelerate_version("<", "1.1.0.dev0"):
+                raise ValueError(
+                    "You are trying to call `.to('cuda')` on a pipeline that has models quantized with `bitsandbytes`. Your current `accelerate` installation does not support it. Please upgrade the installation."
+                )
 
         is_pipeline_device_mapped = self.hf_device_map is not None and len(self.hf_device_map) > 1
         if is_pipeline_device_mapped:
@@ -805,6 +812,7 @@ class DiffusionPipeline(ConfigMixin, PushToHubMixin):
         # in this case they are already instantiated in `kwargs`
         # extract them here
         expected_modules, optional_kwargs = cls._get_signature_keys(pipeline_class)
+        expected_types = pipeline_class._get_signature_types()
         passed_class_obj = {k: kwargs.pop(k) for k in expected_modules if k in kwargs}
         passed_pipe_kwargs = {k: kwargs.pop(k) for k in optional_kwargs if k in kwargs}
         init_dict, unused_kwargs, _ = pipeline_class.extract_init_dict(config_dict, **kwargs)
@@ -827,6 +835,26 @@ class DiffusionPipeline(ConfigMixin, PushToHubMixin):
 
         init_dict = {k: v for k, v in init_dict.items() if load_module(k, v)}
 
+        for key in init_dict.keys():
+            if key not in passed_class_obj:
+                continue
+            if "scheduler" in key:
+                continue
+
+            class_obj = passed_class_obj[key]
+            _expected_class_types = []
+            for expected_type in expected_types[key]:
+                if isinstance(expected_type, enum.EnumMeta):
+                    _expected_class_types.extend(expected_type.__members__.keys())
+                else:
+                    _expected_class_types.append(expected_type.__name__)
+
+            _is_valid_type = class_obj.__class__.__name__ in _expected_class_types
+            if not _is_valid_type:
+                logger.warning(
+                    f"Expected types for {key}: {_expected_class_types}, got {class_obj.__class__.__name__}."
+                )
+
         # Special case: safety_checker must be loaded separately when using `from_flax`
         if from_flax and "safety_checker" in init_dict and "safety_checker" not in passed_class_obj:
             raise NotImplementedError(
@@ -1552,6 +1580,7 @@ class DiffusionPipeline(ConfigMixin, PushToHubMixin):
         """
         return numpy_to_pil(images)
 
+    @torch.compiler.disable
     def progress_bar(self, iterable=None, total=None):
         if not hasattr(self, "_progress_bar_config"):
             self._progress_bar_config = {}

diffusers/pipelines/pixart_alpha/pipeline_pixart_alpha.py

@@ -338,13 +338,6 @@ class PixArtAlphaPipeline(DiffusionPipeline):
         if device is None:
             device = self._execution_device
 
-        if prompt is not None and isinstance(prompt, str):
-            batch_size = 1
-        elif prompt is not None and isinstance(prompt, list):
-            batch_size = len(prompt)
-        else:
-            batch_size = prompt_embeds.shape[0]
-
         # See Section 3.1. of the paper.
         max_length = max_sequence_length
 
@@ -389,12 +382,12 @@ class PixArtAlphaPipeline(DiffusionPipeline):
         # duplicate text embeddings and attention mask for each generation per prompt, using mps friendly method
         prompt_embeds = prompt_embeds.repeat(1, num_images_per_prompt, 1)
         prompt_embeds = prompt_embeds.view(bs_embed * num_images_per_prompt, seq_len, -1)
-        prompt_attention_mask = prompt_attention_mask.view(bs_embed, -1)
-        prompt_attention_mask = prompt_attention_mask.repeat(num_images_per_prompt, 1)
+        prompt_attention_mask = prompt_attention_mask.repeat(1, num_images_per_prompt)
+        prompt_attention_mask = prompt_attention_mask.view(bs_embed * num_images_per_prompt, -1)
 
         # get unconditional embeddings for classifier free guidance
         if do_classifier_free_guidance and negative_prompt_embeds is None:
-            uncond_tokens = [negative_prompt] * batch_size if isinstance(negative_prompt, str) else negative_prompt
+            uncond_tokens = [negative_prompt] * bs_embed if isinstance(negative_prompt, str) else negative_prompt
             uncond_tokens = self._text_preprocessing(uncond_tokens, clean_caption=clean_caption)
             max_length = prompt_embeds.shape[1]
             uncond_input = self.tokenizer(
@@ -421,10 +414,10 @@ class PixArtAlphaPipeline(DiffusionPipeline):
             negative_prompt_embeds = negative_prompt_embeds.to(dtype=dtype, device=device)
 
             negative_prompt_embeds = negative_prompt_embeds.repeat(1, num_images_per_prompt, 1)
-            negative_prompt_embeds = negative_prompt_embeds.view(batch_size * num_images_per_prompt, seq_len, -1)
+            negative_prompt_embeds = negative_prompt_embeds.view(bs_embed * num_images_per_prompt, seq_len, -1)
 
-            negative_prompt_attention_mask = negative_prompt_attention_mask.view(bs_embed, -1)
-            negative_prompt_attention_mask = negative_prompt_attention_mask.repeat(num_images_per_prompt, 1)
+            negative_prompt_attention_mask = negative_prompt_attention_mask.repeat(1, num_images_per_prompt)
+            negative_prompt_attention_mask = negative_prompt_attention_mask.view(bs_embed * num_images_per_prompt, -1)
         else:
             negative_prompt_embeds = None
             negative_prompt_attention_mask = None

diffusers/pipelines/pixart_alpha/pipeline_pixart_sigma.py

@@ -264,13 +264,6 @@ class PixArtSigmaPipeline(DiffusionPipeline):
         if device is None:
             device = self._execution_device
 
