#7535 Update FloatTensor type hints to Tensor (#7883)

* find & replace all FloatTensors to Tensor

* apply formatting

* Update torch.FloatTensor to torch.Tensor in the remaining files

* formatting

* Fix the rest of the places where FloatTensor is used as well as in documentation

* formatting

* Update new file from FloatTensor to Tensor

examples/community/stable_diffusion_xl_reference.py

@@ -190,7 +190,7 @@ class StableDiffusionXLReferencePipeline(StableDiffusionXLPipeline):
         self,
         prompt: Union[str, List[str]] = None,
         prompt_2: Optional[Union[str, List[str]]] = None,
-        ref_image: Union[torch.FloatTensor, PIL.Image.Image] = None,
+        ref_image: Union[torch.Tensor, PIL.Image.Image] = None,
         height: Optional[int] = None,
         width: Optional[int] = None,
         num_inference_steps: int = 50,
@@ -201,14 +201,14 @@ class StableDiffusionXLReferencePipeline(StableDiffusionXLPipeline):
         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,
-        prompt_embeds: Optional[torch.FloatTensor] = None,
-        negative_prompt_embeds: Optional[torch.FloatTensor] = None,
-        pooled_prompt_embeds: Optional[torch.FloatTensor] = None,
-        negative_pooled_prompt_embeds: Optional[torch.FloatTensor] = None,
+        latents: Optional[torch.Tensor] = None,
+        prompt_embeds: Optional[torch.Tensor] = None,
+        negative_prompt_embeds: Optional[torch.Tensor] = None,
+        pooled_prompt_embeds: Optional[torch.Tensor] = None,
+        negative_pooled_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,
         guidance_rescale: float = 0.0,
@@ -335,10 +335,10 @@ class StableDiffusionXLReferencePipeline(StableDiffusionXLPipeline):
 
         def hacked_basic_transformer_inner_forward(
             self,
-            hidden_states: torch.FloatTensor,
-            attention_mask: Optional[torch.FloatTensor] = None,
-            encoder_hidden_states: Optional[torch.FloatTensor] = None,
-            encoder_attention_mask: Optional[torch.FloatTensor] = None,
+            hidden_states: torch.Tensor,
+            attention_mask: Optional[torch.Tensor] = None,
+            encoder_hidden_states: Optional[torch.Tensor] = None,
+            encoder_attention_mask: Optional[torch.Tensor] = None,
             timestep: Optional[torch.LongTensor] = None,
             cross_attention_kwargs: Dict[str, Any] = None,
             class_labels: Optional[torch.LongTensor] = None,
@@ -453,12 +453,12 @@ class StableDiffusionXLReferencePipeline(StableDiffusionXLPipeline):
 
         def hack_CrossAttnDownBlock2D_forward(
             self,
-            hidden_states: torch.FloatTensor,
-            temb: Optional[torch.FloatTensor] = None,
-            encoder_hidden_states: Optional[torch.FloatTensor] = None,
-            attention_mask: Optional[torch.FloatTensor] = None,
+            hidden_states: torch.Tensor,
+            temb: Optional[torch.Tensor] = None,
+            encoder_hidden_states: Optional[torch.Tensor] = None,
+            attention_mask: Optional[torch.Tensor] = None,
             cross_attention_kwargs: Optional[Dict[str, Any]] = None,
-            encoder_attention_mask: Optional[torch.FloatTensor] = None,
+            encoder_attention_mask: Optional[torch.Tensor] = None,
         ):
             eps = 1e-6
 
@@ -549,14 +549,14 @@ class StableDiffusionXLReferencePipeline(StableDiffusionXLPipeline):
 
         def hacked_CrossAttnUpBlock2D_forward(
             self,
-            hidden_states: torch.FloatTensor,
-            res_hidden_states_tuple: Tuple[torch.FloatTensor, ...],
-            temb: Optional[torch.FloatTensor] = None,
-            encoder_hidden_states: Optional[torch.FloatTensor] = None,
+            hidden_states: torch.Tensor,
+            res_hidden_states_tuple: Tuple[torch.Tensor, ...],
+            temb: Optional[torch.Tensor] = None,
+            encoder_hidden_states: Optional[torch.Tensor] = None,
             cross_attention_kwargs: Optional[Dict[str, Any]] = None,
             upsample_size: Optional[int] = None,
-            attention_mask: Optional[torch.FloatTensor] = None,
-            encoder_attention_mask: Optional[torch.FloatTensor] = None,
+            attention_mask: Optional[torch.Tensor] = None,
+            encoder_attention_mask: Optional[torch.Tensor] = None,
         ):
             eps = 1e-6
             # TODO(Patrick, William) - attention mask is not used

examples/community/stable_unclip.py

@@ -191,7 +191,7 @@ class StableUnCLIPPipeline(DiffusionPipeline):
         num_images_per_prompt: int = 1,
         prior_num_inference_steps: int = 25,
         generator: Optional[torch.Generator] = None,
-        prior_latents: Optional[torch.FloatTensor] = None,
+        prior_latents: Optional[torch.Tensor] = None,
         text_model_output: Optional[Union[CLIPTextModelOutput, Tuple]] = None,
         text_attention_mask: Optional[torch.Tensor] = None,
         prior_guidance_scale: float = 4.0,

examples/community/text_inpainting.py

@@ -125,7 +125,7 @@ class TextInpainting(DiffusionPipeline, StableDiffusionMixin):
     def __call__(
         self,
         prompt: Union[str, List[str]],
-        image: Union[torch.FloatTensor, PIL.Image.Image],
+        image: Union[torch.Tensor, PIL.Image.Image],
         text: str,
         height: int = 512,
         width: int = 512,
@@ -135,10 +135,10 @@ class TextInpainting(DiffusionPipeline, StableDiffusionMixin):
         num_images_per_prompt: Optional[int] = 1,
         eta: float = 0.0,
         generator: Optional[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,
     ):
@@ -177,7 +177,7 @@ class TextInpainting(DiffusionPipeline, StableDiffusionMixin):
             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 will ge generated by sampling using the supplied random `generator`.
@@ -189,7 +189,7 @@ class TextInpainting(DiffusionPipeline, StableDiffusionMixin):
                 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.

examples/community/tiled_upscaling.py

@@ -193,8 +193,8 @@ class StableDiffusionTiledUpscalePipeline(StableDiffusionUpscalePipeline):
         num_images_per_prompt: Optional[int] = 1,
         eta: float = 0.0,
         generator: Optional[torch.Generator] = None,
-        latents: Optional[torch.FloatTensor] = None,
-        callback: Optional[Callable[[int, int, torch.FloatTensor], None]] = None,
+        latents: Optional[torch.Tensor] = None,
+        callback: Optional[Callable[[int, int, torch.Tensor], None]] = None,
         callback_steps: int = 1,
         tile_size: int = 128,
         tile_border: int = 32,
@@ -206,7 +206,7 @@ class StableDiffusionTiledUpscalePipeline(StableDiffusionUpscalePipeline):
         Args:
             prompt (`str` or `List[str]`):
                 The prompt or prompts to guide the image generation.
-            image (`PIL.Image.Image` or List[`PIL.Image.Image`] or `torch.FloatTensor`):
+            image (`PIL.Image.Image` or List[`PIL.Image.Image`] or `torch.Tensor`):
                 `Image`, or tensor representing an image batch which will be upscaled. *
             num_inference_steps (`int`, *optional*, defaults to 50):
                 The number of denoising steps. More denoising steps usually lead to a higher quality image at the
@@ -228,7 +228,7 @@ class StableDiffusionTiledUpscalePipeline(StableDiffusionUpscalePipeline):
             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 will ge generated by sampling using the supplied random `generator`.

examples/community/unclip_image_interpolation.py

@@ -207,14 +207,14 @@ class UnCLIPImageInterpolationPipeline(DiffusionPipeline):
     @torch.no_grad()
     def __call__(
         self,
-        image: Optional[Union[List[PIL.Image.Image], torch.FloatTensor]] = None,
+        image: Optional[Union[List[PIL.Image.Image], torch.Tensor]] = None,
         steps: int = 5,
         decoder_num_inference_steps: int = 25,
         super_res_num_inference_steps: int = 7,
         generator: Optional[Union[torch.Generator, List[torch.Generator]]] = None,
         image_embeddings: Optional[torch.Tensor] = None,
-        decoder_latents: Optional[torch.FloatTensor] = None,
-        super_res_latents: Optional[torch.FloatTensor] = None,
+        decoder_latents: Optional[torch.Tensor] = None,
+        super_res_latents: Optional[torch.Tensor] = None,
         decoder_guidance_scale: float = 8.0,
         output_type: Optional[str] = "pil",
         return_dict: bool = True,
@@ -223,7 +223,7 @@ class UnCLIPImageInterpolationPipeline(DiffusionPipeline):
         Function invoked when calling the pipeline for generation.
 
