diff --git a/python/paddle/fluid/tests/unittests/test_adamw_op.py b/python/paddle/fluid/tests/unittests/test_adamw_op.py
index 0a60f4cba09bc6fd8314420c19522533e724ec37..dbeb5a430377f7199ca30220114676651cf530a2 100644
--- a/python/paddle/fluid/tests/unittests/test_adamw_op.py
+++ b/python/paddle/fluid/tests/unittests/test_adamw_op.py
@@ -333,7 +333,7 @@ class TestAdamWOpLayerwiseLR(TestAdamWOp):
             lr_ratio=simple_lr_fun)
 
         loss_ref = np.array(
-            [4.8383293, 3.0854003, 1.33299, -0.418993, -2.171043])
+            [4.8383293, 3.084947, 1.3323904, -0.41943002, -2.1710064])
         for i in range(5):
             a1 = linear1(a)
             out = linear2(a1)
diff --git a/python/paddle/optimizer/adamw.py b/python/paddle/optimizer/adamw.py
index 55aaac8dc4852469a167d7ae0dd65a0d4fa22bc0..5fdcc0cd0d27069ea0d8b1c788dc38ca2b739ec1 100644
--- a/python/paddle/optimizer/adamw.py
+++ b/python/paddle/optimizer/adamw.py
@@ -297,9 +297,8 @@ class AdamW(Adam):
                 moment1, moment2, beta1_pow_acc, beta2_pow_acc, master_weight,
                 'epsilon', self._epsilon, 'lazy_mode', self._lazy_mode,
                 'min_row_size_to_use_multithread', 1000, 'beta1', _beta1,
-                'beta2', _beta2, 'coeff', self._coeff, 'multi_precision',
-                find_master, 'lr_ratio', lr_ratio_)
-
+                'beta2', _beta2, "with_decay", with_decay, 'coeff', self._coeff,
+                'multi_precision', find_master, 'lr_ratio', lr_ratio_)
             return None
 
         inputs = {