diff --git a/test/test_api/test_distributed.py b/test/test_api/test_distributed.py
index d7b06ea700098bfe17295667fc3a60b2c4d67223..b0c699d4ab608541393c48d9d41d029e64ce5100 100644
--- a/test/test_api/test_distributed.py
+++ b/test/test_api/test_distributed.py
@@ -1,7 +1,21 @@
+# Copyright (c) 2020, Huawei Technologies.All rights reserved.
+#
+# Licensed under the BSD 3-Clause License  (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+# https://opensource.org/licenses/BSD-3-Clause
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
 import os
 import torch
-import torch_npu
 import torch.distributed as dist
+import torch_npu
 from torch_npu.testing.testcase import TestCase, run_tests
 
 os.environ["MASTER_ADDR"] = '127.0.0.1'
diff --git a/test/test_api/test_nn_func/test_convolution_functions.py b/test/test_api/test_nn_func/test_convolution_functions.py
index 40582cd3ede26b4429e9e4505d34949d20be688a..96b8b9cd33a1a5d95acde489c542a186f21c7bb1 100644
--- a/test/test_api/test_nn_func/test_convolution_functions.py
+++ b/test/test_api/test_nn_func/test_convolution_functions.py
@@ -12,9 +12,9 @@
 # See the License for the specific language governing permissions and
 # limitations under the License.
 import torch
-import torch_npu
 import torch.nn as nn
 import torch.nn.functional as F
+import torch_npu
 
 from torch_npu.testing.testcase import TestCase, run_tests
 
diff --git a/test/test_api/test_nn_func/test_distance_functions.py b/test/test_api/test_nn_func/test_distance_functions.py
index d1bdeae4b5994faa2a349a6cb8f0e82144f2c383..44f6f83d1bc406d0d176b90c7ab9893ac10e5796 100644
--- a/test/test_api/test_nn_func/test_distance_functions.py
+++ b/test/test_api/test_nn_func/test_distance_functions.py
@@ -14,8 +14,8 @@
 
 import copy
 import torch
-import torch_npu
 import torch.nn.functional as F
+import torch_npu
 
 from torch_npu.testing.testcase import TestCase, run_tests
 
@@ -53,5 +53,4 @@ class TestDistanceFunctions(TestCase):
         self.assertRtolEqual(cpu_output.detach().numpy(), npu_output.detach().cpu().numpy())
 
 if __name__ == "__main__":
-    torch.npu.set_device(0)
     run_tests()
\ No newline at end of file
diff --git a/test/test_api/test_nn_func/test_dropout_functions.py b/test/test_api/test_nn_func/test_dropout_functions.py
index 4b18eed7239f7e590ac31ec73923d106190b6ca0..bdfc615e3c8d9fa39af5b76c6b37a2a0f4595e16 100644
--- a/test/test_api/test_nn_func/test_dropout_functions.py
+++ b/test/test_api/test_nn_func/test_dropout_functions.py
@@ -14,8 +14,8 @@
 
 import copy
 import torch
-import torch_npu
 import torch.nn.functional as F
+import torch_npu
 
 from torch_npu.testing.testcase import TestCase, run_tests
 
@@ -54,5 +54,4 @@ class TestDropoutFunctions(TestCase):
         self.assertRtolEqual(cpu_output.numpy(), npu_output.cpu().numpy())
 
 if __name__ == "__main__":
-    torch.npu.set_device(0)
     run_tests()
\ No newline at end of file
diff --git a/test/test_api/test_nn_func/test_linear_functions.py b/test/test_api/test_nn_func/test_linear_functions.py
index 5a8ac0d7ec549c8bf039b86d681e7c63c4e1126f..237d198eaa707480ed0e0117e42eb4e913896bb3 100644
--- a/test/test_api/test_nn_func/test_linear_functions.py
+++ b/test/test_api/test_nn_func/test_linear_functions.py
@@ -14,8 +14,8 @@
 
 import copy
 import torch
-import torch_npu
 import torch.nn.functional as F
+import torch_npu
 
 from torch_npu.testing.testcase import TestCase, run_tests
 
@@ -50,5 +50,4 @@ class TestLinearFunctions(TestCase):
 
 
 if __name__ == "__main__":
-    torch.npu.set_device(0)
     run_tests()
\ No newline at end of file
diff --git a/test/test_api/test_nn_func/test_loss_functions.py b/test/test_api/test_nn_func/test_loss_functions.py
index 2671e420716934b56d4302131d6feea86453daf8..03046ced122960f1c4ff07828146482d9750a6e3 100644
--- a/test/test_api/test_nn_func/test_loss_functions.py
+++ b/test/test_api/test_nn_func/test_loss_functions.py
@@ -14,8 +14,8 @@
 import copy
 
 import torch
-import torch_npu
 import torch.nn.functional as F
+import torch_npu
 
 
 from torch_npu.testing.testcase import TestCase, run_tests
@@ -69,20 +69,20 @@ class TestLossFunctions(TestCase):
         self.assertRtolEqual(cpu_output.detach().numpy(), npu_output.detach().cpu().numpy())
 
