Fix typos in strings and comments

examples/cogvideo/train_cogvideox_image_to_video_lora.py

@@ -555,7 +555,7 @@ def _load_dataset_from_local_path(self):
 
         if any(not path.is_file() for path in instance_videos):
             raise ValueError(
-                "Expected '--video_column' to be a path to a file in `--instance_data_root` containing line-separated paths to video data but found atleast one path that is not a valid file."
+                "Expected '--video_column' to be a path to a file in `--instance_data_root` containing line-separated paths to video data but found at least one path that is not a valid file."
             )
 
         return instance_prompts, instance_videos

examples/cogvideo/train_cogvideox_lora.py

@@ -539,7 +539,7 @@ def _load_dataset_from_local_path(self):
 
         if any(not path.is_file() for path in instance_videos):
             raise ValueError(
-                "Expected '--video_column' to be a path to a file in `--instance_data_root` containing line-separated paths to video data but found atleast one path that is not a valid file."
+                "Expected '--video_column' to be a path to a file in `--instance_data_root` containing line-separated paths to video data but found at least one path that is not a valid file."
             )
 
         return instance_prompts, instance_videos

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://arxiv.org/abs/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://arxiv.org/abs/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 inpainting 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))
 
@@ -73,7 +73,7 @@ accelerate launch train_multi_subject_dreambooth_inpaint.py \
 
 ## 3. Results
 
-A [![Weights & Biases](https://img.shields.io/badge/Weights%20&%20Biases-Report-blue)](https://wandb.ai/gzguevara/uncategorized/reports/Multi-Subject-Dreambooth-for-Inpainting--Vmlldzo2MzY5NDQ4?accessToken=y0nya2d7baguhbryxaikbfr1203amvn1jsmyl07vk122mrs7tnph037u1nqgse8t) is provided showing the training progress by every 50 steps. Note, the reported  weights & baises run was performed on a A100 GPU with the following stetting:
+A [![Weights & Biases](https://img.shields.io/badge/Weights%20&%20Biases-Report-blue)](https://wandb.ai/gzguevara/uncategorized/reports/Multi-Subject-Dreambooth-for-Inpainting--Vmlldzo2MzY5NDQ4?accessToken=y0nya2d7baguhbryxaikbfr1203amvn1jsmyl07vk122mrs7tnph037u1nqgse8t) is provided showing the training progress by every 50 steps. Note, the reported weights & biases run was performed on a A100 GPU with the following stetting:
 
 ```bash
 accelerate launch train_multi_subject_dreambooth_inpaint.py \

src/diffusers/hooks/faster_cache.py

@@ -146,7 +146,7 @@ class FasterCacheConfig:
     alpha_low_frequency: float = 1.1
     alpha_high_frequency: float = 1.1
 
-    # n as described in CFG-Cache explanation in the paper - dependant on the model
+    # n as described in CFG-Cache explanation in the paper - dependent on the model
     unconditional_batch_skip_range: int = 5
     unconditional_batch_timestep_skip_range: Tuple[int, int] = (-1, 641)
 

src/diffusers/hooks/hooks.py

@@ -43,9 +43,9 @@ def initialize_hook(self, module: torch.nn.Module) -> torch.nn.Module:
         """
         return module
 
-    def deinitalize_hook(self, module: torch.nn.Module) -> torch.nn.Module:
+    def deinitialize_hook(self, module: torch.nn.Module) -> torch.nn.Module:
         r"""
-        Hook that is executed when a model is deinitalized.
+        Hook that is executed when a model is deinitialized.
 
         Args:
             module (`torch.nn.Module`):
@@ -192,7 +192,7 @@ def remove_hook(self, name: str, recurse: bool = True) -> None:
             else:
                 self._fn_refs[index + 1].forward = old_forward
 
-            self._module_ref = hook.deinitalize_hook(self._module_ref)
+            self._module_ref = hook.deinitialize_hook(self._module_ref)
             del self.hooks[name]
             self._hook_order.pop(index)
             self._fn_refs.pop(index)

src/diffusers/hooks/layerwise_casting.py

@@ -60,9 +60,9 @@ def initialize_hook(self, module: torch.nn.Module):
         module.to(dtype=self.storage_dtype, non_blocking=self.non_blocking)
         return module
 
-    def deinitalize_hook(self, module: torch.nn.Module):
+    def deinitialize_hook(self, module: torch.nn.Module):
         raise NotImplementedError(
