From 9cd055e547e9b67a5f9827f8999c38d7eda1bcb8 Mon Sep 17 00:00:00 2001
From: Yuanyuan Chen <cyyever@outlook.com>
Date: Fri, 28 Nov 2025 00:54:57 +0000
Subject: [PATCH] [2/N] Remove unused header inclusion (#165831)

Remove unused header inclusion in JIT code and other locations.

Pull Request resolved: https://github.com/pytorch/pytorch/pull/165831
Approved by: https://github.com/ngimel, https://github.com/albanD
---
 aten/src/ATen/core/DeprecatedTypeProperties.cpp     |  2 --
 aten/src/ATen/core/Formatting.cpp                   |  1 -
 aten/src/ATen/core/NamedRegistrations.cpp           |  2 --
 aten/src/ATen/core/PythonFallbackKernel.cpp         |  1 -
 aten/src/ATen/core/Tensor.cpp                       |  3 ---
 aten/src/ATen/core/VariableFallbackKernel.cpp       |  1 -
 aten/src/ATen/core/Vitals.cpp                       |  1 -
 aten/src/ATen/core/class_type.cpp                   |  2 --
 aten/src/ATen/core/custom_class.cpp                 |  1 -
 aten/src/ATen/core/interned_strings.cpp             |  2 --
 aten/src/ATen/core/tensor_type.cpp                  |  1 -
 aten/src/ATen/core/type.cpp                         |  2 --
 aten/src/ATen/core/union_type.cpp                   |  5 -----
 c10/cuda/CUDADeviceAssertionHost.cpp                |  1 -
 c10/cuda/CUDAException.cpp                          |  1 -
 c10/cuda/CUDAMiscFunctions.cpp                      |  1 -
 torch/csrc/Device.cpp                               |  3 ---
 torch/csrc/Dtype.cpp                                |  4 ----
 torch/csrc/DynamicTypes.cpp                         |  9 ---------
 torch/csrc/Event.cpp                                |  4 ----
 torch/csrc/Layout.cpp                               |  3 ---
 torch/csrc/MemoryFormat.cpp                         |  1 -
 torch/csrc/QScheme.cpp                              |  3 ---
 torch/csrc/Size.cpp                                 |  4 ++--
 torch/csrc/Stream.cpp                               |  3 ---
 torch/csrc/TypeInfo.cpp                             |  4 ----
 .../autograd/functions/recvrpc_backward.cpp         |  1 -
 torch/csrc/distributed/autograd/init.cpp            |  3 ---
 .../autograd/rpc_messages/rpc_with_autograd.cpp     |  1 -
 torch/csrc/distributed/autograd/utils.cpp           |  3 ---
 torch/csrc/jit/api/function_impl.cpp                |  1 -
 torch/csrc/jit/api/module.cpp                       |  7 -------
 torch/csrc/jit/backends/backend_debug_handler.cpp   |  2 --
 torch/csrc/jit/backends/backend_init.cpp            |  2 --
 torch/csrc/jit/backends/nnapi/nnapi_backend_lib.cpp |  1 -
 .../jit/backends/nnapi/nnapi_backend_preprocess.cpp |  2 --
 torch/csrc/jit/codegen/cuda/interface.cpp           |  8 --------
 torch/csrc/jit/codegen/fuser/codegen.cpp            |  4 ----
 torch/csrc/jit/codegen/fuser/compiler.cpp           |  7 -------
 torch/csrc/jit/codegen/fuser/executor.cpp           |  3 ---
 torch/csrc/jit/codegen/fuser/fallback.cpp           |  2 --
 torch/csrc/jit/codegen/fuser/interface.cpp          |  3 ---
 torch/csrc/jit/codegen/onednn/decompose_silu.cpp    |  2 --
 torch/csrc/jit/codegen/onednn/graph_fuser.cpp       |  2 --
 torch/csrc/jit/codegen/onednn/graph_helper.cpp      |  2 --
 torch/csrc/jit/codegen/onednn/graph_rewriter.cpp    |  3 ---
 torch/csrc/jit/codegen/onednn/guard_shape.cpp       |  2 --
 torch/csrc/jit/codegen/onednn/interface.cpp         |  2 --
 torch/csrc/jit/codegen/onednn/kernel.cpp            |  1 -
 torch/csrc/jit/frontend/builtin_functions.cpp       |  2 --
 .../jit/frontend/canonicalize_modified_loop.cpp     |  2 --
 torch/csrc/jit/frontend/error_report.cpp            |  2 --
 torch/csrc/jit/frontend/inline_loop_condition.cpp   |  2 --
 torch/csrc/jit/frontend/ir_emitter.cpp              |  5 -----
 torch/csrc/jit/frontend/lexer.cpp                   |  2 --
 torch/csrc/jit/frontend/schema_matching.cpp         |  1 -
 torch/csrc/jit/frontend/sugared_value.cpp           |  2 --
