[AMP OP&Test]Modify the FP16 and BF16 OpTest of Add_N (#52311)

* adjust defalut tolerance of output and grad

* fix a bug in the grad of OpTest

* fix the type of setting defalut value in optest, both forward and
backward

* add defalut

* fix test_sum_op

* fix test_sum_op test for testing add_n

* modify the add_n op_test

python/paddle/fluid/tests/unittests/test_sum_op.py

@@ -297,14 +297,14 @@ class TestFP16SumOp(TestSumOp):
     def test_check_output(self):
         place = core.CUDAPlace(0)
         if core.is_float16_supported(place):
-            self.check_output_with_place(place, atol=2e-2)
+            self.check_output_with_place(place)
 
     # FIXME: Because of the precision fp16, max_relative_error
     # should be 0.15 here.
     def test_check_grad(self):
         place = core.CUDAPlace(0)
         if core.is_float16_supported(place):
-            self.check_grad(['x0'], 'Out', max_relative_error=0.15)
+            self.check_grad(['x0'], 'Out')
 
 
 def create_test_sum_fp16_class(parent):