diff --git a/python/paddle/distributed/passes/ps_server_pass.py b/python/paddle/distributed/passes/ps_server_pass.py
index c243c0602ff39e6a1abca320eb7b1e62e0b0c0bd..3cc1a14dbcd911c38991cd43481f2114911cbb09 100755
--- a/python/paddle/distributed/passes/ps_server_pass.py
+++ b/python/paddle/distributed/passes/ps_server_pass.py
@@ -69,7 +69,7 @@ class AddLrDecayTablePass(PassBase):
         ] = tensor_table_class
         attrs['tensor_table'] = tensor_table_dict
 
-    def _get_lr_sheduler_program(self, lr_sheduler, lr_decay_steps):
+    def _get_lr_scheduler_program(self, lr_scheduler, lr_decay_steps):
         schedler_decay = [
             'NoamDecay',
             'NaturalExpDecay',
@@ -81,12 +81,12 @@ class AddLrDecayTablePass(PassBase):
         decay_startup_program = paddle.static.Program()
         lr_name = ""
 
-        if isinstance(lr_sheduler, ExponentialDecay):
+        if isinstance(lr_scheduler, ExponentialDecay):
             with paddle.static.program_guard(
                 decay_main_program, decay_startup_program
             ):
                 lr = exponential_decay(
-                    1.0, lr_decay_steps, lr_sheduler.gamma, True
+                    1.0, lr_decay_steps, lr_scheduler.gamma, True
                 )
                 lr_name = lr.name
                 logging.warn(
@@ -96,24 +96,24 @@ class AddLrDecayTablePass(PassBase):
                     "\t strategy.a_sync_configs= { 'lr_decay_steps' : YOUR_DECAY_STEP } \n"
                     % lr_decay_steps
                 )
-        elif isinstance(lr_sheduler, NoamDecay):
+        elif isinstance(lr_scheduler, NoamDecay):
             with paddle.static.program_guard(
                 decay_main_program, decay_startup_program
             ):
                 lr = noam_decay(
-                    lr_sheduler.d_model, lr_sheduler.warmup_steps, 1.0
+                    lr_scheduler.d_model, lr_scheduler.warmup_steps, 1.0
                 )
                 lr_name = lr.name
                 logging.warn(
                     "NoamDecay is set, warmup steps is [ %d ]"
-                    % lr_sheduler.warmup_steps
+                    % lr_scheduler.warmup_steps
                 )
-        elif isinstance(lr_sheduler, NaturalExpDecay):
+        elif isinstance(lr_scheduler, NaturalExpDecay):
             with paddle.static.program_guard(
                 decay_main_program, decay_startup_program
             ):
                 lr = natural_exp_decay(
-                    1.0, lr_decay_steps, lr_sheduler.gamma, True
+                    1.0, lr_decay_steps, lr_scheduler.gamma, True
                 )
                 lr_name = lr.name
                 logging.warn(
@@ -123,12 +123,12 @@ class AddLrDecayTablePass(PassBase):
                     "\t strategy.a_sync_configs= { 'lr_decay_steps' : YOUR_DECAY_STEP } \n"
                     % lr_decay_steps
                 )
-        elif isinstance(lr_sheduler, InverseTimeDecay):
+        elif isinstance(lr_scheduler, InverseTimeDecay):
             with paddle.static.program_guard(
                 decay_main_program, decay_startup_program
             ):
                 lr = inverse_time_decay(
-                    1.0, lr_decay_steps, lr_sheduler.gamma, True
+                    1.0, lr_decay_steps, lr_scheduler.gamma, True
                 )
                 lr_name = lr.name
                 logging.warn(
@@ -149,11 +149,11 @@ class AddLrDecayTablePass(PassBase):
 
     def _apply_single_impl(self, main_program, startup_program, pass_ctx):
         attrs = pass_ctx._attrs
-        if not hasattr(attrs['origin_main_program'], 'lr_sheduler'):
+        if not hasattr(attrs['origin_main_program'], 'lr_scheduler'):
             return
 
         assert isinstance(
-            attrs['origin_main_program'].lr_sheduler, LRScheduler
+            attrs['origin_main_program'].lr_scheduler, LRScheduler
         ), "must be LRScheduler"
 