     def test_ctc_loss(self):
-        log_probs = torch.randn(50, 16, 20).log_softmax(2).detach().requires_grad_()
-        targets = torch.randint(1, 20, (16, 30), dtype=torch.long)
-        input_lengths = torch.full((16,), 50, dtype=torch.long)
-        target_lengths = torch.randint(10,30,(16,), dtype=torch.long)
-
-        npu_log_probs = copy.deepcopy(log_probs).npu()
-        npu_targets = copy.deepcopy(targets).npu().int()
-        npu_input_lengths = copy.deepcopy(input_lengths).npu().int()
-        npu_target_lengths = copy.deepcopy(target_lengths).npu().int()
+        log_probs = torch.randn(50, 16, 20).npu().log_softmax(2)
+        targets = torch.randint(1, 20, (16, 30), dtype=torch.int32).npu()
+        input_lengths = torch.full((16,), 50, dtype=torch.int32).npu()
+        target_lengths = torch.randint(10,30,(16,), dtype=torch.int32).npu()
+
+        cpu_log_probs = copy.deepcopy(log_probs).cpu()
+        cpu_targets = copy.deepcopy(targets).cpu().long()
+        cpu_input_lengths = copy.deepcopy(input_lengths).cpu().long()
+        cpu_target_lengths = copy.deepcopy(target_lengths).cpu().long()
         
-        cpu_output = F.ctc_loss(log_probs, targets, input_lengths, target_lengths)
-        npu_output = F.ctc_loss(npu_log_probs, npu_targets, npu_input_lengths, npu_target_lengths)
+        npu_output = F.ctc_loss(log_probs, targets, input_lengths, target_lengths)
+        cpu_output = F.ctc_loss(cpu_log_probs, cpu_targets, cpu_input_lengths, cpu_target_lengths)
 
-        self.assertRtolEqual(cpu_output.detach().numpy(), npu_output.detach().cpu().numpy())
+        self.assertRtolEqual(cpu_output.numpy(), npu_output.cpu().numpy())
 
     def test_hinge_embedding_loss(self):
         input1 = torch.randn(5, 3)
@@ -210,5 +210,4 @@ class TestLossFunctions(TestCase):
 
 
 if __name__ == "__main__":
-    torch.npu.set_device(0)
     run_tests()
\ No newline at end of file
diff --git a/test/test_api/test_nn_func/test_nonlinear_activation_functions.py b/test/test_api/test_nn_func/test_nonlinear_activation_functions.py
index 811ad4531e8e6ba4d0c715a3a20cb5c8ff3d75f8..22870136de9b60666000351960a4d4852dbf00e2 100644
--- a/test/test_api/test_nn_func/test_nonlinear_activation_functions.py
+++ b/test/test_api/test_nn_func/test_nonlinear_activation_functions.py
@@ -15,8 +15,8 @@
 import copy
 
 import torch
-import torch_npu
 import torch.nn.functional as F
+import torch_npu
 
 from torch_npu.testing.testcase import TestCase, run_tests
 
@@ -210,5 +210,4 @@ class TestNonLiACFunctions(TestCase):
 
 
 if __name__ == "__main__":
-    torch.npu.set_device(0)
     run_tests()
\ No newline at end of file
diff --git a/test/test_api/test_nn_func/test_normalization_functions.py b/test/test_api/test_nn_func/test_normalization_functions.py
index 4f898634d7c624c4cd6d123818d7275ab08c83fc..3e5c9972933b3fb44e184653abc81da2ff35e854 100644
--- a/test/test_api/test_nn_func/test_normalization_functions.py
+++ b/test/test_api/test_nn_func/test_normalization_functions.py
@@ -13,9 +13,9 @@
 # limitations under the License.
 
 import torch
-import torch_npu
 import torch.nn as nn
 import torch.nn.functional as F
+import torch_npu
 
 from torch_npu.testing.testcase import TestCase, run_tests
 
diff --git a/test/test_api/test_nn_func/test_pooling_functions.py b/test/test_api/test_nn_func/test_pooling_functions.py
index c861f85c1eefd77147b341cce56c423f85aea5c5..dce7b7e606dcde9a8ef41715c5b925c22a3d6fad 100644
--- a/test/test_api/test_nn_func/test_pooling_functions.py
+++ b/test/test_api/test_nn_func/test_pooling_functions.py
@@ -11,10 +11,11 @@
 # WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
 # See the License for the specific language governing permissions and
 # limitations under the License.
+
 import torch
-import torch_npu
 import torch.nn as nn
 import torch.nn.functional as F
+import torch_npu
 
 from torch_npu.testing.testcase import TestCase, run_tests
 
diff --git a/test/test_api/test_nn_func/test_sparse_functions.py b/test/test_api/test_nn_func/test_sparse_functions.py
index 79a5230a91cf7a4546fc439eeb7f752d4b0bcdfa..d05b986590684e5b9d4068dd04314b6b91f191d3 100644
--- a/test/test_api/test_nn_func/test_sparse_functions.py
+++ b/test/test_api/test_nn_func/test_sparse_functions.py
@@ -13,9 +13,9 @@
 # limitations under the License.
 
