diff --git a/python/paddle/fluid/contrib/slim/distillation/distillation_strategy.py b/python/paddle/fluid/contrib/slim/distillation/distillation_strategy.py
index 1f11f07a51e713d42cee5e63bd2a9a02d82232f7..2fc6b45183164f135ae3ced08c1900ad526add45 100644
--- a/python/paddle/fluid/contrib/slim/distillation/distillation_strategy.py
+++ b/python/paddle/fluid/contrib/slim/distillation/distillation_strategy.py
@@ -13,7 +13,7 @@
 # limitations under the License.
 
 from ..core.strategy import Strategy
-from ....framework import Program, program_guard
+from ....framework import Program, Variable, program_guard
 from .... import Executor
 import logging
 
@@ -74,8 +74,17 @@ class DistillationStrategy(Strategy):
         startup_program = Program()
         with program_guard(graph.program, startup_program):
             context.distiller_optimizer._name = 'distillation_optimizer'
-            context.distiller_optimizer.minimize(
-                graph.var(graph.out_nodes['loss'])._var)
+
+            # The learning rate variable may be created in other program.
+            # Update information in optimizer to make
+            # learning rate variable being accessible in current program.
+            optimizer = context.distiller_optimizer
+            if isinstance(optimizer._learning_rate, Variable):
+                optimizer._learning_rate_map[
+                    graph.program] = optimizer._learning_rate
+
+            optimizer.minimize(graph.var(graph.out_nodes['loss'])._var)
+
         exe = Executor(context.place)
         exe.run(startup_program, scope=context.scope)
 
diff --git a/python/paddle/fluid/contrib/slim/graph/graph_wrapper.py b/python/paddle/fluid/contrib/slim/graph/graph_wrapper.py
index c208553fd811c7b18f9168b8fcae4da6e5856070..7388ecd3b096fc05d1420b904f2d65d805c3fc53 100644
--- a/python/paddle/fluid/contrib/slim/graph/graph_wrapper.py
+++ b/python/paddle/fluid/contrib/slim/graph/graph_wrapper.py
@@ -402,6 +402,12 @@ class GraphWrapper(object):
             elif 'cost' in graph.out_nodes:
                 target_name = graph.out_nodes['cost']
             target = graph.var(target_name)._var
+            # The learning rate variable may be created in other program.
+            # Update information in optimizer to make
+            # learning rate variable being accessible in current program.
+            if isinstance(optimizer._learning_rate, Variable):
+                optimizer._learning_rate_map[
+                    graph.program] = optimizer._learning_rate
             optimizer.minimize(target, no_grad_set=no_grad_var_names)
 
         exe = Executor(place)
diff --git a/python/paddle/fluid/contrib/slim/tests/test_distillation_strategy.py b/python/paddle/fluid/contrib/slim/tests/test_distillation_strategy.py
index 9b967c0ac7d2bfdab23d4557ef0b7d28f4118ff7..094cc4c6ac8be582fc31d0436e4468d2ebbb235a 100644
--- a/python/paddle/fluid/contrib/slim/tests/test_distillation_strategy.py
+++ b/python/paddle/fluid/contrib/slim/tests/test_distillation_strategy.py
@@ -41,9 +41,11 @@ class TestDistillationStrategy(unittest.TestCase):
 
         cost = fluid.layers.cross_entropy(input=out, label=label)
         avg_cost = fluid.layers.mean(x=cost)
+
         optimizer = fluid.optimizer.Momentum(
             momentum=0.9,
-            learning_rate=0.01,
+            learning_rate=fluid.layers.piecewise_decay(
+                boundaries=[5, 10], values=[0.01, 0.001, 0.0001]),
             regularization=fluid.regularizer.L2Decay(4e-5))
 
         place = fluid.CUDAPlace(0)