         ops = get_optimize_ops(attrs['origin_main_program'])
@@ -161,8 +161,8 @@ class AddLrDecayTablePass(PassBase):
             lr_decay_main_program,
             lr_decay_startup_program,
             lr_name,
-        ) = self._get_lr_sheduler_program(
-            attrs['origin_main_program'].lr_sheduler, attrs['lr_decay_steps']
+        ) = self._get_lr_scheduler_program(
+            attrs['origin_main_program'].lr_scheduler, attrs['lr_decay_steps']
         )
         self._add_tensor_table(
             attrs,
diff --git a/python/paddle/distributed/passes/ps_trainer_pass.py b/python/paddle/distributed/passes/ps_trainer_pass.py
index 72cda1b85573ca53970777e8211894c7b887f3cc..a2db1219a9d6c1ec052b73152675f4a5e0bcb411 100755
--- a/python/paddle/distributed/passes/ps_trainer_pass.py
+++ b/python/paddle/distributed/passes/ps_trainer_pass.py
@@ -612,7 +612,7 @@ class DeleteOptimizesPass(PassBase):
             main_program, remote_optimize_ops, local_optimize_ops
         )
 
-        if hasattr(attrs['origin_main_program'], 'lr_sheduler'):
+        if hasattr(attrs['origin_main_program'], 'lr_scheduler'):
             self._add_lr_var(main_program, attrs)
 
 
diff --git a/python/paddle/fluid/compiler.py b/python/paddle/fluid/compiler.py
index 4843b4e5692349ae1db68fd3a76367700157bbca..acdfa64769ce062eb70f1e498f69198ed12ea011 100644
--- a/python/paddle/fluid/compiler.py
+++ b/python/paddle/fluid/compiler.py
@@ -1235,15 +1235,15 @@ class IpuCompiledProgram:
         convert_pass.apply(self._graph)
         program = framework.Program._construct_from_desc(desc)
 
-        if hasattr(self._program, 'lr_sheduler'):
+        if hasattr(self._program, 'lr_scheduler'):
             # how to share var between two different block ?
-            lr_var_name = self._program.lr_sheduler._var_name
+            lr_var_name = self._program.lr_scheduler._var_name
 
-            program.lr_sheduler = self._program.lr_sheduler
-            # Program.clone will clone lr_sheduler, so i set lr_var as
-            # lr_sheduler attribute
+            program.lr_scheduler = self._program.lr_scheduler
+            # Program.clone will clone lr_scheduler, so i set lr_var as
+            # lr_scheduler attribute
             global_block = self._program.global_block()
-            program.lr_sheduler.lr_var = global_block.vars[lr_var_name]
+            program.lr_scheduler.lr_var = global_block.vars[lr_var_name]
 
         # with popart, we need to support batches_per_step, what means
         # the shape of feed_var and feed_tensor(maybe numpy array) will
diff --git a/python/paddle/fluid/executor.py b/python/paddle/fluid/executor.py
index 3b018f30c64938588bc3b8bc60529dbc19e8dccc..ce1c55bd1716882c2edd2090567e6a2102ac2b9f 100755
--- a/python/paddle/fluid/executor.py
+++ b/python/paddle/fluid/executor.py
@@ -871,8 +871,8 @@ class _ExecutorCache:
             ir_graph = framework.IrGraph(compiled_program._graph)
             converted_program = ir_graph.to_program()
 
-            if hasattr(inner_program, 'lr_sheduler'):
-                converted_program.lr_sheduler = inner_program.lr_sheduler
+            if hasattr(inner_program, 'lr_scheduler'):
+                converted_program.lr_scheduler = inner_program.lr_scheduler
 
             inner_program = converted_program
             # print(f"Program after convert:\n {inner_program}", flush=True)
@@ -1657,17 +1657,17 @@ class Executor:
             )
 
             self._feed_data(program, feed, feed_var_name, scope)
-            if hasattr(program, 'lr_sheduler'):
+            if hasattr(program, 'lr_scheduler'):
                 from paddle.optimizer.lr import LRScheduler
 
                 assert isinstance(
-                    program.lr_sheduler, LRScheduler
+                    program.lr_scheduler, LRScheduler
                 ), "must be LRScheduler"
-                lr_sheduler = program.lr_sheduler
-                lr_value = lr_sheduler()
-                lr_var = program.global_block().vars[lr_sheduler._var_name]
+                lr_scheduler = program.lr_scheduler
+                lr_value = lr_scheduler()
+                lr_var = program.global_block().vars[lr_scheduler._var_name]
                 data = np.array([lr_value]).astype(convert_dtype(lr_var.dtype))
-                tensor = core.get_variable_tensor(scope, lr_sheduler._var_name)
+                tensor = core.get_variable_tensor(scope, lr_scheduler._var_name)
                 # NOTE(dev): `tensor.set(data, self.place)` always call TensorCopySync that is a blocking behavior. So we use `_copy_from` to replace it.
                 cpu_tensor = _as_lodtensor(data, core.CPUPlace())
                 if core.is_cuda_graph_capturing():
@@ -1810,15 +1810,15 @@ class Executor:
             )
 
