diff --git a/docs/cpp/source/index.rst b/docs/cpp/source/index.rst
index 6bf375e05129..572ed031b864 100644
--- a/docs/cpp/source/index.rst
+++ b/docs/cpp/source/index.rst
@@ -61,15 +61,15 @@ a taste of this interface:
   #include <torch/csrc/autograd/variable.h>
   #include <torch/csrc/autograd/function.h>
 
-  at::Tensor a = torch::ones({2, 2}, at::requires_grad());
-  at::Tensor b = torch::randn({2, 2});
+  torch::Tensor a = torch::ones({2, 2}, torch::requires_grad());
+  torch::Tensor b = torch::randn({2, 2});
   auto c = a + b;
   c.backward(); // a.grad() will now hold the gradient of c w.r.t. a.
 
 The ``at::Tensor`` class in ATen is not differentiable by default. To add the
 differentiability of tensors the autograd API provides, you must use tensor
 factory functions from the `torch::` namespace instead of the `at` namespace.
-For example, while a tensor created with `at::ones` will not be differentiable,
+For example, while a tensor created with `torch::ones` will not be differentiable,
 a tensor created with `torch::ones` will be.
 
 C++ Frontend
diff --git a/docs/cpp/source/installing.rst b/docs/cpp/source/installing.rst
index a0f74e7e3bb1..9aabfe4aaa8d 100644
--- a/docs/cpp/source/installing.rst
+++ b/docs/cpp/source/installing.rst
@@ -6,7 +6,7 @@ configuration files required to depend on PyTorch. We call this distribution
 *LibTorch*, and you can download ZIP archives containing the latest LibTorch
 distribution on `our website <https://pytorch.org/get-started/locally/>`_. Below
 is a small example of writing a minimal application that depends on LibTorch
-and uses the ``at::Tensor`` class which comes with the PyTorch C++ API.
+and uses the ``torch::Tensor`` class which comes with the PyTorch C++ API.
 
 Minimal Example
 ---------------
@@ -37,7 +37,7 @@ this:
   target_link_libraries(example-app "${TORCH_LIBRARIES}")
   set_property(TARGET example-app PROPERTY CXX_STANDARD 11)
 
-The implementation of our example will simply create a new `at::Tensor` and
+The implementation of our example will simply create a new `torch::Tensor` and
 print it:
 
 .. code-block:: cpp
@@ -46,7 +46,7 @@ print it:
   #include <iostream>
 
   int main() {
-    at::Tensor tensor = torch::rand({2, 3});
+    torch::Tensor tensor = torch::rand({2, 3});
     std::cout << tensor << std::endl;
   }
 
diff --git a/test/cpp/api/cursor.cpp b/test/cpp/api/cursor.cpp
index e5d50c1fc60d..c66bfca551db 100644
--- a/test/cpp/api/cursor.cpp
+++ b/test/cpp/api/cursor.cpp
@@ -2,7 +2,7 @@
 
 #include <torch/nn/cursor.h>
 #include <torch/nn/module.h>
-#include <torch/tensor.h>
+#include <torch/types.h>
 #include <torch/utils.h>
 
 #include <test/cpp/api/support.h>
diff --git a/test/cpp/api/dataloader.cpp b/test/cpp/api/dataloader.cpp
index 786923698d57..99273b48695a 100644
--- a/test/cpp/api/dataloader.cpp
+++ b/test/cpp/api/dataloader.cpp
@@ -3,7 +3,7 @@
 #include <torch/data.h>
 #include <torch/data/detail/sequencers.h>
 #include <torch/serialize.h>
-#include <torch/tensor.h>
+#include <torch/types.h>
 
 #include <test/cpp/api/support.h>
 
diff --git a/test/cpp/api/integration.cpp b/test/cpp/api/integration.cpp
index b2d10097b239..7338a5cc8035 100644
--- a/test/cpp/api/integration.cpp
+++ b/test/cpp/api/integration.cpp
@@ -7,7 +7,7 @@
 #include <torch/optim/adam.h>
 #include <torch/optim/optimizer.h>
 #include <torch/optim/sgd.h>
-#include <torch/tensor.h>
+#include <torch/types.h>
 #include <torch/utils.h>
 
 #include <test/cpp/api/support.h>
diff --git a/test/cpp/api/jit.cpp b/test/cpp/api/jit.cpp
index 9aa6968df71f..77326bc4e566 100644
--- a/test/cpp/api/jit.cpp
+++ b/test/cpp/api/jit.cpp
@@ -1,7 +1,7 @@
 #include <gtest/gtest.h>
 
 #include <torch/jit.h>
-#include <torch/tensor.h>
+#include <torch/types.h>
 
 #include <string>
 
diff --git a/test/cpp/api/misc.cpp b/test/cpp/api/misc.cpp
index f3f1cc6f4e3e..169ae9095e17 100644
--- a/test/cpp/api/misc.cpp
+++ b/test/cpp/api/misc.cpp
@@ -3,7 +3,7 @@
 #include <torch/csrc/utils/tempfile.h>
 #include <torch/nn/init.h>
 #include <torch/nn/modules/linear.h>
-#include <torch/tensor.h>
+#include <torch/types.h>
 #include <torch/utils.h>
 
 #include <test/cpp/api/support.h>
diff --git a/test/cpp/api/module.cpp b/test/cpp/api/module.cpp
index 0789b42da44b..7612deac5cf5 100644
--- a/test/cpp/api/module.cpp
+++ b/test/cpp/api/module.cpp
@@ -3,7 +3,7 @@
 #include <torch/nn/module.h>
 #include <torch/nn/modules/linear.h>
 #include <torch/nn/modules/rnn.h>
-#include <torch/tensor.h>
+#include <torch/types.h>
 #include <torch/utils.h>
 
 #include <test/cpp/api/support.h>
@@ -54,8 +54,8 @@ TEST_F(ModuleTest, ZeroGrad) {
 TEST_F(ModuleTest, ZeroGradWithUndefined) {
   struct TestModule : torch::nn::Module {
     TestModule() {
-      x = register_parameter("x", torch::ones(5, at::requires_grad()));
-      y = register_parameter("y", torch::ones(5, at::requires_grad()));
+      x = register_parameter("x", torch::ones(5, torch::requires_grad()));
+      y = register_parameter("y", torch::ones(5, torch::requires_grad()));
     }
     torch::Tensor x, y;
   };
@@ -194,7 +194,7 @@ TEST_F(ModuleTest, Conversion_MultiCUDA) {
       ASSERT_EQ(parameter->device().type(), torch::Device::Type::CUDA);
       ASSERT_EQ(parameter->device().index(), 0);
     }
-    module->to({at::kCUDA, 1});
+    module->to({torch::kCUDA, 1});
     for (auto& parameter : module->parameters()) {
       ASSERT_EQ(parameter->device().type(), torch::Device::Type::CUDA);
       ASSERT_EQ(parameter->device().index(), 1);
diff --git a/test/cpp/api/modules.cpp b/test/cpp/api/modules.cpp
index fd9416eb3b9b..3e67a9f26c03 100644
--- a/test/cpp/api/modules.cpp
+++ b/test/cpp/api/modules.cpp
@@ -7,7 +7,7 @@
 #include <torch/nn/modules/embedding.h>
 #include <torch/nn/modules/functional.h>
 #include <torch/nn/modules/linear.h>
-#include <torch/tensor.h>
+#include <torch/types.h>
 #include <torch/utils.h>
 
 #include <test/cpp/api/support.h>
diff --git a/test/cpp/api/optim.cpp b/test/cpp/api/optim.cpp
index 04f3c28411ad..9490960c548e 100644
--- a/test/cpp/api/optim.cpp
+++ b/test/cpp/api/optim.cpp
@@ -5,7 +5,7 @@
 #include <torch/nn/modules/linear.h>
 #include <torch/nn/modules/sequential.h>
 #include <torch/optim.h>
-#include <torch/tensor.h>
+#include <torch/types.h>
 #include <torch/utils.h>
 
 #include <test/cpp/api/optim_baseline.h>
diff --git a/test/cpp/api/optim_baseline.h b/test/cpp/api/optim_baseline.h
index c1e3482a1609..4e192d2699f0 100644
--- a/test/cpp/api/optim_baseline.h
+++ b/test/cpp/api/optim_baseline.h
@@ -1,6 +1,6 @@
 // @generated from test/cpp/api/optim_baseline.py
 
