diffusers 0.27.2__py3-none-any.whl → 0.28.0__py3-none-any.whl

diffusers/pipelines/deepfloyd_if/pipeline_if.py

@@ -12,7 +12,6 @@ from ...models import UNet2DConditionModel
 from ...schedulers import DDPMScheduler
 from ...utils import (
     BACKENDS_MAPPING,
-    is_accelerate_available,
     is_bs4_available,
     is_ftfy_available,
     logging,
@@ -115,6 +114,7 @@ class IFPipeline(DiffusionPipeline, LoraLoaderMixin):
 
     _optional_components = ["tokenizer", "text_encoder", "safety_checker", "feature_extractor", "watermarker"]
     model_cpu_offload_seq = "text_encoder->unet"
+    _exclude_from_cpu_offload = ["watermarker"]
 
     def __init__(
         self,
@@ -156,20 +156,6 @@ class IFPipeline(DiffusionPipeline, LoraLoaderMixin):
         )
         self.register_to_config(requires_safety_checker=requires_safety_checker)
 
-    def remove_all_hooks(self):
-        if is_accelerate_available():
-            from accelerate.hooks import remove_hook_from_module
-        else:
-            raise ImportError("Please install accelerate via `pip install accelerate`")
-
-        for model in [self.text_encoder, self.unet, self.safety_checker]:
-            if model is not None:
-                remove_hook_from_module(model, recurse=True)
-
-        self.unet_offload_hook = None
-        self.text_encoder_offload_hook = None
-        self.final_offload_hook = None
-
     @torch.no_grad()
     def encode_prompt(
         self,
@@ -178,8 +164,8 @@ class IFPipeline(DiffusionPipeline, LoraLoaderMixin):
         num_images_per_prompt: int = 1,
         device: Optional[torch.device] = None,
         negative_prompt: Optional[Union[str, List[str]]] = None,
-        prompt_embeds: Optional[torch.FloatTensor] = None,
-        negative_prompt_embeds: Optional[torch.FloatTensor] = None,
+        prompt_embeds: Optional[torch.Tensor] = None,
+        negative_prompt_embeds: Optional[torch.Tensor] = None,
         clean_caption: bool = False,
     ):
         r"""
@@ -198,10 +184,10 @@ class IFPipeline(DiffusionPipeline, LoraLoaderMixin):
                 The prompt or prompts not to guide the image generation. If not defined, one has to pass
                 `negative_prompt_embeds`. instead. If not defined, one has to pass `negative_prompt_embeds`. instead.
                 Ignored when not using guidance (i.e., ignored if `guidance_scale` is less than `1`).
-            prompt_embeds (`torch.FloatTensor`, *optional*):
+            prompt_embeds (`torch.Tensor`, *optional*):
                 Pre-generated text embeddings. Can be used to easily tweak text inputs, *e.g.* prompt weighting. If not
                 provided, text embeddings will be generated from `prompt` input argument.
-            negative_prompt_embeds (`torch.FloatTensor`, *optional*):
+            negative_prompt_embeds (`torch.Tensor`, *optional*):
                 Pre-generated negative text embeddings. Can be used to easily tweak text inputs, *e.g.* prompt
                 weighting. If not provided, negative_prompt_embeds will be generated from `negative_prompt` input
                 argument.
@@ -335,9 +321,6 @@ class IFPipeline(DiffusionPipeline, LoraLoaderMixin):
             nsfw_detected = None
             watermark_detected = None
 
-            if hasattr(self, "unet_offload_hook") and self.unet_offload_hook is not None:
-                self.unet_offload_hook.offload()
-
         return image, nsfw_detected, watermark_detected
 