 import torch
-import torch_npu
 import torch.nn as nn
 import torch.nn.functional as F
+import torch_npu
 
 from torch_npu.testing.testcase import TestCase, run_tests
 
diff --git a/test/test_api/test_nn_func/test_vision_functions.py b/test/test_api/test_nn_func/test_vision_functions.py
index a12c8328a2eec128566514d1b8063495edbb52b9..7c82eb8ec3c8d565fc7511957a31901b52f8fde2 100644
--- a/test/test_api/test_nn_func/test_vision_functions.py
+++ b/test/test_api/test_nn_func/test_vision_functions.py
@@ -13,8 +13,8 @@
 # limitations under the License.
 
 import torch
-import torch_npu
 import torch.nn.functional as F
+import torch_npu
 
 from torch_npu.testing.testcase import TestCase, run_tests
 
diff --git a/test/test_api/test_torch/test_locally_disabling_gradient_omputation.py b/test/test_api/test_torch/test_locally_disabling_gradient_omputation.py
index 15899c3ca5b9781aa6f2efd48f5d286cee2f345f..8fd41e857161fc04daa93946f24d9d5da064aaa2 100644
--- a/test/test_api/test_torch/test_locally_disabling_gradient_omputation.py
+++ b/test/test_api/test_torch/test_locally_disabling_gradient_omputation.py
@@ -13,10 +13,9 @@
 # limitations under the License.
 
 import torch
-import torch_npu
-
 from torch_npu.testing.testcase import TestCase, run_tests
 from torch.testing._internal.common_utils import freeze_rng_state
+import torch_npu
 
 device = 'npu:0'
 torch.npu.set_device(device)
diff --git a/test/test_api/test_torch/test_math_operations.py b/test/test_api/test_torch/test_math_operations.py
index 873738a7e4e7c189ae2601fff7018cbc670813a5..c9569e28b4dcdaf28b61d4432ed78dabce0b46e5 100644
--- a/test/test_api/test_torch/test_math_operations.py
+++ b/test/test_api/test_torch/test_math_operations.py
@@ -13,12 +13,11 @@
 # limitations under the License.
 
 import copy
-import torch
-import torch_npu
 import numpy as np
-
+import torch
 from torch_npu.testing.testcase import TestCase, run_tests
 from torch.testing._internal.common_utils import freeze_rng_state
+import torch_npu
 
 device = 'npu:0'
 torch.npu.set_device(device)
diff --git a/test/test_api/test_torch/test_serialization.py b/test/test_api/test_torch/test_serialization.py
index 6d00c022e4b9f51555be0d249985d5d8ea37ce87..7b435a8e6ed656fbfaf894cf7919f28429606682 100644
--- a/test/test_api/test_torch/test_serialization.py
+++ b/test/test_api/test_torch/test_serialization.py
@@ -16,9 +16,9 @@ import os
 import tempfile
 import argparse
 import torch
-import torch_npu
 import torch.nn as nn
 import torch.nn.functional as F
+import torch_npu
 
 from torch_npu.testing.testcase import TestCase, run_tests
 
diff --git a/test/test_api/test_torch_nn/test_containers.py b/test/test_api/test_torch_nn/test_containers.py
index 888a3d6b5af71203d3d69564917aea11c43de25a..a55a7350c61ad7d234a87fd371dcfc0efdd38be8 100644
--- a/test/test_api/test_torch_nn/test_containers.py
+++ b/test/test_api/test_torch_nn/test_containers.py
@@ -14,10 +14,10 @@
 
 from collections import OrderedDict
 import torch
-import torch_npu
 import torch.nn as nn
 import torch.nn.functional as F
 from torch.nn import Parameter
+import torch_npu
 
 from torch_npu.testing.testcase import TestCase, run_tests
 
diff --git a/test/test_api/test_torch_nn/test_conv_layers.py b/test/test_api/test_torch_nn/test_conv_layers.py
index 66b71eeb9637905b05ba1ae1b0080082f9c8e639..2640c552bee344b6a23f92e56733a486a2fe5ab8 100644
--- a/test/test_api/test_torch_nn/test_conv_layers.py
+++ b/test/test_api/test_torch_nn/test_conv_layers.py
@@ -13,8 +13,8 @@
 # limitations under the License.
 
 import torch
-import torch_npu
 import torch.nn as nn
+import torch_npu
 
 from torch_npu.testing.testcase import TestCase, run_tests
 
diff --git a/test/test_api/test_torch_nn/test_loss_functions.py b/test/test_api/test_torch_nn/test_loss_functions.py
index 3064d0e3e7fbad55d369e8cbfc9164a0cf18680a..42f6444a039476ca2786d434cdd009d752a529f9 100644
--- a/test/test_api/test_torch_nn/test_loss_functions.py
+++ b/test/test_api/test_torch_nn/test_loss_functions.py
@@ -12,9 +12,9 @@
 # See the License for the specific language governing permissions and
 # limitations under the License.
 import torch
-import torch_npu
 import torch.nn as nn
 import torch.nn.functional as F
+import torch_npu
 
 from torch_npu.testing.testcase import TestCase, run_tests
 
@@ -55,27 +55,27 @@ class TestLossFunctions(TestCase):
         S_min = 10  # Minimum target length, for demonstration purposes
 