-#include <torch/tensor.h>
+#include <torch/types.h>
 
 #include <vector>
 
diff --git a/test/cpp/api/optim_baseline.py b/test/cpp/api/optim_baseline.py
index bc2e06fd6292..a5820becd957 100644
--- a/test/cpp/api/optim_baseline.py
+++ b/test/cpp/api/optim_baseline.py
@@ -9,7 +9,7 @@ import torch.optim
 
 
 HEADER = """
-#include <torch/tensor.h>
+#include <torch/types.h>
 
 #include <vector>
 
diff --git a/test/cpp/api/parallel.cpp b/test/cpp/api/parallel.cpp
index 2b5a772a51b6..d4be5785ed76 100644
--- a/test/cpp/api/parallel.cpp
+++ b/test/cpp/api/parallel.cpp
@@ -5,7 +5,7 @@
 #include <torch/nn/modules/linear.h>
 #include <torch/nn/parallel/data_parallel.h>
 #include <torch/nn/pimpl.h>
-#include <torch/tensor.h>
+#include <torch/types.h>
 
 #include <test/cpp/api/support.h>
 
diff --git a/test/cpp/api/rnn.cpp b/test/cpp/api/rnn.cpp
index e0d511fb0993..c6932bf848de 100644
--- a/test/cpp/api/rnn.cpp
+++ b/test/cpp/api/rnn.cpp
@@ -3,7 +3,7 @@
 #include <torch/nn/modules/linear.h>
 #include <torch/nn/modules/rnn.h>
 #include <torch/optim/adam.h>
-#include <torch/tensor.h>
+#include <torch/types.h>
 #include <torch/utils.h>
 
 #include <test/cpp/api/support.h>
diff --git a/test/cpp/api/sequential.cpp b/test/cpp/api/sequential.cpp
index 19aab60d3194..3a042907765a 100644
--- a/test/cpp/api/sequential.cpp
+++ b/test/cpp/api/sequential.cpp
@@ -7,7 +7,7 @@
 #include <torch/nn/modules/linear.h>
 #include <torch/nn/modules/rnn.h>
 #include <torch/nn/modules/sequential.h>
-#include <torch/tensor.h>
+#include <torch/types.h>
 #include <torch/utils.h>
 
 #include <algorithm>
diff --git a/test/cpp/api/serialize.cpp b/test/cpp/api/serialize.cpp
index 858a6d40a17e..982ed40b10fb 100644
--- a/test/cpp/api/serialize.cpp
+++ b/test/cpp/api/serialize.cpp
@@ -6,7 +6,7 @@
 #include <torch/optim/optimizer.h>
 #include <torch/optim/sgd.h>
 #include <torch/serialize.h>
-#include <torch/tensor.h>
+#include <torch/types.h>
 #include <torch/utils.h>
 
 #include <test/cpp/api/support.h>
diff --git a/test/cpp/api/support.h b/test/cpp/api/support.h
index 3d0ccfa21a1d..0e8c7af4f92b 100644
--- a/test/cpp/api/support.h
+++ b/test/cpp/api/support.h
@@ -5,7 +5,7 @@
 #include <gtest/gtest.h>
 
 #include <torch/nn/cloneable.h>
-#include <torch/tensor.h>
+#include <torch/types.h>
 #include <torch/utils.h>
 
 #include <string>
diff --git a/test/cpp/api/tensor.cpp b/test/cpp/api/tensor.cpp
index 1c1a7804312d..5f2ed0257209 100644
--- a/test/cpp/api/tensor.cpp
+++ b/test/cpp/api/tensor.cpp
@@ -1,6 +1,6 @@
 #include <gtest/gtest.h>
 
-#include <torch/tensor.h>
+#include <torch/types.h>
 
 #include <ATen/ATen.h>
 
diff --git a/test/cpp/api/tensor_options.cpp b/test/cpp/api/tensor_options.cpp
index 319d2ccccf44..1d0780c86a33 100644
--- a/test/cpp/api/tensor_options.cpp
+++ b/test/cpp/api/tensor_options.cpp
@@ -1,6 +1,6 @@
 #include <gtest/gtest.h>
 
-#include <torch/tensor.h>
+#include <torch/types.h>
 
 #include <ATen/Context.h>
 #include <ATen/Functions.h>
diff --git a/test/cpp_extensions/cuda_extension.cpp b/test/cpp_extensions/cuda_extension.cpp
index 9946b4f9cb97..ad7396fe7f45 100644
--- a/test/cpp_extensions/cuda_extension.cpp
+++ b/test/cpp_extensions/cuda_extension.cpp
@@ -5,10 +5,10 @@
 // into one shared library.
 void sigmoid_add_cuda(const float* x, const float* y, float* output, int size);
 
-at::Tensor sigmoid_add(at::Tensor x, at::Tensor y) {
+torch::Tensor sigmoid_add(torch::Tensor x, torch::Tensor y) {
   AT_CHECK(x.type().is_cuda(), "x must be a CUDA tensor");
   AT_CHECK(y.type().is_cuda(), "y must be a CUDA tensor");
-  auto output = at::zeros_like(x);
+  auto output = torch::zeros_like(x);
   sigmoid_add_cuda(
       x.data<float>(), y.data<float>(), output.data<float>(), output.numel());
   return output;
diff --git a/test/cpp_extensions/cudnn_extension.cpp b/test/cpp_extensions/cudnn_extension.cpp
index 498e01a116a1..a626c82b5e6e 100644
--- a/test/cpp_extensions/cudnn_extension.cpp
+++ b/test/cpp_extensions/cudnn_extension.cpp
@@ -1,53 +1,53 @@
 /*
  * CuDNN ReLU extension. Simple function but contains the general structure of
  * most CuDNN extensions:
- * 1) Check arguments. at::check* functions provide a standard way to validate
- * input and provide pretty errors.
- * 2) Create descriptors. Most CuDNN functions require creating and setting a
- * variety of descriptors.
- * 3) Apply the CuDNN function.
- * 4) Destroy your descriptors.
- * 5) Return something (optional).
+ * 1) Check arguments. torch::check* functions provide a standard way to
+ * validate input and provide pretty errors. 2) Create descriptors. Most CuDNN
+ * functions require creating and setting a variety of descriptors. 3) Apply the
+ * CuDNN function. 4) Destroy your descriptors. 5) Return something (optional).
  */
 
 #include <torch/extension.h>
 
-#include <ATen/cudnn/Descriptors.h> // for TensorDescriptor
 #include <ATen/cuda/Exceptions.h> // for CUDNN_CHECK
+#include <ATen/cudnn/Descriptors.h> // for TensorDescriptor
 #include <ATen/cudnn/Handle.h> // for getCudnnHandle
 
 // Name of function in python module and name used for error messages by
-// at::check* functions.
+// torch::check* functions.
 const char* cudnn_relu_name = "cudnn_relu";
 
 // Check arguments to cudnn_relu
-void cudnn_relu_check(const at::Tensor& inputs, const at::Tensor& outputs) {
+void cudnn_relu_check(
+    const torch::Tensor& inputs,
+    const torch::Tensor& outputs) {
   // Create TensorArgs. These record the names and positions of each tensor as a
   // parameter.
-  at::TensorArg arg_inputs(inputs, "inputs", 0);
-  at::TensorArg arg_outputs(outputs, "outputs", 1);
+  torch::TensorArg arg_inputs(inputs, "inputs", 0);
+  torch::TensorArg arg_outputs(outputs, "outputs", 1);
   // Check arguments. No need to return anything. These functions with throw an
   // error if they fail. Messages are populated using information from
   // TensorArgs.
-  at::checkContiguous(cudnn_relu_name, arg_inputs);
-  at::checkScalarType(cudnn_relu_name, arg_inputs, at::kFloat);
-  at::checkBackend(cudnn_relu_name, arg_inputs.tensor, at::Backend::CUDA);
-  at::checkContiguous(cudnn_relu_name, arg_outputs);
-  at::checkScalarType(cudnn_relu_name, arg_outputs, at::kFloat);
-  at::checkBackend(cudnn_relu_name, arg_outputs.tensor, at::Backend::CUDA);
-  at::checkSameSize(cudnn_relu_name, arg_inputs, arg_outputs);
+  torch::checkContiguous(cudnn_relu_name, arg_inputs);
+  torch::checkScalarType(cudnn_relu_name, arg_inputs, torch::kFloat);
+  torch::checkBackend(cudnn_relu_name, arg_inputs.tensor, torch::Backend::CUDA);
+  torch::checkContiguous(cudnn_relu_name, arg_outputs);
+  torch::checkScalarType(cudnn_relu_name, arg_outputs, torch::kFloat);
+  torch::checkBackend(
+      cudnn_relu_name, arg_outputs.tensor, torch::Backend::CUDA);
+  torch::checkSameSize(cudnn_relu_name, arg_inputs, arg_outputs);
 }
 