     # Copied from diffusers.pipelines.stable_diffusion.pipeline_stable_diffusion.StableDiffusionPipeline.prepare_extra_step_kwargs
@@ -566,11 +549,11 @@ class IFPipeline(DiffusionPipeline, LoraLoaderMixin):
         width: Optional[int] = None,
         eta: float = 0.0,
         generator: Optional[Union[torch.Generator, List[torch.Generator]]] = None,
-        prompt_embeds: Optional[torch.FloatTensor] = None,
-        negative_prompt_embeds: Optional[torch.FloatTensor] = None,
+        prompt_embeds: Optional[torch.Tensor] = None,
+        negative_prompt_embeds: 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,
         clean_caption: bool = True,
         cross_attention_kwargs: Optional[Dict[str, Any]] = None,
@@ -610,10 +593,10 @@ class IFPipeline(DiffusionPipeline, LoraLoaderMixin):
             generator (`torch.Generator` or `List[torch.Generator]`, *optional*):
                 One or a list of [torch generator(s)](https://pytorch.org/docs/stable/generated/torch.Generator.html)
                 to make generation deterministic.
-            prompt_embeds (`torch.FloatTensor`, *optional*):
+            prompt_embeds (`torch.Tensor`, *optional*):
                 Pre-generated text embeddings. Can be used to easily tweak text inputs, *e.g.* prompt weighting. If not
                 provided, text embeddings will be generated from `prompt` input argument.
-            negative_prompt_embeds (`torch.FloatTensor`, *optional*):
+            negative_prompt_embeds (`torch.Tensor`, *optional*):
                 Pre-generated negative text embeddings. Can be used to easily tweak text inputs, *e.g.* prompt
                 weighting. If not provided, negative_prompt_embeds will be generated from `negative_prompt` input
                 argument.
@@ -624,7 +607,7 @@ class IFPipeline(DiffusionPipeline, LoraLoaderMixin):
                 Whether or not to return a [`~pipelines.stable_diffusion.IFPipelineOutput`] instead of a plain tuple.
             callback (`Callable`, *optional*):
                 A function that will be called every `callback_steps` steps during inference. The function will be
-                called with the following arguments: `callback(step: int, timestep: int, latents: torch.FloatTensor)`.
+                called with the following arguments: `callback(step: int, timestep: int, latents: torch.Tensor)`.
             callback_steps (`int`, *optional*, defaults to 1):
                 The frequency at which the `callback` function will be called. If not specified, the callback will be
                 called at every step.
@@ -691,6 +674,9 @@ class IFPipeline(DiffusionPipeline, LoraLoaderMixin):
             self.scheduler.set_timesteps(num_inference_steps, device=device)
             timesteps = self.scheduler.timesteps
 
+        if hasattr(self.scheduler, "set_begin_index"):
+            self.scheduler.set_begin_index(0)
+
         # 5. Prepare intermediate images
         intermediate_images = self.prepare_intermediate_images(
             batch_size * num_images_per_prompt,

diffusers/pipelines/deepfloyd_if/pipeline_if_img2img.py

@@ -15,7 +15,6 @@ from ...schedulers import DDPMScheduler
 from ...utils import (
     BACKENDS_MAPPING,
     PIL_INTERPOLATION,
-    is_accelerate_available,
     is_bs4_available,
     is_ftfy_available,
     logging,
@@ -139,6 +138,7 @@ class IFImg2ImgPipeline(DiffusionPipeline, LoraLoaderMixin):
 
     _optional_components = ["tokenizer", "text_encoder", "safety_checker", "feature_extractor", "watermarker"]
     model_cpu_offload_seq = "text_encoder->unet"
+    _exclude_from_cpu_offload = ["watermarker"]
 
     def __init__(
         self,
@@ -180,21 +180,6 @@ class IFImg2ImgPipeline(DiffusionPipeline, LoraLoaderMixin):
         )
         self.register_to_config(requires_safety_checker=requires_safety_checker)
 
-    # Copied from diffusers.pipelines.deepfloyd_if.pipeline_if.IFPipeline.remove_all_hooks
-    def remove_all_hooks(self):
-        if is_accelerate_available():
-            from accelerate.hooks import remove_hook_from_module
-        else:
-            raise ImportError("Please install accelerate via `pip install accelerate`")
-
-        for model in [self.text_encoder, self.unet, self.safety_checker]:
-            if model is not None:
-                remove_hook_from_module(model, recurse=True)
-
-        self.unet_offload_hook = None
-        self.text_encoder_offload_hook = None
-        self.final_offload_hook = None
-
     @torch.no_grad()
     def encode_prompt(
         self,
@@ -203,8 +188,8 @@ class IFImg2ImgPipeline(DiffusionPipeline, LoraLoaderMixin):
         num_images_per_prompt: int = 1,
         device: Optional[torch.device] = None,
         negative_prompt: Optional[Union[str, List[str]]] = None,
-        prompt_embeds: Optional[torch.FloatTensor] = None,
-        negative_prompt_embeds: Optional[torch.FloatTensor] = None,
+        prompt_embeds: Optional[torch.Tensor] = None,
+        negative_prompt_embeds: Optional[torch.Tensor] = None,
         clean_caption: bool = False,
     ):
         r"""
@@ -223,10 +208,10 @@ class IFImg2ImgPipeline(DiffusionPipeline, LoraLoaderMixin):
                 The prompt or prompts not to guide the image generation. If not defined, one has to pass
                 `negative_prompt_embeds`. instead. If not defined, one has to pass `negative_prompt_embeds`. instead.
                 Ignored when not using guidance (i.e., ignored if `guidance_scale` is less than `1`).
-            prompt_embeds (`torch.FloatTensor`, *optional*):
+            prompt_embeds (`torch.Tensor`, *optional*):
                 Pre-generated text embeddings. Can be used to easily tweak text inputs, *e.g.* prompt weighting. If not
                 provided, text embeddings will be generated from `prompt` input argument.
-            negative_prompt_embeds (`torch.FloatTensor`, *optional*):
+            negative_prompt_embeds (`torch.Tensor`, *optional*):
                 Pre-generated negative text embeddings. Can be used to easily tweak text inputs, *e.g.* prompt
                 weighting. If not provided, negative_prompt_embeds will be generated from `negative_prompt` input
                 argument.
@@ -361,9 +346,6 @@ class IFImg2ImgPipeline(DiffusionPipeline, LoraLoaderMixin):
             nsfw_detected = None
             watermark_detected = None
 
