diff --git a/paddle/fluid/prim/api/composite_backward/composite_backward_api.h b/paddle/fluid/prim/api/composite_backward/composite_backward_api.h
index 6473d8501d6d6157e5d79596d91cd2d5e6e1f35b..48641d852537105dbbc4af98dae7410243ed1751 100644
--- a/paddle/fluid/prim/api/composite_backward/composite_backward_api.h
+++ b/paddle/fluid/prim/api/composite_backward/composite_backward_api.h
@@ -88,7 +88,11 @@ void transpose_grad(const Tensor& grad_out,
     std::vector<int> reverse_perm(perm);
     // make origin ranks
     for (int i = 0; i < static_cast<int>(perm.size()); ++i) {
-      reverse_perm[perm[i]] = i;
+      if (perm[i] >= 0) {
+        reverse_perm[perm[i]] = i;
+      } else {
+        reverse_perm[perm[i] + perm.size()] = i;
+      }
     }
     auto grad_x_tmp = transpose<T>(grad_out, reverse_perm);
     set_output<T>(grad_x_tmp, grad_x);
diff --git a/python/paddle/fluid/tests/unittests/CMakeLists.txt b/python/paddle/fluid/tests/unittests/CMakeLists.txt
index 5ae8d37f3f1d9f0daea57c7e2af877e3a5e07739..5229884fb2492d6315ce8339a14b59f5d8f0ef2f 100755
--- a/python/paddle/fluid/tests/unittests/CMakeLists.txt
+++ b/python/paddle/fluid/tests/unittests/CMakeLists.txt
@@ -1217,7 +1217,8 @@ set(TEST_CINN_OPS
     test_elementwise_sub_op
     test_elementwise_div_op
     test_elementwise_mul_op
-    test_gather_nd_op)
+    test_gather_nd_op
+    test_transpose_op)
 
 foreach(TEST_CINN_OPS ${TEST_CINN_OPS})
   if(WITH_CINN)
diff --git a/python/paddle/fluid/tests/unittests/eager_op_test.py b/python/paddle/fluid/tests/unittests/eager_op_test.py
index d2e21863a87e9d081043127a86c97910abda2815..abcb671d47ac082598700bfa49e63ead46a55a1b 100644
--- a/python/paddle/fluid/tests/unittests/eager_op_test.py
+++ b/python/paddle/fluid/tests/unittests/eager_op_test.py
@@ -1392,6 +1392,7 @@ class OpTest(unittest.TestCase):
         inplace_atol=None,
     ):
         core._set_prim_all_enabled(False)
+        core.set_prim_eager_enabled(False)
 
         def find_imperative_actual(target_name, dygraph_outs, place):
             for name in dygraph_outs:
@@ -1982,6 +1983,7 @@ class OpTest(unittest.TestCase):
         numeric_place=None,
     ):
         core._set_prim_all_enabled(False)
+        core.set_prim_eager_enabled(False)
         if check_prim:
             prim_grad_checker = PrimGradChecker(
                 self,
diff --git a/python/paddle/fluid/tests/unittests/mkldnn/test_transpose_mkldnn_op.py b/python/paddle/fluid/tests/unittests/mkldnn/test_transpose_mkldnn_op.py
index 5dd35279d6214936afe4c5c6103f0844b35d4ec1..e1ef8a7a12f3ff7d4859126678a756371ad866d1 100644
--- a/python/paddle/fluid/tests/unittests/mkldnn/test_transpose_mkldnn_op.py
+++ b/python/paddle/fluid/tests/unittests/mkldnn/test_transpose_mkldnn_op.py
@@ -16,10 +16,10 @@ import unittest
 
 import numpy as np
 
-from paddle.fluid.tests.unittests.test_transpose_op import TestTransposeOp
+from paddle.fluid.tests.unittests.op_test import OpTest
 
 
-class TestTransposeMKLDNN(TestTransposeOp):
+class TestTransposeMKLDNN(OpTest):
     def setUp(self):
         self.init_op_type()
         self.initTestCase()
diff --git a/python/paddle/fluid/tests/unittests/test_transpose_op.py b/python/paddle/fluid/tests/unittests/test_transpose_op.py
index 4ec40ea9eb87e9d345d6617b00f0b768e3815d38..a8cf2b2c92615661541d7ad8b44037f6392eb1a9 100644
--- a/python/paddle/fluid/tests/unittests/test_transpose_op.py
+++ b/python/paddle/fluid/tests/unittests/test_transpose_op.py
@@ -32,6 +32,7 @@ class TestTransposeOp(OpTest):
         self.init_op_type()
         self.initTestCase()
         self.python_api = paddle.transpose
+        self.prim_op_type = "prim"
         self.inputs = {'X': np.random.random(self.shape).astype("float64")}
         self.attrs = {
             'axis': list(self.axis),
@@ -50,7 +51,7 @@ class TestTransposeOp(OpTest):
         self.check_output(no_check_set=['XShape'])
 