-void cudnn_relu(const at::Tensor& inputs, const at::Tensor& outputs) {
+void cudnn_relu(const torch::Tensor& inputs, const torch::Tensor& outputs) {
   // Most CuDNN extensions will follow a similar pattern.
   // Step 1: Check inputs. This will throw an error if inputs are invalid, so no
   // need to check return codes here.
   cudnn_relu_check(inputs, outputs);
   // Step 2: Create descriptors
-  cudnnHandle_t cuDnn = at::native::getCudnnHandle();
+  cudnnHandle_t cuDnn = torch::native::getCudnnHandle();
   // Note: 4 is minimum dim for a TensorDescriptor. Input and output are same
   // size and type and contiguous, so one descriptor is sufficient.
-  at::native::TensorDescriptor input_tensor_desc(inputs, 4);
+  torch::native::TensorDescriptor input_tensor_desc(inputs, 4);
   cudnnActivationDescriptor_t activationDesc;
   // Note: Always check return value of cudnn functions using CUDNN_CHECK
   AT_CUDNN_CHECK(cudnnCreateActivationDescriptor(&activationDesc));
diff --git a/test/cpp_extensions/doubler.h b/test/cpp_extensions/doubler.h
index afd4a007019d..4edd4de8fd6e 100644
--- a/test/cpp_extensions/doubler.h
+++ b/test/cpp_extensions/doubler.h
@@ -3,15 +3,15 @@
 struct Doubler {
   Doubler(int A, int B) {
     tensor_ =
-        torch::ones({A, B}, torch::dtype(torch::kDouble).requires_grad(true));
+        torch::ones({A, B}, torch::dtype(torch::kFloat64).requires_grad(true));
   }
-  at::Tensor forward() {
+  torch::Tensor forward() {
     return tensor_ * 2;
   }
-  at::Tensor get() const {
+  torch::Tensor get() const {
     return tensor_;
   }
 
  private:
-  at::Tensor tensor_;
+  torch::Tensor tensor_;
 };
diff --git a/test/cpp_extensions/extension.cpp b/test/cpp_extensions/extension.cpp
index e217b210860f..8aff0dd018a4 100644
--- a/test/cpp_extensions/extension.cpp
+++ b/test/cpp_extensions/extension.cpp
@@ -1,26 +1,26 @@
 #include <torch/extension.h>
 
-at::Tensor sigmoid_add(at::Tensor x, at::Tensor y) {
+torch::Tensor sigmoid_add(torch::Tensor x, torch::Tensor y) {
   return x.sigmoid() + y.sigmoid();
 }
 
 struct MatrixMultiplier {
   MatrixMultiplier(int A, int B) {
     tensor_ =
-        torch::ones({A, B}, torch::dtype(torch::kDouble).requires_grad(true));
+        torch::ones({A, B}, torch::dtype(torch::kFloat64).requires_grad(true));
   }
-  at::Tensor forward(at::Tensor weights) {
+  torch::Tensor forward(torch::Tensor weights) {
     return tensor_.mm(weights);
   }
-  at::Tensor get() const {
+  torch::Tensor get() const {
     return tensor_;
   }
 
  private:
-  at::Tensor tensor_;
+  torch::Tensor tensor_;
 };
 
-bool function_taking_optional(c10::optional<at::Tensor> tensor) {
+bool function_taking_optional(c10::optional<torch::Tensor> tensor) {
   return tensor.has_value();
 }
 
diff --git a/test/custom_operator/op.cpp b/test/custom_operator/op.cpp
index f63742091f18..c26bdb02325e 100644
--- a/test/custom_operator/op.cpp
+++ b/test/custom_operator/op.cpp
@@ -5,11 +5,11 @@
 #include <cstddef>
 #include <vector>
 
-std::vector<at::Tensor> custom_op(
-    at::Tensor tensor,
+std::vector<torch::Tensor> custom_op(
+    torch::Tensor tensor,
     double scalar,
     int64_t repeat) {
-  std::vector<at::Tensor> output;
+  std::vector<torch::Tensor> output;
   output.reserve(repeat);
   for (int64_t i = 0; i < repeat; ++i) {
     output.push_back(tensor * scalar);
diff --git a/test/custom_operator/op.h b/test/custom_operator/op.h
index a05798ce52a2..b18a48edc10b 100644
--- a/test/custom_operator/op.h
+++ b/test/custom_operator/op.h
@@ -15,7 +15,7 @@
 #  endif
 // clang-format on
 
-CUSTOM_OP_API std::vector<at::Tensor> custom_op(
-    at::Tensor tensor,
+CUSTOM_OP_API std::vector<torch::Tensor> custom_op(
+    torch::Tensor tensor,
     double scalar,
     int64_t repeat);
diff --git a/test/custom_operator/test_custom_ops.cpp b/test/custom_operator/test_custom_ops.cpp
index 645afec5c378..8f35e3d5bd14 100644
--- a/test/custom_operator/test_custom_ops.cpp
+++ b/test/custom_operator/test_custom_ops.cpp
@@ -33,7 +33,7 @@ void get_operator_from_registry_and_execute() {
   torch::jit::Stack stack;
   torch::jit::push(stack, torch::ones(5), 2.0, 3);
   op->getOperation()(stack);
-  std::vector<at::Tensor> output;
+  std::vector<torch::Tensor> output;
   torch::jit::pop(stack, output);
 
   const auto manual = custom_op(torch::ones(5), 2.0, 3);
@@ -99,19 +99,19 @@ void test_move_to_device(const std::string& path_to_exported_script_module) {
       torch::jit::load(path_to_exported_script_module);
   AT_ASSERT(module != nullptr);
 
-  helpers::check_all_parameters(*module, [](const at::Tensor& tensor) {
+  helpers::check_all_parameters(*module, [](const torch::Tensor& tensor) {
     return tensor.device().is_cpu();
   });
 
-  module->to(at::kCUDA);
+  module->to(torch::kCUDA);
 
-  helpers::check_all_parameters(*module, [](const at::Tensor& tensor) {
+  helpers::check_all_parameters(*module, [](const torch::Tensor& tensor) {
     return tensor.device().is_cuda();
   });
 
-  module->to(at::kCPU);
+  module->to(torch::kCPU);
 
-  helpers::check_all_parameters(*module, [](const at::Tensor& tensor) {
+  helpers::check_all_parameters(*module, [](const torch::Tensor& tensor) {
     return tensor.device().is_cpu();
   });
 }
@@ -121,16 +121,16 @@ void test_move_to_dtype(const std::string& path_to_exported_script_module) {
       torch::jit::load(path_to_exported_script_module);
   AT_ASSERT(module != nullptr);
 
-  module->to(at::kInt);
+  module->to(torch::kInt);
 
-  helpers::check_all_parameters(*module, [](const at::Tensor& tensor) {
-    return tensor.dtype() == at::kInt;
+  helpers::check_all_parameters(*module, [](const torch::Tensor& tensor) {
+    return tensor.dtype() == torch::kInt;
   });
 
-  module->to(at::kDouble);
+  module->to(torch::kDouble);
 
-  helpers::check_all_parameters(*module, [](const at::Tensor& tensor) {
-    return tensor.dtype() == at::kDouble;
+  helpers::check_all_parameters(*module, [](const torch::Tensor& tensor) {
+    return tensor.dtype() == torch::kDouble;
   });
 }
 
@@ -147,7 +147,7 @@ int main(int argc, const char* argv[]) {
   test_argument_checking_for_serialized_modules(path_to_exported_script_module);
   test_move_to_dtype(path_to_exported_script_module);
 