 torch/csrc/jit/frontend/tracer.cpp                  |  8 --------
 torch/csrc/jit/frontend/versioned_symbols.cpp       |  3 ---
 torch/csrc/jit/ir/alias_analysis.cpp                |  1 -
 torch/csrc/jit/ir/constants.cpp                     |  4 ----
 torch/csrc/jit/ir/ir.cpp                            |  3 ---
 torch/csrc/jit/ir/irparser.cpp                      |  2 --
 torch/csrc/jit/ir/node_hashing.cpp                  |  3 ---
 torch/csrc/jit/ir/type_hashing.cpp                  |  3 ---
 torch/csrc/jit/jit_log.cpp                          |  1 -
 torch/csrc/jit/jit_opt_limit.cpp                    |  4 ----
 torch/csrc/jit/mobile/compatibility/backport.cpp    |  2 --
 .../jit/mobile/compatibility/backport_manager.cpp   |  2 --
 torch/csrc/jit/mobile/function.cpp                  |  2 --
 torch/csrc/jit/mobile/import.cpp                    |  2 --
 torch/csrc/jit/mobile/interpreter.cpp               |  2 --
 torch/csrc/jit/mobile/module.cpp                    |  3 ---
 torch/csrc/jit/mobile/parse_bytecode.cpp            |  1 -
 torch/csrc/jit/mobile/register_ops_common_utils.cpp |  1 -
 torch/csrc/jit/mobile/train/export_data.cpp         |  1 -
 torch/csrc/jit/mobile/train/optim/sgd.cpp           |  3 ---
 torch/csrc/jit/mobile/train/sequential.cpp          |  1 -
 torch/csrc/jit/mobile/upgrader_mobile.cpp           |  1 -
 torch/csrc/jit/operator_upgraders/utils.cpp         |  3 +--
 torch/csrc/jit/passes/autocast.cpp                  |  1 -
 torch/csrc/jit/passes/bailout_graph.cpp             |  3 ---
 torch/csrc/jit/passes/check_strict_fusion.cpp       |  1 -
 .../jit/passes/common_subexpression_elimination.cpp |  2 --
 torch/csrc/jit/passes/concat_opt.cpp                |  2 --
 torch/csrc/jit/passes/constant_propagation.cpp      |  2 --
 torch/csrc/jit/passes/create_autodiff_subgraphs.cpp |  2 --
 torch/csrc/jit/passes/create_functional_graphs.cpp  |  1 -
 .../dbr_quantization/remove_redundant_aliases.cpp   |  1 -
 torch/csrc/jit/passes/dtype_analysis.cpp            |  5 -----
 torch/csrc/jit/passes/erase_number_types.cpp        |  1 -
 torch/csrc/jit/passes/fixup_trace_scope_blocks.cpp  |  3 ---
 torch/csrc/jit/passes/fold_conv_bn.cpp              |  1 -
 torch/csrc/jit/passes/frozen_concat_linear.cpp      |  6 ------
 .../csrc/jit/passes/frozen_conv_add_relu_fusion.cpp |  6 ------
 .../jit/passes/frozen_conv_add_relu_fusion_cuda.cpp |  2 --
 torch/csrc/jit/passes/frozen_conv_folding.cpp       |  2 --
 .../csrc/jit/passes/frozen_graph_optimizations.cpp  |  4 ----
 torch/csrc/jit/passes/frozen_linear_transpose.cpp   |  3 ---
 torch/csrc/jit/passes/fuse_relu.cpp                 |  1 -
 torch/csrc/jit/passes/graph_fuser.cpp               |  4 ----
 torch/csrc/jit/passes/guard_elimination.cpp         |  3 ---
 torch/csrc/jit/passes/hoist_conv_packed_params.cpp  |  2 --
 torch/csrc/jit/passes/inliner.cpp                   |  2 --
 torch/csrc/jit/passes/insert_guards.cpp             |  1 -
 torch/csrc/jit/passes/integer_value_refinement.cpp  |  1 -
 torch/csrc/jit/passes/liveness.cpp                  |  2 --
 torch/csrc/jit/passes/lower_tuples.cpp              |  1 -
 torch/csrc/jit/passes/metal_rewrite.cpp             |  4 ----
 torch/csrc/jit/passes/mkldnn_rewrite.cpp            |  1 -
 torch/csrc/jit/passes/normalize_ops.cpp             |  2 --
 torch/csrc/jit/passes/onnx.cpp                      |  3 ---
 torch/csrc/jit/passes/onnx/constant_map.cpp         |  2 --
 torch/csrc/jit/passes/onnx/eval_peephole.cpp        |  2 --
 .../csrc/jit/passes/onnx/fixup_onnx_controlflow.cpp |  2 --
 torch/csrc/jit/passes/onnx/helper.cpp               |  2 --
 .../csrc/jit/passes/onnx/list_model_parameters.cpp  |  2 --
 .../autograd_function_process.cpp                   |  2 --
 .../pattern_conversion/pattern_encapsulation.cpp    |  3 ---
 torch/csrc/jit/passes/onnx/preprocess_for_onnx.cpp  |  1 -
 torch/csrc/jit/passes/peephole.cpp                  |  3 ---
 torch/csrc/jit/passes/peephole_alias_sensitive.cpp  |  6 ------
