[docs] add xpu device check (#34684)

* add XPU path

* use accelerate API

* Update docs/source/en/tasks/semantic_segmentation.md

Co-authored-by: Steven Liu <59462357+stevhliu@users.noreply.github.com>

* update more places with accelerate API

---------

Co-authored-by: Steven Liu <59462357+stevhliu@users.noreply.github.com>

docs/source/en/generation_strategies.md

@@ -508,10 +508,11 @@ See the following examples for DoLa decoding with the 32-layer LLaMA-7B model.
 ```python
 >>> from transformers import AutoTokenizer, AutoModelForCausalLM, set_seed
 >>> import torch
+>>> from accelerate.test_utils.testing import get_backend
 
 >>> tokenizer = AutoTokenizer.from_pretrained("huggyllama/llama-7b")
 >>> model = AutoModelForCausalLM.from_pretrained("huggyllama/llama-7b", torch_dtype=torch.float16)
->>> device = 'cuda' if torch.cuda.is_available() else 'cpu'
+>>> device, _, _ = get_backend() # automatically detects the underlying device type (CUDA, CPU, XPU, MPS, etc.)
 >>> model.to(device)
 >>> set_seed(42)
 

docs/source/en/tasks/idefics.md

@@ -386,9 +386,9 @@ The use and prompting for the conversational use is very similar to using the ba
 ```py
 >>> import torch
 >>> from transformers import IdeficsForVisionText2Text, AutoProcessor
+>>> from accelerate.test_utils.testing import get_backend
 
->>> device = "cuda" if torch.cuda.is_available() else "cpu"
-
+>>> device, _, _ = get_backend() # automatically detects the underlying device type (CUDA, CPU, XPU, MPS, etc.)
 >>> checkpoint = "HuggingFaceM4/idefics-9b-instruct"
 >>> model = IdeficsForVisionText2Text.from_pretrained(checkpoint, torch_dtype=torch.bfloat16).to(device)
 >>> processor = AutoProcessor.from_pretrained(checkpoint)

docs/source/en/tasks/image_captioning.md

@@ -256,8 +256,9 @@ image
 Prepare image for the model.
 
 ```python
-device = "cuda" if torch.cuda.is_available() else "cpu"
-
+from accelerate.test_utils.testing import get_backend
+# automatically detects the underlying device type (CUDA, CPU, XPU, MPS, etc.)
+device, _, _ = get_backend()
 inputs = processor(images=image, return_tensors="pt").to(device)
 pixel_values = inputs.pixel_values
 ```

docs/source/en/tasks/image_feature_extraction.md

@@ -43,8 +43,9 @@ Let's see the pipeline in action. First, initialize the pipeline. If you don't p
 ```python
 import torch
 from transformers import pipeline
-
-DEVICE = torch.device('cuda' if torch.cuda.is_available() else 'cpu')
+from accelerate.test_utils.testing import get_backend
+# automatically detects the underlying device type (CUDA, CPU, XPU, MPS, etc.)
+DEVICE, _, _ = get_backend()
 pipe = pipeline(task="image-feature-extraction", model_name="google/vit-base-patch16-384", device=DEVICE, pool=True)
 ```
 

docs/source/en/tasks/image_to_image.md

@@ -37,8 +37,9 @@ We can now initialize the pipeline with a [Swin2SR model](https://huggingface.co
 ```python
 from transformers import pipeline
 import torch
-
-device = torch.device('cuda' if torch.cuda.is_available() else 'cpu')
+from accelerate.test_utils.testing import get_backend
+# automatically detects the underlying device type (CUDA, CPU, XPU, MPS, etc.)
+device, _, _ = get_backend()
 pipe = pipeline(task="image-to-image", model="caidas/swin2SR-lightweight-x2-64", device=device)
 ```
 

docs/source/en/tasks/knowledge_distillation_for_image_classification.md

@@ -58,7 +58,7 @@ from transformers import TrainingArguments, Trainer
 import torch
 import torch.nn as nn
 import torch.nn.functional as F
-
+from accelerate.test_utils.testing import get_backend
 
 class ImageDistilTrainer(Trainer):
     def __init__(self, teacher_model=None, student_model=None, temperature=None, lambda_param=None,  *args, **kwargs):
@@ -66,7 +66,7 @@ class ImageDistilTrainer(Trainer):
         self.teacher = teacher_model
         self.student = student_model
         self.loss_function = nn.KLDivLoss(reduction="batchmean")
-        device = torch.device('cuda' if torch.cuda.is_available() else 'cpu')
+        device, _, _ = get_backend() # automatically detects the underlying device type (CUDA, CPU, XPU, MPS, etc.)
         self.teacher.to(device)
         self.teacher.eval()
         self.temperature = temperature