-  if (at::globalContext().getNumGPUs() > 0) {
+  if (torch::globalContext().getNumGPUs() > 0) {
     test_move_to_device(path_to_exported_script_module);
   }
 
diff --git a/test/test_cpp_extensions.py b/test/test_cpp_extensions.py
index 185356b04187..0d44cdc35649 100755
--- a/test/test_cpp_extensions.py
+++ b/test/test_cpp_extensions.py
@@ -149,7 +149,7 @@ class TestCppExtension(common.TestCase):
 
     def test_inline_jit_compile_extension_with_functions_as_list(self):
         cpp_source = '''
-        at::Tensor tanh_add(at::Tensor x, at::Tensor y) {
+        torch::Tensor tanh_add(torch::Tensor x, torch::Tensor y) {
           return x.tanh() + y.tanh();
         }
         '''
@@ -170,7 +170,7 @@ class TestCppExtension(common.TestCase):
 
     def test_inline_jit_compile_extension_with_functions_as_dict(self):
         cpp_source = '''
-        at::Tensor tanh_add(at::Tensor x, at::Tensor y) {
+        torch::Tensor tanh_add(torch::Tensor x, torch::Tensor y) {
           return x.tanh() + y.tanh();
         }
         '''
@@ -186,14 +186,14 @@ class TestCppExtension(common.TestCase):
 
     def test_inline_jit_compile_extension_multiple_sources_and_no_functions(self):
         cpp_source1 = '''
-        at::Tensor sin_add(at::Tensor x, at::Tensor y) {
+        torch::Tensor sin_add(torch::Tensor x, torch::Tensor y) {
           return x.sin() + y.sin();
         }
         '''
 
         cpp_source2 = '''
         #include <torch/extension.h>
-        at::Tensor sin_add(at::Tensor x, at::Tensor y);
+        torch::Tensor sin_add(torch::Tensor x, torch::Tensor y);
         PYBIND11_MODULE(TORCH_EXTENSION_NAME, m) {
           m.def("sin_add", &sin_add, "sin(x) + sin(y)");
         }
@@ -224,8 +224,8 @@ class TestCppExtension(common.TestCase):
           }
         }
 
-        at::Tensor cos_add(at::Tensor x, at::Tensor y) {
-          auto output = at::zeros_like(x);
+        torch::Tensor cos_add(torch::Tensor x, torch::Tensor y) {
+          auto output = torch::zeros_like(x);
           const int threads = 1024;
           const int blocks = (output.numel() + threads - 1) / threads;
           cos_add_kernel<<<blocks, threads>>>(x.data<float>(), y.data<float>(), output.data<float>(), output.numel());
@@ -234,7 +234,7 @@ class TestCppExtension(common.TestCase):
         '''
 
         # Here, the C++ source need only declare the function signature.
-        cpp_source = 'at::Tensor cos_add(at::Tensor x, at::Tensor y);'
+        cpp_source = 'torch::Tensor cos_add(torch::Tensor x, torch::Tensor y);'
 
         module = torch.utils.cpp_extension.load_inline(
             name='inline_jit_extension_cuda',
@@ -258,7 +258,7 @@ class TestCppExtension(common.TestCase):
 
     def test_lenient_flag_handling_in_jit_extensions(self):
         cpp_source = '''
-        at::Tensor tanh_add(at::Tensor x, at::Tensor y) {
+        torch::Tensor tanh_add(torch::Tensor x, torch::Tensor y) {
           return x.tanh() + y.tanh();
         }
         '''
@@ -303,8 +303,8 @@ class TestCppExtension(common.TestCase):
             }
         }
 
-        at::Tensor half_test(at::Tensor input) {
-            auto output = at::empty(1, input.options().dtype(at::kFloat));
+        torch::Tensor half_test(torch::Tensor input) {
+            auto output = torch::empty(1, input.options().dtype(torch::kFloat));
             AT_DISPATCH_FLOATING_TYPES_AND_HALF(input.type(), "half_test", [&] {
                 half_test_kernel<scalar_t><<<1, 1>>>(
                     input.data<scalar_t>(),
@@ -316,7 +316,7 @@ class TestCppExtension(common.TestCase):
 
         module = torch.utils.cpp_extension.load_inline(
             name='half_test_extension',
-            cpp_sources='at::Tensor half_test(at::Tensor input);',
+            cpp_sources='torch::Tensor half_test(torch::Tensor input);',
             cuda_sources=cuda_source,
             functions=['half_test'],
             verbose=True)
diff --git a/torch/csrc/api/include/torch/all.h b/torch/csrc/api/include/torch/all.h
index dcd2367873de..bdce7585682f 100644
--- a/torch/csrc/api/include/torch/all.h
+++ b/torch/csrc/api/include/torch/all.h
@@ -6,5 +6,5 @@
 #include <torch/nn.h>
 #include <torch/optim.h>
 #include <torch/serialize.h>
-#include <torch/tensor.h>
+#include <torch/types.h>
 #include <torch/utils.h>
diff --git a/torch/csrc/api/include/torch/data/dataloader.h b/torch/csrc/api/include/torch/data/dataloader.h
index 62bcf382ca1a..95a588ec7027 100644
--- a/torch/csrc/api/include/torch/data/dataloader.h
+++ b/torch/csrc/api/include/torch/data/dataloader.h
@@ -6,7 +6,7 @@
 #include <torch/data/iterator.h>
 #include <torch/data/samplers/random.h>
 #include <torch/data/worker_exception.h>
-#include <torch/tensor.h>
+#include <torch/types.h>
 
 #include <torch/csrc/utils/memory.h>
 #include <torch/csrc/utils/variadic.h>
diff --git a/torch/csrc/api/include/torch/data/dataloader_options.h b/torch/csrc/api/include/torch/data/dataloader_options.h
index 305ddd040e8b..e16e2a59bbef 100644
--- a/torch/csrc/api/include/torch/data/dataloader_options.h
+++ b/torch/csrc/api/include/torch/data/dataloader_options.h
@@ -1,7 +1,7 @@
 #pragma once
 
 #include <torch/arg.h>
-#include <torch/tensor.h>
+#include <torch/types.h>
 
 #include <chrono>
 #include <cstddef>
diff --git a/torch/csrc/api/include/torch/data/datasets/base.h b/torch/csrc/api/include/torch/data/datasets/base.h
index b6a486dd705b..e31618ffd49c 100644
--- a/torch/csrc/api/include/torch/data/datasets/base.h
+++ b/torch/csrc/api/include/torch/data/datasets/base.h
@@ -1,7 +1,7 @@
 #pragma once
 
 #include <torch/data/example.h>
-#include <torch/tensor.h>
+#include <torch/types.h>
 
 #include <ATen/core/ArrayRef.h>
 
diff --git a/torch/csrc/api/include/torch/data/datasets/map.h b/torch/csrc/api/include/torch/data/datasets/map.h
index 07d0f4400a39..0429ced62a03 100644
--- a/torch/csrc/api/include/torch/data/datasets/map.h
+++ b/torch/csrc/api/include/torch/data/datasets/map.h
@@ -1,7 +1,7 @@
 #pragma once
 
 #include <torch/data/datasets/base.h>
-#include <torch/tensor.h>
+#include <torch/types.h>
 
 #include <ATen/core/ArrayRef.h>
 
diff --git a/torch/csrc/api/include/torch/data/datasets/mnist.h b/torch/csrc/api/include/torch/data/datasets/mnist.h
index ddca9f463d88..de8c410415d1 100644
--- a/torch/csrc/api/include/torch/data/datasets/mnist.h
+++ b/torch/csrc/api/include/torch/data/datasets/mnist.h
@@ -2,7 +2,7 @@
 
 #include <torch/data/datasets/base.h>
 #include <torch/data/example.h>
-#include <torch/tensor.h>
+#include <torch/types.h>
 
 #include <cstddef>
 #include <string>
diff --git a/torch/csrc/api/include/torch/data/datasets/tensor.h b/torch/csrc/api/include/torch/data/datasets/tensor.h
index 8b83ef481e17..3059dfcf8108 100644
--- a/torch/csrc/api/include/torch/data/datasets/tensor.h
+++ b/torch/csrc/api/include/torch/data/datasets/tensor.h
@@ -2,7 +2,7 @@
 
 #include <torch/data/datasets/base.h>
 #include <torch/data/example.h>
-#include <torch/tensor.h>
+#include <torch/types.h>
 
 #include <cstddef>
 #include <vector>
diff --git a/torch/csrc/api/include/torch/data/detail/data_shuttle.h b/torch/csrc/api/include/torch/data/detail/data_shuttle.h
index 37a4b3657ff6..4f087a01f433 100644
--- a/torch/csrc/api/include/torch/data/detail/data_shuttle.h
+++ b/torch/csrc/api/include/torch/data/detail/data_shuttle.h
@@ -1,7 +1,7 @@
 #pragma once
 