         Args:
-            image (`List[PIL.Image.Image]` or `torch.FloatTensor`):
+            image (`List[PIL.Image.Image]` or `torch.Tensor`):
                 The images to use for the image interpolation. Only accepts a list of two PIL Images or If you provide a tensor, it needs to comply with the
                 configuration of
                 [this](https://huggingface.co/fusing/karlo-image-variations-diffusers/blob/main/feature_extractor/preprocessor_config.json)
@@ -242,9 +242,9 @@ class UnCLIPImageInterpolationPipeline(DiffusionPipeline):
             image_embeddings (`torch.Tensor`, *optional*):
                 Pre-defined image embeddings that can be derived from the image encoder. Pre-defined image embeddings
                 can be passed for tasks like image interpolations. `image` can the be left to `None`.
-            decoder_latents (`torch.FloatTensor` of shape (batch size, channels, height, width), *optional*):
+            decoder_latents (`torch.Tensor` of shape (batch size, channels, height, width), *optional*):
                 Pre-generated noisy latents to be used as inputs for the decoder.
-            super_res_latents (`torch.FloatTensor` of shape (batch size, channels, super res height, super res width), *optional*):
+            super_res_latents (`torch.Tensor` of shape (batch size, channels, super res height, super res width), *optional*):
                 Pre-generated noisy latents to be used as inputs for the decoder.
             decoder_guidance_scale (`float`, *optional*, defaults to 4.0):
                 Guidance scale as defined in [Classifier-Free Diffusion Guidance](https://arxiv.org/abs/2207.12598).
@@ -272,19 +272,19 @@ class UnCLIPImageInterpolationPipeline(DiffusionPipeline):
                 raise AssertionError(
                     f"Expected 'image' List to contain PIL.Image.Image, but passed 'image' contents are {type(image[0])} and {type(image[1])}"
                 )
-        elif isinstance(image, torch.FloatTensor):
+        elif isinstance(image, torch.Tensor):
             if image.shape[0] != 2:
                 raise AssertionError(
-                    f"Expected 'image' to be torch.FloatTensor of shape 2 in 0th dimension, but passed 'image' size is {image.shape[0]}"
+                    f"Expected 'image' to be torch.Tensor of shape 2 in 0th dimension, but passed 'image' size is {image.shape[0]}"
                 )
         elif isinstance(image_embeddings, torch.Tensor):
             if image_embeddings.shape[0] != 2:
                 raise AssertionError(
-                    f"Expected 'image_embeddings' to be torch.FloatTensor of shape 2 in 0th dimension, but passed 'image_embeddings' shape is {image_embeddings.shape[0]}"
+                    f"Expected 'image_embeddings' to be torch.Tensor of shape 2 in 0th dimension, but passed 'image_embeddings' shape is {image_embeddings.shape[0]}"
                 )
         else:
             raise AssertionError(
-                f"Expected 'image' or 'image_embeddings' to be not None with types List[PIL.Image] or Torch.FloatTensor respectively. Received {type(image)} and {type(image_embeddings)} repsectively"
+                f"Expected 'image' or 'image_embeddings' to be not None with types List[PIL.Image] or torch.Tensor respectively. Received {type(image)} and {type(image_embeddings)} repsectively"
             )
 
         original_image_embeddings = self._encode_image(

examples/community/wildcard_stable_diffusion.py

@@ -166,10 +166,10 @@ class WildcardStableDiffusionPipeline(DiffusionPipeline, StableDiffusionMixin):
         num_images_per_prompt: Optional[int] = 1,
         eta: float = 0.0,
         generator: Optional[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,
         wildcard_option_dict: Dict[str, List[str]] = {},
         wildcard_files: List[str] = [],
@@ -206,7 +206,7 @@ class WildcardStableDiffusionPipeline(DiffusionPipeline, StableDiffusionMixin):
             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 will ge generated by sampling using the supplied random `generator`.
@@ -218,7 +218,7 @@ class WildcardStableDiffusionPipeline(DiffusionPipeline, StableDiffusionMixin):
                 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.

examples/consistency_distillation/train_lcm_distill_lora_sd_wds.py

@@ -336,7 +336,7 @@ def guidance_scale_embedding(w, embedding_dim=512, dtype=torch.float32):
             data type of the generated embeddings
 
     Returns:
-        `torch.FloatTensor`: Embedding vectors with shape `(len(timesteps), embedding_dim)`
+        `torch.Tensor`: Embedding vectors with shape `(len(timesteps), embedding_dim)`
     """
     assert len(w.shape) == 1
     w = w * 1000.0

examples/consistency_distillation/train_lcm_distill_sd_wds.py

@@ -314,7 +314,7 @@ def guidance_scale_embedding(w, embedding_dim=512, dtype=torch.float32):
             data type of the generated embeddings
 
     Returns:
-        `torch.FloatTensor`: Embedding vectors with shape `(len(timesteps), embedding_dim)`
+        `torch.Tensor`: Embedding vectors with shape `(len(timesteps), embedding_dim)`
     """
     assert len(w.shape) == 1
     w = w * 1000.0

examples/consistency_distillation/train_lcm_distill_sdxl_wds.py

@@ -406,7 +406,7 @@ def guidance_scale_embedding(w, embedding_dim=512, dtype=torch.float32):
             data type of the generated embeddings
 
     Returns:
-        `torch.FloatTensor`: Embedding vectors with shape `(len(timesteps), embedding_dim)`
+        `torch.Tensor`: Embedding vectors with shape `(len(timesteps), embedding_dim)`
     """
     assert len(w.shape) == 1
     w = w * 1000.0

examples/research_projects/consistency_training/train_cm_ct_unconditional.py

@@ -126,7 +126,7 @@ def get_karras_sigmas(
     return sigmas
 
 
-def get_discretized_lognormal_weights(noise_levels: torch.FloatTensor, p_mean: float = -1.1, p_std: float = 2.0):
+def get_discretized_lognormal_weights(noise_levels: torch.Tensor, p_mean: float = -1.1, p_std: float = 2.0):
     """
     Calculates the unnormalized weights for a 1D array of noise level sigma_i based on the discretized lognormal"
     " distribution used in the iCT paper (given in Equation 10).
@@ -137,14 +137,14 @@ def get_discretized_lognormal_weights(noise_levels: torch.FloatTensor, p_mean: f
     return weights
 
 
-def get_loss_weighting_schedule(noise_levels: torch.FloatTensor):
+def get_loss_weighting_schedule(noise_levels: torch.Tensor):
     """
     Calculates the loss weighting schedule lambda given a set of noise levels.
     """
     return 1.0 / (noise_levels[1:] - noise_levels[:-1])
 