          # Initialize random batch of input vectors, for *size = (T,N,C)
-        input1 = torch.randn(T, N, C).log_softmax(2).detach()
-        ninput = input1.npu()
+        input1 = torch.randn(T, N, C).npu().log_softmax(2).detach()
+        cinput = input1.cpu()
         input1.requires_grad_(True)
-        ninput.requires_grad_(True)
+        cinput.requires_grad_(True)
          # Initialize random batch of targets (0 = blank, 1:C = classes)
-        target = torch.randint(low=1, high=C, size=(N, S), dtype=torch.long)
+        target = torch.randint(low=1, high=C, size=(N, S), dtype=torch.int32).npu()
 
-        input_lengths = torch.full(size=(N,), fill_value=T, dtype=torch.long)
-        target_lengths = torch.randint(low=S_min, high=S, size=(N,), dtype=torch.long)
-        ctc_loss = nn.CTCLoss()
+        input_lengths = torch.full(size=(N,), fill_value=T, dtype=torch.int32).npu()
+        target_lengths = torch.randint(low=S_min, high=S, size=(N,), dtype=torch.int32).npu()
+        ctc_loss = nn.CTCLoss().npu()
         loss = ctc_loss(input1, target, input_lengths, target_lengths)
         loss.backward(torch.ones_like(loss))
 
-        ntarget = target.npu()
-        ninput_lengths = input_lengths.npu()
-        ntarget_lengths = target_lengths.npu()
-        nctc_loss = ctc_loss.npu()
-        nloss = nctc_loss(ninput, ntarget, ninput_lengths, ntarget_lengths)
-        nloss.backward(torch.ones_like(nloss))
-        self.assertRtolEqual(loss.detach().numpy(), nloss.detach().cpu().numpy())
-        self.assertRtolEqual(input1.grad.numpy(), ninput.grad.cpu().numpy())
+        ctarget = target.cpu().long()
+        cinput_lengths = input_lengths.cpu().long()
+        ctarget_lengths = target_lengths.cpu().long()
+        cctc_loss = ctc_loss.cpu()
+        closs = cctc_loss(cinput, ctarget, cinput_lengths, ctarget_lengths)
+        closs.backward(torch.ones_like(closs))
+        self.assertRtolEqual(loss.detach().cpu().numpy(), closs.detach().numpy())
+        self.assertRtolEqual(input1.grad.cpu().numpy(), cinput.grad.numpy())
 
     def test_NLLLoss(self):
         m = nn.LogSoftmax(dim=1)
@@ -211,5 +211,4 @@ class TestLossFunctions(TestCase):
 
 
 if __name__ == "__main__":
-    torch.npu.set_device(0)
     run_tests()
\ No newline at end of file
diff --git a/test/test_api/test_torch_nn/test_nonlinear_activations.py b/test/test_api/test_torch_nn/test_nonlinear_activations.py
index c5556115a2bdb806b88b6f93710021b3e9e33c3d..2b6be84d1082075264a1cef71ede56c4b1a26528 100644
--- a/test/test_api/test_torch_nn/test_nonlinear_activations.py
+++ b/test/test_api/test_torch_nn/test_nonlinear_activations.py
@@ -13,8 +13,8 @@
 # limitations under the License.
 
 import torch
-import torch_npu
 import torch.nn as nn
+import torch_npu
 
 from torch_npu.testing.testcase import TestCase, run_tests
 
@@ -148,6 +148,5 @@ class TestOtherNonLinearActivations(TestCase):
         
         
 if __name__ == "__main__":
-    torch.npu.set_device(0)
     run_tests()
 
diff --git a/test/test_api/test_torch_nn/test_normalization_layers.py b/test/test_api/test_torch_nn/test_normalization_layers.py
index e2f1d73cb18bf5a3f3cce93ad049c2dee7852900..0b8651e2053a8165261ad7eef4ed5d9258722dea 100644
--- a/test/test_api/test_torch_nn/test_normalization_layers.py
+++ b/test/test_api/test_torch_nn/test_normalization_layers.py
@@ -13,8 +13,8 @@
 # limitations under the License.
 
 import torch
-import torch_npu
 import torch.nn as nn
+import torch_npu
 
 from torch_npu.testing.testcase import TestCase, run_tests
 
@@ -89,6 +89,5 @@ class TestNormalizationLayers(TestCase):
 
         
 if __name__ == "__main__":
-    torch.npu.set_device(0)
     run_tests()
 
diff --git a/test/test_api/test_torch_nn/test_padding_layers.py b/test/test_api/test_torch_nn/test_padding_layers.py
index c4f1b61614a2457d352c51ecc86471b340b5693c..a3f0178cc717a0693e1a5544eb50263029673700 100644
--- a/test/test_api/test_torch_nn/test_padding_layers.py
+++ b/test/test_api/test_torch_nn/test_padding_layers.py
@@ -13,8 +13,8 @@
 # limitations under the License.
 