 #include <torch/data/detail/queue.h>
-#include <torch/tensor.h>
+#include <torch/types.h>
 
 #include <c10/util/Exception.h>
 #include <c10/util/Optional.h>
diff --git a/torch/csrc/api/include/torch/data/detail/queue.h b/torch/csrc/api/include/torch/data/detail/queue.h
index 9918b7b4fb54..332914bb1405 100644
--- a/torch/csrc/api/include/torch/data/detail/queue.h
+++ b/torch/csrc/api/include/torch/data/detail/queue.h
@@ -1,6 +1,6 @@
 #pragma once
 
-#include <torch/tensor.h>
+#include <torch/types.h>
 
 #include <c10/util/Exception.h>
 
diff --git a/torch/csrc/api/include/torch/data/detail/sequencers.h b/torch/csrc/api/include/torch/data/detail/sequencers.h
index 6da98a972d0c..f8c26f7acb1a 100644
--- a/torch/csrc/api/include/torch/data/detail/sequencers.h
+++ b/torch/csrc/api/include/torch/data/detail/sequencers.h
@@ -1,6 +1,6 @@
 #pragma once
 
-#include <torch/tensor.h>
+#include <torch/types.h>
 
 #include <algorithm>
 #include <cstddef>
diff --git a/torch/csrc/api/include/torch/data/example.h b/torch/csrc/api/include/torch/data/example.h
index dfd1749baf07..b43ef2ca1955 100644
--- a/torch/csrc/api/include/torch/data/example.h
+++ b/torch/csrc/api/include/torch/data/example.h
@@ -1,6 +1,6 @@
 #pragma once
 
-#include <torch/tensor.h>
+#include <torch/types.h>
 
 namespace torch {
 namespace data {
diff --git a/torch/csrc/api/include/torch/data/iterator.h b/torch/csrc/api/include/torch/data/iterator.h
index 7196c89df40d..bf5b7429bf48 100644
--- a/torch/csrc/api/include/torch/data/iterator.h
+++ b/torch/csrc/api/include/torch/data/iterator.h
@@ -1,7 +1,7 @@
 #pragma once
 
 #include <torch/csrc/utils/variadic.h>
-#include <torch/tensor.h>
+#include <torch/types.h>
 
 #include <c10/util/Exception.h>
 
diff --git a/torch/csrc/api/include/torch/data/samplers/base.h b/torch/csrc/api/include/torch/data/samplers/base.h
index c74a2acf0d58..2c7d6b4c53de 100644
--- a/torch/csrc/api/include/torch/data/samplers/base.h
+++ b/torch/csrc/api/include/torch/data/samplers/base.h
@@ -1,6 +1,6 @@
 #pragma once
 
-#include <torch/tensor.h>
+#include <torch/types.h>
 
 #include <cstddef>
 #include <vector>
diff --git a/torch/csrc/api/include/torch/data/samplers/random.h b/torch/csrc/api/include/torch/data/samplers/random.h
index 97f47f980fa6..cb0de8890e70 100644
--- a/torch/csrc/api/include/torch/data/samplers/random.h
+++ b/torch/csrc/api/include/torch/data/samplers/random.h
@@ -1,7 +1,7 @@
 #pragma once
 
 #include <torch/data/samplers/base.h>
-#include <torch/tensor.h>
+#include <torch/types.h>
 
 #include <cstddef>
 #include <vector>
diff --git a/torch/csrc/api/include/torch/data/samplers/sequential.h b/torch/csrc/api/include/torch/data/samplers/sequential.h
index e870b39589f8..21dc012d2252 100644
--- a/torch/csrc/api/include/torch/data/samplers/sequential.h
+++ b/torch/csrc/api/include/torch/data/samplers/sequential.h
@@ -1,7 +1,7 @@
 #pragma once
 
 #include <torch/data/samplers/base.h>
-#include <torch/tensor.h>
+#include <torch/types.h>
 
 #include <cstddef>
 #include <vector>
diff --git a/torch/csrc/api/include/torch/data/samplers/stream.h b/torch/csrc/api/include/torch/data/samplers/stream.h
index de3d5fa7ad09..68148a3f9362 100644
--- a/torch/csrc/api/include/torch/data/samplers/stream.h
+++ b/torch/csrc/api/include/torch/data/samplers/stream.h
@@ -2,7 +2,7 @@
 
 #include <torch/data/samplers/base.h>
 #include <torch/data/samplers/custom_batch_request.h>
-#include <torch/tensor.h>
+#include <torch/types.h>
 
 #include <cstddef>
 
diff --git a/torch/csrc/api/include/torch/data/transforms/base.h b/torch/csrc/api/include/torch/data/transforms/base.h
index 19d21b7875fb..0bc1f2ea7b14 100644
--- a/torch/csrc/api/include/torch/data/transforms/base.h
+++ b/torch/csrc/api/include/torch/data/transforms/base.h
@@ -1,6 +1,6 @@
 #pragma once
 
-#include <torch/tensor.h>
+#include <torch/types.h>
 
 #include <utility>
 #include <vector>
diff --git a/torch/csrc/api/include/torch/data/transforms/stack.h b/torch/csrc/api/include/torch/data/transforms/stack.h
index 83beb4ee4594..4be1bd920b71 100644
--- a/torch/csrc/api/include/torch/data/transforms/stack.h
+++ b/torch/csrc/api/include/torch/data/transforms/stack.h
@@ -2,7 +2,7 @@
 
 #include <torch/data/example.h>
 #include <torch/data/transforms/collate.h>
-#include <torch/tensor.h>
+#include <torch/types.h>
 
 #include <utility>
 #include <vector>
diff --git a/torch/csrc/api/include/torch/data/transforms/tensor.h b/torch/csrc/api/include/torch/data/transforms/tensor.h
index 8c053ad9e0e3..c1fed2011636 100644
--- a/torch/csrc/api/include/torch/data/transforms/tensor.h
+++ b/torch/csrc/api/include/torch/data/transforms/tensor.h
@@ -2,7 +2,7 @@
 
 #include <torch/data/example.h>
 #include <torch/data/transforms/base.h>
-#include <torch/tensor.h>
+#include <torch/types.h>
 
 #include <functional>
 #include <utility>
diff --git a/torch/csrc/api/include/torch/detail/static.h b/torch/csrc/api/include/torch/detail/static.h
index b5a3e542aaaa..c85fc7fff4b4 100644
--- a/torch/csrc/api/include/torch/detail/static.h
+++ b/torch/csrc/api/include/torch/detail/static.h
@@ -1,7 +1,7 @@
 #pragma once
 
 #include <torch/csrc/utils/variadic.h>
-#include <torch/tensor.h>
+#include <torch/types.h>
 
 #include <cstdint>
 #include <type_traits>
diff --git a/torch/csrc/api/include/torch/nn/cloneable.h b/torch/csrc/api/include/torch/nn/cloneable.h
index b4840fee8f75..86b3308fe9d7 100644
--- a/torch/csrc/api/include/torch/nn/cloneable.h
+++ b/torch/csrc/api/include/torch/nn/cloneable.h
@@ -1,7 +1,7 @@
 #pragma once
 
 #include <torch/nn/module.h>
-#include <torch/tensor.h>
+#include <torch/types.h>
 #include <torch/utils.h>
 
 #include <ATen/OptionsGuard.h>
diff --git a/torch/csrc/api/include/torch/nn/cursor.h b/torch/csrc/api/include/torch/nn/cursor.h
index 072c4e20a9ab..0e3989e80c45 100644
--- a/torch/csrc/api/include/torch/nn/cursor.h
+++ b/torch/csrc/api/include/torch/nn/cursor.h
@@ -1,6 +1,6 @@
 #pragma once
 
-#include <torch/tensor.h>
+#include <torch/types.h>
 
 #include <cstddef>
 #include <iterator>
diff --git a/torch/csrc/api/include/torch/nn/init.h b/torch/csrc/api/include/torch/nn/init.h
index 0fd68df5bbd0..9bc7b8fe8a25 100644
--- a/torch/csrc/api/include/torch/nn/init.h
+++ b/torch/csrc/api/include/torch/nn/init.h
@@ -1,6 +1,6 @@
 #pragma once
 