-            "LayerwiseCastingHook does not support deinitalization. A model once enabled with layerwise casting will "
+            "LayerwiseCastingHook does not support deinitialization. A model once enabled with layerwise casting will "
             "have casted its weights to a lower precision dtype for storage. Casting this back to the original dtype "
             "will lead to precision loss, which might have an impact on the model's generation quality. The model should "
             "be re-initialized and loaded in the original dtype."

src/diffusers/loaders/peft.py

@@ -250,7 +250,7 @@ def load_lora_adapter(
 
             rank = {}
             for key, val in state_dict.items():
-                # Cannot figure out rank from lora layers that don't have atleast 2 dimensions.
+                # Cannot figure out rank from lora layers that don't have at least 2 dimensions.
                 # Bias layers in LoRA only have a single dimension
                 if "lora_B" in key and val.ndim > 1:
                     # Check out https://github.com/huggingface/peft/pull/2419 for the `^` symbol.

src/diffusers/models/autoencoders/autoencoder_kl.py

@@ -63,7 +63,7 @@ class AutoencoderKL(ModelMixin, ConfigMixin, FromOriginalModelMixin, PeftAdapter
             Synthesis with Latent Diffusion Models](https://arxiv.org/abs/2112.10752) paper.
         force_upcast (`bool`, *optional*, default to `True`):
             If enabled it will force the VAE to run in float32 for high image resolution pipelines, such as SD-XL. VAE
-            can be fine-tuned / trained to a lower range without loosing too much precision in which case
+            can be fine-tuned / trained to a lower range without losing too much precision in which case
             `force_upcast` can be set to `False` - see: https://huggingface.co/madebyollin/sdxl-vae-fp16-fix
         mid_block_add_attention (`bool`, *optional*, default to `True`):
             If enabled, the mid_block of the Encoder and Decoder will have attention blocks. If set to false, the

src/diffusers/models/autoencoders/autoencoder_kl_allegro.py

@@ -715,7 +715,7 @@ class AutoencoderKLAllegro(ModelMixin, ConfigMixin):
             Synthesis with Latent Diffusion Models](https://arxiv.org/abs/2112.10752) paper.
         force_upcast (`bool`, default to `True`):
             If enabled it will force the VAE to run in float32 for high image resolution pipelines, such as SD-XL. VAE
-            can be fine-tuned / trained to a lower range without loosing too much precision in which case
+            can be fine-tuned / trained to a lower range without losing too much precision in which case
             `force_upcast` can be set to `False` - see: https://huggingface.co/madebyollin/sdxl-vae-fp16-fix
     """
 

src/diffusers/models/autoencoders/autoencoder_kl_cogvideox.py

@@ -983,7 +983,7 @@ class AutoencoderKLCogVideoX(ModelMixin, ConfigMixin, FromOriginalModelMixin):
             Synthesis with Latent Diffusion Models](https://arxiv.org/abs/2112.10752) paper.
         force_upcast (`bool`, *optional*, default to `True`):
             If enabled it will force the VAE to run in float32 for high image resolution pipelines, such as SD-XL. VAE
-            can be fine-tuned / trained to a lower range without loosing too much precision in which case
+            can be fine-tuned / trained to a lower range without losing too much precision in which case
             `force_upcast` can be set to `False` - see: https://huggingface.co/madebyollin/sdxl-vae-fp16-fix
     """
 

src/diffusers/models/autoencoders/autoencoder_kl_temporal_decoder.py

@@ -161,7 +161,7 @@ class AutoencoderKLTemporalDecoder(ModelMixin, ConfigMixin):
             Synthesis with Latent Diffusion Models](https://arxiv.org/abs/2112.10752) paper.
         force_upcast (`bool`, *optional*, default to `True`):
             If enabled it will force the VAE to run in float32 for high image resolution pipelines, such as SD-XL. VAE
-            can be fine-tuned / trained to a lower range without loosing too much precision in which case
+            can be fine-tuned / trained to a lower range without losing too much precision in which case
             `force_upcast` can be set to `False` - see: https://huggingface.co/madebyollin/sdxl-vae-fp16-fix
     """
 

src/diffusers/pipelines/consisid/consisid_utils.py

@@ -166,7 +166,7 @@ def process_face_embeddings(