-        if prompt is not None and isinstance(prompt, str):
-            batch_size = 1
-        elif prompt is not None and isinstance(prompt, list):
-            batch_size = len(prompt)
-        else:
-            batch_size = prompt_embeds.shape[0]
-
         # See Section 3.1. of the paper.
         max_length = max_sequence_length
 
@@ -315,12 +308,12 @@ class PixArtSigmaPipeline(DiffusionPipeline):
         # duplicate text embeddings and attention mask for each generation per prompt, using mps friendly method
         prompt_embeds = prompt_embeds.repeat(1, num_images_per_prompt, 1)
         prompt_embeds = prompt_embeds.view(bs_embed * num_images_per_prompt, seq_len, -1)
-        prompt_attention_mask = prompt_attention_mask.view(bs_embed, -1)
-        prompt_attention_mask = prompt_attention_mask.repeat(num_images_per_prompt, 1)
+        prompt_attention_mask = prompt_attention_mask.repeat(1, num_images_per_prompt)
+        prompt_attention_mask = prompt_attention_mask.view(bs_embed * num_images_per_prompt, -1)
 
         # get unconditional embeddings for classifier free guidance
         if do_classifier_free_guidance and negative_prompt_embeds is None:
-            uncond_tokens = [negative_prompt] * batch_size if isinstance(negative_prompt, str) else negative_prompt
+            uncond_tokens = [negative_prompt] * bs_embed if isinstance(negative_prompt, str) else negative_prompt
             uncond_tokens = self._text_preprocessing(uncond_tokens, clean_caption=clean_caption)
             max_length = prompt_embeds.shape[1]
             uncond_input = self.tokenizer(
@@ -347,10 +340,10 @@ class PixArtSigmaPipeline(DiffusionPipeline):
             negative_prompt_embeds = negative_prompt_embeds.to(dtype=dtype, device=device)
 
             negative_prompt_embeds = negative_prompt_embeds.repeat(1, num_images_per_prompt, 1)
-            negative_prompt_embeds = negative_prompt_embeds.view(batch_size * num_images_per_prompt, seq_len, -1)
+            negative_prompt_embeds = negative_prompt_embeds.view(bs_embed * num_images_per_prompt, seq_len, -1)
 
-            negative_prompt_attention_mask = negative_prompt_attention_mask.view(bs_embed, -1)
-            negative_prompt_attention_mask = negative_prompt_attention_mask.repeat(num_images_per_prompt, 1)
+            negative_prompt_attention_mask = negative_prompt_attention_mask.repeat(1, num_images_per_prompt)
+            negative_prompt_attention_mask = negative_prompt_attention_mask.view(bs_embed * num_images_per_prompt, -1)
         else:
             negative_prompt_embeds = None
             negative_prompt_attention_mask = None

diffusers/pipelines/sana/__init__.py

@@ -0,0 +1,47 @@
+from typing import TYPE_CHECKING
+
+from ...utils import (
+    DIFFUSERS_SLOW_IMPORT,
+    OptionalDependencyNotAvailable,
+    _LazyModule,
+    get_objects_from_module,
+    is_torch_available,
+    is_transformers_available,
+)
+
+
+_dummy_objects = {}
+_import_structure = {}
+
+
+try:
+    if not (is_transformers_available() and is_torch_available()):
+        raise OptionalDependencyNotAvailable()
+except OptionalDependencyNotAvailable:
+    from ...utils import dummy_torch_and_transformers_objects  # noqa F403
+
+    _dummy_objects.update(get_objects_from_module(dummy_torch_and_transformers_objects))
+else:
+    _import_structure["pipeline_sana"] = ["SanaPipeline"]
+
+if TYPE_CHECKING or DIFFUSERS_SLOW_IMPORT:
+    try:
+        if not (is_transformers_available() and is_torch_available()):
+            raise OptionalDependencyNotAvailable()
+
+    except OptionalDependencyNotAvailable:
+        from ...utils.dummy_torch_and_transformers_objects import *
+    else:
+        from .pipeline_sana import SanaPipeline
+else:
+    import sys
+
+    sys.modules[__name__] = _LazyModule(
+        __name__,
+        globals()["__file__"],
+        _import_structure,
+        module_spec=__spec__,
+    )
+
+    for name, value in _dummy_objects.items():
+        setattr(sys.modules[__name__], name, value)

diffusers/pipelines/sana/pipeline_output.py

@@ -0,0 +1,21 @@
+from dataclasses import dataclass
+from typing import List, Union
+
+import numpy as np
+import PIL.Image
+
+from ...utils import BaseOutput
+
+
+@dataclass
+class SanaPipelineOutput(BaseOutput):
+    """
+    Output class for Sana pipelines.
+
+    Args:
+        images (`List[PIL.Image.Image]` or `np.ndarray`)
+            List of denoised PIL images of length `batch_size` or numpy array of shape `(batch_size, height, width,
+            num_channels)`. PIL images or numpy array present the denoised images of the diffusion pipeline.
+    """
+
+    images: Union[List[PIL.Image.Image], np.ndarray]