 
-def add_noise(original_samples: torch.FloatTensor, noise: torch.FloatTensor, timesteps: torch.FloatTensor):
+def add_noise(original_samples: torch.Tensor, noise: torch.Tensor, timesteps: torch.Tensor):
     # Make sure timesteps (Karras sigmas) have the same device and dtype as original_samples
     sigmas = timesteps.to(device=original_samples.device, dtype=original_samples.dtype)
     while len(sigmas.shape) < len(original_samples.shape):

examples/research_projects/geodiff/geodiff_molecule_conformation.ipynb

@@ -737,11 +737,11 @@
         "class MoleculeGNNOutput(BaseOutput):\n",
         "    \"\"\"\n",
         "    Args:\n",
-        "        sample (`torch.FloatTensor` of shape `(batch_size, num_channels, height, width)`):\n",
+        "        sample (`torch.Tensor` of shape `(batch_size, num_channels, height, width)`):\n",
         "            Hidden states output. Output of last layer of model.\n",
         "    \"\"\"\n",
         "\n",
-        "    sample: torch.FloatTensor\n",
+        "    sample: torch.Tensor\n",
         "\n",
         "\n",
         "class MultiLayerPerceptron(nn.Module):\n",
@@ -1354,7 +1354,7 @@
         "        r\"\"\"\n",
         "        Args:\n",
         "            sample: packed torch geometric object\n",
-        "            timestep (`torch.FloatTensor` or `float` or `int): TODO verify type and shape (batch) timesteps\n",
+        "            timestep (`torch.Tensor` or `float` or `int): TODO verify type and shape (batch) timesteps\n",
         "            return_dict (`bool`, *optional*, defaults to `True`):\n",
         "                Whether or not to return a [`~models.molecule_gnn.MoleculeGNNOutput`] instead of a plain tuple.\n",
         "        Returns:\n",
@@ -1404,7 +1404,7 @@
         "        if not return_dict:\n",
         "            return (-eps_pos,)\n",
         "\n",
-        "        return MoleculeGNNOutput(sample=torch.FloatTensor(-eps_pos).to(pos.device))"
+        "        return MoleculeGNNOutput(sample=torch.Tensor(-eps_pos).to(pos.device))"
       ],
       "metadata": {
         "id": "MCeZA1qQXzoK"

examples/research_projects/promptdiffusion/pipeline_prompt_diffusion.py

@@ -279,8 +279,8 @@ class PromptDiffusionPipeline(DiffusionPipeline, TextualInversionLoaderMixin, Lo
         num_images_per_prompt,
         do_classifier_free_guidance,
         negative_prompt=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,
         lora_scale: Optional[float] = None,
         **kwargs,
     ):
@@ -312,8 +312,8 @@ class PromptDiffusionPipeline(DiffusionPipeline, TextualInversionLoaderMixin, Lo
         num_images_per_prompt,
         do_classifier_free_guidance,
         negative_prompt=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,
         lora_scale: Optional[float] = None,
         clip_skip: Optional[int] = None,
     ):
@@ -333,10 +333,10 @@ class PromptDiffusionPipeline(DiffusionPipeline, TextualInversionLoaderMixin, Lo
                 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
                 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.
@@ -852,7 +852,7 @@ class PromptDiffusionPipeline(DiffusionPipeline, TextualInversionLoaderMixin, Lo
                 data type of the generated embeddings
 
         Returns:
-            `torch.FloatTensor`: Embedding vectors with shape `(len(timesteps), embedding_dim)`
+            `torch.Tensor`: Embedding vectors with shape `(len(timesteps), embedding_dim)`
         """
         assert len(w.shape) == 1
         w = w * 1000.0
@@ -906,9 +906,9 @@ class PromptDiffusionPipeline(DiffusionPipeline, TextualInversionLoaderMixin, Lo
         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,
-        prompt_embeds: Optional[torch.FloatTensor] = None,
-        negative_prompt_embeds: Optional[torch.FloatTensor] = None,
+        latents: Optional[torch.Tensor] = None,
+        prompt_embeds: Optional[torch.Tensor] = None,
+        negative_prompt_embeds: Optional[torch.Tensor] = None,
         ip_adapter_image: Optional[PipelineImageInput] = None,
         output_type: Optional[str] = "pil",
         return_dict: bool = True,
@@ -928,10 +928,10 @@ class PromptDiffusionPipeline(DiffusionPipeline, TextualInversionLoaderMixin, Lo
         Args:
             prompt (`str` or `List[str]`, *optional*):
                 The prompt or prompts to guide image generation. If not defined, you need to pass `prompt_embeds`.
-            image (`torch.FloatTensor`, `PIL.Image.Image`, `np.ndarray`, `List[torch.FloatTensor]`, `List[PIL.Image.Image]`, `List[np.ndarray]`,:
-                    `List[List[torch.FloatTensor]]`, `List[List[np.ndarray]]` or `List[List[PIL.Image.Image]]`):
+            image (`torch.Tensor`, `PIL.Image.Image`, `np.ndarray`, `List[torch.Tensor]`, `List[PIL.Image.Image]`, `List[np.ndarray]`,:
+                    `List[List[torch.Tensor]]`, `List[List[np.ndarray]]` or `List[List[PIL.Image.Image]]`):
                 The ControlNet input condition to provide guidance to the `unet` for generation. If the type is
-                specified as `torch.FloatTensor`, it is passed to ControlNet as is. `PIL.Image.Image` can also be
+                specified as `torch.Tensor`, it is passed to ControlNet as is. `PIL.Image.Image` can also be
                 accepted as an image. The dimensions of the output image defaults to `image`'s dimensions. If height
                 and/or width are passed, `image` is resized accordingly. If multiple ControlNets are specified in
                 `init`, images must be passed as a list such that each element of the list can be correctly batched for
@@ -963,14 +963,14 @@ class PromptDiffusionPipeline(DiffusionPipeline, TextualInversionLoaderMixin, Lo
             generator (`torch.Generator` or `List[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`.
-            prompt_embeds (`torch.FloatTensor`, *optional*):
+            prompt_embeds (`torch.Tensor`, *optional*):
                 Pre-generated text embeddings. Can be used to easily tweak text inputs (prompt weighting). If not
                 provided, text embeddings are generated from the `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 (prompt weighting). If
                 not provided, `negative_prompt_embeds` are generated from the `negative_prompt` input argument.
             ip_adapter_image: (`PipelineImageInput`, *optional*): Optional image input to work with IP Adapters.
@@ -981,7 +981,7 @@ class PromptDiffusionPipeline(DiffusionPipeline, TextualInversionLoaderMixin, Lo
                 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.

examples/research_projects/promptdiffusion/promptdiffusioncontrolnet.py

@@ -181,11 +181,11 @@ class PromptDiffusionControlNetModel(ControlNetModel):
 
     def forward(
         self,
-        sample: torch.FloatTensor,
+        sample: torch.Tensor,
         timestep: Union[torch.Tensor, float, int],
         encoder_hidden_states: torch.Tensor,
-        controlnet_cond: torch.FloatTensor,
-        controlnet_query_cond: torch.FloatTensor,
+        controlnet_cond: torch.Tensor,
+        controlnet_query_cond: torch.Tensor,
         conditioning_scale: float = 1.0,
         class_labels: Optional[torch.Tensor] = None,
         timestep_cond: Optional[torch.Tensor] = None,
@@ -194,20 +194,20 @@ class PromptDiffusionControlNetModel(ControlNetModel):
         cross_attention_kwargs: Optional[Dict[str, Any]] = None,
         guess_mode: bool = False,
         return_dict: bool = True,
-    ) -> Union[ControlNetOutput, Tuple[Tuple[torch.FloatTensor, ...], torch.FloatTensor]]:
+    ) -> Union[ControlNetOutput, Tuple[Tuple[torch.Tensor, ...], torch.Tensor]]:
         """
         The [`~PromptDiffusionControlNetModel`] forward method.
 
         Args:
-            sample (`torch.FloatTensor`):
+            sample (`torch.Tensor`):
                 The noisy input tensor.
             timestep (`Union[torch.Tensor, float, int]`):
                 The number of timesteps to denoise an input.
             encoder_hidden_states (`torch.Tensor`):
                 The encoder hidden states.
-            controlnet_cond (`torch.FloatTensor`):
+            controlnet_cond (`torch.Tensor`):
                 The conditional input tensor of shape `(batch_size, sequence_length, hidden_size)`.
-            controlnet_query_cond (`torch.FloatTensor`):
+            controlnet_query_cond (`torch.Tensor`):
                 The conditional input tensor of shape `(batch_size, sequence_length, hidden_size)`.
             conditioning_scale (`float`, defaults to `1.0`):
                 The scale factor for ControlNet outputs.

examples/research_projects/rdm/pipeline_rdm.py

@@ -163,11 +163,11 @@ class RDMPipeline(DiffusionPipeline, StableDiffusionMixin):
         num_images_per_prompt: Optional[int] = 1,
         eta: float = 0.0,
         generator: Optional[torch.Generator] = None,
-        latents: Optional[torch.FloatTensor] = None,
-        prompt_embeds: Optional[torch.FloatTensor] = None,
+        latents: Optional[torch.Tensor] = None,
+        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: Optional[int] = 1,
         knn: Optional[int] = 10,
         **kwargs,
@@ -199,11 +199,11 @@ class RDMPipeline(DiffusionPipeline, StableDiffusionMixin):
             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 will ge generated by sampling using the supplied random `generator`.
-            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.
             output_type (`str`, *optional*, defaults to `"pil"`):
@@ -213,7 +213,7 @@ class RDMPipeline(DiffusionPipeline, StableDiffusionMixin):
                 Whether or not to return a [`~pipeline_utils.ImagePipelineOutput`] 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.

examples/research_projects/rdm/retriever.py

@@ -20,7 +20,7 @@ def normalize_images(images: List[Image.Image]):
     return images
 
 
-def preprocess_images(images: List[np.array], feature_extractor: CLIPFeatureExtractor) -> torch.FloatTensor:
+def preprocess_images(images: List[np.array], feature_extractor: CLIPFeatureExtractor) -> torch.Tensor:
     """
     Preprocesses a list of images into a batch of tensors.
 