         self._feed_data(program, feed, feed_var_name, scope)
-        if hasattr(program, 'lr_sheduler'):
+        if hasattr(program, 'lr_schedulerr'):
             assert isinstance(
-                program.lr_sheduler, LRScheduler
+                program.lr_scheduler, LRScheduler
             ), "must be LRScheduler"
-            lr_sheduler = program.lr_sheduler
-            lr_value = lr_sheduler()
-            lr_var = program.global_block().vars[lr_sheduler._var_name]
+            lr_scheduler = program.lr_scheduler
+            lr_value = lr_scheduler()
+            lr_var = program.global_block().vars[lr_scheduler._var_name]
             data = np.array([lr_value]).astype(convert_dtype(lr_var.dtype))
-            tensor = core.get_variable_tensor(scope, lr_sheduler._var_name)
+            tensor = core.get_variable_tensor(scope, lr_scheduler._var_name)
             tensor.set(data, self.place)
 
         if not use_program_cache:
@@ -2588,14 +2588,14 @@ class Executor:
 
         from paddle.optimizer.lr import LRScheduler
 
-        if hasattr(program, 'lr_sheduler'):
-            lr_sheduler = program.lr_sheduler
-            assert isinstance(lr_sheduler, LRScheduler), "must be LRScheduler"
-            lr_value = lr_sheduler()
-            lr_var = program.global_block().vars[lr_sheduler._var_name]
+        if hasattr(program, 'lr_scheduler'):
+            lr_scheduler = program.lr_scheduler
+            assert isinstance(lr_scheduler, LRScheduler), "must be LRScheduler"
+            lr_value = lr_scheduler()
+            lr_var = program.global_block().vars[lr_scheduler._var_name]
             data = np.array([lr_value]).astype(convert_dtype(lr_var.dtype))
             tensor = core.get_variable_tensor(
-                cached_scope, lr_sheduler._var_name
+                cached_scope, lr_scheduler._var_name
             )
             tensor.set(data, self.place)
 
@@ -2732,13 +2732,13 @@ class Executor:
 
         from paddle.optimizer.lr import LRScheduler
 
-        if hasattr(program, 'lr_sheduler'):
-            lr_sheduler = program.lr_sheduler
-            assert isinstance(lr_sheduler, LRScheduler), "must be LRScheduler"
-            lr_value = lr_sheduler()
-            lr_var = program.global_block().vars[lr_sheduler._var_name]
+        if hasattr(program, 'lr_scheduler'):
+            lr_scheduler = program.lr_scheduler
+            assert isinstance(lr_scheduler, LRScheduler), "must be LRScheduler"
+            lr_value = lr_scheduler()
+            lr_var = program.global_block().vars[lr_scheduler._var_name]
             data = np.array([lr_value]).astype(convert_dtype(lr_var.dtype))
-            tensor = core.get_variable_tensor(scope, lr_sheduler._var_name)
+            tensor = core.get_variable_tensor(scope, lr_scheduler._var_name)
             tensor.set(data, self.place)
 
         self._default_executor.run_from_dataset(trainer_instance)
diff --git a/python/paddle/fluid/framework.py b/python/paddle/fluid/framework.py
index 43c23eb003e2ee2b86df2a61522315344e39a810..cb57fe793737547c07366249c9375672cf0dc1e4 100644
--- a/python/paddle/fluid/framework.py
+++ b/python/paddle/fluid/framework.py
@@ -6080,8 +6080,8 @@ class Program:
             p._current_role = self._current_role
             p.__op_role_var = self.__op_role_var
             p._appending_grad_times = self._appending_grad_times
-            if hasattr(self, 'lr_sheduler'):
-                p.lr_sheduler = self.lr_sheduler
+            if hasattr(self, 'lr_scheduler'):
+                p.lr_scheduler = self.lr_scheduler
 
