fix some typos

examples/research_projects/geodiff/geodiff_molecule_conformation.ipynb

@@ -1760,7 +1760,7 @@
     "clip_local = None\n",
     "clip_pos = None\n",
     "\n",
-    "# constands for data handling\n",
+    "# constants for data handling\n",
     "save_traj = False\n",
     "save_data = False\n",
     "output_dir = \"/content/\""

examples/research_projects/multi_subject_dreambooth_inpainting/README.md

@@ -2,7 +2,7 @@
 
 Please note that this project is not actively maintained. However, you can open an issue and tag @gzguevara.
 
-[DreamBooth](https://huggingface.co/papers/2208.12242) is a method to personalize text2image models like stable diffusion given just a few(3~5) images of a subject. This project consists of **two parts**. Training Stable Diffusion for inpainting requieres prompt-image-mask pairs. The Unet of inpainiting models have 5 additional input channels (4 for the encoded masked-image and 1 for the mask itself).
+[DreamBooth](https://huggingface.co/papers/2208.12242) is a method to personalize text2image models like stable diffusion given just a few(3~5) images of a subject. This project consists of **two parts**. Training Stable Diffusion for inpainting requires prompt-image-mask pairs. The Unet of inpainiting models have 5 additional input channels (4 for the encoded masked-image and 1 for the mask itself).
 
 **The first part**, the `multi_inpaint_dataset.ipynb` notebook, demonstrates how make a 🤗 dataset of prompt-image-mask pairs. You can, however, skip the first part and move straight to the second part with the example datasets in this project. ([cat toy dataset masked](https://huggingface.co/datasets/gzguevara/cat_toy_masked), [mr. potato head dataset masked](https://huggingface.co/datasets/gzguevara/mr_potato_head_masked))
 

src/diffusers/guiders/frequency_decoupled_guidance.py

@@ -61,7 +61,7 @@ def project(v0: torch.Tensor, v1: torch.Tensor, upcast_to_double: bool = True) -
 def build_image_from_pyramid(pyramid: List[torch.Tensor]) -> torch.Tensor:
     """
     Recovers the data space latents from the Laplacian pyramid frequency space. Implementation from the paper
-    (Algorihtm 2).
+    (Algorithm 2).
     """
     # pyramid shapes: [[B, C, H, W], [B, C, H/2, W/2], ...]
     img = pyramid[-1]

src/diffusers/hooks/faster_cache.py

@@ -54,11 +54,11 @@ class FasterCacheConfig:
     Attributes:
         spatial_attention_block_skip_range (`int`, defaults to `2`):
             Calculate the attention states every `N` iterations. If this is set to `N`, the attention computation will
-            be skipped `N - 1` times (i.e., cached attention states will be re-used) before computing the new attention
+            be skipped `N - 1` times (i.e., cached attention states will be reused) before computing the new attention
             states again.
         temporal_attention_block_skip_range (`int`, *optional*, defaults to `None`):
             Calculate the attention states every `N` iterations. If this is set to `N`, the attention computation will
-            be skipped `N - 1` times (i.e., cached attention states will be re-used) before computing the new attention
+            be skipped `N - 1` times (i.e., cached attention states will be reused) before computing the new attention
             states again.
         spatial_attention_timestep_skip_range (`Tuple[float, float]`, defaults to `(-1, 681)`):
             The timestep range within which the spatial attention computation can be skipped without a significant loss
@@ -90,7 +90,7 @@ class FasterCacheConfig:
             from the conditional branch outputs.
         unconditional_batch_skip_range (`int`, defaults to `5`):
             Process the unconditional branch every `N` iterations. If this is set to `N`, the unconditional branch
-            computation will be skipped `N - 1` times (i.e., cached unconditional branch states will be re-used) before
+            computation will be skipped `N - 1` times (i.e., cached unconditional branch states will be reused) before
             computing the new unconditional branch states again.
         unconditional_batch_timestep_skip_range (`Tuple[float, float]`, defaults to `(-1, 641)`):
             The timestep range within which the unconditional branch computation can be skipped without a significant

