diff --git a/paddle/fluid/operators/squeeze_op_npu.cc b/paddle/fluid/operators/squeeze_op_npu.cc
new file mode 100644
index 0000000000000000000000000000000000000000..344aaaa4baac26ea584b054d5c60d2566ebf3569
--- /dev/null
+++ b/paddle/fluid/operators/squeeze_op_npu.cc
@@ -0,0 +1,45 @@
+/* Copyright (c) 2021 PaddlePaddle Authors. All Rights Reserved.
+
+Licensed under the Apache License, Version 2.0 (the "License");
+you may not use this file except in compliance with the License.
+You may obtain a copy of the License at
+
+    http://www.apache.org/licenses/LICENSE-2.0
+
+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. */
+
+#ifdef PADDLE_WITH_ASCEND_CL
+#include <memory>
+#include <string>
+
+#include "paddle/fluid/operators/squeeze_op.h"
+#include "paddle/fluid/operators/npu_op_runner.h"
+
+namespace ops = paddle::operators;
+namespace plat = paddle::platform;
+
+REGISTER_OP_NPU_KERNEL(
+    squeeze,
+    ops::SqueezeKernel<plat::NPUDeviceContext, float>,
+    ops::SqueezeKernel<plat::NPUDeviceContext, double>,
+    ops::SqueezeKernel<plat::NPUDeviceContext, plat::float16>,
+    ops::SqueezeKernel<plat::NPUDeviceContext, bool>,
+    ops::SqueezeKernel<plat::NPUDeviceContext, int>,
+    ops::SqueezeKernel<plat::NPUDeviceContext, uint8_t>,
+    ops::SqueezeKernel<plat::NPUDeviceContext, int8_t>,
+    ops::SqueezeKernel<plat::NPUDeviceContext, int64_t>);
+REGISTER_OP_NPU_KERNEL(
+    squeeze2,
+    ops::SqueezeKernel<plat::NPUDeviceContext, float>,
+    ops::SqueezeKernel<plat::NPUDeviceContext, double>,
+    ops::SqueezeKernel<plat::NPUDeviceContext, plat::float16>,
+    ops::SqueezeKernel<plat::NPUDeviceContext, bool>,
+    ops::SqueezeKernel<plat::NPUDeviceContext, int>,
+    ops::SqueezeKernel<plat::NPUDeviceContext, uint8_t>,
+    ops::SqueezeKernel<plat::NPUDeviceContext, int8_t>,
+    ops::SqueezeKernel<plat::NPUDeviceContext, int64_t>);
+#endif
diff --git a/paddle/fluid/operators/squeeze_op_npu_test.cc b/paddle/fluid/operators/squeeze_op_npu_test.cc
new file mode 100644
index 0000000000000000000000000000000000000000..9b0464fc2234bba71645bbe586e30d580eca0c19
--- /dev/null
+++ b/paddle/fluid/operators/squeeze_op_npu_test.cc
@@ -0,0 +1,92 @@
+/* Copyright (c) 2021 PaddlePaddle Authors. All Rights Reserved.
+
+Licensed under the Apache License, Version 2.0 (the "License");
+you may not use this file except in compliance with the License.
