diffusers 0.27.2__py3-none-any.whl → 0.28.1__py3-none-any.whl

diffusers/pipelines/deprecated/versatile_diffusion/pipeline_versatile_diffusion_image_variation.py

@@ -197,7 +197,7 @@ class VersatileDiffusionImageVariationPipeline(DiffusionPipeline):
             and not isinstance(image, list)
         ):
             raise ValueError(
-                "`image` has to be of type `torch.FloatTensor` or `PIL.Image.Image` or `List[PIL.Image.Image]` but is"
+                "`image` has to be of type `torch.Tensor` or `PIL.Image.Image` or `List[PIL.Image.Image]` but is"
                 f" {type(image)}"
             )
 
@@ -214,7 +214,12 @@ class VersatileDiffusionImageVariationPipeline(DiffusionPipeline):
 
     # Copied from diffusers.pipelines.stable_diffusion.pipeline_stable_diffusion.StableDiffusionPipeline.prepare_latents
     def prepare_latents(self, batch_size, num_channels_latents, height, width, dtype, device, generator, latents=None):
-        shape = (batch_size, num_channels_latents, height // self.vae_scale_factor, width // self.vae_scale_factor)
+        shape = (
+            batch_size,
+            num_channels_latents,
+            int(height) // self.vae_scale_factor,
+            int(width) // self.vae_scale_factor,
+        )
         if isinstance(generator, list) and len(generator) != batch_size:
             raise ValueError(
                 f"You have passed a list of generators of length {len(generator)}, but requested an effective batch"
@@ -242,10 +247,10 @@ class VersatileDiffusionImageVariationPipeline(DiffusionPipeline):
         num_images_per_prompt: Optional[int] = 1,
         eta: float = 0.0,
         generator: Optional[Union[torch.Generator, List[torch.Generator]]] = None,
-        latents: Optional[torch.FloatTensor] = None,
+        latents: Optional[torch.Tensor] = None,
         output_type: Optional[str] = "pil",
         return_dict: bool = True,
-        callback: Optional[Callable[[int, int, torch.FloatTensor], None]] = None,
+        callback: Optional[Callable[[int, int, torch.Tensor], None]] = None,
         callback_steps: int = 1,
         **kwargs,
     ):
@@ -276,7 +281,7 @@ class VersatileDiffusionImageVariationPipeline(DiffusionPipeline):
             generator (`torch.Generator`, *optional*):
                 A [`torch.Generator`](https://pytorch.org/docs/stable/generated/torch.Generator.html) to make
                 generation deterministic.
-            latents (`torch.FloatTensor`, *optional*):
+            latents (`torch.Tensor`, *optional*):
                 Pre-generated noisy latents sampled from a Gaussian distribution, to be used as inputs for image
                 generation. Can be used to tweak the same generation with different prompts. If not provided, a latents
                 tensor is generated by sampling using the supplied random `generator`.
@@ -287,7 +292,7 @@ class VersatileDiffusionImageVariationPipeline(DiffusionPipeline):
                 plain tuple.
             callback (`Callable`, *optional*):
                 A function that calls every `callback_steps` steps during inference. The function is called with the
-                following arguments: `callback(step: int, timestep: int, latents: torch.FloatTensor)`.
+                following arguments: `callback(step: int, timestep: int, latents: torch.Tensor)`.
             callback_steps (`int`, *optional*, defaults to 1):
                 The frequency at which the `callback` function is called. If not specified, the callback is called at
                 every step.

diffusers/pipelines/deprecated/versatile_diffusion/pipeline_versatile_diffusion_text_to_image.py

@@ -300,7 +300,12 @@ class VersatileDiffusionTextToImagePipeline(DiffusionPipeline):
 
