Tensor location is already handled (#14224)

in `base.py` not in subclasses.

src/transformers/pipelines/text_classification.py

@@ -145,10 +145,7 @@ class TextClassificationPipeline(Pipeline):
                 function_to_apply = ClassificationFunction.NONE
 
         outputs = model_outputs["logits"][0]
-        if self.framework == "pt":
-            outputs = outputs.cpu().numpy()
-        else:
-            outputs = outputs.numpy()
+        outputs = outputs.numpy()
 
         if function_to_apply == ClassificationFunction.SIGMOID:
             scores = sigmoid(outputs)