diff --git a/aten/src/ATen/native/cpu/SoftMaxKernel.cpp b/aten/src/ATen/native/cpu/SoftMaxKernel.cpp
index 317647123d4c..979e41985075 100644
--- a/aten/src/ATen/native/cpu/SoftMaxKernel.cpp
+++ b/aten/src/ATen/native/cpu/SoftMaxKernel.cpp
@@ -647,8 +647,8 @@ _vec_softmax(
   parallel_for(
       0, outer_size * inner_size, 0, [&](int64_t begin, int64_t end) {
         int64_t idx = begin;
-        std::unique_ptr<float[]> temp_vec_input(new float[dim_size*vectorized_step]());
-        std::unique_ptr<float[]> temp_vec_output(new float[dim_size*vectorized_step]());
+        auto temp_vec_input = std::make_unique<float[]>(dim_size * vectorized_step);
+        auto temp_vec_output = std::make_unique<float[]>(dim_size * vectorized_step);
         float* temp_vec_input_data = temp_vec_input.get();
         float* temp_vec_output_data = temp_vec_output.get();
         while (idx < end) {
diff --git a/aten/src/ATen/native/cudnn/Conv_v7.cpp b/aten/src/ATen/native/cudnn/Conv_v7.cpp
index 4d869e5679f8..7e64af0c6636 100644
--- a/aten/src/ATen/native/cudnn/Conv_v7.cpp
+++ b/aten/src/ATen/native/cudnn/Conv_v7.cpp
@@ -285,7 +285,7 @@ struct algorithm_search<cudnnConvolutionFwdAlgoPerf_t> {
         sizeof(algos) / sizeof(algos[0]) == num_algos,
         "Missing cuDNN convolution forward algorithms");
     int perf_count;
-    std::unique_ptr<perf_t[]> perf_results(new perf_t[num_algos]);
+    auto perf_results = std::make_unique<perf_t[]>(num_algos);
     if (!benchmark) {
       AT_CUDNN_CHECK_WITH_SHAPES(
           cudnnGetConvolutionForwardAlgorithm_v7(
@@ -369,7 +369,7 @@ struct algorithm_search<cudnnConvolutionBwdDataAlgoPerf_t> {
         sizeof(algos) / sizeof(algos[0]) == num_algos,
         "Missing cuDNN convolution backward data algorithms.");
     int perf_count;
-    std::unique_ptr<perf_t[]> perf_results(new perf_t[num_algos]);
+    auto perf_results = std::make_unique<perf_t[]>(num_algos);
     if (!benchmark) {
       AT_CUDNN_CHECK_WITH_SHAPES(
           cudnnGetConvolutionBackwardDataAlgorithm_v7(
@@ -456,7 +456,7 @@ struct algorithm_search<cudnnConvolutionBwdFilterAlgoPerf_t> {
     static_assert(
         sizeof(algos) / sizeof(algos[0]) == num_algos,
         "Missing cuDNN convolution backward filter algorithms.");
-    std::unique_ptr<perf_t[]> perf_results(new perf_t[num_algos]);
+    auto perf_results = std::make_unique<perf_t[]>(num_algos);
     int perf_count;
     if (!benchmark) {
       AT_CUDNN_CHECK_WITH_SHAPES(
diff --git a/aten/src/ATen/native/quantized/cpu/UpSampleNearest2d.cpp b/aten/src/ATen/native/quantized/cpu/UpSampleNearest2d.cpp
index e9e32e43ae02..a8f1c8a7d00f 100644
--- a/aten/src/ATen/native/quantized/cpu/UpSampleNearest2d.cpp
+++ b/aten/src/ATen/native/quantized/cpu/UpSampleNearest2d.cpp
@@ -53,7 +53,7 @@ static void upsample_nearest2d_out_frame(
     return;
   }
 
-  std::unique_ptr<int64_t []> input_offset_arr(new int64_t[output_width]);
+  auto input_offset_arr = std::make_unique<int64_t[]>(output_width);
   int64_t* input_offset = input_offset_arr.get();
 
   for (const auto w2 : c10::irange(output_width)) {
diff --git a/aten/src/ATen/native/sparse/cuda/SparseCUDATensorMath.cu b/aten/src/ATen/native/sparse/cuda/SparseCUDATensorMath.cu
index fe0ddd087dd3..8013dc989189 100644
--- a/aten/src/ATen/native/sparse/cuda/SparseCUDATensorMath.cu
+++ b/aten/src/ATen/native/sparse/cuda/SparseCUDATensorMath.cu
@@ -800,7 +800,7 @@ Tensor& bmm_out_sparse_cuda(const SparseTensor& self, const Tensor& mat2, Tensor
   Tensor indices_dim1 = indices[1].to(ScalarType::Int);
   Tensor indices_dim2 = indices[2].to(ScalarType::Int);
 