@@ -29,7 +29,7 @@ def preprocess_images(images: List[np.array], feature_extractor: CLIPFeatureExtr
             A list of images to preprocess.
 
     Returns:
-        :obj:`torch.FloatTensor`: A batch of tensors.
+        :obj:`torch.Tensor`: A batch of tensors.
     """
     images = [np.array(image) for image in images]
     images = [(image + 1.0) / 2.0 for image in images]

scripts/convert_music_spectrogram_to_diffusers.py

@@ -17,109 +17,99 @@ MODEL = "base_with_context"
 
 
 def load_notes_encoder(weights, model):
-    model.token_embedder.weight = nn.Parameter(torch.FloatTensor(weights["token_embedder"]["embedding"]))
-    model.position_encoding.weight = nn.Parameter(
-        torch.FloatTensor(weights["Embed_0"]["embedding"]), requires_grad=False
-    )
+    model.token_embedder.weight = nn.Parameter(torch.Tensor(weights["token_embedder"]["embedding"]))
+    model.position_encoding.weight = nn.Parameter(torch.Tensor(weights["Embed_0"]["embedding"]), requires_grad=False)
     for lyr_num, lyr in enumerate(model.encoders):
         ly_weight = weights[f"layers_{lyr_num}"]
-        lyr.layer[0].layer_norm.weight = nn.Parameter(
-            torch.FloatTensor(ly_weight["pre_attention_layer_norm"]["scale"])
-        )
+        lyr.layer[0].layer_norm.weight = nn.Parameter(torch.Tensor(ly_weight["pre_attention_layer_norm"]["scale"]))
 
         attention_weights = ly_weight["attention"]
-        lyr.layer[0].SelfAttention.q.weight = nn.Parameter(torch.FloatTensor(attention_weights["query"]["kernel"].T))
-        lyr.layer[0].SelfAttention.k.weight = nn.Parameter(torch.FloatTensor(attention_weights["key"]["kernel"].T))
-        lyr.layer[0].SelfAttention.v.weight = nn.Parameter(torch.FloatTensor(attention_weights["value"]["kernel"].T))
-        lyr.layer[0].SelfAttention.o.weight = nn.Parameter(torch.FloatTensor(attention_weights["out"]["kernel"].T))
+        lyr.layer[0].SelfAttention.q.weight = nn.Parameter(torch.Tensor(attention_weights["query"]["kernel"].T))
+        lyr.layer[0].SelfAttention.k.weight = nn.Parameter(torch.Tensor(attention_weights["key"]["kernel"].T))
+        lyr.layer[0].SelfAttention.v.weight = nn.Parameter(torch.Tensor(attention_weights["value"]["kernel"].T))
+        lyr.layer[0].SelfAttention.o.weight = nn.Parameter(torch.Tensor(attention_weights["out"]["kernel"].T))
 
-        lyr.layer[1].layer_norm.weight = nn.Parameter(torch.FloatTensor(ly_weight["pre_mlp_layer_norm"]["scale"]))
+        lyr.layer[1].layer_norm.weight = nn.Parameter(torch.Tensor(ly_weight["pre_mlp_layer_norm"]["scale"]))
 
-        lyr.layer[1].DenseReluDense.wi_0.weight = nn.Parameter(torch.FloatTensor(ly_weight["mlp"]["wi_0"]["kernel"].T))
-        lyr.layer[1].DenseReluDense.wi_1.weight = nn.Parameter(torch.FloatTensor(ly_weight["mlp"]["wi_1"]["kernel"].T))
-        lyr.layer[1].DenseReluDense.wo.weight = nn.Parameter(torch.FloatTensor(ly_weight["mlp"]["wo"]["kernel"].T))
+        lyr.layer[1].DenseReluDense.wi_0.weight = nn.Parameter(torch.Tensor(ly_weight["mlp"]["wi_0"]["kernel"].T))
+        lyr.layer[1].DenseReluDense.wi_1.weight = nn.Parameter(torch.Tensor(ly_weight["mlp"]["wi_1"]["kernel"].T))
+        lyr.layer[1].DenseReluDense.wo.weight = nn.Parameter(torch.Tensor(ly_weight["mlp"]["wo"]["kernel"].T))
 
-    model.layer_norm.weight = nn.Parameter(torch.FloatTensor(weights["encoder_norm"]["scale"]))
+    model.layer_norm.weight = nn.Parameter(torch.Tensor(weights["encoder_norm"]["scale"]))
     return model
 
 
 def load_continuous_encoder(weights, model):
-    model.input_proj.weight = nn.Parameter(torch.FloatTensor(weights["input_proj"]["kernel"].T))
+    model.input_proj.weight = nn.Parameter(torch.Tensor(weights["input_proj"]["kernel"].T))
 
-    model.position_encoding.weight = nn.Parameter(
-        torch.FloatTensor(weights["Embed_0"]["embedding"]), requires_grad=False
-    )
+    model.position_encoding.weight = nn.Parameter(torch.Tensor(weights["Embed_0"]["embedding"]), requires_grad=False)
 
     for lyr_num, lyr in enumerate(model.encoders):
         ly_weight = weights[f"layers_{lyr_num}"]
         attention_weights = ly_weight["attention"]
 
-        lyr.layer[0].SelfAttention.q.weight = nn.Parameter(torch.FloatTensor(attention_weights["query"]["kernel"].T))
-        lyr.layer[0].SelfAttention.k.weight = nn.Parameter(torch.FloatTensor(attention_weights["key"]["kernel"].T))
-        lyr.layer[0].SelfAttention.v.weight = nn.Parameter(torch.FloatTensor(attention_weights["value"]["kernel"].T))
-        lyr.layer[0].SelfAttention.o.weight = nn.Parameter(torch.FloatTensor(attention_weights["out"]["kernel"].T))
-        lyr.layer[0].layer_norm.weight = nn.Parameter(
-            torch.FloatTensor(ly_weight["pre_attention_layer_norm"]["scale"])
-        )
+        lyr.layer[0].SelfAttention.q.weight = nn.Parameter(torch.Tensor(attention_weights["query"]["kernel"].T))
+        lyr.layer[0].SelfAttention.k.weight = nn.Parameter(torch.Tensor(attention_weights["key"]["kernel"].T))
+        lyr.layer[0].SelfAttention.v.weight = nn.Parameter(torch.Tensor(attention_weights["value"]["kernel"].T))
+        lyr.layer[0].SelfAttention.o.weight = nn.Parameter(torch.Tensor(attention_weights["out"]["kernel"].T))
+        lyr.layer[0].layer_norm.weight = nn.Parameter(torch.Tensor(ly_weight["pre_attention_layer_norm"]["scale"]))
 
-        lyr.layer[1].DenseReluDense.wi_0.weight = nn.Parameter(torch.FloatTensor(ly_weight["mlp"]["wi_0"]["kernel"].T))
-        lyr.layer[1].DenseReluDense.wi_1.weight = nn.Parameter(torch.FloatTensor(ly_weight["mlp"]["wi_1"]["kernel"].T))
-        lyr.layer[1].DenseReluDense.wo.weight = nn.Parameter(torch.FloatTensor(ly_weight["mlp"]["wo"]["kernel"].T))
-        lyr.layer[1].layer_norm.weight = nn.Parameter(torch.FloatTensor(ly_weight["pre_mlp_layer_norm"]["scale"]))
+        lyr.layer[1].DenseReluDense.wi_0.weight = nn.Parameter(torch.Tensor(ly_weight["mlp"]["wi_0"]["kernel"].T))
+        lyr.layer[1].DenseReluDense.wi_1.weight = nn.Parameter(torch.Tensor(ly_weight["mlp"]["wi_1"]["kernel"].T))
+        lyr.layer[1].DenseReluDense.wo.weight = nn.Parameter(torch.Tensor(ly_weight["mlp"]["wo"]["kernel"].T))
+        lyr.layer[1].layer_norm.weight = nn.Parameter(torch.Tensor(ly_weight["pre_mlp_layer_norm"]["scale"]))
 