src/diffusers/hooks/pyramid_attention_broadcast.py

@@ -45,15 +45,15 @@ class PyramidAttentionBroadcastConfig:
         spatial_attention_block_skip_range (`int`, *optional*, defaults to `None`):
             The number of times a specific spatial attention broadcast is skipped before computing the attention states
             to re-use. If this is set to the value `N`, the attention computation will be skipped `N - 1` times (i.e.,
-            old attention states will be re-used) before computing the new attention states again.
+            old attention states will be reused) before computing the new attention states again.
         temporal_attention_block_skip_range (`int`, *optional*, defaults to `None`):
             The number of times a specific temporal attention broadcast is skipped before computing the attention
             states to re-use. If this is set to the value `N`, the attention computation will be skipped `N - 1` times
-            (i.e., old attention states will be re-used) before computing the new attention states again.
+            (i.e., old attention states will be reused) before computing the new attention states again.
         cross_attention_block_skip_range (`int`, *optional*, defaults to `None`):
             The number of times a specific cross-attention broadcast is skipped before computing the attention states
             to re-use. If this is set to the value `N`, the attention computation will be skipped `N - 1` times (i.e.,
-            old attention states will be re-used) before computing the new attention states again.
+            old attention states will be reused) before computing the new attention states again.
         spatial_attention_timestep_skip_range (`Tuple[int, int]`, defaults to `(100, 800)`):
             The range of timesteps to skip in the spatial attention layer. The attention computations will be
             conditionally skipped if the current timestep is within the specified range.
@@ -305,7 +305,7 @@ def _apply_pyramid_attention_broadcast_hook(
         block_skip_range (`int`):
             The number of times a specific attention broadcast is skipped before computing the attention states to
             re-use. If this is set to the value `N`, the attention computation will be skipped `N - 1` times (i.e., old
-            attention states will be re-used) before computing the new attention states again.
+            attention states will be reused) before computing the new attention states again.
         current_timestep_callback (`Callable[[], int]`):
             A callback function that returns the current inference timestep.
     """

src/diffusers/modular_pipelines/flux/denoise.py

@@ -220,7 +220,7 @@ def description(self) -> str:
         return (
             "Denoise step that iteratively denoise the latents. \n"
             "Its loop logic is defined in `FluxDenoiseLoopWrapper.__call__` method \n"
-            "At each iteration, it runs blocks defined in `sub_blocks` sequencially:\n"
+            "At each iteration, it runs blocks defined in `sub_blocks` sequentially:\n"
             " - `FluxLoopDenoiser`\n"
             " - `FluxLoopAfterDenoiser`\n"
             "This block supports both text2image and img2img tasks."

src/diffusers/modular_pipelines/modular_pipeline.py

@@ -229,7 +229,7 @@ class ModularPipelineBlocks(ConfigMixin, PushToHubMixin):
     Base class for all Pipeline Blocks: PipelineBlock, AutoPipelineBlocks, SequentialPipelineBlocks,
     LoopSequentialPipelineBlocks
 
-    [`ModularPipelineBlocks`] provides method to load and save the defination of pipeline blocks.
+    [`ModularPipelineBlocks`] provides method to load and save the definition of pipeline blocks.
 
     <Tip warning={true}>
 
@@ -1418,7 +1418,7 @@ def set_progress_bar_config(self, **kwargs):
 # YiYi TODO:
 # 1. look into the serialization of modular_model_index.json, make sure the items are properly ordered like model_index.json (currently a mess)
 # 2. do we need ConfigSpec? the are basically just key/val kwargs
-# 3. imnprove docstring and potentially add validator for methods where we accpet kwargs to be passed to from_pretrained/save_pretrained/load_components()
+# 3. imnprove docstring and potentially add validator for methods where we accept kwargs to be passed to from_pretrained/save_pretrained/load_components()
 class ModularPipeline(ConfigMixin, PushToHubMixin):
     """
     Base class for all Modular pipelines.

src/diffusers/modular_pipelines/node_utils.py

@@ -384,14 +384,14 @@ def __init__(self, blocks, category=DEFAULT_CATEGORY, label=None, **kwargs):
         # pass or create a default param dict for each input
         # e.g. for prompt,
         #       prompt = {
-        #               "name": "text_input", # the name of the input in node defination, could be different from the input name in diffusers
+        #               "name": "text_input", # the name of the input in node definition, could be different from the input name in diffusers
         #               "label": "Prompt",
         #               "type": "string",
         #               "default": "a bear sitting in a chair drinking a milkshake",
         #               "display": "textarea"}
         # if type is not specified, it'll be a "custom" param of its own type
         # e.g. you can pass ModularNode(scheduler = {name :"scheduler"})
-        #  it will get this spec in node defination {"scheduler": {"label": "Scheduler", "type": "scheduler", "display": "input"}}
+        #  it will get this spec in node definition {"scheduler": {"label": "Scheduler", "type": "scheduler", "display": "input"}}
         #  name can be a dict, in that case, it is part of a "dict" input in mellon nodes, e.g. text_encoder= {name: {"text_encoders": "text_encoder"}}
         inputs = self.blocks.inputs + self.blocks.intermediate_inputs
         for inp in inputs:

