Fix doctest CI (#21166)

* fix
Co-authored-by: ydshieh <ydshieh@users.noreply.github.com>

src/transformers/utils/doc.py

@@ -256,7 +256,7 @@ PT_SEQUENCE_CLASSIFICATION_SAMPLE = r"""
     >>> with torch.no_grad():
     ...     logits = model(**inputs).logits
 
-    >>> predicted_class_ids = torch.arange(0, logits.shape[-1])[torch.sigmoid(logits).squeeze() > 0.5]
+    >>> predicted_class_ids = torch.arange(0, logits.shape[-1])[torch.sigmoid(logits).squeeze(dim=0) > 0.5]
 
     >>> # To train a model on `num_labels` classes, you can pass `num_labels=num_labels` to `.from_pretrained(...)`
     >>> num_labels = len(model.config.id2label)
@@ -264,7 +264,9 @@ PT_SEQUENCE_CLASSIFICATION_SAMPLE = r"""
     ...     "{checkpoint}", num_labels=num_labels, problem_type="multi_label_classification"
     ... )
 
-    >>> labels = torch.nn.functional.one_hot(torch.tensor(predicted_class_ids), num_classes=num_labels).to(torch.float)
+    >>> labels = torch.sum(
+    ...     torch.nn.functional.one_hot(predicted_class_ids[None, :].clone(), num_classes=num_labels), dim=1
+    ... ).to(torch.float)
     >>> loss = model(**inputs, labels=labels).loss
     ```
 """