docs/source/en/tasks/mask_generation.md

@@ -125,9 +125,9 @@ the processor.
 ```python
 from transformers import SamModel, SamProcessor
 import torch
-
-device = torch.device('cuda' if torch.cuda.is_available() else 'cpu')
-
+from accelerate.test_utils.testing import get_backend
+# automatically detects the underlying device type (CUDA, CPU, XPU, MPS, etc.)
+device, _, _ = get_backend()
 model = SamModel.from_pretrained("facebook/sam-vit-base").to(device)
 processor = SamProcessor.from_pretrained("facebook/sam-vit-base")
 ```

docs/source/en/tasks/monocular_depth_estimation.md

@@ -53,8 +53,9 @@ Instantiate a pipeline from a [checkpoint on the Hugging Face Hub](https://huggi
 ```py
 >>> from transformers import pipeline
 >>> import torch
-
->>> device = "cuda" if torch.cuda.is_available() else "cpu"
+>>> from accelerate.test_utils.testing import get_backend
+# automatically detects the underlying device type (CUDA, CPU, XPU, MPS, etc.)
+>>> device, _, _ = get_backend()
 >>> checkpoint = "depth-anything/Depth-Anything-V2-base-hf"
 >>> pipe = pipeline("depth-estimation", model=checkpoint, device=device)
 ```

docs/source/en/tasks/object_detection.md

@@ -1488,7 +1488,9 @@ Now that you have finetuned a model, evaluated it, and uploaded it to the Huggin
 
 Load model and image processor from the Hugging Face Hub (skip to use already trained in this session):
 ```py
->>> device = "cuda"
+>>> from accelerate.test_utils.testing import get_backend
+# automatically detects the underlying device type (CUDA, CPU, XPU, MPS, etc.)
+>>> device, _, _ = get_backend()
 >>> model_repo = "qubvel-hf/detr_finetuned_cppe5"
 
 >>> image_processor = AutoImageProcessor.from_pretrained(model_repo)

docs/source/en/tasks/semantic_segmentation.md

@@ -689,7 +689,9 @@ Reload the dataset and load an image for inference.
 We will now see how to infer without a pipeline. Process the image with an image processor and place the `pixel_values` on a GPU:
 
 ```py
->>> device = torch.device("cuda" if torch.cuda.is_available() else "cpu")  # use GPU if available, otherwise use a CPU
+>>> from accelerate.test_utils.testing import get_backend
+# automatically detects the underlying device type (CUDA, CPU, XPU, MPS, etc.)
+>>> device, _, _ = get_backend()
 >>> encoding = image_processor(image, return_tensors="pt")
 >>> pixel_values = encoding.pixel_values.to(device)
 ```

docs/source/en/tasks/text-to-speech.md

@@ -282,10 +282,10 @@ containing the corresponding speaker embedding.
 >>> import os
 >>> import torch
 >>> from speechbrain.inference.classifiers import EncoderClassifier
+>>> from accelerate.test_utils.testing import get_backend
 
 >>> spk_model_name = "speechbrain/spkrec-xvect-voxceleb"
-
->>> device = "cuda" if torch.cuda.is_available() else "cpu"
+>>> device, _, _ = get_backend() # automatically detects the underlying device type (CUDA, CPU, XPU, MPS, etc.)
 >>> speaker_model = EncoderClassifier.from_hparams(
 ...     source=spk_model_name,
 ...     run_opts={"device": device},

docs/source/en/tasks/visual_question_answering.md

@@ -363,10 +363,11 @@ GPU, if available, which we didn't need to do earlier when training, as [`Traine
 ```py
 >>> from transformers import AutoProcessor, Blip2ForConditionalGeneration
 >>> import torch
+>>> from accelerate.test_utils.testing import get_backend
 
 >>> processor = AutoProcessor.from_pretrained("Salesforce/blip2-opt-2.7b")
 >>> model = Blip2ForConditionalGeneration.from_pretrained("Salesforce/blip2-opt-2.7b", torch_dtype=torch.float16)
->>> device = "cuda" if torch.cuda.is_available() else "cpu"
+>>> device, _, _ = get_backend() # automatically detects the underlying device type (CUDA, CPU, XPU, MPS, etc.)
 >>> model.to(device)
 ```