             # NOTE(zhiqiu): we sync the cloned program, to update its program by
             # its desc.
diff --git a/python/paddle/fluid/optimizer.py b/python/paddle/fluid/optimizer.py
index ded8883ffdb533093c1c77162f0093af3a71662b..d3c3f72d63990dbe694037c095ebc389d72739be 100755
--- a/python/paddle/fluid/optimizer.py
+++ b/python/paddle/fluid/optimizer.py
@@ -389,7 +389,7 @@ class Optimizer:
                     dtype='float32' if self._dtype is None else self._dtype,
                 )
                 main_prog = framework.default_main_program()
-                main_prog.lr_sheduler = self._learning_rate
+                main_prog.lr_scheduler = self._learning_rate
                 main_prog.lr_var = lr_var
                 self._learning_rate_map[
                     framework.default_main_program()
diff --git a/python/paddle/fluid/tests/unittests/ipu/test_lr_sheduler_ipu.py b/python/paddle/fluid/tests/unittests/ipu/test_lr_sheduler_ipu.py
index 91c94ce48dda1ee8d6296c2d13f6b513797816f4..567b9ef84a5a1d72ea2bb21c24deed2de3164a6c 100644
--- a/python/paddle/fluid/tests/unittests/ipu/test_lr_sheduler_ipu.py
+++ b/python/paddle/fluid/tests/unittests/ipu/test_lr_sheduler_ipu.py
@@ -67,8 +67,8 @@ class TestConvNet(IPUOpTest):
 
         result = []
         for _ in range(100):
-            if hasattr(program, "lr_sheduler"):
-                program.lr_sheduler.step()
+            if hasattr(program, "lr_scheduler"):
+                program.lr_scheduler.step()
             loss_res = exe.run(
                 program, feed=self.feed, fetch_list=self.fetch_list
             )
diff --git a/python/paddle/fluid/tests/unittests/test_dist_base.py b/python/paddle/fluid/tests/unittests/test_dist_base.py
index 54abe60f013f2fe6b0b013528d4bddeaaf762e8f..3e157b201f580d4d013a88d026d8557dfe0fa589 100755
--- a/python/paddle/fluid/tests/unittests/test_dist_base.py
+++ b/python/paddle/fluid/tests/unittests/test_dist_base.py
@@ -132,13 +132,13 @@ class TestDistRunnerBase:
 
     @staticmethod
     def get_lr_scheduler(program):
-        lr_sheduler = None
-        if hasattr(program, 'lr_sheduler'):
+        lr_scheduler = None
+        if hasattr(program, 'lr_scheduler'):
             from paddle.optimizer.lr import LRScheduler
 
-            lr_sheduler = program.lr_sheduler
-            assert isinstance(lr_sheduler, LRScheduler), "must be LRScheduler"
-        return lr_sheduler
+            lr_scheduler = program.lr_scheduler
+            assert isinstance(lr_scheduler, LRScheduler), "must be LRScheduler"
+        return lr_scheduler
 
     def run_pserver(self, args):
         self.lr = args.lr
@@ -196,14 +196,14 @@ class TestDistRunnerBase:
         out_losses = []
 
         main_program = fluid.default_main_program()
-        lr_sheduler = self.get_lr_scheduler(main_program)
+        lr_scheduler = self.get_lr_scheduler(main_program)
         for i in range(RUN_STEP):
             loss = exe.run(main_program, fetch_list=[avg_cost])
             loss = loss[0] if loss else None
             out_losses.append(loss)
             print_to_err(type(self).__name__, "run step %d finished" % i)
-            if lr_sheduler is not None:
-                lr_sheduler.step()
+            if lr_scheduler is not None:
+                lr_scheduler.step()
 
         data_loader.reset()
         print_to_err(type(self).__name__, "trainer run finished")
diff --git a/python/paddle/fluid/tests/unittests/test_newprofiler.py b/python/paddle/fluid/tests/unittests/test_newprofiler.py
index 18c948a3658eca9608e05e015edc1d402c53de77..7f2009b1a884acd1012beddc9d924de5b75aa566 100755
--- a/python/paddle/fluid/tests/unittests/test_newprofiler.py
+++ b/python/paddle/fluid/tests/unittests/test_newprofiler.py
@@ -92,7 +92,7 @@ class TestProfiler(unittest.TestCase):
                 y = x / 2.0
                 prof.step()
 
-        def my_sheduler(num_step):
+        def my_scheduler(num_step):
             if num_step % 5 < 2:
                 return profiler.ProfilerState.RECORD_AND_RETURN
             elif num_step % 5 < 3:
@@ -102,7 +102,7 @@ class TestProfiler(unittest.TestCase):
             else:
                 return profiler.ProfilerState.CLOSED
 