-#include <torch/tensor.h>
+#include <torch/types.h>
 
 namespace torch {
 namespace nn {
diff --git a/torch/csrc/api/include/torch/nn/module.h b/torch/csrc/api/include/torch/nn/module.h
index 9fe9eaf19bbf..49b8fe300cdc 100644
--- a/torch/csrc/api/include/torch/nn/module.h
+++ b/torch/csrc/api/include/torch/nn/module.h
@@ -4,7 +4,7 @@
 #include <torch/nn/cursor.h>
 #include <torch/nn/pimpl.h>
 #include <torch/serialize/archive.h>
-#include <torch/tensor.h>
+#include <torch/types.h>
 
 #include <ATen/ATen.h>
 
diff --git a/torch/csrc/api/include/torch/nn/modules/any.h b/torch/csrc/api/include/torch/nn/modules/any.h
index 4983ad711a02..769b662c9ef4 100644
--- a/torch/csrc/api/include/torch/nn/modules/any.h
+++ b/torch/csrc/api/include/torch/nn/modules/any.h
@@ -3,7 +3,7 @@
 #include <torch/detail/static.h>
 #include <torch/nn/module.h>
 #include <torch/nn/pimpl.h>
-#include <torch/tensor.h>
+#include <torch/types.h>
 
 #include <torch/csrc/autograd/variable.h>
 #include <torch/csrc/utils/memory.h>
diff --git a/torch/csrc/api/include/torch/nn/modules/batchnorm.h b/torch/csrc/api/include/torch/nn/modules/batchnorm.h
index ff5cebc7ed6b..c0ee0a316ef1 100644
--- a/torch/csrc/api/include/torch/nn/modules/batchnorm.h
+++ b/torch/csrc/api/include/torch/nn/modules/batchnorm.h
@@ -2,7 +2,7 @@
 
 #include <torch/nn/cloneable.h>
 #include <torch/nn/pimpl.h>
-#include <torch/tensor.h>
+#include <torch/types.h>
 
 #include <cstdint>
 
diff --git a/torch/csrc/api/include/torch/nn/modules/conv.h b/torch/csrc/api/include/torch/nn/modules/conv.h
index c6e6a2392dbe..d558ba454e64 100644
--- a/torch/csrc/api/include/torch/nn/modules/conv.h
+++ b/torch/csrc/api/include/torch/nn/modules/conv.h
@@ -3,7 +3,7 @@
 #include <torch/expanding_array.h>
 #include <torch/nn/cloneable.h>
 #include <torch/nn/pimpl.h>
-#include <torch/tensor.h>
+#include <torch/types.h>
 
 #include <cstddef>
 #include <vector>
diff --git a/torch/csrc/api/include/torch/nn/modules/dropout.h b/torch/csrc/api/include/torch/nn/modules/dropout.h
index 48b89642864e..06468493a121 100644
--- a/torch/csrc/api/include/torch/nn/modules/dropout.h
+++ b/torch/csrc/api/include/torch/nn/modules/dropout.h
@@ -2,7 +2,7 @@
 
 #include <torch/nn/cloneable.h>
 #include <torch/nn/pimpl.h>
-#include <torch/tensor.h>
+#include <torch/types.h>
 
 #include <cstddef>
 #include <vector>
diff --git a/torch/csrc/api/include/torch/nn/modules/embedding.h b/torch/csrc/api/include/torch/nn/modules/embedding.h
index c2a701eb3ede..c1293730842d 100644
--- a/torch/csrc/api/include/torch/nn/modules/embedding.h
+++ b/torch/csrc/api/include/torch/nn/modules/embedding.h
@@ -2,7 +2,7 @@
 
 #include <torch/nn/cloneable.h>
 #include <torch/nn/pimpl.h>
-#include <torch/tensor.h>
+#include <torch/types.h>
 
 #include <cstddef>
 #include <vector>
diff --git a/torch/csrc/api/include/torch/nn/modules/functional.h b/torch/csrc/api/include/torch/nn/modules/functional.h
index 98f72963686f..962bee17db7a 100644
--- a/torch/csrc/api/include/torch/nn/modules/functional.h
+++ b/torch/csrc/api/include/torch/nn/modules/functional.h
@@ -3,7 +3,7 @@
 #include <torch/csrc/utils/variadic.h>
 #include <torch/nn/cloneable.h>
 #include <torch/nn/pimpl.h>
-#include <torch/tensor.h>
+#include <torch/types.h>
 
 #include <functional>
 #include <utility>
diff --git a/torch/csrc/api/include/torch/nn/modules/linear.h b/torch/csrc/api/include/torch/nn/modules/linear.h
index 3e8044f57158..99170f025b7a 100644
--- a/torch/csrc/api/include/torch/nn/modules/linear.h
+++ b/torch/csrc/api/include/torch/nn/modules/linear.h
@@ -3,7 +3,7 @@
 #include <torch/nn/cloneable.h>
 #include <torch/nn/module.h>
 #include <torch/nn/pimpl.h>
-#include <torch/tensor.h>
+#include <torch/types.h>
 
 #include <cstddef>
 #include <vector>
diff --git a/torch/csrc/api/include/torch/nn/modules/rnn.h b/torch/csrc/api/include/torch/nn/modules/rnn.h
index ba87afbd0111..96496976068e 100644
--- a/torch/csrc/api/include/torch/nn/modules/rnn.h
+++ b/torch/csrc/api/include/torch/nn/modules/rnn.h
@@ -3,7 +3,7 @@
 #include <torch/nn/cloneable.h>
 #include <torch/nn/modules/dropout.h>
 #include <torch/nn/pimpl.h>
-#include <torch/tensor.h>
+#include <torch/types.h>
 
 #include <ATen/ATen.h>
 #include <c10/util/Exception.h>
diff --git a/torch/csrc/api/include/torch/nn/modules/sequential.h b/torch/csrc/api/include/torch/nn/modules/sequential.h
index a5e3690f7c31..779e32ad2514 100644
--- a/torch/csrc/api/include/torch/nn/modules/sequential.h
+++ b/torch/csrc/api/include/torch/nn/modules/sequential.h
@@ -5,7 +5,7 @@
 #include <torch/nn/module.h>
 #include <torch/nn/modules/any.h>
 #include <torch/nn/pimpl.h>
-#include <torch/tensor.h>
+#include <torch/types.h>
 
 #include <c10/util/Exception.h>
 
diff --git a/torch/csrc/api/include/torch/nn/parallel/data_parallel.h b/torch/csrc/api/include/torch/nn/parallel/data_parallel.h
index 01b7dfdee3fb..adbab17b2288 100644
--- a/torch/csrc/api/include/torch/nn/parallel/data_parallel.h
+++ b/torch/csrc/api/include/torch/nn/parallel/data_parallel.h
@@ -3,7 +3,7 @@
 #include <torch/cuda.h>
 #include <torch/nn/module.h>
 #include <torch/nn/pimpl.h>
-#include <torch/tensor.h>
+#include <torch/types.h>
 
 #include <torch/csrc/autograd/functions/comm.h>
 #include <torch/csrc/cuda/comm.h>
diff --git a/torch/csrc/api/include/torch/nn/pimpl.h b/torch/csrc/api/include/torch/nn/pimpl.h
index 41a172f895f6..778ee5e68004 100644
--- a/torch/csrc/api/include/torch/nn/pimpl.h
+++ b/torch/csrc/api/include/torch/nn/pimpl.h
@@ -2,7 +2,7 @@
 
 #include <torch/arg.h>
 #include <torch/serialize/archive.h>
-#include <torch/tensor.h>
+#include <torch/types.h>
 
 #include <torch/csrc/utils/variadic.h>
 
diff --git a/torch/csrc/api/include/torch/optim/adagrad.h b/torch/csrc/api/include/torch/optim/adagrad.h
index ebfaf9daf41c..87bc29b137a5 100644
--- a/torch/csrc/api/include/torch/optim/adagrad.h
+++ b/torch/csrc/api/include/torch/optim/adagrad.h
@@ -3,7 +3,7 @@
 #include <torch/nn/pimpl.h>
 #include <torch/optim/optimizer.h>
 #include <torch/optim/serialize.h>
-#include <torch/tensor.h>
+#include <torch/types.h>
 
 #include <utility>
 #include <vector>
diff --git a/torch/csrc/api/include/torch/optim/rmsprop.h b/torch/csrc/api/include/torch/optim/rmsprop.h
index 6643d045c258..ca2f45bcb8b7 100644
--- a/torch/csrc/api/include/torch/optim/rmsprop.h
+++ b/torch/csrc/api/include/torch/optim/rmsprop.h
@@ -4,7 +4,7 @@
 #include <torch/nn/module.h>
 #include <torch/optim/optimizer.h>
 #include <torch/optim/serialize.h>
-#include <torch/tensor.h>
+#include <torch/types.h>
 