-    model.layer_norm.weight = nn.Parameter(torch.FloatTensor(weights["encoder_norm"]["scale"]))
+    model.layer_norm.weight = nn.Parameter(torch.Tensor(weights["encoder_norm"]["scale"]))
 
     return model
 
 
 def load_decoder(weights, model):
-    model.conditioning_emb[0].weight = nn.Parameter(torch.FloatTensor(weights["time_emb_dense0"]["kernel"].T))
-    model.conditioning_emb[2].weight = nn.Parameter(torch.FloatTensor(weights["time_emb_dense1"]["kernel"].T))
+    model.conditioning_emb[0].weight = nn.Parameter(torch.Tensor(weights["time_emb_dense0"]["kernel"].T))
+    model.conditioning_emb[2].weight = nn.Parameter(torch.Tensor(weights["time_emb_dense1"]["kernel"].T))
 
-    model.position_encoding.weight = nn.Parameter(
-        torch.FloatTensor(weights["Embed_0"]["embedding"]), requires_grad=False
-    )
+    model.position_encoding.weight = nn.Parameter(torch.Tensor(weights["Embed_0"]["embedding"]), requires_grad=False)
 
     model.continuous_inputs_projection.weight = nn.Parameter(
-        torch.FloatTensor(weights["continuous_inputs_projection"]["kernel"].T)
+        torch.Tensor(weights["continuous_inputs_projection"]["kernel"].T)
     )
 
     for lyr_num, lyr in enumerate(model.decoders):
         ly_weight = weights[f"layers_{lyr_num}"]
         lyr.layer[0].layer_norm.weight = nn.Parameter(
-            torch.FloatTensor(ly_weight["pre_self_attention_layer_norm"]["scale"])
+            torch.Tensor(ly_weight["pre_self_attention_layer_norm"]["scale"])
         )
 
         lyr.layer[0].FiLMLayer.scale_bias.weight = nn.Parameter(
-            torch.FloatTensor(ly_weight["FiLMLayer_0"]["DenseGeneral_0"]["kernel"].T)
+            torch.Tensor(ly_weight["FiLMLayer_0"]["DenseGeneral_0"]["kernel"].T)
         )
 
         attention_weights = ly_weight["self_attention"]
-        lyr.layer[0].attention.to_q.weight = nn.Parameter(torch.FloatTensor(attention_weights["query"]["kernel"].T))
-        lyr.layer[0].attention.to_k.weight = nn.Parameter(torch.FloatTensor(attention_weights["key"]["kernel"].T))
-        lyr.layer[0].attention.to_v.weight = nn.Parameter(torch.FloatTensor(attention_weights["value"]["kernel"].T))
-        lyr.layer[0].attention.to_out[0].weight = nn.Parameter(torch.FloatTensor(attention_weights["out"]["kernel"].T))
+        lyr.layer[0].attention.to_q.weight = nn.Parameter(torch.Tensor(attention_weights["query"]["kernel"].T))
+        lyr.layer[0].attention.to_k.weight = nn.Parameter(torch.Tensor(attention_weights["key"]["kernel"].T))
+        lyr.layer[0].attention.to_v.weight = nn.Parameter(torch.Tensor(attention_weights["value"]["kernel"].T))
+        lyr.layer[0].attention.to_out[0].weight = nn.Parameter(torch.Tensor(attention_weights["out"]["kernel"].T))
 
         attention_weights = ly_weight["MultiHeadDotProductAttention_0"]
-        lyr.layer[1].attention.to_q.weight = nn.Parameter(torch.FloatTensor(attention_weights["query"]["kernel"].T))
-        lyr.layer[1].attention.to_k.weight = nn.Parameter(torch.FloatTensor(attention_weights["key"]["kernel"].T))
-        lyr.layer[1].attention.to_v.weight = nn.Parameter(torch.FloatTensor(attention_weights["value"]["kernel"].T))
-        lyr.layer[1].attention.to_out[0].weight = nn.Parameter(torch.FloatTensor(attention_weights["out"]["kernel"].T))
+        lyr.layer[1].attention.to_q.weight = nn.Parameter(torch.Tensor(attention_weights["query"]["kernel"].T))
+        lyr.layer[1].attention.to_k.weight = nn.Parameter(torch.Tensor(attention_weights["key"]["kernel"].T))
+        lyr.layer[1].attention.to_v.weight = nn.Parameter(torch.Tensor(attention_weights["value"]["kernel"].T))
+        lyr.layer[1].attention.to_out[0].weight = nn.Parameter(torch.Tensor(attention_weights["out"]["kernel"].T))
         lyr.layer[1].layer_norm.weight = nn.Parameter(
-            torch.FloatTensor(ly_weight["pre_cross_attention_layer_norm"]["scale"])
+            torch.Tensor(ly_weight["pre_cross_attention_layer_norm"]["scale"])
         )
 
-        lyr.layer[2].layer_norm.weight = nn.Parameter(torch.FloatTensor(ly_weight["pre_mlp_layer_norm"]["scale"]))
+        lyr.layer[2].layer_norm.weight = nn.Parameter(torch.Tensor(ly_weight["pre_mlp_layer_norm"]["scale"]))
         lyr.layer[2].film.scale_bias.weight = nn.Parameter(
-            torch.FloatTensor(ly_weight["FiLMLayer_1"]["DenseGeneral_0"]["kernel"].T)
+            torch.Tensor(ly_weight["FiLMLayer_1"]["DenseGeneral_0"]["kernel"].T)
         )
-        lyr.layer[2].DenseReluDense.wi_0.weight = nn.Parameter(torch.FloatTensor(ly_weight["mlp"]["wi_0"]["kernel"].T))
-        lyr.layer[2].DenseReluDense.wi_1.weight = nn.Parameter(torch.FloatTensor(ly_weight["mlp"]["wi_1"]["kernel"].T))
-        lyr.layer[2].DenseReluDense.wo.weight = nn.Parameter(torch.FloatTensor(ly_weight["mlp"]["wo"]["kernel"].T))
+        lyr.layer[2].DenseReluDense.wi_0.weight = nn.Parameter(torch.Tensor(ly_weight["mlp"]["wi_0"]["kernel"].T))
+        lyr.layer[2].DenseReluDense.wi_1.weight = nn.Parameter(torch.Tensor(ly_weight["mlp"]["wi_1"]["kernel"].T))
+        lyr.layer[2].DenseReluDense.wo.weight = nn.Parameter(torch.Tensor(ly_weight["mlp"]["wo"]["kernel"].T))
 
-    model.decoder_norm.weight = nn.Parameter(torch.FloatTensor(weights["decoder_norm"]["scale"]))
+    model.decoder_norm.weight = nn.Parameter(torch.Tensor(weights["decoder_norm"]["scale"]))
 
-    model.spec_out.weight = nn.Parameter(torch.FloatTensor(weights["spec_out_dense"]["kernel"].T))
+    model.spec_out.weight = nn.Parameter(torch.Tensor(weights["spec_out_dense"]["kernel"].T))
 
     return model
 

src/diffusers/models/attention.py

@@ -282,15 +282,15 @@ class BasicTransformerBlock(nn.Module):
 
     def forward(
         self,
-        hidden_states: torch.FloatTensor,
-        attention_mask: Optional[torch.FloatTensor] = None,
-        encoder_hidden_states: Optional[torch.FloatTensor] = None,
-        encoder_attention_mask: Optional[torch.FloatTensor] = None,
+        hidden_states: torch.Tensor,
+        attention_mask: Optional[torch.Tensor] = None,
+        encoder_hidden_states: Optional[torch.Tensor] = None,
+        encoder_attention_mask: Optional[torch.Tensor] = None,
         timestep: Optional[torch.LongTensor] = None,
         cross_attention_kwargs: Dict[str, Any] = None,
         class_labels: Optional[torch.LongTensor] = None,
         added_cond_kwargs: Optional[Dict[str, torch.Tensor]] = None,
-    ) -> torch.FloatTensor:
+    ) -> torch.Tensor:
         if cross_attention_kwargs is not None:
             if cross_attention_kwargs.get("scale", None) is not None:
                 logger.warning("Passing `scale` to `cross_attention_kwargs` is deprecated. `scale` will be ignored.")
@@ -477,10 +477,10 @@ class TemporalBasicTransformerBlock(nn.Module):
 