-        def my_sheduler1(num_step):
+        def my_scheduler1(num_step):
             if num_step % 5 < 2:
                 return profiler.ProfilerState.RECORD
             elif num_step % 5 < 3:
@@ -124,7 +124,7 @@ class TestProfiler(unittest.TestCase):
         prof = None
         with profiler.Profiler(
             targets=[profiler.ProfilerTarget.CPU],
-            scheduler=my_sheduler,
+            scheduler=my_scheduler,
             on_trace_ready=my_trace_back,
         ) as prof:
             for i in range(5):
@@ -132,7 +132,7 @@ class TestProfiler(unittest.TestCase):
                 prof.step()
         prof = None
         with profiler.Profiler(
-            targets=[profiler.ProfilerTarget.CPU], scheduler=my_sheduler1
+            targets=[profiler.ProfilerTarget.CPU], scheduler=my_scheduler1
         ) as prof:
             for i in range(5):
                 y = x / 2.0
diff --git a/python/paddle/incubate/distributed/fleet/parameter_server/ir/public.py b/python/paddle/incubate/distributed/fleet/parameter_server/ir/public.py
index 7d8e09d9b17cce99b405e9c631fd24599d8c2b22..16a93ab6b6fe3712e145e5c3e497b187e952ccd7 100755
--- a/python/paddle/incubate/distributed/fleet/parameter_server/ir/public.py
+++ b/python/paddle/incubate/distributed/fleet/parameter_server/ir/public.py
@@ -1362,11 +1362,11 @@ def _get_optimize_ops(_program):
 
 
 def _add_lr_decay_table_pass(main_program, compiled_config, lr_decay_steps):
-    if hasattr(compiled_config.origin_main_program, 'lr_sheduler'):
+    if hasattr(compiled_config.origin_main_program, 'lr_scheduler'):
         from paddle.optimizer.lr import LRScheduler
 
         assert isinstance(
-            compiled_config.origin_main_program.lr_sheduler, LRScheduler
+            compiled_config.origin_main_program.lr_scheduler, LRScheduler
         ), "must be LRScheduler"
         ops = _get_optimize_ops(compiled_config.origin_main_program)
         lr_param_dict = _get_lr_param_dict(ops)
@@ -1374,8 +1374,8 @@ def _add_lr_decay_table_pass(main_program, compiled_config, lr_decay_steps):
             lr_decay_main_program,
             lr_decay_startup_program,
             lr_name,
-        ) = _get_lr_sheduler_program(
-            compiled_config.origin_main_program.lr_sheduler,
+        ) = _get_lr_scheduler_program(
+            compiled_config.origin_main_program.lr_scheduler,
             lr_param_dict,
             lr_decay_steps,
         )
@@ -1399,7 +1399,7 @@ def _get_lr_param_dict(opt_ops):
     return lr_param_dict
 
 
-def _get_lr_sheduler_program(lr_sheduler, lr_param_dict, lr_decay_steps):
+def _get_lr_scheduler_program(lr_scheduler, lr_param_dict, lr_decay_steps):
     schedler_decay = [
         'NoamDecay',
         'NaturalExpDecay',
@@ -1424,11 +1424,13 @@ def _get_lr_sheduler_program(lr_sheduler, lr_param_dict, lr_decay_steps):
     decay_startup_program = paddle.static.Program()
     lr_name = ""
 