     # Copied from diffusers.pipelines.stable_diffusion.pipeline_stable_diffusion.StableDiffusionPipeline.prepare_latents
     def prepare_latents(self, batch_size, num_channels_latents, height, width, dtype, device, generator, latents=None):
-        shape = (batch_size, num_channels_latents, height // self.vae_scale_factor, width // self.vae_scale_factor)
+        shape = (
+            batch_size,
+            num_channels_latents,
+            int(height) // self.vae_scale_factor,
+            int(width) // self.vae_scale_factor,
+        )
         if isinstance(generator, list) and len(generator) != batch_size:
             raise ValueError(
                 f"You have passed a list of generators of length {len(generator)}, but requested an effective batch"
@@ -328,10 +333,10 @@ class VersatileDiffusionTextToImagePipeline(DiffusionPipeline):
         num_images_per_prompt: Optional[int] = 1,
         eta: float = 0.0,
         generator: Optional[Union[torch.Generator, List[torch.Generator]]] = None,
-        latents: Optional[torch.FloatTensor] = None,
+        latents: Optional[torch.Tensor] = None,
         output_type: Optional[str] = "pil",
         return_dict: bool = True,
-        callback: Optional[Callable[[int, int, torch.FloatTensor], None]] = None,
+        callback: Optional[Callable[[int, int, torch.Tensor], None]] = None,
         callback_steps: int = 1,
         **kwargs,
     ):
@@ -362,7 +367,7 @@ class VersatileDiffusionTextToImagePipeline(DiffusionPipeline):
             generator (`torch.Generator`, *optional*):
                 A [`torch.Generator`](https://pytorch.org/docs/stable/generated/torch.Generator.html) to make
                 generation deterministic.
-            latents (`torch.FloatTensor`, *optional*):
+            latents (`torch.Tensor`, *optional*):
                 Pre-generated noisy latents sampled from a Gaussian distribution, to be used as inputs for image
                 generation. Can be used to tweak the same generation with different prompts. If not provided, a latents
                 tensor is generated by sampling using the supplied random `generator`.
@@ -373,7 +378,7 @@ class VersatileDiffusionTextToImagePipeline(DiffusionPipeline):
                 plain tuple.
             callback (`Callable`, *optional*):
                 A function that calls every `callback_steps` steps during inference. The function is called with the
-                following arguments: `callback(step: int, timestep: int, latents: torch.FloatTensor)`.
+                following arguments: `callback(step: int, timestep: int, latents: torch.Tensor)`.
             callback_steps (`int`, *optional*, defaults to 1):
                 The frequency at which the `callback` function is called. If not specified, the callback is called at
                 every step.

diffusers/pipelines/deprecated/vq_diffusion/pipeline_vq_diffusion.py

@@ -169,10 +169,10 @@ class VQDiffusionPipeline(DiffusionPipeline):
         truncation_rate: float = 1.0,
         num_images_per_prompt: int = 1,
         generator: Optional[Union[torch.Generator, List[torch.Generator]]] = None,
-        latents: Optional[torch.FloatTensor] = None,
+        latents: Optional[torch.Tensor] = None,
         output_type: Optional[str] = "pil",
         return_dict: bool = True,
-        callback: Optional[Callable[[int, int, torch.FloatTensor], None]] = None,
+        callback: Optional[Callable[[int, int, torch.Tensor], None]] = None,
         callback_steps: int = 1,
     ) -> Union[ImagePipelineOutput, Tuple]:
         """
@@ -196,7 +196,7 @@ class VQDiffusionPipeline(DiffusionPipeline):
             generator (`torch.Generator`, *optional*):
                 A [`torch.Generator`](https://pytorch.org/docs/stable/generated/torch.Generator.html) to make
                 generation deterministic.
-            latents (`torch.FloatTensor` of shape (batch), *optional*):
+            latents (`torch.Tensor` of shape (batch), *optional*):
                 Pre-generated noisy latents sampled from a Gaussian distribution, to be used as inputs for image
                 generation. Must be valid embedding indices.If not provided, a latents tensor will be generated of
                 completely masked latent pixels.
@@ -206,7 +206,7 @@ class VQDiffusionPipeline(DiffusionPipeline):
                 Whether or not to return a [`~pipelines.ImagePipelineOutput`] instead of a plain tuple.
             callback (`Callable`, *optional*):
                 A function that calls every `callback_steps` steps during inference. The function is called with the
-                following arguments: `callback(step: int, timestep: int, latents: torch.FloatTensor)`.
+                following arguments: `callback(step: int, timestep: int, latents: torch.Tensor)`.
             callback_steps (`int`, *optional*, defaults to 1):
                 The frequency at which the `callback` function is called. If not specified, the callback is called at
                 every step.
@@ -301,7 +301,7 @@ class VQDiffusionPipeline(DiffusionPipeline):
 
         return ImagePipelineOutput(images=image)
 