     def test_check_grad(self):
-        self.check_grad(['X'], 'Out')
+        self.check_grad(['X'], 'Out', check_prim=True)
 
     def initTestCase(self):
         self.shape = (3, 40)
@@ -118,12 +119,44 @@ class TestCase9(TestTransposeOp):
 
 
 class TestCase10(TestTransposeOp):
+    def setUp(self):
+        self.init_op_type()
+        self.initTestCase()
+        self.python_api = paddle.transpose
+        self.prim_op_type = "prim"
+        self.enable_cinn = False
+        self.inputs = {'X': np.random.random(self.shape).astype("float64")}
+        self.attrs = {
+            'axis': list(self.axis),
+            'use_mkldnn': self.use_mkldnn,
+        }
+        self.outputs = {
+            'XShape': np.random.random(self.shape).astype("float64"),
+            'Out': self.inputs['X'].transpose(self.axis),
+        }
+
     def initTestCase(self):
         self.shape = (10, 8, 2)
         self.axis = (-1, 1, -3)
 
 
 class TestCase_ZeroDim(TestTransposeOp):
+    def setUp(self):
+        self.init_op_type()
+        self.initTestCase()
+        self.python_api = paddle.transpose
+        self.prim_op_type = "prim"
+        self.enable_cinn = False
+        self.inputs = {'X': np.random.random(self.shape).astype("float64")}
+        self.attrs = {
+            'axis': list(self.axis),
+            'use_mkldnn': self.use_mkldnn,
+        }
+        self.outputs = {
+            'XShape': np.random.random(self.shape).astype("float64"),
+            'Out': self.inputs['X'].transpose(self.axis),
+        }
+
     def initTestCase(self):
         self.shape = ()
         self.axis = ()
@@ -134,6 +167,7 @@ class TestAutoTuneTransposeOp(OpTest):
         self.init_op_type()
         self.initTestCase()
         self.python_api = paddle.transpose
+        self.prim_op_type = "prim"
         self.inputs = {'X': np.random.random(self.shape).astype("float64")}
         self.attrs = {
             'axis': list(self.axis),
@@ -160,7 +194,7 @@ class TestAutoTuneTransposeOp(OpTest):
         fluid.core.disable_autotune()
 
     def test_check_grad(self):
-        self.check_grad(['X'], 'Out')
+        self.check_grad(['X'], 'Out', check_prim=True)
 
 
 class TestAutoTuneTransposeBF16Op(OpTest):
@@ -169,6 +203,8 @@ class TestAutoTuneTransposeBF16Op(OpTest):
         self.initTestCase()
         self.dtype = np.uint16
         self.python_api = paddle.transpose
+        self.prim_op_type = "prim"
+        self.enable_cinn = False
         x = np.random.random(self.shape).astype("float32")
         self.inputs = {'X': convert_float_to_uint16(x)}
         self.attrs = {
@@ -198,7 +234,7 @@ class TestAutoTuneTransposeBF16Op(OpTest):
         fluid.core.disable_autotune()
 
     def test_check_grad(self):
-        self.check_grad(['X'], 'Out')
+        self.check_grad(['X'], 'Out', check_prim=True)
 
 
 class TestTransposeBF16Op(OpTest):
@@ -206,6 +242,8 @@ class TestTransposeBF16Op(OpTest):
         self.init_op_type()
         self.initTestCase()
         self.dtype = np.uint16
+        self.prim_op_type = "prim"
+        self.enable_cinn = False
         self.python_api = paddle.transpose
         x = np.random.random(self.shape).astype("float32")
 
diff --git a/python/paddle/tensor/linalg.py b/python/paddle/tensor/linalg.py
index 309efc63eb76ff1789c02bac7ed3126034924243..c9b47df4d1c5e7e1ed090c942d2e8ec69d4940cd 100644
--- a/python/paddle/tensor/linalg.py
+++ b/python/paddle/tensor/linalg.py
@@ -99,6 +99,7 @@ def transpose(x, perm, name=None):
                 'float64',
                 'int32',
                 'int64',
+                'uint16',
                 'complex64',
                 'complex128',
             ],
diff --git a/python/paddle/tensor/manipulation.py b/python/paddle/tensor/manipulation.py
index 3f8b35087f1cc8cf24806cd06ddbab860ac91de3..bf0a52e5724d3cb74948f7d37cb241a42e61fa18 100644
--- a/python/paddle/tensor/manipulation.py
+++ b/python/paddle/tensor/manipulation.py
@@ -482,6 +482,7 @@ def transpose(x, perm, name=None):
                 'float64',
                 'int32',
                 'int64',
+                'uint16',
                 'complex64',
                 'complex128',
             ],