+You may obtain a copy of the License at
+
+    http://www.apache.org/licenses/LICENSE-2.0
+
+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. */
+
+#ifndef _WIN32
+#include <unistd.h>
+#endif
+
+#include <string>
+#include <thread>  // NOLINT
+#include <vector>
+
+#include "gtest/gtest.h"
+#include "paddle/fluid/framework/op_registry.h"
+#include "paddle/fluid/framework/operator.h"
+#include "paddle/fluid/framework/program_desc.h"
+#include "paddle/fluid/operators/dropout_op.h"
+#include "paddle/fluid/operators/math/math_function.h"
+#include "paddle/fluid/string/printf.h"
+
+namespace f = paddle::framework;
+namespace p = paddle::platform;
+namespace m = paddle::operators::math;
+
+USE_OP(squeeze);
+USE_OP_DEVICE_KERNEL(squeeze, NPU);
+
+template <typename T>
+void Compare(f::Scope* scope, const p::DeviceContext& ctx) {
+  // init
+  auto x = scope->Var("X");
+  auto tensor_x = x->GetMutable<f::LoDTensor>();
+
+  int dim0 = 1;
+  int dim1 = 10;
+  int dim2 = 1;
+
+  std::vector<T> init;
+  for (int64_t i = 0; i < dim0 * dim1 * dim2; ++i) {
+    init.push_back(static_cast<T>(0.1));
+  }
+
+  TensorFromVector(init, ctx, tensor_x);
+  tensor_x->Resize({dim0, dim1, dim2});
+
+  ctx.Wait();
+
+  // run
+  auto place = ctx.GetPlace();
+  auto out = scope->Var("Out");
+  auto tensor_out = out->GetMutable<f::LoDTensor>();
+
+  std::vector<int> axis;
+  axis.push_back(2);
+  f::AttributeMap attrs = {{"axes", axis}};
+
+  auto op =
+      f::OpRegistry::CreateOp("squeeze", {{"X", {"X"}}},
+                              {{"Out", {"Out"}}}, attrs);
+
+  op->Run(*scope, place);
+  ctx.Wait();
+
+  EXPECT_EQ((uint32_t)tensor_out->dims().size(), uint32_t(2));
+  EXPECT_EQ((uint32_t)tensor_out->dims()[0], uint32_t(dim0));
+  EXPECT_EQ((uint32_t)tensor_out->dims()[1], uint32_t(dim1));
+  
+  std::vector<T> out_vec;
+  TensorToVector(*tensor_out, ctx, &out_vec);
+  for (uint32_t i = 0; i < out_vec.size(); i++) {
+    EXPECT_EQ(out_vec[i], static_cast<T>(0.1));
+  }
+
+  ctx.Wait();
+}
+
+TEST(squeeze, NPU_fp32) {
+  f::Scope scope;
+  p::NPUDeviceContext ctx(p::NPUPlace(0));
+  Compare<float>(&scope, ctx);
+}
+
diff --git a/paddle/fluid/operators/unsqueeze_op_npu.cc b/paddle/fluid/operators/unsqueeze_op_npu.cc
new file mode 100644
index 0000000000000000000000000000000000000000..c37fc44b2e26a95f0c2533860f34c3a54c916039
--- /dev/null
+++ b/paddle/fluid/operators/unsqueeze_op_npu.cc
@@ -0,0 +1,44 @@
+/* Copyright (c) 2021 PaddlePaddle Authors. All Rights Reserved.
+
+Licensed under the Apache License, Version 2.0 (the "License");
+you may not use this file except in compliance with the License.
+You may obtain a copy of the License at
+
+    http://www.apache.org/licenses/LICENSE-2.0
+
+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. */
+
+#ifdef PADDLE_WITH_ASCEND_CL
+#include <memory>
+#include <string>
+
+#include "paddle/fluid/operators/unsqueeze_op.h"
+#include "paddle/fluid/operators/npu_op_runner.h"
+
+namespace ops = paddle::operators;
+namespace plat = paddle::platform;
+
+REGISTER_OP_NPU_KERNEL(
+    unsqueeze,
+    ops::UnsqueezeKernel<plat::NPUDeviceContext, float>,
+    ops::UnsqueezeKernel<plat::NPUDeviceContext, double>,
+    ops::UnsqueezeKernel<plat::NPUDeviceContext, plat::float16>,
+    ops::UnsqueezeKernel<plat::NPUDeviceContext, bool>,
+    ops::UnsqueezeKernel<plat::NPUDeviceContext, int>,
+    ops::UnsqueezeKernel<plat::NPUDeviceContext, int8_t>,
+    ops::UnsqueezeKernel<plat::NPUDeviceContext, int64_t>);
+REGISTER_OP_NPU_KERNEL(
+    unsqueeze2,
+    ops::UnsqueezeKernel<plat::NPUDeviceContext, float>,
+    ops::UnsqueezeKernel<plat::NPUDeviceContext, double>,
+    ops::UnsqueezeKernel<plat::NPUDeviceContext, plat::float16>,
+    ops::UnsqueezeKernel<plat::NPUDeviceContext, bool>,
+    ops::UnsqueezeKernel<plat::NPUDeviceContext, int>,
+    ops::UnsqueezeKernel<plat::NPUDeviceContext, int8_t>,
+    ops::UnsqueezeKernel<plat::NPUDeviceContext, int64_t>);
+#endif
+
diff --git a/paddle/fluid/operators/unsqueeze_op_npu_test.cc b/paddle/fluid/operators/unsqueeze_op_npu_test.cc
new file mode 100644
index 0000000000000000000000000000000000000000..6b53cc328a1dea6e6259bf9c83ffb8971a7b263c
--- /dev/null
+++ b/paddle/fluid/operators/unsqueeze_op_npu_test.cc
@@ -0,0 +1,92 @@
+/* Copyright (c) 2021 PaddlePaddle Authors. All Rights Reserved.
