Support 0 dimension tensor modification (#7250)

ppdet/modeling/tests/test_yolov3_loss.py

@@ -356,10 +356,7 @@ class TestYolov3LossOp(unittest.TestCase):
             x, t, gtbox, anchor, self.downsample_ratio, self.scale_x_y)
         for k in yolo_loss2:
             self.assertAlmostEqual(
-                yolo_loss1[k].numpy()[0],
-                yolo_loss2[k].numpy()[0],
-                delta=1e-2,
-                msg=k)
+                float(yolo_loss1[k]), float(yolo_loss2[k]), delta=1e-2, msg=k)
 
 
 class TestYolov3LossNoGTScore(TestYolov3LossOp):