add `reduce_sum` op into amp black list (#33960)

* reduce sum op default fp32, add into amp black list

* reduce_sum default fp32 can avoid return inf when the sum value large than 65504

python/paddle/fluid/contrib/mixed_precision/fp16_lists.py

@@ -99,6 +99,8 @@ black_list = {
     # fp16 is slower than fp32, though fp16 is supported.
     'lookup_table',
     'lookup_table_v2',
+    # default fp32 can avoid return inf when the sum value large than 65504
+    'reduce_sum',
 }
 
 # This set contains two types of ops. All ops supported fp16 calculation. One 

python/paddle/fluid/dygraph/amp/auto_cast.py

@@ -47,6 +47,8 @@ BLACK_LIST = {
     'sigmoid_cross_entropy_with_logits',
     'cross_entropy',
     'cross_entropy2',
+    # default fp32 can avoid return inf when the sum value large than 65504
+    'reduce_sum',
 }
 
 AMP_RELATED_FLAGS = [