+
+Licensed under the Apache License, Version 2.0 (the "License");
+you may not use this file except in compliance with the License.
+You may obtain a copy of the License at
+
+    http://www.apache.org/licenses/LICENSE-2.0
+
+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. */
+
+#ifndef _WIN32
+#include <unistd.h>
+#endif
+
+#include <string>
+#include <thread>  // NOLINT
+#include <vector>
+
+#include "gtest/gtest.h"
+#include "paddle/fluid/framework/op_registry.h"
+#include "paddle/fluid/framework/operator.h"
+#include "paddle/fluid/framework/program_desc.h"
+#include "paddle/fluid/operators/dropout_op.h"
+#include "paddle/fluid/operators/math/math_function.h"
+#include "paddle/fluid/string/printf.h"
+
+namespace f = paddle::framework;
+namespace p = paddle::platform;
+namespace m = paddle::operators::math;
+
+USE_OP(unsqueeze);
+USE_OP_DEVICE_KERNEL(unsqueeze, NPU);
+
+template <typename T>
+void Compare(f::Scope* scope, const p::DeviceContext& ctx) {
+  // init
+  auto x = scope->Var("X");
+  auto tensor_x = x->GetMutable<f::LoDTensor>();
+
+  int dim0 = 5;
+  int dim1 = 10;
+
+  std::vector<T> init;
+  for (int64_t i = 0; i < dim0 * dim1; ++i) {
+    init.push_back(static_cast<T>(0.1));
+  }
+
+  TensorFromVector(init, ctx, tensor_x);
+  tensor_x->Resize({dim0, dim1});
+
+  ctx.Wait();
+
+  // run
+  auto place = ctx.GetPlace();
+  auto out = scope->Var("Out");
+  auto tensor_out = out->GetMutable<f::LoDTensor>();
+
+  std::vector<int> axis;
+  axis.push_back(1);
+  f::AttributeMap attrs = {{"axes", axis}};
+
+  auto op =
+      f::OpRegistry::CreateOp("unsqueeze", {{"X", {"X"}}},
+                              {{"Out", {"Out"}}}, attrs);
+
+  op->Run(*scope, place);
+  ctx.Wait();
+
+  EXPECT_EQ((uint32_t)tensor_out->dims().size(), uint32_t(3));
+  EXPECT_EQ((uint32_t)tensor_out->dims()[0], uint32_t(5));
+  EXPECT_EQ((uint32_t)tensor_out->dims()[1], uint32_t(1));
+  EXPECT_EQ((uint32_t)tensor_out->dims()[2], uint32_t(10));
+
+  std::vector<T> out_vec;
+  TensorToVector(*tensor_out, ctx, &out_vec);
+  for (uint32_t i = 0; i < out_vec.size(); i++) {
+    EXPECT_EQ(out_vec[i], static_cast<T>(0.1));
+  }
+
+  ctx.Wait();
+}
+
+TEST(unsqueeze, NPU_fp32) {
+  f::Scope scope;
+  p::NPUDeviceContext ctx(p::NPUPlace(0));
+  Compare<float>(&scope, ctx);
+}
+