         raise RuntimeError("facexlib align face fail")
     align_face = face_helper_1.cropped_faces[0]  # (512, 512, 3)  # RGB
 
-    # incase insightface didn't detect face
+    # in case insightface didn't detect face
     if id_ante_embedding is None:
         logger.warning("Failed to detect face using insightface. Extracting embedding with align face")
         id_ante_embedding = face_helper_2.get_feat(align_face)

tests/lora/utils.py

@@ -1090,7 +1090,7 @@ def test_simple_inference_with_text_denoiser_block_scale(self):
     def test_simple_inference_with_text_denoiser_multi_adapter_block_lora(self):
         """
         Tests a simple inference with lora attached to text encoder and unet, attaches
-        multiple adapters and set differnt weights for different blocks (i.e. block lora)
+        multiple adapters and set different weights for different blocks (i.e. block lora)
         """
         for scheduler_cls in self.scheduler_classes:
             components, text_lora_config, denoiser_lora_config = self.get_dummy_components(scheduler_cls)
@@ -1638,7 +1638,7 @@ def test_simple_inference_with_text_lora_denoiser_fused_multi(
 
             pipe.fuse_lora(components=self.pipeline_class._lora_loadable_modules, adapter_names=["adapter-1"])
 
-            # Fusing should still keep the LoRA layers so outpout should remain the same
+            # Fusing should still keep the LoRA layers so output should remain the same
             outputs_lora_1_fused = pipe(**inputs, generator=torch.manual_seed(0))[0]
 
             self.assertTrue(

tests/pipelines/cogvideo/test_cogvideox_image2video.py

@@ -270,7 +270,7 @@ def test_vae_tiling(self, expected_diff_max: float = 0.3):
         generator_device = "cpu"
         components = self.get_dummy_components()
 
-        # The reason to modify it this way is because I2V Transformer limits the generation to resolutions used during initalization.
+        # The reason to modify it this way is because I2V Transformer limits the generation to resolutions used during initialization.
         # This limitation comes from using learned positional embeddings which cannot be generated on-the-fly like sincos or RoPE embeddings.
         # See the if-statement on "self.use_learned_positional_embeddings" in diffusers/models/embeddings.py
         components["transformer"] = CogVideoXTransformer3DModel.from_config(

tests/pipelines/consisid/test_consisid.py

@@ -279,7 +279,7 @@ def test_vae_tiling(self, expected_diff_max: float = 0.4):
         generator_device = "cpu"
         components = self.get_dummy_components()
 
-        # The reason to modify it this way is because ConsisID Transformer limits the generation to resolutions used during initalization.
+        # The reason to modify it this way is because ConsisID Transformer limits the generation to resolutions used during initialization.
         # This limitation comes from using learned positional embeddings which cannot be generated on-the-fly like sincos or RoPE embeddings.
         # See the if-statement on "self.use_learned_positional_embeddings" in diffusers/models/embeddings.py
         components["transformer"] = ConsisIDTransformer3DModel.from_config(

tests/pipelines/kolors/test_kolors_img2img.py

@@ -155,6 +155,6 @@ def test_inference_batch_single_identical(self):
     def test_float16_inference(self):
         super().test_float16_inference(expected_max_diff=7e-2)
 
-    @unittest.skip("Test not supported because kolors img2img doesn't take pooled embeds as inputs unline kolors t2i.")
+    @unittest.skip("Test not supported because kolors img2img doesn't take pooled embeds as inputs unlike kolors t2i.")
     def test_encode_prompt_works_in_isolation(self):
         pass

tests/pipelines/pag/test_pag_pixart_sigma.py

@@ -254,7 +254,7 @@ def test_attention_slicing_forward_pass(
             assert_mean_pixel_difference(to_np(output_with_slicing1[0]), to_np(output_without_slicing[0]))
             assert_mean_pixel_difference(to_np(output_with_slicing2[0]), to_np(output_without_slicing[0]))
 
-    # Because we have `pag_applied_layers` we cannot direcly apply
+    # Because we have `pag_applied_layers` we cannot directly apply
     # `set_default_attn_processor`
     def test_dict_tuple_outputs_equivalent(self, expected_slice=None, expected_max_difference=1e-4):
         components = self.get_dummy_components()

tests/pipelines/stable_unclip/test_stable_unclip.py