 #include <functional>
 #include <memory>
diff --git a/torch/csrc/api/include/torch/optim/serialize.h b/torch/csrc/api/include/torch/optim/serialize.h
index 1c85fa74e006..0aec75c610dd 100644
--- a/torch/csrc/api/include/torch/optim/serialize.h
+++ b/torch/csrc/api/include/torch/optim/serialize.h
@@ -1,7 +1,7 @@
 #pragma once
 
 #include <torch/serialize/archive.h>
-#include <torch/tensor.h>
+#include <torch/types.h>
 
 #include <cstddef>
 #include <cstdint>
diff --git a/torch/csrc/api/include/torch/optim/sgd.h b/torch/csrc/api/include/torch/optim/sgd.h
index e5b8a4fc9b90..7988586b02cd 100644
--- a/torch/csrc/api/include/torch/optim/sgd.h
+++ b/torch/csrc/api/include/torch/optim/sgd.h
@@ -3,7 +3,7 @@
 #include <torch/arg.h>
 #include <torch/nn/module.h>
 #include <torch/optim/optimizer.h>
-#include <torch/tensor.h>
+#include <torch/types.h>
 
 #include <cstddef>
 #include <utility>
diff --git a/torch/csrc/api/include/torch/python.h b/torch/csrc/api/include/torch/python.h
index e35db625b40b..892f1ad7451d 100644
--- a/torch/csrc/api/include/torch/python.h
+++ b/torch/csrc/api/include/torch/python.h
@@ -1,11 +1,11 @@
 #pragma once
 
 #include <torch/detail/static.h>
-#include <torch/tensor.h>
+#include <torch/types.h>
 
 #include <torch/csrc/python_headers.h>
 #include <torch/csrc/utils/pybind.h>
-#include <torch/tensor.h>
+#include <torch/types.h>
 
 #include <iterator>
 #include <string>
diff --git a/torch/csrc/api/include/torch/tensor.h b/torch/csrc/api/include/torch/types.h
similarity index 95%
rename from torch/csrc/api/include/torch/tensor.h
rename to torch/csrc/api/include/torch/types.h
index 89e6ddde8243..185989619b6c 100644
--- a/torch/csrc/api/include/torch/tensor.h
+++ b/torch/csrc/api/include/torch/types.h
@@ -13,9 +13,6 @@ using namespace at; // NOLINT
 using c10::optional;
 using c10::nullopt;
 
-using c10::optional;
-using c10::nullopt;
-
 using Dtype = at::ScalarType;
 
 /// Fixed width dtypes.
diff --git a/torch/csrc/api/src/data/datasets/mnist.cpp b/torch/csrc/api/src/data/datasets/mnist.cpp
index d2c1c5b27999..0ef3fbcb2d1c 100644
--- a/torch/csrc/api/src/data/datasets/mnist.cpp
+++ b/torch/csrc/api/src/data/datasets/mnist.cpp
@@ -1,7 +1,7 @@
 #include <torch/data/datasets/mnist.h>
 
 #include <torch/data/example.h>
-#include <torch/tensor.h>
+#include <torch/types.h>
 
 #include <c10/util/Exception.h>
 
diff --git a/torch/csrc/api/src/data/samplers/random.cpp b/torch/csrc/api/src/data/samplers/random.cpp
index 5cc16d987cbf..4ea975b035be 100644
--- a/torch/csrc/api/src/data/samplers/random.cpp
+++ b/torch/csrc/api/src/data/samplers/random.cpp
@@ -1,6 +1,6 @@
 #include <torch/data/samplers/random.h>
 #include <torch/serialize/archive.h>
-#include <torch/tensor.h>
+#include <torch/types.h>
 
 #include <algorithm>
 #include <cstddef>
diff --git a/torch/csrc/api/src/data/samplers/sequential.cpp b/torch/csrc/api/src/data/samplers/sequential.cpp
index 83260bf98df4..3072346115c0 100644
--- a/torch/csrc/api/src/data/samplers/sequential.cpp
+++ b/torch/csrc/api/src/data/samplers/sequential.cpp
@@ -1,6 +1,6 @@
 #include <torch/data/samplers/sequential.h>
 #include <torch/serialize/archive.h>
-#include <torch/tensor.h>
+#include <torch/types.h>
 
 #include <algorithm>
 #include <cstddef>
diff --git a/torch/csrc/api/src/data/samplers/stream.cpp b/torch/csrc/api/src/data/samplers/stream.cpp
index 5e41fb5d0ad6..2ac175541384 100644
--- a/torch/csrc/api/src/data/samplers/stream.cpp
+++ b/torch/csrc/api/src/data/samplers/stream.cpp
@@ -1,6 +1,6 @@
 #include <torch/data/samplers/stream.h>
 #include <torch/serialize/archive.h>
-#include <torch/tensor.h>
+#include <torch/types.h>
 
 #include <c10/util/Exception.h>
 
diff --git a/torch/csrc/api/src/nn/cursor.cpp b/torch/csrc/api/src/nn/cursor.cpp
index 4cf65145ed97..c3fe075456fe 100644
--- a/torch/csrc/api/src/nn/cursor.cpp
+++ b/torch/csrc/api/src/nn/cursor.cpp
@@ -1,7 +1,7 @@
 #include <torch/nn/cursor.h>
 
 #include <torch/nn/module.h>
-#include <torch/tensor.h>
+#include <torch/types.h>
 
 #include <algorithm>
 #include <cstdint>
diff --git a/torch/csrc/api/src/nn/init.cpp b/torch/csrc/api/src/nn/init.cpp
index f6f977fe4104..a2f49bb98c35 100644
--- a/torch/csrc/api/src/nn/init.cpp
+++ b/torch/csrc/api/src/nn/init.cpp
@@ -1,6 +1,6 @@
 #include <torch/nn/init.h>
 
-#include <torch/tensor.h>
+#include <torch/types.h>
 #include <torch/utils.h>
 
 #include <ATen/ATen.h>
diff --git a/torch/csrc/api/src/nn/modules/batchnorm.cpp b/torch/csrc/api/src/nn/modules/batchnorm.cpp
index 4896dcfb6a98..a8aa886f4a1d 100644
--- a/torch/csrc/api/src/nn/modules/batchnorm.cpp
+++ b/torch/csrc/api/src/nn/modules/batchnorm.cpp
@@ -1,7 +1,7 @@
 #include <torch/nn/modules/batchnorm.h>
 
 #include <torch/cuda.h>
-#include <torch/tensor.h>
+#include <torch/types.h>
 
 #include <c10/util/Exception.h>
 
diff --git a/torch/csrc/api/src/nn/modules/conv.cpp b/torch/csrc/api/src/nn/modules/conv.cpp
index e4c2ff35050a..14c804e269c9 100644
--- a/torch/csrc/api/src/nn/modules/conv.cpp
+++ b/torch/csrc/api/src/nn/modules/conv.cpp
@@ -1,7 +1,7 @@
 #include <torch/nn/modules/conv.h>
 
 #include <torch/expanding_array.h>
-#include <torch/tensor.h>
+#include <torch/types.h>
 #include <torch/utils.h>
 
 #include <cmath>
diff --git a/torch/csrc/api/src/nn/modules/dropout.cpp b/torch/csrc/api/src/nn/modules/dropout.cpp
index f980a4c7e058..b7887f4f04a8 100644
--- a/torch/csrc/api/src/nn/modules/dropout.cpp
+++ b/torch/csrc/api/src/nn/modules/dropout.cpp
@@ -1,6 +1,6 @@
 #include <torch/nn/modules/dropout.h>
 
-#include <torch/tensor.h>
+#include <torch/types.h>
 
 #include <c10/util/Exception.h>
 
diff --git a/torch/csrc/api/src/nn/modules/embedding.cpp b/torch/csrc/api/src/nn/modules/embedding.cpp
index 3911da67cbe4..d09afe057b84 100644
--- a/torch/csrc/api/src/nn/modules/embedding.cpp
+++ b/torch/csrc/api/src/nn/modules/embedding.cpp
@@ -1,6 +1,6 @@
 #include <torch/nn/modules/embedding.h>
 