 import torch
-import torch_npu
 import torch.nn as nn
+import torch_npu
 
 from torch_npu.testing.testcase import TestCase, run_tests
 
diff --git a/test/test_api/test_torch_nn/test_pooling_layers.py b/test/test_api/test_torch_nn/test_pooling_layers.py
index 7226a6122c0119d063ffd55e9c773148e918cff6..0386150bc52850f94a7f4e6cb44aa4ce695ffd3c 100644
--- a/test/test_api/test_torch_nn/test_pooling_layers.py
+++ b/test/test_api/test_torch_nn/test_pooling_layers.py
@@ -13,8 +13,8 @@
 # limitations under the License.
 
 import torch
-import torch_npu
 import torch.nn as nn
+import torch_npu
 
 from torch_npu.testing.testcase import TestCase, run_tests
 
@@ -117,6 +117,4 @@ class TestPoolingLayers(TestCase):
         
         
 if __name__ == "__main__":
-    torch.npu.set_device(0)
-    run_tests()
-
+    run_tests()
\ No newline at end of file
diff --git a/test/test_api/test_torch_nn/test_recurrent_layers.py b/test/test_api/test_torch_nn/test_recurrent_layers.py
index 54c6174dc8d66ed6eb0eb575534dbbcef7ec5b0d..be1ae2b1cb25b516923a0a3a0df68584ff05a46d 100644
--- a/test/test_api/test_torch_nn/test_recurrent_layers.py
+++ b/test/test_api/test_torch_nn/test_recurrent_layers.py
@@ -13,8 +13,8 @@
 # limitations under the License.
 
 import torch
-import torch_npu
 import torch.nn as nn
+import torch_npu
 
 from torch_npu.testing.testcase import TestCase, run_tests
 
diff --git a/test/test_api/test_torch_nn/test_sparse_layers.py b/test/test_api/test_torch_nn/test_sparse_layers.py
index bdae319bc185eaddaab4e3bc7b25dbf1b870b7c8..7aa98c261e9fc12f6e722c6b2a2cb01ce7578555 100644
--- a/test/test_api/test_torch_nn/test_sparse_layers.py
+++ b/test/test_api/test_torch_nn/test_sparse_layers.py
@@ -12,8 +12,8 @@
 # See the License for the specific language governing permissions and
 # limitations under the License.
 import torch
-import torch_npu
 import torch.nn as nn
+import torch_npu
 
 from torch_npu.testing.testcase import TestCase, run_tests
 
@@ -27,5 +27,4 @@ class TestSparseLayers(TestCase):
 
 
 if __name__ == "__main__":
-    torch.npu.set_device(0)
     run_tests()
\ No newline at end of file
diff --git a/test/test_api/test_torch_nn/test_transformer_layers.py b/test/test_api/test_torch_nn/test_transformer_layers.py
index cb6a67943a5bc3ecebeb4982a680bc8fccc8692a..7c4a20dc18a6346a8598c6168f23b5e42edbc407 100644
--- a/test/test_api/test_torch_nn/test_transformer_layers.py
+++ b/test/test_api/test_torch_nn/test_transformer_layers.py
@@ -13,8 +13,8 @@
 # limitations under the License.
 
 import torch
-import torch_npu
 import torch.nn as nn
+import torch_npu
 
 from torch_npu.testing.testcase import TestCase, run_tests
 
diff --git a/test/test_api/test_torch_nn/test_utilities.py b/test/test_api/test_torch_nn/test_utilities.py
index d634ee32dcda681dda0e6f9350d340936fea1f94..b6eaae8691e508c8764c5e784613f7b9d1a42698 100644
--- a/test/test_api/test_torch_nn/test_utilities.py
+++ b/test/test_api/test_torch_nn/test_utilities.py
@@ -15,9 +15,9 @@
 from collections import OrderedDict
 
 import torch
-import torch_npu
 import torch.nn as nn
 from torch.nn.utils import prune
+import torch_npu
 
 from torch_npu.testing.testcase import TestCase, run_tests
 
diff --git a/test/test_api/test_torch_nn/test_vision_layers.py b/test/test_api/test_torch_nn/test_vision_layers.py
index ddda4e4873be689ecf16877c289a3c2f089d00c9..adece30c9ff3ff16056479ca680ff6acd473e598 100644
--- a/test/test_api/test_torch_nn/test_vision_layers.py
+++ b/test/test_api/test_torch_nn/test_vision_layers.py
@@ -12,8 +12,8 @@
 # See the License for the specific language governing permissions and
 # limitations under the License.
 import torch
-import torch_npu
 import torch.nn as nn
+import torch_npu
 
 from torch_npu.testing.testcase import TestCase, run_tests
 
diff --git a/test/test_api/test_torch_npu.py b/test/test_api/test_torch_npu.py
index 0f0d1c95257deee834ed99ed24d15ca5b2ec6781..8c0559011d25686b1fad6a133caa018529b25db0 100644
--- a/test/test_api/test_torch_npu.py
+++ b/test/test_api/test_torch_npu.py
@@ -1,3 +1,17 @@
+# Copyright (c) 2020, Huawei Technologies.All rights reserved.
+#
+# Licensed under the BSD 3-Clause License  (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+# https://opensource.org/licenses/BSD-3-Clause
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
 import contextlib
 import collections
 
diff --git a/test/test_optimized_lib/test_activations.py b/test/test_optimized_lib/test_activations.py
index 95057ca59ead861ce69ef89e551d66c9aaefe16a..ce8cb72e192ed5261874dc7136f4397c422de312 100644
--- a/test/test_optimized_lib/test_activations.py
+++ b/test/test_optimized_lib/test_activations.py
@@ -13,13 +13,13 @@
 # See the License for the specific language governing permissions and
 # limitations under the License.
 