@@ -217,7 +217,7 @@ def test_stable_unclip(self):
         pipe.enable_sequential_cpu_offload()
 
         generator = torch.Generator(device="cpu").manual_seed(0)
-        output = pipe("anime turle", generator=generator, output_type="np")
+        output = pipe("anime turtle", generator=generator, output_type="np")
 
         image = output.images[0]
 

tests/pipelines/stable_unclip/test_stable_unclip_img2img.py

@@ -246,7 +246,7 @@ def test_stable_unclip_l_img2img(self):
         pipe.enable_sequential_cpu_offload()
 
         generator = torch.Generator(device="cpu").manual_seed(0)
-        output = pipe(input_image, "anime turle", generator=generator, output_type="np")
+        output = pipe(input_image, "anime turtle", generator=generator, output_type="np")
 
         image = output.images[0]
 
@@ -273,7 +273,7 @@ def test_stable_unclip_h_img2img(self):
         pipe.enable_sequential_cpu_offload()
 
         generator = torch.Generator(device="cpu").manual_seed(0)
-        output = pipe(input_image, "anime turle", generator=generator, output_type="np")
+        output = pipe(input_image, "anime turtle", generator=generator, output_type="np")
 
         image = output.images[0]
 

tests/pipelines/test_pipelines_common.py

@@ -2096,11 +2096,11 @@ def test_encode_prompt_works_in_isolation(self, extra_required_param_value_dict=
         with torch.no_grad():
             encoded_prompt_outputs = pipe_with_just_text_encoder.encode_prompt(**encode_prompt_inputs)
 
-        # Programatically determine the reutrn names of `encode_prompt.`
-        ast_vistor = ReturnNameVisitor()
-        encode_prompt_tree = ast_vistor.get_ast_tree(cls=self.pipeline_class)
-        ast_vistor.visit(encode_prompt_tree)
-        prompt_embed_kwargs = ast_vistor.return_names
+        # Programmatically determine the return names of `encode_prompt.`
+        ast_visitor = ReturnNameVisitor()
+        encode_prompt_tree = ast_visitor.get_ast_tree(cls=self.pipeline_class)
+        ast_visitor.visit(encode_prompt_tree)
+        prompt_embed_kwargs = ast_visitor.return_names
         prompt_embeds_kwargs = dict(zip(prompt_embed_kwargs, encoded_prompt_outputs))
 
         # Pack the outputs of `encode_prompt`.

tests/quantization/bnb/test_4bit.py

@@ -205,7 +205,7 @@ def test_model_memory_usage(self):
 
     def test_original_dtype(self):
         r"""
-        A simple test to check if the model succesfully stores the original dtype
+        A simple test to check if the model successfully stores the original dtype
         """
         self.assertTrue("_pre_quantization_dtype" in self.model_4bit.config)
         self.assertFalse("_pre_quantization_dtype" in self.model_fp16.config)

tests/quantization/bnb/test_mixed_int8.py

@@ -195,7 +195,7 @@ def test_model_memory_usage(self):
 
     def test_original_dtype(self):
         r"""
-        A simple test to check if the model succesfully stores the original dtype
+        A simple test to check if the model successfully stores the original dtype
         """
         self.assertTrue("_pre_quantization_dtype" in self.model_8bit.config)
         self.assertFalse("_pre_quantization_dtype" in self.model_fp16.config)

tests/single_file/test_model_autoencoder_dc_single_file.py

@@ -95,7 +95,7 @@ def test_single_file_in_type_variant_components(self):
         # `in` variant checkpoints require passing in a `config` parameter
         # in order to set the scaling factor correctly.
         # `in` and `mix` variants have the same keys and we cannot automatically infer a scaling factor.
-        # We default to using teh `mix` config
+        # We default to using the `mix` config
         repo_id = "mit-han-lab/dc-ae-f128c512-in-1.0-diffusers"
         ckpt_path = "https://huggingface.co/mit-han-lab/dc-ae-f128c512-in-1.0/blob/main/model.safetensors"
 

utils/custom_init_isort.py

@@ -252,7 +252,7 @@ def sort_imports(file: str, check_only: bool = True):
         code, start_prompt="_import_structure = {", end_prompt="if TYPE_CHECKING:"
     )
 
-    # We ignore block 0 (everything untils start_prompt) and the last block (everything after end_prompt).
+    # We ignore block 0 (everything until start_prompt) and the last block (everything after end_prompt).
     for block_idx in range(1, len(main_blocks) - 1):
         # Check if the block contains some `_import_structure`s thingy to sort.
         block = main_blocks[block_idx]