diff --git a/python/paddle/distributed/fleet/meta_parallel/sharding/group_sharded_optimizer_stage2.py b/python/paddle/distributed/fleet/meta_parallel/sharding/group_sharded_optimizer_stage2.py
index 230ad8ade034e764d3d6fd86056c5035eb44feb6..b1a572d4edfc30d9fdccc45b1b056ef7411cf44d 100644
--- a/python/paddle/distributed/fleet/meta_parallel/sharding/group_sharded_optimizer_stage2.py
+++ b/python/paddle/distributed/fleet/meta_parallel/sharding/group_sharded_optimizer_stage2.py
@@ -580,6 +580,13 @@ class GroupShardedOptimizerStage2(Optimizer):
 
         return __impl__
 
+    def set_lr(self, lr):
+        super().set_lr(lr)
+        self._optim.set_lr(lr)
+
+    def get_lr(self):
+        return self._optim.get_lr()
+
     @paddle.autograd.no_grad()
     def _broadcast_params_overlap_forward(self):
         # Exchange all the shards with the other ranks,
diff --git a/python/paddle/fluid/tests/unittests/collective/fleet/dygraph_group_sharded_api_eager.py b/python/paddle/fluid/tests/unittests/collective/fleet/dygraph_group_sharded_api_eager.py
index 4d9f43db7fa6c3e94d7c7553baeb6fa0dafc8ef4..ef3bf9df182058df5f4cc92ad3aa7dd12ae96bb2 100644
--- a/python/paddle/fluid/tests/unittests/collective/fleet/dygraph_group_sharded_api_eager.py
+++ b/python/paddle/fluid/tests/unittests/collective/fleet/dygraph_group_sharded_api_eager.py
@@ -100,6 +100,10 @@ def train_mlp(
         dp_group=dp_group,
     )
 
+    # just for test_coverage.
+    if shard_level == "os_g":
+        optimizer.set_lr(optimizer.get_lr())
+
     train_reader = paddle.batch(
         reader_decorator(), batch_size=batch_size, drop_last=True
     )