-#include <torch/tensor.h>
+#include <torch/types.h>
 #include <torch/utils.h>
 
 #include <cstddef>
diff --git a/torch/csrc/api/src/nn/modules/functional.cpp b/torch/csrc/api/src/nn/modules/functional.cpp
index 1820e8f4df23..e35a2e60d2af 100644
--- a/torch/csrc/api/src/nn/modules/functional.cpp
+++ b/torch/csrc/api/src/nn/modules/functional.cpp
@@ -1,6 +1,6 @@
 #include <torch/nn/modules/functional.h>
 
-#include <torch/tensor.h>
+#include <torch/types.h>
 
 #include <functional>
 #include <utility>
diff --git a/torch/csrc/api/src/nn/modules/linear.cpp b/torch/csrc/api/src/nn/modules/linear.cpp
index d95d6b3e90bf..311558311a2d 100644
--- a/torch/csrc/api/src/nn/modules/linear.cpp
+++ b/torch/csrc/api/src/nn/modules/linear.cpp
@@ -1,6 +1,6 @@
 #include <torch/nn/modules/linear.h>
 
-#include <torch/tensor.h>
+#include <torch/types.h>
 #include <torch/utils.h>
 
 #include <cmath>
diff --git a/torch/csrc/api/src/nn/modules/rnn.cpp b/torch/csrc/api/src/nn/modules/rnn.cpp
index c828c49f6866..a8b49eae2aab 100644
--- a/torch/csrc/api/src/nn/modules/rnn.cpp
+++ b/torch/csrc/api/src/nn/modules/rnn.cpp
@@ -1,7 +1,7 @@
 #include <torch/nn/modules/rnn.h>
 
 #include <torch/nn/modules/dropout.h>
-#include <torch/tensor.h>
+#include <torch/types.h>
 #include <torch/utils.h>
 
 #include <c10/util/Exception.h>
diff --git a/torch/csrc/api/src/optim/optimizer.cpp b/torch/csrc/api/src/optim/optimizer.cpp
index 297a1341b87f..fe89c75690ac 100644
--- a/torch/csrc/api/src/optim/optimizer.cpp
+++ b/torch/csrc/api/src/optim/optimizer.cpp
@@ -3,7 +3,7 @@
 #include <torch/csrc/autograd/generated/variable_factories.h>
 #include <torch/nn/cursor.h>
 #include <torch/serialize/archive.h>
-#include <torch/tensor.h>
+#include <torch/types.h>
 
 #include <string>
 #include <utility>
diff --git a/torch/csrc/api/src/optim/serialize.cpp b/torch/csrc/api/src/optim/serialize.cpp
index 24f9096c6ac3..808b604a2484 100644
--- a/torch/csrc/api/src/optim/serialize.cpp
+++ b/torch/csrc/api/src/optim/serialize.cpp
@@ -1,7 +1,7 @@
 #include <torch/optim/serialize.h>
 
 #include <torch/serialize/archive.h>
-#include <torch/tensor.h>
+#include <torch/types.h>
 
 #include <cstddef>
 #include <cstdint>
diff --git a/torch/csrc/api/src/optim/sgd.cpp b/torch/csrc/api/src/optim/sgd.cpp
index 8d03872b0d18..0a927b78a13e 100644
--- a/torch/csrc/api/src/optim/sgd.cpp
+++ b/torch/csrc/api/src/optim/sgd.cpp
@@ -4,7 +4,7 @@
 #include <torch/nn/pimpl.h>
 #include <torch/optim/optimizer.h>
 #include <torch/optim/serialize.h>
-#include <torch/tensor.h>
+#include <torch/types.h>
 #include <torch/utils.h>
 
 #include <ATen/ATen.h>
diff --git a/torch/csrc/api/src/serialize/input-archive.cpp b/torch/csrc/api/src/serialize/input-archive.cpp
index 9d7b83c8168c..1e55182383f8 100644
--- a/torch/csrc/api/src/serialize/input-archive.cpp
+++ b/torch/csrc/api/src/serialize/input-archive.cpp
@@ -1,6 +1,6 @@
 #include <torch/serialize/input-archive.h>
 
-#include <torch/tensor.h>
+#include <torch/types.h>
 #include <torch/utils.h>
 
 #include <torch/csrc/jit/import.h>
diff --git a/torch/csrc/api/src/serialize/output-archive.cpp b/torch/csrc/api/src/serialize/output-archive.cpp
index 7b516265c870..5dd8cd644eb2 100644
--- a/torch/csrc/api/src/serialize/output-archive.cpp
+++ b/torch/csrc/api/src/serialize/output-archive.cpp
@@ -1,6 +1,6 @@
 #include <torch/serialize/output-archive.h>
 
-#include <torch/tensor.h>
+#include <torch/types.h>
 #include <torch/utils.h>
 
 #include <torch/csrc/jit/export.h>
diff --git a/torch/csrc/api/src/serialize/tensor.cpp b/torch/csrc/api/src/serialize/tensor.cpp
index 9fc92cf779e0..7b67d7f07261 100644
--- a/torch/csrc/api/src/serialize/tensor.cpp
+++ b/torch/csrc/api/src/serialize/tensor.cpp
@@ -1,4 +1,4 @@
-#include <torch/tensor.h>
+#include <torch/types.h>
 #include <torch/serialize/archive.h>
 
 namespace torch {
diff --git a/torch/csrc/jit/type.h b/torch/csrc/jit/type.h
index a236e0aa985f..0d6f728690d9 100644
--- a/torch/csrc/jit/type.h
+++ b/torch/csrc/jit/type.h
@@ -423,7 +423,6 @@ private:
 template<TypeKind K, typename T>
 struct SingleElementType : public Type {
   static const TypeKind Kind = K;
-  static constexpr bool is_singleton = true;
   TypePtr getElementType() const {
     return elem;
   }
@@ -488,9 +487,6 @@ struct FutureType;
 using FutureTypePtr = std::shared_ptr<FutureType>;
 
 struct TORCH_API FutureType : public Type {
-  // It's not exactly a singleton, but there should be exactly once instance of
-  // Future[T] for every T
-  static constexpr bool is_singleton = true;
   friend struct Type;
   template<typename ... T>
   static FutureTypePtr create(TypePtr elem) {
diff --git a/torch/script.h b/torch/script.h
index 982d3c9170c9..8c8cc5f563f6 100644
--- a/torch/script.h
+++ b/torch/script.h
@@ -1,5 +1,6 @@
 #pragma once
 
+#include <torch/csrc/api/include/torch/types.h>
 #include <torch/csrc/autograd/generated/variable_factories.h>
 #include <torch/csrc/jit/custom_operator.h>
 #include <torch/csrc/jit/import.h>
diff --git a/torch/utils/cpp_extension.py b/torch/utils/cpp_extension.py
index 28f0d6c08902..0b715d1010ad 100644
--- a/torch/utils/cpp_extension.py
+++ b/torch/utils/cpp_extension.py
@@ -633,13 +633,13 @@ def load_inline(name,
     as its docstring.
 
     The sources in ``cuda_sources`` are concatenated into a separate ``.cu``
-    file and  prepended with ``ATen/ATen.h``, ``cuda.h`` and ``cuda_runtime.h``
-    includes. The ``.cpp`` and ``.cu`` files are compiled separately, but
-    ultimately linked into a single library. Note that no bindings are
-    generated for functions in ``cuda_sources`` per  se. To bind to a CUDA
-    kernel, you must create a C++ function that calls it, and either declare or
-    define this C++ function in one of the ``cpp_sources`` (and include its
-    name in ``functions``).
+    file and  prepended with ``torch/types.h``, ``cuda.h`` and
+    ``cuda_runtime.h`` includes. The ``.cpp`` and ``.cu`` files are compiled
+    separately, but ultimately linked into a single library. Note that no
+    bindings are generated for functions in ``cuda_sources`` per  se. To bind
+    to a CUDA kernel, you must create a C++ function that calls it, and either
+    declare or define this C++ function in one of the ``cpp_sources`` (and
+    include its name in ``functions``).
 
     See :func:`load` for a description of arguments omitted below.
 
@@ -702,7 +702,7 @@ def load_inline(name,
     sources = [cpp_source_path]
 
     if cuda_sources:
-        cuda_sources.insert(0, '#include <ATen/ATen.h>')
+        cuda_sources.insert(0, '#include <torch/types.h>')
         cuda_sources.insert(1, '#include <cuda.h>')
         cuda_sources.insert(2, '#include <cuda_runtime.h>')