 torch/csrc/jit/passes/peephole_list_idioms.cpp      |  3 ---
 torch/csrc/jit/passes/peephole_non_tensor.cpp       |  1 -
 torch/csrc/jit/passes/prepack_folding.cpp           |  1 -
 .../jit/passes/quantization/insert_observers.cpp    |  2 --
 torch/csrc/jit/passes/refine_tuple_types.cpp        |  2 --
 torch/csrc/jit/passes/remove_redundant_profiles.cpp |  2 --
 .../jit/passes/replacement_of_old_operators.cpp     |  1 -
 torch/csrc/jit/passes/requires_grad_analysis.cpp    |  1 -
 torch/csrc/jit/passes/restore_mutation.cpp          |  2 --
 torch/csrc/jit/passes/shape_analysis.cpp            |  3 ---
 torch/csrc/jit/passes/symbolic_shape_analysis.cpp   |  4 ----
 .../jit/passes/symbolic_shape_runtime_fusion.cpp    |  1 -
 torch/csrc/jit/passes/tensorexpr_fuser.cpp          |  2 --
 .../update_differentiable_graph_requires_grad.cpp   |  1 -
 torch/csrc/jit/passes/utils/memory_dag.cpp          |  1 -
 torch/csrc/jit/passes/utils/subgraph_utils.cpp      |  1 -
 torch/csrc/jit/passes/vulkan_rewrite.cpp            |  1 -
 torch/csrc/jit/python/pybind_utils.cpp              |  2 --
 torch/csrc/jit/python/python_custom_class.cpp       |  2 --
 torch/csrc/jit/python/python_dict.cpp               |  1 -
 torch/csrc/jit/python/python_interpreter.cpp        | 13 -------------
 torch/csrc/jit/python/python_ir.cpp                 |  4 ----
 torch/csrc/jit/python/python_sugared_value.cpp      |  4 ----
 torch/csrc/jit/python/python_tracer.cpp             |  1 -
 torch/csrc/jit/runtime/autodiff.cpp                 |  1 -
 torch/csrc/jit/runtime/decomposition_registry.cpp   |  1 -
 .../jit/runtime/decomposition_registry_util.cpp     |  3 ---
 torch/csrc/jit/runtime/graph_executor.cpp           |  1 -
 torch/csrc/jit/runtime/interpreter.cpp              |  8 --------
 torch/csrc/jit/runtime/jit_trace.cpp                |  6 ------
 torch/csrc/jit/runtime/operator.cpp                 |  1 -
 .../jit/runtime/profiling_graph_executor_impl.cpp   |  3 ---
 torch/csrc/jit/runtime/profiling_record.cpp         |  1 -
 torch/csrc/jit/runtime/register_c10_ops.cpp         |  3 ---
 torch/csrc/jit/runtime/register_cuda_ops.cpp        |  1 -
 torch/csrc/jit/runtime/register_distributed_ops.cpp |  3 ---
 .../csrc/jit/runtime/register_prim_ops_fulljit.cpp  | 11 -----------
 torch/csrc/jit/runtime/register_special_ops.cpp     |  3 ---
 torch/csrc/jit/runtime/script_profile.cpp           |  1 -
 .../runtime/serialized_shape_function_registry.cpp  |  3 ---
 torch/csrc/jit/runtime/static/fusion.cpp            |  1 -
 torch/csrc/jit/runtime/static/impl.cpp              |  3 ---
 torch/csrc/jit/runtime/static/memory_planner.cpp    |  3 ---
 torch/csrc/jit/runtime/static/native_ops.cpp        |  5 -----
 torch/csrc/jit/runtime/static/passes.cpp            |  2 --
 torch/csrc/jit/runtime/symbolic_shape_registry.cpp  |  1 -
 .../jit/runtime/symbolic_shape_registry_util.cpp    |  5 -----
 torch/csrc/jit/testing/file_check.cpp               |  1 -
 torch/csrc/python_dimname.cpp                       |  1 -
 torch/csrc/utils/cpp_stacktraces.cpp                |  3 ---
 torch/csrc/utils/device_lazy_init.cpp               |  1 -
 torch/csrc/utils/disable_torch_function.cpp         |  1 -
 torch/csrc/utils/init.cpp                           |  3 ---
 torch/csrc/utils/object_ptr.cpp                     |  2 --
 torch/csrc/utils/python_dispatch.cpp                |  8 --------
 torch/csrc/utils/tensor_apply.cpp                   |  2 --
 torch/csrc/utils/tensor_dtypes.cpp                  |  1 -
 torch/csrc/utils/tensor_layouts.cpp                 |  3 ---
 torch/csrc/utils/tensor_list.cpp                    |  2 --
 torch/csrc/utils/tensor_memoryformats.cpp           |  2 --
 torch/csrc/utils/tensor_qschemes.cpp                |  2 --
 torch/csrc/utils/tensor_types.cpp                   |  2 --
 torch/csrc/utils/throughput_benchmark.cpp           |  2 --
 185 files changed, 3 insertions(+), 462 deletions(-)

