Fix #2941 (#4109)

* Fix of issue #2941

Reshaped score array to avoid `numpy` ValueError.

* Update src/transformers/pipelines.py

* Update src/transformers/pipelines.py
Co-authored-by: Julien Chaumond <chaumond@gmail.com>

src/transformers/pipelines.py

@@ -656,8 +656,8 @@ class TextClassificationPipeline(Pipeline):
 
     def __call__(self, *args, **kwargs):
         outputs = super().__call__(*args, **kwargs)
-        scores = np.exp(outputs) / np.exp(outputs).sum(-1)
-        return [{"label": self.model.config.id2label[item.argmax()], "score": item.max()} for item in scores]
+        scores = np.exp(outputs) / np.exp(outputs).sum(-1, keepdims=True)
+        return [{"label": self.model.config.id2label[item.argmax()], "score": item.max().item()} for item in scores]
 
 
 class FillMaskPipeline(Pipeline):