-    def truncate(self, log_p_x_0: torch.FloatTensor, truncation_rate: float) -> torch.FloatTensor:
+    def truncate(self, log_p_x_0: torch.Tensor, truncation_rate: float) -> torch.Tensor:
         """
         Truncates `log_p_x_0` such that for each column vector, the total cumulative probability is `truncation_rate`
         The lowest probabilities that would increase the cumulative probability above `truncation_rate` are set to

diffusers/pipelines/dit/pipeline_dit.py

@@ -22,7 +22,7 @@ from typing import Dict, List, Optional, Tuple, Union
 
 import torch
 
-from ...models import AutoencoderKL, Transformer2DModel
+from ...models import AutoencoderKL, DiTTransformer2DModel
 from ...schedulers import KarrasDiffusionSchedulers
 from ...utils.torch_utils import randn_tensor
 from ..pipeline_utils import DiffusionPipeline, ImagePipelineOutput
@@ -36,8 +36,8 @@ class DiTPipeline(DiffusionPipeline):
     implemented for all pipelines (downloading, saving, running on a particular device, etc.).
 
     Parameters:
-        transformer ([`Transformer2DModel`]):
-            A class conditioned `Transformer2DModel` to denoise the encoded image latents.
+        transformer ([`DiTTransformer2DModel`]):
+            A class conditioned `DiTTransformer2DModel` to denoise the encoded image latents.
         vae ([`AutoencoderKL`]):
             Variational Auto-Encoder (VAE) model to encode and decode images to and from latent representations.
         scheduler ([`DDIMScheduler`]):
@@ -48,7 +48,7 @@ class DiTPipeline(DiffusionPipeline):
 
     def __init__(
         self,
-        transformer: Transformer2DModel,
+        transformer: DiTTransformer2DModel,
         vae: AutoencoderKL,
         scheduler: KarrasDiffusionSchedulers,
         id2label: Optional[Dict[int, str]] = None,
@@ -227,6 +227,9 @@ class DiTPipeline(DiffusionPipeline):
         if output_type == "pil":
             samples = self.numpy_to_pil(samples)
 
+        # Offload all models
+        self.maybe_free_model_hooks()
+
         if not return_dict:
             return (samples,)
 