-  std::unique_ptr<int64_t[]> mat_el_end_indices_host(new int64_t[num_matrices]);
+  auto mat_el_end_indices_host = std::make_unique<int64_t[]>(num_matrices);
 
   {
     auto& allocator = *::c10::cuda::CUDACachingAllocator::get();
diff --git a/functorch/csrc/dim/minpybind.h b/functorch/csrc/dim/minpybind.h
index e1ac428864a9..ceced399b40d 100644
--- a/functorch/csrc/dim/minpybind.h
+++ b/functorch/csrc/dim/minpybind.h
@@ -602,7 +602,7 @@ struct vector_args {
             _PyArg_ParseStackAndKeywords((PyObject*const*)args, nargs, kwnames.ptr(), _parser, &dummy, &dummy, &dummy, &dummy, &dummy);
 #else
             _PyArg_Parser* _parser = new _PyArg_Parser{NULL, &names_buf[0], fname_cstr, 0};
-            std::unique_ptr<PyObject*[]> buf(new PyObject*[names.size()]);
+            auto buf = std::make_unique<PyObject*[]>(names.size());
             _PyArg_UnpackKeywords((PyObject*const*)args, nargs, NULL, kwnames.ptr(), _parser, required, (Py_ssize_t)values.size() - kwonly, 0, &buf[0]);
 #endif
             throw exception_set();
diff --git a/torch/_inductor/codegen/cpp_micro_gemm.py b/torch/_inductor/codegen/cpp_micro_gemm.py
index 13d946863425..67c725b3a53d 100644
--- a/torch/_inductor/codegen/cpp_micro_gemm.py
+++ b/torch/_inductor/codegen/cpp_micro_gemm.py
@@ -195,7 +195,7 @@ inline void {{kernel_name}}(
     ALLOCATE_WEIGHT_BUFFER = r"""
     {%- if is_msvc_compiler %}
     // MSVC doesn't support stack-allocated dynamic-sized arrays, so using heap memory here.
-    std::unique_ptr<{{buffer_dtype}}[]> heap_deq_b_buf_ptr(new {{buffer_dtype}}[{{buffer_size}}]);
+    auto heap_deq_b_buf_ptr = std::make_unique<{{buffer_dtype}}[]>({{buffer_size}});
     {{buffer_dtype}}* {{buffer_name}} = heap_deq_b_buf_ptr.get();
     {%- else %}
     // It's safe to use a stack-allocated array since the blocking strategy would
diff --git a/torch/csrc/distributed/c10d/ProcessGroupGloo.cpp b/torch/csrc/distributed/c10d/ProcessGroupGloo.cpp
index 9625bbfdde35..1391067c7ec4 100644
--- a/torch/csrc/distributed/c10d/ProcessGroupGloo.cpp
+++ b/torch/csrc/distributed/c10d/ProcessGroupGloo.cpp
@@ -528,7 +528,7 @@ std::shared_ptr<::gloo::transport::Device> ProcessGroupGloo::
   // use. Note: if the hostname does not resolve to an address (e.g.
   // because of misconfigured /etc/hosts file), this will not work.
   const auto hostNameMax = sysconf(_SC_HOST_NAME_MAX);
-  auto hostname = std::unique_ptr<char[]>(new char[hostNameMax]);
+  auto hostname = std::make_unique<char[]>(hostNameMax);
   auto rv = gethostname(hostname.get(), hostNameMax);
   if (rv != 0) {
     C10_THROW_ERROR(DistBackendError, c10::utils::str_error(errno));
diff --git a/torch/csrc/serialization.cpp b/torch/csrc/serialization.cpp
index b2bcabc363c1..539106ec02a3 100644
--- a/torch/csrc/serialization.cpp
+++ b/torch/csrc/serialization.cpp
@@ -359,7 +359,7 @@ c10::intrusive_ptr<c10::StorageImpl> THPStorage_readFileRaw(
     data = static_cast<uint8_t*>(storage->mutable_data());
   } else {
     // NOLINTNEXTLINE(cppcoreguidelines-avoid-c-arrays,modernize-avoid-c-arrays)
-    cpu_data = std::unique_ptr<char[]>(new char[nbytes]);
+    cpu_data = std::make_unique<char[]>(nbytes);
     data = (uint8_t*)cpu_data.get();
   }
 
diff --git a/torch/library.h b/torch/library.h
index ea3f2183a23d..f906e04ddecf 100644
--- a/torch/library.h
+++ b/torch/library.h
@@ -926,7 +926,7 @@ class TorchLibraryInit final {
             }
 
       void initialize() {
-        lib = std::unique_ptr<Library>(new Library(kind, ns, key, file, line));
+        lib = std::make_unique<Library>(kind, ns, key, file, line);
         init_function(*lib);
       }
 };