-    if isinstance(lr_sheduler, ExponentialDecay):
+    if isinstance(lr_scheduler, ExponentialDecay):
         with paddle.static.program_guard(
             decay_main_program, decay_startup_program
         ):
-            lr = exponential_decay(1.0, lr_decay_steps, lr_sheduler.gamma, True)
+            lr = exponential_decay(
+                1.0, lr_decay_steps, lr_scheduler.gamma, True
+            )
             lr_name = lr.name
             logging.warn(
                 "ExponentialDecay is set, staircase = True, global learning rate decay step is [ %d ], Change decay steps as follow: \n"
@@ -1437,21 +1439,25 @@ def _get_lr_sheduler_program(lr_sheduler, lr_param_dict, lr_decay_steps):
                 "\t strategy.a_sync_configs= { 'lr_decay_steps' : YOUR_DECAY_STEP } \n"
                 % lr_decay_steps
             )
-    elif isinstance(lr_sheduler, NoamDecay):
+    elif isinstance(lr_scheduler, NoamDecay):
         with paddle.static.program_guard(
             decay_main_program, decay_startup_program
         ):
-            lr = noam_decay(lr_sheduler.d_model, lr_sheduler.warmup_steps, 1.0)
+            lr = noam_decay(
+                lr_scheduler.d_model, lr_scheduler.warmup_steps, 1.0
+            )
             lr_name = lr.name
             logging.warn(
                 "NoamDecay is set, warmup steps is [ %d ]"
-                % lr_sheduler.warmup_steps
+                % lr_scheduler.warmup_steps
             )
-    elif isinstance(lr_sheduler, NaturalExpDecay):
+    elif isinstance(lr_scheduler, NaturalExpDecay):
         with paddle.static.program_guard(
             decay_main_program, decay_startup_program
         ):
-            lr = natural_exp_decay(1.0, lr_decay_steps, lr_sheduler.gamma, True)
+            lr = natural_exp_decay(
+                1.0, lr_decay_steps, lr_scheduler.gamma, True
+            )
             lr_name = lr.name
             logging.warn(
                 "NaturalExpDecay is set, staircase = True, global learning rate decay step is [ %d ], Change decay steps as follow: \n"
@@ -1460,12 +1466,12 @@ def _get_lr_sheduler_program(lr_sheduler, lr_param_dict, lr_decay_steps):
                 "\t strategy.a_sync_configs= { 'lr_decay_steps' : YOUR_DECAY_STEP } \n"
                 % lr_decay_steps
             )
-    elif isinstance(lr_sheduler, InverseTimeDecay):
+    elif isinstance(lr_scheduler, InverseTimeDecay):
         with paddle.static.program_guard(
             decay_main_program, decay_startup_program
         ):
             lr = inverse_time_decay(
-                1.0, lr_decay_steps, lr_sheduler.gamma, True
+                1.0, lr_decay_steps, lr_scheduler.gamma, True
             )
             lr_name = lr.name
             logging.warn(
diff --git a/python/paddle/incubate/distributed/fleet/parameter_server/ir/trainer_pass.py b/python/paddle/incubate/distributed/fleet/parameter_server/ir/trainer_pass.py
index 67d32480761c47541dcd1305efe04aa711c5d7c9..3d5ec4c1bf76e8f8b470078f971f7691a50024dd 100644
--- a/python/paddle/incubate/distributed/fleet/parameter_server/ir/trainer_pass.py
+++ b/python/paddle/incubate/distributed/fleet/parameter_server/ir/trainer_pass.py
@@ -93,7 +93,7 @@ def delete_optimizer_pass(program, config):
     optimizer_ops.extend(lr_ops)
     _delete_optimizer_op_and_vars(program, optimizer_ops)
 
-    if hasattr(config.origin_main_program, 'lr_sheduler'):
+    if hasattr(config.origin_main_program, 'lr_scheduler'):
         _add_lr_var(program, config)
 
     return program
diff --git a/python/paddle/jit/dy2static/partial_program.py b/python/paddle/jit/dy2static/partial_program.py
index f46c3a65cfeb34ed333b2acdf637533f532c0c22..e2abec10c08235396735aab4cb7502eeaf00b095 100644
--- a/python/paddle/jit/dy2static/partial_program.py
+++ b/python/paddle/jit/dy2static/partial_program.py
@@ -1127,8 +1127,10 @@ def add_build_strategy_for(
         )
         ir_graph = framework.IrGraph(compiled_program._graph)
         builded_program = ir_graph.to_program()
-        if hasattr(compiled_program._program, 'lr_sheduler'):
-            builded_program.lr_sheduler = compiled_program._program.lr_sheduler
+        if hasattr(compiled_program._program, 'lr_scheduler'):
+            builded_program.lr_scheduler = (
+                compiled_program._program.lr_scheduler
+            )
     else:
         # can't just create a new program, we need copy the vardesc.
         builded_program = paddle.static.Program()
diff --git a/python/paddle/optimizer/optimizer.py b/python/paddle/optimizer/optimizer.py
index 72924ed692fceaeb26f61bfc38279d5291446d49..d7bfc0b22963205fa201efa63529b97128328ab0 100644
--- a/python/paddle/optimizer/optimizer.py
+++ b/python/paddle/optimizer/optimizer.py
@@ -444,7 +444,7 @@ class Optimizer:
                     dtype=_lr_dtype,
                 )
                 main_prog = framework.default_main_program()
-                main_prog.lr_sheduler = self._learning_rate
+                main_prog.lr_scheduler = self._learning_rate
                 main_prog.lr_var = lr_var
 
                 self._learning_rate_map[