-            if hasattr(self, "unet_offload_hook") and self.unet_offload_hook is not None:
-                self.unet_offload_hook.offload()
-
         return image, nsfw_detected, watermark_detected
 
     # Copied from diffusers.pipelines.deepfloyd_if.pipeline_if.IFPipeline.prepare_extra_step_kwargs
@@ -439,7 +421,7 @@ class IFImg2ImgPipeline(DiffusionPipeline, LoraLoaderMixin):
             and not isinstance(check_image_type, np.ndarray)
         ):
             raise ValueError(
-                "`image` has to be of type `torch.FloatTensor`, `PIL.Image.Image`, `np.ndarray`, or List[...] but is"
+                "`image` has to be of type `torch.Tensor`, `PIL.Image.Image`, `np.ndarray`, or List[...] but is"
                 f" {type(check_image_type)}"
             )
 
@@ -613,7 +595,7 @@ class IFImg2ImgPipeline(DiffusionPipeline, LoraLoaderMixin):
 
             for image_ in image:
                 image_ = image_.convert("RGB")
-                image_ = resize(image_, self.unet.sample_size)
+                image_ = resize(image_, self.unet.config.sample_size)
                 image_ = np.array(image_)
                 image_ = image_.astype(np.float32)
                 image_ = image_ / 127.5 - 1
@@ -633,12 +615,15 @@ class IFImg2ImgPipeline(DiffusionPipeline, LoraLoaderMixin):
 
         return image
 
+    # Copied from diffusers.pipelines.stable_diffusion.pipeline_stable_diffusion_img2img.StableDiffusionImg2ImgPipeline.get_timesteps
     def get_timesteps(self, num_inference_steps, strength):
         # get the original timestep using init_timestep
         init_timestep = min(int(num_inference_steps * strength), num_inference_steps)
 
         t_start = max(num_inference_steps - init_timestep, 0)
-        timesteps = self.scheduler.timesteps[t_start:]
+        timesteps = self.scheduler.timesteps[t_start * self.scheduler.order :]
+        if hasattr(self.scheduler, "set_begin_index"):
+            self.scheduler.set_begin_index(t_start * self.scheduler.order)
 