+import numpy as np
 import torch
-import torch_npu
 import torch.nn.functional as F
-import numpy as np
+
+import torch_npu
 from torch_npu.testing.testcase import TestCase, run_tests
 from torch_npu.testing.common_utils import create_common_tensor
-
 from torch_npu.contrib.module import Mish, SiLU
 
 class TestActivations(TestCase):
diff --git a/test/test_optimized_lib/test_index_op.py b/test/test_optimized_lib/test_index_op.py
index 5716e4847c7a0d4ad11960f23aba81d5ada0b4f4..dbb11bcb2b03cdb142a6343d9e272687dee0b13f 100644
--- a/test/test_optimized_lib/test_index_op.py
+++ b/test/test_optimized_lib/test_index_op.py
@@ -13,9 +13,9 @@
 # See the License for the specific language governing permissions and
 # limitations under the License.
 
+import numpy as np
 import torch
 import torch_npu
-import numpy as np
 
 from torch_npu.testing.testcase import TestCase, run_tests
 from torch_npu.testing.common_utils import create_common_tensor
@@ -35,7 +35,7 @@ class TestIndexOp(TestCase):
 
     def test_npu_index_op(self):
         dtype_list = [np.float16, np.float32]
-        format_list = [-1]
+        format_list = [-1, 0, 2]
         shape_list = [
             [2, 3, 7, 7],
             [1, 2, 3, 6, 6],
diff --git a/torch_npu/contrib/function/anchor_generator.py b/torch_npu/contrib/function/anchor_generator.py
index 9cb4cf85a6e4eb28bd8759e51c8e179327ac2848..2c1d8b180b05bd231d9eb5344c8923cd03bfaed8 100644
--- a/torch_npu/contrib/function/anchor_generator.py
+++ b/torch_npu/contrib/function/anchor_generator.py
@@ -18,6 +18,7 @@ import torch_npu
 def box_dtype_check(box):
     if box not in [torch.float, torch.half]:
         return box.float()
+    return box
 
 def npu_single_level_responsible_flags(featmap_size,
                                        gt_bboxes,
diff --git a/torch_npu/contrib/function/bbox_coder.py b/torch_npu/contrib/function/bbox_coder.py
index 6c8929595d843190ac9581dc4806d71cef3311fe..0c39da0a8507c0938b8239753b3e3b4cac601707 100644
--- a/torch_npu/contrib/function/bbox_coder.py
+++ b/torch_npu/contrib/function/bbox_coder.py
@@ -18,11 +18,13 @@ import torch_npu
 def box_dtype_check(box):
     if box not in [torch.float, torch.half]:
         return box.float()
+    return box
 
 
 def stride_dtype_check(stride):
     if stride not in [torch.int]:
         return stride.int()
+    return stride
 
 
 def npu_bbox_coder_encode_yolo(bboxes, gt_bboxes, stride):
diff --git a/torch_npu/contrib/function/iou.py b/torch_npu/contrib/function/iou.py
index 48ee1b650570a56858ad61134f5afa3b3f78e2af..d97fa3b6579b96194876c7537be0835cbc014858 100644
--- a/torch_npu/contrib/function/iou.py
+++ b/torch_npu/contrib/function/iou.py
@@ -18,6 +18,7 @@ import torch_npu
 def box_dtype_check(box):
     if box not in [torch.float, torch.half]:
         return box.float()
+    return box
 
 
 def npu_iou(boxes1,
diff --git a/torch_npu/contrib/module/activations.py b/torch_npu/contrib/module/activations.py
index 2776d2a4769bce0ed979ef902cc544e111277b60..8f27383d8cba6d21918274b9749b7359a19cc0bc 100644
--- a/torch_npu/contrib/module/activations.py
+++ b/torch_npu/contrib/module/activations.py
@@ -13,8 +13,8 @@
 # limitations under the License.
 