     def forward(
         self,
-        hidden_states: torch.FloatTensor,
+        hidden_states: torch.Tensor,
         num_frames: int,
-        encoder_hidden_states: Optional[torch.FloatTensor] = None,
-    ) -> torch.FloatTensor:
+        encoder_hidden_states: Optional[torch.Tensor] = None,
+    ) -> torch.Tensor:
         # Notice that normalization is always applied before the real computation in the following blocks.
         # 0. Self-Attention
         batch_size = hidden_states.shape[0]

src/diffusers/models/attention_processor.py

@@ -503,9 +503,9 @@ class Attention(nn.Module):
 
     def forward(
         self,
-        hidden_states: torch.FloatTensor,
-        encoder_hidden_states: Optional[torch.FloatTensor] = None,
-        attention_mask: Optional[torch.FloatTensor] = None,
+        hidden_states: torch.Tensor,
+        encoder_hidden_states: Optional[torch.Tensor] = None,
+        attention_mask: Optional[torch.Tensor] = None,
         **cross_attention_kwargs,
     ) -> torch.Tensor:
         r"""
@@ -751,10 +751,10 @@ class AttnProcessor:
     def __call__(
         self,
         attn: Attention,
-        hidden_states: torch.FloatTensor,
-        encoder_hidden_states: Optional[torch.FloatTensor] = None,
-        attention_mask: Optional[torch.FloatTensor] = None,
-        temb: Optional[torch.FloatTensor] = None,
+        hidden_states: torch.Tensor,
+        encoder_hidden_states: Optional[torch.Tensor] = None,
+        attention_mask: Optional[torch.Tensor] = None,
+        temb: Optional[torch.Tensor] = None,
         *args,
         **kwargs,
     ) -> torch.Tensor:
@@ -863,9 +863,9 @@ class CustomDiffusionAttnProcessor(nn.Module):
     def __call__(
         self,
         attn: Attention,
-        hidden_states: torch.FloatTensor,
-        encoder_hidden_states: Optional[torch.FloatTensor] = None,
-        attention_mask: Optional[torch.FloatTensor] = None,
+        hidden_states: torch.Tensor,
+        encoder_hidden_states: Optional[torch.Tensor] = None,
+        attention_mask: Optional[torch.Tensor] = None,
     ) -> torch.Tensor:
         batch_size, sequence_length, _ = hidden_states.shape
         attention_mask = attn.prepare_attention_mask(attention_mask, sequence_length, batch_size)
@@ -928,9 +928,9 @@ class AttnAddedKVProcessor:
     def __call__(
         self,
         attn: Attention,
-        hidden_states: torch.FloatTensor,
-        encoder_hidden_states: Optional[torch.FloatTensor] = None,
-        attention_mask: Optional[torch.FloatTensor] = None,
+        hidden_states: torch.Tensor,
+        encoder_hidden_states: Optional[torch.Tensor] = None,
+        attention_mask: Optional[torch.Tensor] = None,
         *args,
         **kwargs,
     ) -> torch.Tensor:
@@ -1001,9 +1001,9 @@ class AttnAddedKVProcessor2_0:
     def __call__(
         self,
         attn: Attention,
-        hidden_states: torch.FloatTensor,
-        encoder_hidden_states: Optional[torch.FloatTensor] = None,
-        attention_mask: Optional[torch.FloatTensor] = None,
+        hidden_states: torch.Tensor,
+        encoder_hidden_states: Optional[torch.Tensor] = None,
+        attention_mask: Optional[torch.Tensor] = None,
         *args,
         **kwargs,
     ) -> torch.Tensor:
@@ -1080,9 +1080,9 @@ class XFormersAttnAddedKVProcessor:
     def __call__(
         self,
         attn: Attention,
-        hidden_states: torch.FloatTensor,
-        encoder_hidden_states: Optional[torch.FloatTensor] = None,
-        attention_mask: Optional[torch.FloatTensor] = None,
+        hidden_states: torch.Tensor,
+        encoder_hidden_states: Optional[torch.Tensor] = None,
+        attention_mask: Optional[torch.Tensor] = None,
     ) -> torch.Tensor:
         residual = hidden_states
         hidden_states = hidden_states.view(hidden_states.shape[0], hidden_states.shape[1], -1).transpose(1, 2)
@@ -1151,13 +1151,13 @@ class XFormersAttnProcessor:
     def __call__(
         self,
         attn: Attention,
-        hidden_states: torch.FloatTensor,
-        encoder_hidden_states: Optional[torch.FloatTensor] = None,
-        attention_mask: Optional[torch.FloatTensor] = None,
-        temb: Optional[torch.FloatTensor] = None,
+        hidden_states: torch.Tensor,
+        encoder_hidden_states: Optional[torch.Tensor] = None,
+        attention_mask: Optional[torch.Tensor] = None,
+        temb: Optional[torch.Tensor] = None,
         *args,
         **kwargs,
-    ) -> torch.FloatTensor:
+    ) -> torch.Tensor:
         if len(args) > 0 or kwargs.get("scale", None) is not None:
             deprecation_message = "The `scale` argument is deprecated and will be ignored. Please remove it, as passing it will raise an error in the future. `scale` should directly be passed while calling the underlying pipeline component i.e., via `cross_attention_kwargs`."
             deprecate("scale", "1.0.0", deprecation_message)
@@ -1243,13 +1243,13 @@ class AttnProcessorNPU:
     def __call__(
         self,
         attn: Attention,
-        hidden_states: torch.FloatTensor,
-        encoder_hidden_states: Optional[torch.FloatTensor] = None,
-        attention_mask: Optional[torch.FloatTensor] = None,
-        temb: Optional[torch.FloatTensor] = None,
+        hidden_states: torch.Tensor,
+        encoder_hidden_states: Optional[torch.Tensor] = None,
+        attention_mask: Optional[torch.Tensor] = None,
+        temb: Optional[torch.Tensor] = None,
         *args,
         **kwargs,
-    ) -> torch.FloatTensor:
+    ) -> torch.Tensor:
         if len(args) > 0 or kwargs.get("scale", None) is not None:
             deprecation_message = "The `scale` argument is deprecated and will be ignored. Please remove it, as passing it will raise an error in the future. `scale` should directly be passed while calling the underlying pipeline component i.e., via `cross_attention_kwargs`."
             deprecate("scale", "1.0.0", deprecation_message)
@@ -1349,13 +1349,13 @@ class AttnProcessor2_0:
     def __call__(
         self,
         attn: Attention,
-        hidden_states: torch.FloatTensor,
-        encoder_hidden_states: Optional[torch.FloatTensor] = None,
-        attention_mask: Optional[torch.FloatTensor] = None,
-        temb: Optional[torch.FloatTensor] = None,
+        hidden_states: torch.Tensor,
+        encoder_hidden_states: Optional[torch.Tensor] = None,
+        attention_mask: Optional[torch.Tensor] = None,
+        temb: Optional[torch.Tensor] = None,
         *args,
         **kwargs,
-    ) -> torch.FloatTensor:
+    ) -> torch.Tensor:
         if len(args) > 0 or kwargs.get("scale", None) is not None:
             deprecation_message = "The `scale` argument is deprecated and will be ignored. Please remove it, as passing it will raise an error in the future. `scale` should directly be passed while calling the underlying pipeline component i.e., via `cross_attention_kwargs`."
             deprecate("scale", "1.0.0", deprecation_message)
@@ -1448,13 +1448,13 @@ class FusedAttnProcessor2_0:
     def __call__(
         self,
         attn: Attention,
-        hidden_states: torch.FloatTensor,
-        encoder_hidden_states: Optional[torch.FloatTensor] = None,
-        attention_mask: Optional[torch.FloatTensor] = None,
-        temb: Optional[torch.FloatTensor] = None,
+        hidden_states: torch.Tensor,
+        encoder_hidden_states: Optional[torch.Tensor] = None,
+        attention_mask: Optional[torch.Tensor] = None,
+        temb: Optional[torch.Tensor] = None,
         *args,
         **kwargs,
-    ) -> torch.FloatTensor:
+    ) -> torch.Tensor:
         if len(args) > 0 or kwargs.get("scale", None) is not None:
             deprecation_message = "The `scale` argument is deprecated and will be ignored. Please remove it, as passing it will raise an error in the future. `scale` should directly be passed while calling the underlying pipeline component i.e., via `cross_attention_kwargs`."
             deprecate("scale", "1.0.0", deprecation_message)
@@ -1581,10 +1581,10 @@ class CustomDiffusionXFormersAttnProcessor(nn.Module):
     def __call__(
         self,
         attn: Attention,
-        hidden_states: torch.FloatTensor,
-        encoder_hidden_states: Optional[torch.FloatTensor] = None,
-        attention_mask: Optional[torch.FloatTensor] = None,
-    ) -> torch.FloatTensor:
+        hidden_states: torch.Tensor,
+        encoder_hidden_states: Optional[torch.Tensor] = None,
+        attention_mask: Optional[torch.Tensor] = None,
+    ) -> torch.Tensor:
         batch_size, sequence_length, _ = (
             hidden_states.shape if encoder_hidden_states is None else encoder_hidden_states.shape
         )
@@ -1692,10 +1692,10 @@ class CustomDiffusionAttnProcessor2_0(nn.Module):
     def __call__(
         self,
         attn: Attention,
-        hidden_states: torch.FloatTensor,
-        encoder_hidden_states: Optional[torch.FloatTensor] = None,
-        attention_mask: Optional[torch.FloatTensor] = None,
-    ) -> torch.FloatTensor:
+        hidden_states: torch.Tensor,
+        encoder_hidden_states: Optional[torch.Tensor] = None,
+        attention_mask: Optional[torch.Tensor] = None,
+    ) -> torch.Tensor:
         batch_size, sequence_length, _ = hidden_states.shape
         attention_mask = attn.prepare_attention_mask(attention_mask, sequence_length, batch_size)
         if self.train_q_out:
@@ -1773,10 +1773,10 @@ class SlicedAttnProcessor:
     def __call__(
         self,
         attn: Attention,
-        hidden_states: torch.FloatTensor,
-        encoder_hidden_states: Optional[torch.FloatTensor] = None,
-        attention_mask: Optional[torch.FloatTensor] = None,
-    ) -> torch.FloatTensor:
+        hidden_states: torch.Tensor,
+        encoder_hidden_states: Optional[torch.Tensor] = None,
+        attention_mask: Optional[torch.Tensor] = None,
+    ) -> torch.Tensor:
         residual = hidden_states
 