         return timesteps, num_inference_steps - t_start
 
@@ -680,11 +665,11 @@ class IFImg2ImgPipeline(DiffusionPipeline, LoraLoaderMixin):
         num_images_per_prompt: Optional[int] = 1,
         eta: float = 0.0,
         generator: Optional[Union[torch.Generator, List[torch.Generator]]] = None,
-        prompt_embeds: Optional[torch.FloatTensor] = None,
-        negative_prompt_embeds: Optional[torch.FloatTensor] = None,
+        prompt_embeds: Optional[torch.Tensor] = None,
+        negative_prompt_embeds: 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,
         clean_caption: bool = True,
         cross_attention_kwargs: Optional[Dict[str, Any]] = None,
@@ -696,7 +681,7 @@ class IFImg2ImgPipeline(DiffusionPipeline, LoraLoaderMixin):
             prompt (`str` or `List[str]`, *optional*):
                 The prompt or prompts to guide the image generation. If not defined, one has to pass `prompt_embeds`.
                 instead.
-            image (`torch.FloatTensor` or `PIL.Image.Image`):
+            image (`torch.Tensor` or `PIL.Image.Image`):
                 `Image`, or tensor representing an image batch, that will be used as the starting point for the
                 process.
             strength (`float`, *optional*, defaults to 0.7):
@@ -729,10 +714,10 @@ class IFImg2ImgPipeline(DiffusionPipeline, LoraLoaderMixin):
             generator (`torch.Generator` or `List[torch.Generator]`, *optional*):
                 One or a list of [torch generator(s)](https://pytorch.org/docs/stable/generated/torch.Generator.html)
                 to make generation deterministic.
-            prompt_embeds (`torch.FloatTensor`, *optional*):
+            prompt_embeds (`torch.Tensor`, *optional*):
                 Pre-generated text embeddings. Can be used to easily tweak text inputs, *e.g.* prompt weighting. If not
                 provided, text embeddings will be generated from `prompt` input argument.
-            negative_prompt_embeds (`torch.FloatTensor`, *optional*):
+            negative_prompt_embeds (`torch.Tensor`, *optional*):
                 Pre-generated negative text embeddings. Can be used to easily tweak text inputs, *e.g.* prompt
                 weighting. If not provided, negative_prompt_embeds will be generated from `negative_prompt` input
                 argument.
@@ -743,7 +728,7 @@ class IFImg2ImgPipeline(DiffusionPipeline, LoraLoaderMixin):
                 Whether or not to return a [`~pipelines.stable_diffusion.IFPipelineOutput`] instead of a plain tuple.
             callback (`Callable`, *optional*):
                 A function that will be called every `callback_steps` steps during inference. The function will be
-                called with the following arguments: `callback(step: int, timestep: int, latents: torch.FloatTensor)`.
+                called with the following arguments: `callback(step: int, timestep: int, latents: torch.Tensor)`.
             callback_steps (`int`, *optional*, defaults to 1):
                 The frequency at which the `callback` function will be called. If not specified, the callback will be
                 called at every step.

diffusers/pipelines/deepfloyd_if/pipeline_if_img2img_superresolution.py

@@ -16,7 +16,6 @@ from ...schedulers import DDPMScheduler
 from ...utils import (
     BACKENDS_MAPPING,
     PIL_INTERPOLATION,
-    is_accelerate_available,
     is_bs4_available,
     is_ftfy_available,
     logging,
@@ -143,6 +142,7 @@ class IFImg2ImgSuperResolutionPipeline(DiffusionPipeline, LoraLoaderMixin):
 
     _optional_components = ["tokenizer", "text_encoder", "safety_checker", "feature_extractor"]
     model_cpu_offload_seq = "text_encoder->unet"
+    _exclude_from_cpu_offload = ["watermarker"]
 
     def __init__(
         self,
@@ -191,21 +191,6 @@ class IFImg2ImgSuperResolutionPipeline(DiffusionPipeline, LoraLoaderMixin):
         )
         self.register_to_config(requires_safety_checker=requires_safety_checker)
 
-    # Copied from diffusers.pipelines.deepfloyd_if.pipeline_if.IFPipeline.remove_all_hooks
-    def remove_all_hooks(self):
-        if is_accelerate_available():
-            from accelerate.hooks import remove_hook_from_module
-        else:
-            raise ImportError("Please install accelerate via `pip install accelerate`")
-
-        for model in [self.text_encoder, self.unet, self.safety_checker]:
-            if model is not None:
-                remove_hook_from_module(model, recurse=True)
-
-        self.unet_offload_hook = None
-        self.text_encoder_offload_hook = None
-        self.final_offload_hook = None
-
     # Copied from diffusers.pipelines.deepfloyd_if.pipeline_if.IFPipeline._text_preprocessing
     def _text_preprocessing(self, text, clean_caption=False):
         if clean_caption and not is_bs4_available():
@@ -355,8 +340,8 @@ class IFImg2ImgSuperResolutionPipeline(DiffusionPipeline, LoraLoaderMixin):
         num_images_per_prompt: int = 1,
         device: Optional[torch.device] = None,
         negative_prompt: Optional[Union[str, List[str]]] = None,
-        prompt_embeds: Optional[torch.FloatTensor] = None,
-        negative_prompt_embeds: Optional[torch.FloatTensor] = None,
+        prompt_embeds: Optional[torch.Tensor] = None,
+        negative_prompt_embeds: Optional[torch.Tensor] = None,
         clean_caption: bool = False,
     ):
         r"""
@@ -375,10 +360,10 @@ class IFImg2ImgSuperResolutionPipeline(DiffusionPipeline, LoraLoaderMixin):
                 The prompt or prompts not to guide the image generation. If not defined, one has to pass
                 `negative_prompt_embeds`. instead. If not defined, one has to pass `negative_prompt_embeds`. instead.
                 Ignored when not using guidance (i.e., ignored if `guidance_scale` is less than `1`).
-            prompt_embeds (`torch.FloatTensor`, *optional*):
+            prompt_embeds (`torch.Tensor`, *optional*):
                 Pre-generated text embeddings. Can be used to easily tweak text inputs, *e.g.* prompt weighting. If not
                 provided, text embeddings will be generated from `prompt` input argument.
-            negative_prompt_embeds (`torch.FloatTensor`, *optional*):
+            negative_prompt_embeds (`torch.Tensor`, *optional*):
                 Pre-generated negative text embeddings. Can be used to easily tweak text inputs, *e.g.* prompt
                 weighting. If not provided, negative_prompt_embeds will be generated from `negative_prompt` input
                 argument.
@@ -513,9 +498,6 @@ class IFImg2ImgSuperResolutionPipeline(DiffusionPipeline, LoraLoaderMixin):
             nsfw_detected = None
             watermark_detected = None
 
