diff --git a/demo/mnist/api_train.py b/demo/mnist/api_train.py
index 924bd39a5053085c03895f74e50943b80e243d42..f301da382ff8a5bc16d9c18b956f78566ed4894f 100644
--- a/demo/mnist/api_train.py
+++ b/demo/mnist/api_train.py
@@ -12,7 +12,6 @@ import paddle.trainer.PyDataProvider2 as dp
 import numpy as np
 import random
 from mnist_util import read_from_mnist
-import paddle.trainer_config_helpers.config_parser_utils as config_parser_utils
 from paddle.trainer_config_helpers import *
 
 
@@ -80,14 +79,13 @@ def main():
     # enable_types = [value, gradient, momentum, etc]
     # For each optimizer(SGD, Adam), GradientMachine should enable different
     # buffers.
-    opt_config_proto = config_parser_utils.parse_optimizer_config(
-        optimizer_config)
+    opt_config_proto = parse_optimizer_config(optimizer_config)
     opt_config = api.OptimizationConfig.createFromProto(opt_config_proto)
     _temp_optimizer_ = api.ParameterOptimizer.create(opt_config)
     enable_types = _temp_optimizer_.getParameterTypes()
 
     # Create Simple Gradient Machine.
-    model_config = config_parser_utils.parse_network_config(network_config)
+    model_config = parse_network_config(network_config)
     m = api.GradientMachine.createFromConfigProto(
         model_config, api.CREATE_MODE_NORMAL, enable_types)