         input_ndim = hidden_states.ndim
@@ -1860,11 +1860,11 @@ class SlicedAttnAddedKVProcessor:
     def __call__(
         self,
         attn: "Attention",
-        hidden_states: torch.FloatTensor,
-        encoder_hidden_states: Optional[torch.FloatTensor] = None,
-        attention_mask: Optional[torch.FloatTensor] = None,
-        temb: Optional[torch.FloatTensor] = None,
-    ) -> torch.FloatTensor:
+        hidden_states: torch.Tensor,
+        encoder_hidden_states: Optional[torch.Tensor] = None,
+        attention_mask: Optional[torch.Tensor] = None,
+        temb: Optional[torch.Tensor] = None,
+    ) -> torch.Tensor:
         residual = hidden_states
 
         if attn.spatial_norm is not None:
@@ -1957,7 +1957,7 @@ class SpatialNorm(nn.Module):
         self.conv_y = nn.Conv2d(zq_channels, f_channels, kernel_size=1, stride=1, padding=0)
         self.conv_b = nn.Conv2d(zq_channels, f_channels, kernel_size=1, stride=1, padding=0)
 
-    def forward(self, f: torch.FloatTensor, zq: torch.FloatTensor) -> torch.FloatTensor:
+    def forward(self, f: torch.Tensor, zq: torch.Tensor) -> torch.Tensor:
         f_size = f.shape[-2:]
         zq = F.interpolate(zq, size=f_size, mode="nearest")
         norm_f = self.norm_layer(f)
@@ -2003,7 +2003,7 @@ class LoRAAttnProcessor(nn.Module):
         self.to_v_lora = LoRALinearLayer(cross_attention_dim or v_hidden_size, v_hidden_size, v_rank, network_alpha)
         self.to_out_lora = LoRALinearLayer(out_hidden_size, out_hidden_size, out_rank, network_alpha)
 
-    def __call__(self, attn: Attention, hidden_states: torch.FloatTensor, **kwargs) -> torch.FloatTensor:
+    def __call__(self, attn: Attention, hidden_states: torch.Tensor, **kwargs) -> torch.Tensor:
         self_cls_name = self.__class__.__name__
         deprecate(
             self_cls_name,
@@ -2064,7 +2064,7 @@ class LoRAAttnProcessor2_0(nn.Module):
         self.to_v_lora = LoRALinearLayer(cross_attention_dim or v_hidden_size, v_hidden_size, v_rank, network_alpha)
         self.to_out_lora = LoRALinearLayer(out_hidden_size, out_hidden_size, out_rank, network_alpha)
 
-    def __call__(self, attn: Attention, hidden_states: torch.FloatTensor, **kwargs) -> torch.FloatTensor:
+    def __call__(self, attn: Attention, hidden_states: torch.Tensor, **kwargs) -> torch.Tensor:
         self_cls_name = self.__class__.__name__
         deprecate(
             self_cls_name,
@@ -2143,7 +2143,7 @@ class LoRAXFormersAttnProcessor(nn.Module):
         self.to_v_lora = LoRALinearLayer(cross_attention_dim or v_hidden_size, v_hidden_size, v_rank, network_alpha)
         self.to_out_lora = LoRALinearLayer(out_hidden_size, out_hidden_size, out_rank, network_alpha)
 
-    def __call__(self, attn: Attention, hidden_states: torch.FloatTensor, **kwargs) -> torch.FloatTensor:
+    def __call__(self, attn: Attention, hidden_states: torch.Tensor, **kwargs) -> torch.Tensor:
         self_cls_name = self.__class__.__name__
         deprecate(
             self_cls_name,
@@ -2202,7 +2202,7 @@ class LoRAAttnAddedKVProcessor(nn.Module):
         self.to_v_lora = LoRALinearLayer(hidden_size, hidden_size, rank, network_alpha)
         self.to_out_lora = LoRALinearLayer(hidden_size, hidden_size, rank, network_alpha)
 
-    def __call__(self, attn: Attention, hidden_states: torch.FloatTensor, **kwargs) -> torch.FloatTensor:
+    def __call__(self, attn: Attention, hidden_states: torch.Tensor, **kwargs) -> torch.Tensor:
         self_cls_name = self.__class__.__name__
         deprecate(
             self_cls_name,
@@ -2264,12 +2264,12 @@ class IPAdapterAttnProcessor(nn.Module):
     def __call__(
         self,
         attn: Attention,
-        hidden_states: torch.FloatTensor,
-        encoder_hidden_states: Optional[torch.FloatTensor] = None,
-        attention_mask: Optional[torch.FloatTensor] = None,
-        temb: Optional[torch.FloatTensor] = None,
+        hidden_states: torch.Tensor,
+        encoder_hidden_states: Optional[torch.Tensor] = None,
+        attention_mask: Optional[torch.Tensor] = None,
+        temb: Optional[torch.Tensor] = None,
         scale: float = 1.0,
-        ip_adapter_masks: Optional[torch.FloatTensor] = None,
+        ip_adapter_masks: Optional[torch.Tensor] = None,
     ):
         residual = hidden_states
 
@@ -2467,12 +2467,12 @@ class IPAdapterAttnProcessor2_0(torch.nn.Module):
     def __call__(
         self,
         attn: Attention,
-        hidden_states: torch.FloatTensor,
-        encoder_hidden_states: Optional[torch.FloatTensor] = None,
-        attention_mask: Optional[torch.FloatTensor] = None,
-        temb: Optional[torch.FloatTensor] = None,
+        hidden_states: torch.Tensor,
+        encoder_hidden_states: Optional[torch.Tensor] = None,
+        attention_mask: Optional[torch.Tensor] = None,
+        temb: Optional[torch.Tensor] = None,
         scale: float = 1.0,
-        ip_adapter_masks: Optional[torch.FloatTensor] = None,
+        ip_adapter_masks: Optional[torch.Tensor] = None,
     ):
         residual = hidden_states
 

src/diffusers/models/autoencoders/autoencoder_asym_kl.py

@@ -112,9 +112,7 @@ class AsymmetricAutoencoderKL(ModelMixin, ConfigMixin):
         self.register_to_config(force_upcast=False)
 