src/diffusers/modular_pipelines/stable_diffusion_xl/denoise.py

@@ -695,7 +695,7 @@ def description(self) -> str:
         return (
             "Denoise step that iteratively denoise the latents. \n"
             "Its loop logic is defined in `StableDiffusionXLDenoiseLoopWrapper.__call__` method \n"
-            "At each iteration, it runs blocks defined in `sub_blocks` sequencially:\n"
+            "At each iteration, it runs blocks defined in `sub_blocks` sequentially:\n"
             " - `StableDiffusionXLLoopBeforeDenoiser`\n"
             " - `StableDiffusionXLLoopDenoiser`\n"
             " - `StableDiffusionXLLoopAfterDenoiser`\n"
@@ -717,7 +717,7 @@ def description(self) -> str:
         return (
             "Denoise step that iteratively denoise the latents with controlnet. \n"
             "Its loop logic is defined in  `StableDiffusionXLDenoiseLoopWrapper.__call__` method \n"
-            "At each iteration, it runs blocks defined in `sub_blocks` sequencially:\n"
+            "At each iteration, it runs blocks defined in `sub_blocks` sequentially:\n"
             " - `StableDiffusionXLLoopBeforeDenoiser`\n"
             " - `StableDiffusionXLControlNetLoopDenoiser`\n"
             " - `StableDiffusionXLLoopAfterDenoiser`\n"
@@ -739,7 +739,7 @@ def description(self) -> str:
         return (
             "Denoise step that iteratively denoise the latents(for inpainting task only). \n"
             "Its loop logic is defined in `StableDiffusionXLDenoiseLoopWrapper.__call__` method \n"
-            "At each iteration, it runs blocks defined in `sub_blocks` sequencially:\n"
+            "At each iteration, it runs blocks defined in `sub_blocks` sequentially:\n"
             " - `StableDiffusionXLInpaintLoopBeforeDenoiser`\n"
             " - `StableDiffusionXLLoopDenoiser`\n"
             " - `StableDiffusionXLInpaintLoopAfterDenoiser`\n"
@@ -761,7 +761,7 @@ def description(self) -> str:
         return (
             "Denoise step that iteratively denoise the latents(for inpainting task only) with controlnet. \n"
             "Its loop logic is defined in `StableDiffusionXLDenoiseLoopWrapper.__call__` method \n"
-            "At each iteration, it runs blocks defined in `sub_blocks` sequencially:\n"
+            "At each iteration, it runs blocks defined in `sub_blocks` sequentially:\n"
             " - `StableDiffusionXLInpaintLoopBeforeDenoiser`\n"
             " - `StableDiffusionXLControlNetLoopDenoiser`\n"
             " - `StableDiffusionXLInpaintLoopAfterDenoiser`\n"

src/diffusers/modular_pipelines/wan/denoise.py

@@ -253,7 +253,7 @@ def description(self) -> str:
         return (
             "Denoise step that iteratively denoise the latents. \n"
             "Its loop logic is defined in `WanDenoiseLoopWrapper.__call__` method \n"
-            "At each iteration, it runs blocks defined in `sub_blocks` sequencially:\n"
+            "At each iteration, it runs blocks defined in `sub_blocks` sequentially:\n"
             " - `WanLoopDenoiser`\n"
             " - `WanLoopAfterDenoiser`\n"
             "This block supports both text2vid tasks."

src/diffusers/pipelines/pipeline_loading_utils.py

@@ -613,7 +613,7 @@ def _assign_components_to_devices(
 
 
 def _get_final_device_map(device_map, pipeline_class, passed_class_obj, init_dict, library, max_memory, **kwargs):
-    # TODO: seperate out different device_map methods when it gets to it.
+    # TODO: separate out different device_map methods when it gets to it.
     if device_map != "balanced":
         return device_map
     # To avoid circular import problem.