-            if hasattr(self, "unet_offload_hook") and self.unet_offload_hook is not None:
-                self.unet_offload_hook.offload()
-
         return image, nsfw_detected, watermark_detected
 
     # Copied from diffusers.pipelines.deepfloyd_if.pipeline_if.IFPipeline.prepare_extra_step_kwargs
@@ -594,7 +576,7 @@ class IFImg2ImgSuperResolutionPipeline(DiffusionPipeline, LoraLoaderMixin):
             and not isinstance(check_image_type, np.ndarray)
         ):
             raise ValueError(
-                "`image` has to be of type `torch.FloatTensor`, `PIL.Image.Image`, `np.ndarray`, or List[...] but is"
+                "`image` has to be of type `torch.Tensor`, `PIL.Image.Image`, `np.ndarray`, or List[...] but is"
                 f" {type(check_image_type)}"
             )
 
@@ -625,7 +607,7 @@ class IFImg2ImgSuperResolutionPipeline(DiffusionPipeline, LoraLoaderMixin):
             and not isinstance(check_image_type, np.ndarray)
         ):
             raise ValueError(
-                "`original_image` has to be of type `torch.FloatTensor`, `PIL.Image.Image`, `np.ndarray`, or List[...] but is"
+                "`original_image` has to be of type `torch.Tensor`, `PIL.Image.Image`, `np.ndarray`, or List[...] but is"
                 f" {type(check_image_type)}"
             )
 
@@ -662,7 +644,7 @@ class IFImg2ImgSuperResolutionPipeline(DiffusionPipeline, LoraLoaderMixin):
 
             for image_ in image:
                 image_ = image_.convert("RGB")
-                image_ = resize(image_, self.unet.sample_size)
+                image_ = resize(image_, self.unet.config.sample_size)
                 image_ = np.array(image_)
                 image_ = image_.astype(np.float32)
                 image_ = image_ / 127.5 - 1
@@ -714,13 +696,15 @@ class IFImg2ImgSuperResolutionPipeline(DiffusionPipeline, LoraLoaderMixin):
 
         return image
 
-    # Copied from diffusers.pipelines.deepfloyd_if.pipeline_if_img2img.IFImg2ImgPipeline.get_timesteps
+    # Copied from diffusers.pipelines.stable_diffusion.pipeline_stable_diffusion_img2img.StableDiffusionImg2ImgPipeline.get_timesteps
     def get_timesteps(self, num_inference_steps, strength):
         # get the original timestep using init_timestep
         init_timestep = min(int(num_inference_steps * strength), num_inference_steps)
 
         t_start = max(num_inference_steps - init_timestep, 0)
-        timesteps = self.scheduler.timesteps[t_start:]
+        timesteps = self.scheduler.timesteps[t_start * self.scheduler.order :]
+        if hasattr(self.scheduler, "set_begin_index"):
+            self.scheduler.set_begin_index(t_start * self.scheduler.order)
 