     @apply_forward_hook
-    def encode(
-        self, x: torch.FloatTensor, return_dict: bool = True
-    ) -> Union[AutoencoderKLOutput, Tuple[torch.FloatTensor]]:
+    def encode(self, x: torch.Tensor, return_dict: bool = True) -> Union[AutoencoderKLOutput, Tuple[torch.Tensor]]:
         h = self.encoder(x)
         moments = self.quant_conv(h)
         posterior = DiagonalGaussianDistribution(moments)
@@ -126,11 +124,11 @@ class AsymmetricAutoencoderKL(ModelMixin, ConfigMixin):
 
     def _decode(
         self,
-        z: torch.FloatTensor,
-        image: Optional[torch.FloatTensor] = None,
-        mask: Optional[torch.FloatTensor] = None,
+        z: torch.Tensor,
+        image: Optional[torch.Tensor] = None,
+        mask: Optional[torch.Tensor] = None,
         return_dict: bool = True,
-    ) -> Union[DecoderOutput, Tuple[torch.FloatTensor]]:
+    ) -> Union[DecoderOutput, Tuple[torch.Tensor]]:
         z = self.post_quant_conv(z)
         dec = self.decoder(z, image, mask)
 
@@ -142,12 +140,12 @@ class AsymmetricAutoencoderKL(ModelMixin, ConfigMixin):
     @apply_forward_hook
     def decode(
         self,
-        z: torch.FloatTensor,
+        z: torch.Tensor,
         generator: Optional[torch.Generator] = None,
-        image: Optional[torch.FloatTensor] = None,
-        mask: Optional[torch.FloatTensor] = None,
+        image: Optional[torch.Tensor] = None,
+        mask: Optional[torch.Tensor] = None,
         return_dict: bool = True,
-    ) -> Union[DecoderOutput, Tuple[torch.FloatTensor]]:
+    ) -> Union[DecoderOutput, Tuple[torch.Tensor]]:
         decoded = self._decode(z, image, mask).sample
 
         if not return_dict:
@@ -157,16 +155,16 @@ class AsymmetricAutoencoderKL(ModelMixin, ConfigMixin):
 
     def forward(
         self,
-        sample: torch.FloatTensor,
-        mask: Optional[torch.FloatTensor] = None,
+        sample: torch.Tensor,
+        mask: Optional[torch.Tensor] = None,
         sample_posterior: bool = False,
         return_dict: bool = True,
         generator: Optional[torch.Generator] = None,
-    ) -> Union[DecoderOutput, Tuple[torch.FloatTensor]]:
+    ) -> Union[DecoderOutput, Tuple[torch.Tensor]]:
         r"""
         Args:
-            sample (`torch.FloatTensor`): Input sample.
-            mask (`torch.FloatTensor`, *optional*, defaults to `None`): Optional inpainting mask.
+            sample (`torch.Tensor`): Input sample.
+            mask (`torch.Tensor`, *optional*, defaults to `None`): Optional inpainting mask.
             sample_posterior (`bool`, *optional*, defaults to `False`):
                 Whether to sample from the posterior.
             return_dict (`bool`, *optional*, defaults to `True`):

src/diffusers/models/autoencoders/autoencoder_kl.py

@@ -237,13 +237,13 @@ class AutoencoderKL(ModelMixin, ConfigMixin, FromOriginalModelMixin):
 
     @apply_forward_hook
     def encode(
-        self, x: torch.FloatTensor, return_dict: bool = True
+        self, x: torch.Tensor, return_dict: bool = True
     ) -> Union[AutoencoderKLOutput, Tuple[DiagonalGaussianDistribution]]:
         """
         Encode a batch of images into latents.
 
         Args:
-            x (`torch.FloatTensor`): Input batch of images.
+            x (`torch.Tensor`): Input batch of images.
             return_dict (`bool`, *optional*, defaults to `True`):
                 Whether to return a [`~models.autoencoder_kl.AutoencoderKLOutput`] instead of a plain tuple.
 
@@ -268,7 +268,7 @@ class AutoencoderKL(ModelMixin, ConfigMixin, FromOriginalModelMixin):
 
         return AutoencoderKLOutput(latent_dist=posterior)
 
-    def _decode(self, z: torch.FloatTensor, return_dict: bool = True) -> Union[DecoderOutput, torch.FloatTensor]:
+    def _decode(self, z: torch.Tensor, return_dict: bool = True) -> Union[DecoderOutput, torch.Tensor]:
         if self.use_tiling and (z.shape[-1] > self.tile_latent_min_size or z.shape[-2] > self.tile_latent_min_size):
             return self.tiled_decode(z, return_dict=return_dict)
 
@@ -281,14 +281,12 @@ class AutoencoderKL(ModelMixin, ConfigMixin, FromOriginalModelMixin):
         return DecoderOutput(sample=dec)
 
     @apply_forward_hook
-    def decode(
-        self, z: torch.FloatTensor, return_dict: bool = True, generator=None
-    ) -> Union[DecoderOutput, torch.FloatTensor]:
+    def decode(self, z: torch.Tensor, return_dict: bool = True, generator=None) -> Union[DecoderOutput, torch.Tensor]:
         """
         Decode a batch of images.
 
         Args:
-            z (`torch.FloatTensor`): Input batch of latent vectors.
+            z (`torch.Tensor`): Input batch of latent vectors.
             return_dict (`bool`, *optional*, defaults to `True`):
                 Whether to return a [`~models.vae.DecoderOutput`] instead of a plain tuple.
 
@@ -321,7 +319,7 @@ class AutoencoderKL(ModelMixin, ConfigMixin, FromOriginalModelMixin):
             b[:, :, :, x] = a[:, :, :, -blend_extent + x] * (1 - x / blend_extent) + b[:, :, :, x] * (x / blend_extent)
         return b
 
-    def tiled_encode(self, x: torch.FloatTensor, return_dict: bool = True) -> AutoencoderKLOutput:
+    def tiled_encode(self, x: torch.Tensor, return_dict: bool = True) -> AutoencoderKLOutput:
         r"""Encode a batch of images using a tiled encoder.
 
         When this option is enabled, the VAE will split the input tensor into tiles to compute encoding in several
@@ -331,7 +329,7 @@ class AutoencoderKL(ModelMixin, ConfigMixin, FromOriginalModelMixin):
         output, but they should be much less noticeable.
 
         Args:
-            x (`torch.FloatTensor`): Input batch of images.
+            x (`torch.Tensor`): Input batch of images.
             return_dict (`bool`, *optional*, defaults to `True`):
                 Whether or not to return a [`~models.autoencoder_kl.AutoencoderKLOutput`] instead of a plain tuple.
 
@@ -375,12 +373,12 @@ class AutoencoderKL(ModelMixin, ConfigMixin, FromOriginalModelMixin):
 
         return AutoencoderKLOutput(latent_dist=posterior)
 
-    def tiled_decode(self, z: torch.FloatTensor, return_dict: bool = True) -> Union[DecoderOutput, torch.FloatTensor]:
+    def tiled_decode(self, z: torch.Tensor, return_dict: bool = True) -> Union[DecoderOutput, torch.Tensor]:
         r"""
         Decode a batch of images using a tiled decoder.
 
         Args:
-            z (`torch.FloatTensor`): Input batch of latent vectors.
+            z (`torch.Tensor`): Input batch of latent vectors.
             return_dict (`bool`, *optional*, defaults to `True`):
                 Whether or not to return a [`~models.vae.DecoderOutput`] instead of a plain tuple.
 
@@ -425,14 +423,14 @@ class AutoencoderKL(ModelMixin, ConfigMixin, FromOriginalModelMixin):
 
     def forward(
         self,
-        sample: torch.FloatTensor,
+        sample: torch.Tensor,
         sample_posterior: bool = False,
         return_dict: bool = True,
         generator: Optional[torch.Generator] = None,
-    ) -> Union[DecoderOutput, torch.FloatTensor]:
+    ) -> Union[DecoderOutput, torch.Tensor]:
         r"""
         Args:
-            sample (`torch.FloatTensor`): Input sample.
+            sample (`torch.Tensor`): Input sample.
             sample_posterior (`bool`, *optional*, defaults to `False`):
                 Whether to sample from the posterior.
             return_dict (`bool`, *optional*, defaults to `True`):