 import torch
-import torch_npu
 import torch.nn as nn
+import torch_npu
 
 class Mish(nn.Module):
     def __init__(self):
diff --git a/torch_npu/contrib/module/channel_shuffle.py b/torch_npu/contrib/module/channel_shuffle.py
index 07bc37b1d6b83d77ecc8134ca536c30f2f5f196c..0a9c859e8ef46121977048d8daa95f1d1829a89c 100644
--- a/torch_npu/contrib/module/channel_shuffle.py
+++ b/torch_npu/contrib/module/channel_shuffle.py
@@ -14,8 +14,8 @@
 
 import numpy as np
 import torch
-import torch_npu
 import torch.nn as nn
+import torch_npu
 
 
 class ChannelShuffle(nn.Module):
diff --git a/torch_npu/contrib/module/crossentropy.py b/torch_npu/contrib/module/crossentropy.py
index 5cf2932b3808fc25b6ac5b34b8836bee038d9d66..f7593d625468a3277bef5645827d249b4b4d3b00 100644
--- a/torch_npu/contrib/module/crossentropy.py
+++ b/torch_npu/contrib/module/crossentropy.py
@@ -13,8 +13,8 @@
 # limitations under the License.
 
 import torch
-import torch_npu
 import torch.nn as nn
+import torch_npu
 
 
 class LabelSmoothingCrossEntropy(nn.Module):
diff --git a/torch_npu/contrib/module/deform_conv.py b/torch_npu/contrib/module/deform_conv.py
index 47fde7719d9f7d3d40ad306c630fca56da7c9404..c273bf33058f66b15618be7c1ae2542a0e197c28 100644
--- a/torch_npu/contrib/module/deform_conv.py
+++ b/torch_npu/contrib/module/deform_conv.py
@@ -14,10 +14,10 @@
 
 import math
 import torch
-import torch_npu
 import torch.nn as nn
 from torch.autograd import Function
 from torch.nn.modules.utils import _pair, _single
+import torch_npu
 
 
 class ModulatedDeformConv2dFunction(Function):
@@ -86,7 +86,61 @@ class ModulatedDeformConv2dFunction(Function):
 
 
 class ModulatedDeformConv(nn.Module):
-
+    r"""Applies an NPU based Modulated Deformable 2D convolution operation.
+
+    Paper link:
+    [Deformable ConvNets v2: More Deformable, Better Results](https://arxiv.org/abs/1811.11168)
+
+    Reference implementation link:
+    https://github.com/open-mmlab/mmcv/blob/master/mmcv/ops/modulated_deform_conv.py
+
+    The implementation of this ModulatedDeformConv is mainly based
+    on the implementation of mmcv for design and reconstruction.
+
+    In ModulatedDeformConvFunction, the forward and backward are customized,
+    and the input tensor is reconstructed ito match the NPU based function.
+
+    It is worth mentioning that DeformConv(DCNv1) is also implemented
+    by setting modulated==False. Due to the difference between input
+    and initialization, there is no additional implementation here.
+
+    .. note::
+        ModulatedDeformConv only implements operations under fp32 data types.
+        Notice, weight and bias in conv_offset must be initialized to 0.
+
+    Args:
+        in_channels (int): Number of channels in the input image.
+        out_channels (int): Number of channels produced by the convolution.
+        kernel_size(int, tuple): Size of the convolving kernel.
+        stride(int, tuple): Stride of the convolution. Default: 1.
+        padding (int or tuple): Zero-padding added to both sides of the input.
+            Default: 0.
+        dilation (int or tuple): Spacing between kernel elements. Default: 1.
+        groups (int): Number of blocked connections from input.
+            channels to output channels. Default: 1.
+        deform_groups (int): Number of deformable group partitions.
+        bias (bool): If True, adds a learnable bias to the output. Default: False.
+        pack (bool): If True, conv_offset and mask will be included in this module. Default: True.
+
+    Examples::
+        >>> m = ModulatedDeformConv(32, 32, 1)
+        >>> input_tensor = torch.randn(2, 32, 5, 5)
+        >>> output = m(input_tensor)
+
+
+        >>> x = torch.randn(2, 32, 7, 7)
+        >>> model = ModulatedDeformConv(32, 32, 3, 2, 1)
+
+        >>> torch.npu.set_device(0)
+        >>> x = x.npu()
+        >>> model = model.npu()
+
+        >>> o = model(x)
+        >>> l = o.sum()
+        >>> l.backward()
+        >>> print(l)
+    """
+    
     def __init__(self,
                  in_channels,
                  out_channels,
@@ -99,64 +153,7 @@ class ModulatedDeformConv(nn.Module):
                  bias=True,
                  pack=True,
                  ):
-
-        r"""Applies an NPU based Modulated Deformable 2D convolution operation.
-
-        Paper link:
-        [Deformable ConvNets v2: More Deformable, Better Results](https://arxiv.org/abs/1811.11168)
-
-        Reference implementation link:
-        https://github.com/open-mmlab/mmcv/blob/master/mmcv/ops/modulated_deform_conv.py
-
-        The implementation of this ModulatedDeformConv is mainly based
-        on the implementation of mmcv for design and reconstruction.
-
-        In ModulatedDeformConvFunction, the forward and backward are customized,
-        and the input tensor is reconstructed ito match the NPU based function.
-
-        It is worth mentioning that DeformConv(DCNv1) is also implemented
-        by setting modulated==False. Due to the difference between input
-        and initialization, there is no additional implementation here.
-
-        .. note::
-            ModulatedDeformConv only implements operations under fp32 data types.
-            Notice, weight and bias in conv_offset must be initialized to 0.
-
-        Args:
-            in_channels (int): Number of channels in the input image.
-            out_channels (int): Number of channels produced by the convolution.
-            kernel_size(int, tuple): Size of the convolving kernel.
-            stride(int, tuple): Stride of the convolution. Default: 1.
-            padding (int or tuple): Zero-padding added to both sides of the input.
-                Default: 0.
-            dilation (int or tuple): Spacing between kernel elements. Default: 1.
-            groups (int): Number of blocked connections from input.
-                channels to output channels. Default: 1.
-            deform_groups (int): Number of deformable group partitions.
-            bias (bool): If True, adds a learnable bias to the output. Default: False.
-            pack (bool): If True, conv_offset and mask will be included in this module. Default: True.
-
-        Examples::
-            >>> m = ModulatedDeformConv(32, 32, 1)
-            >>> input_tensor = torch.randn(2, 32, 5, 5)
-            >>> output = m(input_tensor)
-
-
-            >>> x = torch.randn(2, 32, 7, 7)
-            >>> model = ModulatedDeformConv(32, 32, 3, 2, 1)
-
-            >>> torch.npu.set_device(0)
-            >>> x = x.npu()
-            >>> model = model.npu()
-
-            >>> o = model(x)
-            >>> l = o.sum()
-            >>> l.backward()
-            >>> print(l)
-        """
-
         super(ModulatedDeformConv, self).__init__()
-
         self.in_channels = in_channels
         self.out_channels = out_channels
         self.kernel_size = _pair(kernel_size)
@@ -167,14 +164,12 @@ class ModulatedDeformConv(nn.Module):
         self.deformable_groups = deformable_groups
         self.with_bias = bias
         self.pack = pack
-
         self.weight = nn.Parameter(
             torch.Tensor(out_channels, in_channels // groups, *self.kernel_size))
         if bias:
             self.bias = nn.Parameter(torch.Tensor(out_channels))
         else:
             self.bias = torch.zeros(self.weight.shape[0])
-
         if self.pack:
             self.conv_offset = nn.Conv2d(
                 self.in_channels,
@@ -184,7 +179,6 @@ class ModulatedDeformConv(nn.Module):
                 stride=_pair(self.stride),
                 padding=_pair(self.padding),
                 bias=True)
-
         self.split_num = self.deformable_groups * 2 * self.kernel_size[0] * self.kernel_size[1]
         sort_index_for_npu = list(range(self.split_num))
         sort_index_for_npu_fp = sort_index_for_npu[1::2] + sort_index_for_npu[::2]
@@ -193,7 +187,6 @@ class ModulatedDeformConv(nn.Module):
         self.sort_index_for_npu_fp = torch.IntTensor(sort_index_for_npu_fp)
         self.sort_index_for_npu_bp = torch.IntTensor(sort_index_for_npu_bp)
         self.sort_index_for_npu_todevice = False
-
         self.init_param()
 
