Fix more typos (#40627)

Fix typos

Signed-off-by: Yuanyuan Chen <[email protected]>

Author: cyyever

docs/source/en/kv_cache.md

@@ -102,7 +102,7 @@ You may want to consider offloading if you have a small GPU and you're getting o
 Offloading is available for both [`DynamicCache`] and [`StaticCache`]. You can enable it by configuring `cache_implementation="offloaded"` for the dynamic version, or `cache_implementation="offloaded_static"` for the static version, in either [`GenerationConfig`] or [`~GenerationMixin.generate`].
 Additionally, you can also instantiate your own [`DynamicCache`] or [`StaticCache`] with the `offloading=True` option, and pass this cache in `generate` or your model's `forward` (for example, `past_key_values=DynamicCache(config=model.config, offloading=True)` for a dynamic cache).
 
-Note that the 2 [`Cache`] classes mentionned above have an additional option when instantiating them directly, `offload_only_non_sliding`.
+Note that the 2 [`Cache`] classes mentioned above have an additional option when instantiating them directly, `offload_only_non_sliding`.
 This additional argument decides if the layers using sliding window/chunk attention (if any), will be offloaded as well. Since
 these layers are usually short anyway, it may be better to avoid offloading them, as offloading may incur a speed penalty. By default, this option is `False` for [`DynamicCache`], and `True` for [`StaticCache`].
 

docs/source/en/model_doc/gptsan-japanese.md

@@ -50,7 +50,7 @@ The `generate()` method can be used to generate text using GPTSAN-Japanese model
 >>> model = AutoModel.from_pretrained("Tanrei/GPTSAN-japanese").to(device)
 >>> x_tok = tokenizer("は、", prefix_text="織田信長", return_tensors="pt")
 >>> torch.manual_seed(0)
->>> gen_tok = model.generate(x_tok.input_ids.to(model.device), token_type_ids=x_tok.token_type_ids.to(mdoel.device), max_new_tokens=20)
+>>> gen_tok = model.generate(x_tok.input_ids.to(model.device), token_type_ids=x_tok.token_type_ids.to(model.device), max_new_tokens=20)
 >>> tokenizer.decode(gen_tok[0])
 '織田信長は、2004年に『戦国BASARA』のために、豊臣秀吉'
 ```

docs/source/en/tasks/visual_document_retrieval.md

@@ -79,7 +79,7 @@ Index the images offline, and during inference, return the query text embeddings
 Store the image and image embeddings by writing them to the dataset with [`~datasets.Dataset.map`] as shown below. Add an `embeddings` column that contains the indexed embeddings. ColPali embeddings take up a lot of storage, so remove them from the accelerator and store them in the CPU as NumPy vectors.
 
 ```python
-ds_with_embeddings = dataset.map(lambda example: {'embeddings': model(**processor(images=example["image"]).to(devide), return_tensors="pt").embeddings.to(torch.float32).detach().cpu().numpy()})
+ds_with_embeddings = dataset.map(lambda example: {'embeddings': model(**processor(images=example["image"]).to(device), return_tensors="pt").embeddings.to(torch.float32).detach().cpu().numpy()})
 ```
 
 For online inference, create a function to search the image embeddings in batches and retrieve the k-most relevant images. The function below returns the indices in the dataset and their scores for a given indexed dataset, text embeddings, number of top results, and the batch size.

examples/modular-transformers/modeling_new_task_model.py

@@ -232,7 +232,7 @@ def get_placeholder_mask(
         self, input_ids: torch.LongTensor, inputs_embeds: torch.FloatTensor, image_features: torch.FloatTensor
     ):
         """
-        Obtains multimodal placeholdr mask from `input_ids` or `inputs_embeds`, and checks that the placeholder token count is
+        Obtains multimodal placeholder mask from `input_ids` or `inputs_embeds`, and checks that the placeholder token count is
         equal to the length of multimodal features. If the lengths are different, an error is raised.
         """
         if input_ids is None:
@@ -406,7 +406,7 @@ def get_decoder(self):
     def get_image_features(self, pixel_values):
         return self.model.get_image_features(pixel_values)
 
-    # Make modules available throught conditional class for BC
+    # Make modules available through conditional class for BC
     @property
     def language_model(self):
         return self.model.language_model