         return timesteps, num_inference_steps - t_start
 
@@ -751,7 +735,7 @@ class IFImg2ImgSuperResolutionPipeline(DiffusionPipeline, LoraLoaderMixin):
     @replace_example_docstring(EXAMPLE_DOC_STRING)
     def __call__(
         self,
-        image: Union[PIL.Image.Image, np.ndarray, torch.FloatTensor],
+        image: Union[PIL.Image.Image, np.ndarray, torch.Tensor],
         original_image: Union[
             PIL.Image.Image, torch.Tensor, np.ndarray, List[PIL.Image.Image], List[torch.Tensor], List[np.ndarray]
         ] = None,
@@ -764,11 +748,11 @@ class IFImg2ImgSuperResolutionPipeline(DiffusionPipeline, LoraLoaderMixin):
         num_images_per_prompt: Optional[int] = 1,
         eta: float = 0.0,
         generator: Optional[Union[torch.Generator, List[torch.Generator]]] = None,
-        prompt_embeds: Optional[torch.FloatTensor] = None,
-        negative_prompt_embeds: Optional[torch.FloatTensor] = None,
+        prompt_embeds: Optional[torch.Tensor] = None,
+        negative_prompt_embeds: 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,
         cross_attention_kwargs: Optional[Dict[str, Any]] = None,
         noise_level: int = 250,
@@ -778,10 +762,10 @@ class IFImg2ImgSuperResolutionPipeline(DiffusionPipeline, LoraLoaderMixin):
         Function invoked when calling the pipeline for generation.
 
         Args:
-            image (`torch.FloatTensor` or `PIL.Image.Image`):
+            image (`torch.Tensor` or `PIL.Image.Image`):
                 `Image`, or tensor representing an image batch, that will be used as the starting point for the
                 process.
-            original_image (`torch.FloatTensor` or `PIL.Image.Image`):
+            original_image (`torch.Tensor` or `PIL.Image.Image`):
                 The original image that `image` was varied from.
             strength (`float`, *optional*, defaults to 0.8):
                 Conceptually, indicates how much to transform the reference `image`. Must be between 0 and 1. `image`
@@ -816,10 +800,10 @@ class IFImg2ImgSuperResolutionPipeline(DiffusionPipeline, LoraLoaderMixin):
             generator (`torch.Generator` or `List[torch.Generator]`, *optional*):
                 One or a list of [torch generator(s)](https://pytorch.org/docs/stable/generated/torch.Generator.html)
                 to make generation deterministic.
-            prompt_embeds (`torch.FloatTensor`, *optional*):
+            prompt_embeds (`torch.Tensor`, *optional*):
                 Pre-generated text embeddings. Can be used to easily tweak text inputs, *e.g.* prompt weighting. If not
                 provided, text embeddings will be generated from `prompt` input argument.
-            negative_prompt_embeds (`torch.FloatTensor`, *optional*):
+            negative_prompt_embeds (`torch.Tensor`, *optional*):
                 Pre-generated negative text embeddings. Can be used to easily tweak text inputs, *e.g.* prompt
                 weighting. If not provided, negative_prompt_embeds will be generated from `negative_prompt` input
                 argument.
@@ -830,7 +814,7 @@ class IFImg2ImgSuperResolutionPipeline(DiffusionPipeline, LoraLoaderMixin):
                 Whether or not to return a [`~pipelines.stable_diffusion.IFPipelineOutput`] instead of a plain tuple.
             callback (`Callable`, *optional*):
                 A function that will be called every `callback_steps` steps during inference. The function will be
-                called with the following arguments: `callback(step: int, timestep: int, latents: torch.FloatTensor)`.
+                called with the following arguments: `callback(step: int, timestep: int, latents: torch.Tensor)`.
             callback_steps (`int`, *optional*, defaults to 1):
                 The frequency at which the `callback` function will be called. If not specified, the callback will be
                 called at every step.
@@ -1010,8 +994,6 @@ class IFImg2ImgSuperResolutionPipeline(DiffusionPipeline, LoraLoaderMixin):
             nsfw_detected = None
             watermark_detected = None
 
-            if hasattr(self, "unet_offload_hook") and self.unet_offload_hook is not None:
-                self.unet_offload_hook.offload()
         else:
             # 10. Post-processing
             image = (image / 2 + 0.5).clamp(0, 1)

diffusers/pipelines/deepfloyd_if/pipeline_if_inpainting.py

@@ -15,7 +15,6 @@ from ...schedulers import DDPMScheduler
 from ...utils import (
     BACKENDS_MAPPING,
     PIL_INTERPOLATION,
-    is_accelerate_available,
     is_bs4_available,
     is_ftfy_available,
     logging,
@@ -142,6 +141,7 @@ class IFInpaintingPipeline(DiffusionPipeline, LoraLoaderMixin):
 
     _optional_components = ["tokenizer", "text_encoder", "safety_checker", "feature_extractor", "watermarker"]
     model_cpu_offload_seq = "text_encoder->unet"
+    _exclude_from_cpu_offload = ["watermarker"]
 
     def __init__(
         self,
@@ -183,21 +183,6 @@ class IFInpaintingPipeline(DiffusionPipeline, LoraLoaderMixin):
         )
         self.register_to_config(requires_safety_checker=requires_safety_checker)
 