     def init_param(self):
diff --git a/torch_npu/contrib/module/ps_roi_pooling.py b/torch_npu/contrib/module/ps_roi_pooling.py
index 05b708eab7034776b0d6e24e2c291090ecdda36f..5296ab2ac0933f50d995d8868deacae0d89f5333 100644
--- a/torch_npu/contrib/module/ps_roi_pooling.py
+++ b/torch_npu/contrib/module/ps_roi_pooling.py
@@ -13,8 +13,8 @@
 # See the License for the specific language governing permissions and
 # limitations under the License.
 import torch
-import torch_npu
 from torch import nn
+import torch_npu
 
 class PSROIPool(nn.Module):
     def __init__(self, pooled_height=7, pooled_width=7, spatial_scale=1 / 16.0, group_size=7, output_dim=22):
diff --git a/torch_npu/contrib/module/roi_align.py b/torch_npu/contrib/module/roi_align.py
index 8aeb656d161af9cbc7cd8aa1617067f3320abe4d..0d5c15839f0e38900a3cb8a54cd707a722f1d9ab 100644
--- a/torch_npu/contrib/module/roi_align.py
+++ b/torch_npu/contrib/module/roi_align.py
@@ -14,12 +14,11 @@
 # See the License for the specific language governing permissions and
 # limitations under the License.
 import torch
-import torch_npu
 from torch import nn
-
 from torch.nn.modules.utils import _pair
 from torch.autograd import Function
 from torch.autograd.function import once_differentiable
+import torch_npu
 
 
 class _ROIAlign(Function):
diff --git a/torch_npu/utils/serialization.py b/torch_npu/utils/serialization.py
index c9a2c5ce77bad65bf73a90dd27c921ec55e9b45e..cbfcdde23a3f0e7e65a97331d8f782ace38f7fe7 100644
--- a/torch_npu/utils/serialization.py
+++ b/torch_npu/utils/serialization.py
@@ -17,10 +17,10 @@
 import pickle
 import argparse
 import torch
-import torch_npu
 import torch.nn as nn
 import torch.serialization as se
 from torch._six import container_abcs, string_classes
+import torch_npu
 
 DEFAULT_PROTOCOL = 2