diff --git a/python/paddle/fluid/tests/unittests/test_adamw_op.py b/python/paddle/fluid/tests/unittests/test_adamw_op.py
index ce01ca042c123d..fdc3682f4b1bfb 100644
--- a/python/paddle/fluid/tests/unittests/test_adamw_op.py
+++ b/python/paddle/fluid/tests/unittests/test_adamw_op.py
@@ -147,5 +147,33 @@ def test_adamw_op_dygraph(self):
             adam.clear_gradients()
 
 
+class TestAdamWOpGroupWithLR(TestAdamWOp):
+    def test_adamw_op_dygraph(self):
+        paddle.disable_static()
+        value = np.arange(26).reshape(2, 13).astype("float32")
+        a = paddle.to_tensor(value)
+        linear_1 = paddle.nn.Linear(13, 5)
+        linear_2 = paddle.nn.Linear(5, 3)
+        adam = paddle.optimizer.AdamW(
+            learning_rate=paddle.optimizer.lr.PiecewiseDecay(
+                boundaries=[3, 6], values=[0.1, 0.2, 0.3]),
+            parameters=[{
+                'params': linear_1.parameters(),
+                'learning_rate': 0.1,
+            }, {
+                'params': linear_2.parameters(),
+                'weight_decay': 0.001,
+            }],
+            apply_decay_param_fun=lambda name: True,
+            weight_decay=0.01)
+
+        for _ in range(2):
+            out = linear_1(a)
+            out = linear_2(out)
+            out.backward()
+            adam.step()
+            adam.clear_gradients()
+
+
 if __name__ == "__main__":
     unittest.main()
diff --git a/python/paddle/optimizer/adamw.py b/python/paddle/optimizer/adamw.py
index d39de0ae7683f5..f830a9096caa75 100644
--- a/python/paddle/optimizer/adamw.py
+++ b/python/paddle/optimizer/adamw.py
@@ -185,10 +185,9 @@ def _append_decoupled_weight_decay(self, block, param_and_grad):
         Raises:
             Exception: The type of coeff and parameter is not consistent.
         """
-        if not isinstance(param_and_grad, dict):
-            param, grad = param_and_grad
-        else:
-            param, grad = self._update_param_group(param_and_grad)
+        if isinstance(param_and_grad, dict):
+            param_and_grad = self._update_param_group(param_and_grad)
+        param, grad = param_and_grad
 
         if self._apply_decay_param_fun is not None \
                 and not self._apply_decay_param_fun(param.name):
diff --git a/python/paddle/optimizer/optimizer.py b/python/paddle/optimizer/optimizer.py
index bc48adbdc09caf..e76ea622148fb9 100644
--- a/python/paddle/optimizer/optimizer.py
+++ b/python/paddle/optimizer/optimizer.py
@@ -206,7 +206,6 @@ def __init__(self,
         self._param_device_map = dict()
         self.clear_gradients = self.clear_grad
         self._default_dict = {
-            'learning_rate': self._learning_rate,
             'weight_decay': self.regularization,
             'grad_clip': self._grad_clip
         }
@@ -1190,7 +1189,8 @@ def _add_param_group(self, param_group):
             else:
                 regularization = weight_decay
             param.regularizer = regularization
-            param.optimize_attr['learning_rate'] = param_group['learning_rate']
+            param.optimize_attr['learning_rate'] = param_group.get(
+                'learning_rate', 1.)
 
         self._param_groups.append(param_group)