-    # Copied from diffusers.pipelines.deepfloyd_if.pipeline_if.IFPipeline.remove_all_hooks
-    def remove_all_hooks(self):
-        if is_accelerate_available():
-            from accelerate.hooks import remove_hook_from_module
-        else:
-            raise ImportError("Please install accelerate via `pip install accelerate`")
-
-        for model in [self.text_encoder, self.unet, self.safety_checker]:
-            if model is not None:
-                remove_hook_from_module(model, recurse=True)
-
-        self.unet_offload_hook = None
-        self.text_encoder_offload_hook = None
-        self.final_offload_hook = None
-
     @torch.no_grad()
     # Copied from diffusers.pipelines.deepfloyd_if.pipeline_if.IFPipeline.encode_prompt
     def encode_prompt(
@@ -207,8 +192,8 @@ class IFInpaintingPipeline(DiffusionPipeline, LoraLoaderMixin):
         num_images_per_prompt: int = 1,
         device: Optional[torch.device] = None,
         negative_prompt: Optional[Union[str, List[str]]] = None,
-        prompt_embeds: Optional[torch.FloatTensor] = None,
-        negative_prompt_embeds: Optional[torch.FloatTensor] = None,
+        prompt_embeds: Optional[torch.Tensor] = None,
+        negative_prompt_embeds: Optional[torch.Tensor] = None,
         clean_caption: bool = False,
     ):
         r"""
@@ -227,10 +212,10 @@ class IFInpaintingPipeline(DiffusionPipeline, LoraLoaderMixin):
                 The prompt or prompts not to guide the image generation. If not defined, one has to pass
                 `negative_prompt_embeds`. instead. If not defined, one has to pass `negative_prompt_embeds`. instead.
                 Ignored when not using guidance (i.e., ignored if `guidance_scale` is less than `1`).
-            prompt_embeds (`torch.FloatTensor`, *optional*):
+            prompt_embeds (`torch.Tensor`, *optional*):
                 Pre-generated text embeddings. Can be used to easily tweak text inputs, *e.g.* prompt weighting. If not
                 provided, text embeddings will be generated from `prompt` input argument.
-            negative_prompt_embeds (`torch.FloatTensor`, *optional*):
+            negative_prompt_embeds (`torch.Tensor`, *optional*):
                 Pre-generated negative text embeddings. Can be used to easily tweak text inputs, *e.g.* prompt
                 weighting. If not provided, negative_prompt_embeds will be generated from `negative_prompt` input
                 argument.
@@ -365,9 +350,6 @@ class IFInpaintingPipeline(DiffusionPipeline, LoraLoaderMixin):
             nsfw_detected = None
             watermark_detected = None
 
-            if hasattr(self, "unet_offload_hook") and self.unet_offload_hook is not None:
-                self.unet_offload_hook.offload()
-
         return image, nsfw_detected, watermark_detected
 
     # Copied from diffusers.pipelines.deepfloyd_if.pipeline_if.IFPipeline.prepare_extra_step_kwargs
@@ -446,7 +428,7 @@ class IFInpaintingPipeline(DiffusionPipeline, LoraLoaderMixin):
             and not isinstance(check_image_type, np.ndarray)
         ):
             raise ValueError(
-                "`image` has to be of type `torch.FloatTensor`, `PIL.Image.Image`, `np.ndarray`, or List[...] but is"
+                "`image` has to be of type `torch.Tensor`, `PIL.Image.Image`, `np.ndarray`, or List[...] but is"
                 f" {type(check_image_type)}"
             )
 
@@ -477,7 +459,7 @@ class IFInpaintingPipeline(DiffusionPipeline, LoraLoaderMixin):
             and not isinstance(check_image_type, np.ndarray)
         ):
             raise ValueError(
-                "`mask_image` has to be of type `torch.FloatTensor`, `PIL.Image.Image`, `np.ndarray`, or List[...] but is"
+                "`mask_image` has to be of type `torch.Tensor`, `PIL.Image.Image`, `np.ndarray`, or List[...] but is"
                 f" {type(check_image_type)}"
             )
 
