diff --git a/paddle/fluid/inference/api/mkldnn_quantizer_config.cc b/paddle/fluid/inference/api/mkldnn_quantizer_config.cc
index 9adc1150b59f9dea1beb2e6cf4bb8e76aa8b32f6..cfb5c2e7af3ca55a112bab401d396f7925c5f579 100644
--- a/paddle/fluid/inference/api/mkldnn_quantizer_config.cc
+++ b/paddle/fluid/inference/api/mkldnn_quantizer_config.cc
@@ -35,8 +35,8 @@ MkldnnQuantizerConfig::MkldnnQuantizerConfig() {
   rules_["prior_box"]["Boxes"] = ScaleAlgo::NONE;
   rules_["prior_box"]["Variances"] = ScaleAlgo::NONE;
 
-  rules_["transpose"]["X"] = ScaleAlgo::KL;
-  rules_["transpose"]["Out"] = ScaleAlgo::KL;
+  rules_["transpose2"]["X"] = ScaleAlgo::KL;
+  rules_["transpose2"]["Out"] = ScaleAlgo::KL;
 }
 
 ScaleAlgo MkldnnQuantizerConfig::scale_algo(
diff --git a/paddle/fluid/operators/mkldnn/dequantize_mkldnn_op.cc b/paddle/fluid/operators/mkldnn/dequantize_mkldnn_op.cc
index accc9a9d71ffccf2812d57a7516eaf7e0f83275c..0e203ef0010e47b25f2014fd61dd86e02b68b387 100644
--- a/paddle/fluid/operators/mkldnn/dequantize_mkldnn_op.cc
+++ b/paddle/fluid/operators/mkldnn/dequantize_mkldnn_op.cc
@@ -86,8 +86,9 @@ class DeQuantOpKernel : public framework::OpKernel<T> {
       std::shared_ptr<primitive::at> src_memory_p =
           std::shared_ptr<primitive::at>(new primitive::at(*src_memory));
 
-      auto dst_md = platform::MKLDNNMemDesc({dst_tz}, memory::data_type::f32,
-                                            memory::format::nchw);
+      auto dst_md = platform::MKLDNNMemDesc(
+          {dst_tz}, memory::data_type::f32,
+          platform::MKLDNNFormatForSize(dst_tz.size(), memory::format::nchw));
       auto dst_pd = mkldnn::memory::primitive_desc(dst_md, engine);
       dst_memory = std::make_shared<mkldnn::memory>(
           dst_pd, to_void_cast<float>(output_data));
diff --git a/python/paddle/fluid/tests/unittests/mkldnn/test_transpose_int8_mkldnn_op.py b/python/paddle/fluid/tests/unittests/mkldnn/test_transpose_int8_mkldnn_op.py
index a8127bcc781378fa5ef4a189a0b14d079a793946..3a7f620a1003b89d2f31f0fad0a6caf7bb6fa499 100644
--- a/python/paddle/fluid/tests/unittests/mkldnn/test_transpose_int8_mkldnn_op.py
+++ b/python/paddle/fluid/tests/unittests/mkldnn/test_transpose_int8_mkldnn_op.py
@@ -16,6 +16,7 @@ from __future__ import print_function
 
 import unittest
 import numpy as np
+import paddle.fluid.core as core
 from paddle.fluid.tests.unittests.op_test import OpTest
 from mkldnn_op_test import format_reorder
 
@@ -26,10 +27,11 @@ class TestTransposeOp(OpTest):
         self.initTestCase()
         self.initInputData()
         self.use_mkldnn = True
+        self._cpu_only = True
         self.axis = (0, 2, 3, 1)
 
         self.inputs = {
-            'X': format_reorder(self.input_data, self.shape)
+            'X': format_reorder(self.input_data, self.shape).astype(np.int8)
         }  #transform data format to 'NHWC' for INT8 transpose specially.
 
         self.attrs = {
@@ -38,7 +40,7 @@ class TestTransposeOp(OpTest):
         }
 
         self.outputs = {
-            'XShape': np.random.random(self.shape).astype('int8'),
+            'XShape': np.random.random(self.shape).astype(np.int8),
             'Out': self.inputs['X'].transpose(self.axis)
         }
 
@@ -46,14 +48,15 @@ class TestTransposeOp(OpTest):
         self.op_type = "transpose2"
 
     def test_check_output(self):
-        self.check_output(no_check_set=['XShape'])
+        self.check_output_with_place(
+            core.CPUPlace(), 1e-5, no_check_set=['XShape'])
 
     def initTestCase(self):
         self.shape = (2, 3, 4, 5)
 
     def initInputData(self):
         self.input_data = (
-            np.random.randint(0, 100, self.shape) - 50).astype('int8')
+            np.random.randint(0, 100, self.shape) - 50).astype(np.int8)
 
 
 class TestINT8Case(TestTransposeOp):
@@ -62,7 +65,7 @@ class TestINT8Case(TestTransposeOp):
 
     def initInputData(self):
         self.input_data = (
-            np.random.randint(0, 100, self.shape) - 50).astype('int8')
+            np.random.randint(0, 100, self.shape) - 50).astype(np.int8)
 
 
 class TestUINT8Case(TestTransposeOp):
@@ -71,7 +74,7 @@ class TestUINT8Case(TestTransposeOp):
 
     def initDataType(self):
         self.input_data = (np.random.randint(0, 100,
-                                             self.shape)).astype('uint8')
+                                             self.shape)).astype(np.uint8)
 
 
 if __name__ == '__main__':