diff --git a/aten/src/ATen/core/DeprecatedTypeProperties.cpp b/aten/src/ATen/core/DeprecatedTypeProperties.cpp
index a97a6828571..369556aad91 100644
--- a/aten/src/ATen/core/DeprecatedTypeProperties.cpp
+++ b/aten/src/ATen/core/DeprecatedTypeProperties.cpp
@@ -1,7 +1,5 @@
 #include <ATen/core/DeprecatedTypeProperties.h>
 
-#include <ATen/core/LegacyTypeDispatch.h>
-#include <ATen/core/Tensor.h>
 #include <ATen/core/UnsafeFromTH.h>
 
 namespace at {
diff --git a/aten/src/ATen/core/Formatting.cpp b/aten/src/ATen/core/Formatting.cpp
index eddd5e4b4d6..62b16a83e52 100644
--- a/aten/src/ATen/core/Formatting.cpp
+++ b/aten/src/ATen/core/Formatting.cpp
@@ -9,7 +9,6 @@
 #include <iostream>
 #include <iterator>
 #include <string>
-#include <tuple>
 
 namespace c10 {
 std::ostream& operator<<(std::ostream& out, Backend b) {
diff --git a/aten/src/ATen/core/NamedRegistrations.cpp b/aten/src/ATen/core/NamedRegistrations.cpp
index b78a563b673..fc2193e70cb 100644
--- a/aten/src/ATen/core/NamedRegistrations.cpp
+++ b/aten/src/ATen/core/NamedRegistrations.cpp
@@ -1,7 +1,5 @@
 #include <torch/library.h>
 
-#include <ATen/core/boxing/KernelFunction.h>
-
 using torch::CppFunction;
 