@@ -654,7 +636,7 @@ class IFInpaintingPipeline(DiffusionPipeline, LoraLoaderMixin):
 
             for image_ in image:
                 image_ = image_.convert("RGB")
-                image_ = resize(image_, self.unet.sample_size)
+                image_ = resize(image_, self.unet.config.sample_size)
                 image_ = np.array(image_)
                 image_ = image_.astype(np.float32)
                 image_ = image_ / 127.5 - 1
@@ -701,7 +683,7 @@ class IFInpaintingPipeline(DiffusionPipeline, LoraLoaderMixin):
 
             for mask_image_ in mask_image:
                 mask_image_ = mask_image_.convert("L")
-                mask_image_ = resize(mask_image_, self.unet.sample_size)
+                mask_image_ = resize(mask_image_, self.unet.config.sample_size)
                 mask_image_ = np.array(mask_image_)
                 mask_image_ = mask_image_[None, None, :]
                 new_mask_image.append(mask_image_)
@@ -723,13 +705,15 @@ class IFInpaintingPipeline(DiffusionPipeline, LoraLoaderMixin):
 
         return mask_image
 
-    # Copied from diffusers.pipelines.deepfloyd_if.pipeline_if_img2img.IFImg2ImgPipeline.get_timesteps
+    # Copied from diffusers.pipelines.stable_diffusion.pipeline_stable_diffusion_img2img.StableDiffusionImg2ImgPipeline.get_timesteps
     def get_timesteps(self, num_inference_steps, strength):
         # get the original timestep using init_timestep
         init_timestep = min(int(num_inference_steps * strength), num_inference_steps)
 
         t_start = max(num_inference_steps - init_timestep, 0)
-        timesteps = self.scheduler.timesteps[t_start:]
+        timesteps = self.scheduler.timesteps[t_start * self.scheduler.order :]
+        if hasattr(self.scheduler, "set_begin_index"):
+            self.scheduler.set_begin_index(t_start * self.scheduler.order)
 
         return timesteps, num_inference_steps - t_start
 
@@ -776,11 +760,11 @@ class IFInpaintingPipeline(DiffusionPipeline, LoraLoaderMixin):
         num_images_per_prompt: Optional[int] = 1,
         eta: float = 0.0,
         generator: Optional[Union[torch.Generator, List[torch.Generator]]] = None,
-        prompt_embeds: Optional[torch.FloatTensor] = None,
-        negative_prompt_embeds: Optional[torch.FloatTensor] = None,
+        prompt_embeds: Optional[torch.Tensor] = None,
+        negative_prompt_embeds: 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,
         clean_caption: bool = True,
         cross_attention_kwargs: Optional[Dict[str, Any]] = None,
@@ -792,7 +776,7 @@ class IFInpaintingPipeline(DiffusionPipeline, LoraLoaderMixin):
             prompt (`str` or `List[str]`, *optional*):
                 The prompt or prompts to guide the image generation. If not defined, one has to pass `prompt_embeds`.
                 instead.
-            image (`torch.FloatTensor` or `PIL.Image.Image`):
+            image (`torch.Tensor` or `PIL.Image.Image`):
                 `Image`, or tensor representing an image batch, that will be used as the starting point for the
                 process.
             mask_image (`PIL.Image.Image`):
@@ -830,10 +814,10 @@ class IFInpaintingPipeline(DiffusionPipeline, LoraLoaderMixin):
             generator (`torch.Generator` or `List[torch.Generator]`, *optional*):
                 One or a list of [torch generator(s)](https://pytorch.org/docs/stable/generated/torch.Generator.html)
                 to make generation deterministic.
-            prompt_embeds (`torch.FloatTensor`, *optional*):
+            prompt_embeds (`torch.Tensor`, *optional*):
                 Pre-generated text embeddings. Can be used to easily tweak text inputs, *e.g.* prompt weighting. If not
                 provided, text embeddings will be generated from `prompt` input argument.
-            negative_prompt_embeds (`torch.FloatTensor`, *optional*):
+            negative_prompt_embeds (`torch.Tensor`, *optional*):
                 Pre-generated negative text embeddings. Can be used to easily tweak text inputs, *e.g.* prompt
                 weighting. If not provided, negative_prompt_embeds will be generated from `negative_prompt` input
                 argument.
@@ -844,7 +828,7 @@ class IFInpaintingPipeline(DiffusionPipeline, LoraLoaderMixin):
                 Whether or not to return a [`~pipelines.stable_diffusion.IFPipelineOutput`] instead of a plain tuple.
             callback (`Callable`, *optional*):
                 A function that will be called every `callback_steps` steps during inference. The function will be
-                called with the following arguments: `callback(step: int, timestep: int, latents: torch.FloatTensor)`.
+                called with the following arguments: `callback(step: int, timestep: int, latents: torch.Tensor)`.
             callback_steps (`int`, *optional*, defaults to 1):
                 The frequency at which the `callback` function will be called. If not specified, the callback will be
                 called at every step.