diffusers/pipelines/free_init_utils.py

@@ -41,20 +41,20 @@ class FreeInitMixin:
             num_iters (`int`, *optional*, defaults to `3`):
                 Number of FreeInit noise re-initialization iterations.
             use_fast_sampling (`bool`, *optional*, defaults to `False`):
-                Whether or not to speedup sampling procedure at the cost of probably lower quality results. Enables
-                the "Coarse-to-Fine Sampling" strategy, as mentioned in the paper, if set to `True`.
+                Whether or not to speedup sampling procedure at the cost of probably lower quality results. Enables the
+                "Coarse-to-Fine Sampling" strategy, as mentioned in the paper, if set to `True`.
             method (`str`, *optional*, defaults to `butterworth`):
-                Must be one of `butterworth`, `ideal` or `gaussian` to use as the filtering method for the
-                FreeInit low pass filter.
+                Must be one of `butterworth`, `ideal` or `gaussian` to use as the filtering method for the FreeInit low
+                pass filter.
             order (`int`, *optional*, defaults to `4`):
                 Order of the filter used in `butterworth` method. Larger values lead to `ideal` method behaviour
                 whereas lower values lead to `gaussian` method behaviour.
             spatial_stop_frequency (`float`, *optional*, defaults to `0.25`):
-                Normalized stop frequency for spatial dimensions. Must be between 0 to 1. Referred to as `d_s` in
-                the original implementation.
+                Normalized stop frequency for spatial dimensions. Must be between 0 to 1. Referred to as `d_s` in the
+                original implementation.
             temporal_stop_frequency (`float`, *optional*, defaults to `0.25`):
-                Normalized stop frequency for temporal dimensions. Must be between 0 to 1. Referred to as `d_t` in
-                the original implementation.
+                Normalized stop frequency for temporal dimensions. Must be between 0 to 1. Referred to as `d_t` in the
+                original implementation.
         """
         self._free_init_num_iters = num_iters
         self._free_init_use_fast_sampling = use_fast_sampling
@@ -146,39 +146,40 @@ class FreeInitMixin:
     ):
         if free_init_iteration == 0:
             self._free_init_initial_noise = latents.detach().clone()
-            return latents, self.scheduler.timesteps
-
-        latent_shape = latents.shape
-
-        free_init_filter_shape = (1, *latent_shape[1:])
-        free_init_freq_filter = self._get_free_init_freq_filter(
-            shape=free_init_filter_shape,
-            device=device,
-            filter_type=self._free_init_method,
-            order=self._free_init_order,
-            spatial_stop_frequency=self._free_init_spatial_stop_frequency,
-            temporal_stop_frequency=self._free_init_temporal_stop_frequency,
-        )
-
-        current_diffuse_timestep = self.scheduler.config.num_train_timesteps - 1
-        diffuse_timesteps = torch.full((latent_shape[0],), current_diffuse_timestep).long()
-
-        z_t = self.scheduler.add_noise(
-            original_samples=latents, noise=self._free_init_initial_noise, timesteps=diffuse_timesteps.to(device)
-        ).to(dtype=torch.float32)
-
-        z_rand = randn_tensor(
-            shape=latent_shape,
-            generator=generator,
-            device=device,
-            dtype=torch.float32,
-        )
-        latents = self._apply_freq_filter(z_t, z_rand, low_pass_filter=free_init_freq_filter)
-        latents = latents.to(dtype)
+        else:
+            latent_shape = latents.shape
+
+            free_init_filter_shape = (1, *latent_shape[1:])
+            free_init_freq_filter = self._get_free_init_freq_filter(
+                shape=free_init_filter_shape,
+                device=device,
+                filter_type=self._free_init_method,
+                order=self._free_init_order,
+                spatial_stop_frequency=self._free_init_spatial_stop_frequency,
+                temporal_stop_frequency=self._free_init_temporal_stop_frequency,
+            )
+
+            current_diffuse_timestep = self.scheduler.config.num_train_timesteps - 1
+            diffuse_timesteps = torch.full((latent_shape[0],), current_diffuse_timestep).long()
+
+            z_t = self.scheduler.add_noise(
+                original_samples=latents, noise=self._free_init_initial_noise, timesteps=diffuse_timesteps.to(device)
+            ).to(dtype=torch.float32)
+
+            z_rand = randn_tensor(
+                shape=latent_shape,
+                generator=generator,
+                device=device,
+                dtype=torch.float32,
+            )
+            latents = self._apply_freq_filter(z_t, z_rand, low_pass_filter=free_init_freq_filter)
+            latents = latents.to(dtype)
 
         # Coarse-to-Fine Sampling for faster inference (can lead to lower quality)
         if self._free_init_use_fast_sampling:
-            num_inference_steps = int(num_inference_steps / self._free_init_num_iters * (free_init_iteration + 1))
+            num_inference_steps = max(
+                1, int(num_inference_steps / self._free_init_num_iters * (free_init_iteration + 1))
+            )
             self.scheduler.set_timesteps(num_inference_steps, device=device)
 
         return latents, self.scheduler.timesteps

diffusers/pipelines/hunyuandit/__init__.py

@@ -0,0 +1,48 @@
+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_hunyuandit"] = ["HunyuanDiTPipeline"]
+
+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_hunyuandit import HunyuanDiTPipeline
+
+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)