 TORCH_LIBRARY_IMPL(_, Named, m) {
diff --git a/aten/src/ATen/core/PythonFallbackKernel.cpp b/aten/src/ATen/core/PythonFallbackKernel.cpp
index 39f4e7cb697..7b2b32531f0 100644
--- a/aten/src/ATen/core/PythonFallbackKernel.cpp
+++ b/aten/src/ATen/core/PythonFallbackKernel.cpp
@@ -1,7 +1,6 @@
 #include <c10/core/impl/TorchDispatchModeTLS.h>
 #include <c10/core/impl/PythonDispatcherTLS.h>
 #include <ATen/core/PythonFallbackKernel.h>
-#include <c10/core/SafePyObject.h>
 #include <ATen/record_function.h>
 
 namespace {
diff --git a/aten/src/ATen/core/Tensor.cpp b/aten/src/ATen/core/Tensor.cpp
index 090e77e7037..70907a60b65 100644
--- a/aten/src/ATen/core/Tensor.cpp
+++ b/aten/src/ATen/core/Tensor.cpp
@@ -1,8 +1,5 @@
 #include <ATen/core/Tensor.h>
-#include <ATen/core/Formatting.h>
 #include <ATen/core/VariableHooksInterface.h>
-#include <ATen/core/LegacyTypeDispatch.h>
-#include <ATen/FunctionalTensorWrapper.h>
 
 #ifndef AT_PER_OPERATOR_HEADERS
 #include <ATen/MethodOperators.h>
diff --git a/aten/src/ATen/core/VariableFallbackKernel.cpp b/aten/src/ATen/core/VariableFallbackKernel.cpp
index dad3f090bb1..94422df4045 100644
--- a/aten/src/ATen/core/VariableFallbackKernel.cpp
+++ b/aten/src/ATen/core/VariableFallbackKernel.cpp
@@ -1,4 +1,3 @@
-#include <ATen/core/LegacyTypeDispatch.h>
 #include <ATen/core/dispatch/Dispatcher.h>
 #include <ATen/core/VariableHooksInterface.h>
 #include <torch/library.h>
diff --git a/aten/src/ATen/core/Vitals.cpp b/aten/src/ATen/core/Vitals.cpp
index ac1ee45d583..db58c038305 100644
--- a/aten/src/ATen/core/Vitals.cpp
+++ b/aten/src/ATen/core/Vitals.cpp
@@ -1,6 +1,5 @@
 #include <ATen/core/Vitals.h>
 #include <c10/util/env.h>
-#include <cstdlib>
 #include <iostream>
 
 namespace at::vitals {
diff --git a/aten/src/ATen/core/class_type.cpp b/aten/src/ATen/core/class_type.cpp
index a65124e8097..ec1dba6192a 100644
--- a/aten/src/ATen/core/class_type.cpp
+++ b/aten/src/ATen/core/class_type.cpp
@@ -1,12 +1,10 @@
 #include <ATen/core/class_type.h>
 
 #include <ATen/core/Dict.h>
-#include <ATen/core/Tensor.h>
 #include <ATen/core/function_schema.h>
 #include <ATen/core/ivalue.h>
 #include <c10/macros/Macros.h>
 #include <c10/util/irange.h>
-#include <ATen/core/grad_mode.h>
 #include <ATen/core/function.h>
 
 namespace c10 {
diff --git a/aten/src/ATen/core/custom_class.cpp b/aten/src/ATen/core/custom_class.cpp
index 2c9cc465466..820d27097c4 100644
--- a/aten/src/ATen/core/custom_class.cpp
+++ b/aten/src/ATen/core/custom_class.cpp
@@ -2,7 +2,6 @@
 #include <ATen/core/functional.h>
 #include <ATen/core/jit_type.h>
 #include <ATen/core/type_factory.h>
-#include <ATen/record_function.h>
 #include <c10/util/flat_hash_map.h>
 #include <torch/custom_class.h>
 #include <torch/custom_class_detail.h>
diff --git a/aten/src/ATen/core/interned_strings.cpp b/aten/src/ATen/core/interned_strings.cpp
index 799f6821bb9..018ee82fe32 100644
--- a/aten/src/ATen/core/interned_strings.cpp
+++ b/aten/src/ATen/core/interned_strings.cpp
@@ -2,12 +2,10 @@
 #undef TORCH_ASSERT_ONLY_METHOD_OPERATORS
 
 #include <ATen/core/interned_strings.h>
-#include <cstdint>
 #include <cstring>
 #include <mutex>
 #include <sstream>
 #include <string>
-#include <c10/util/Exception.h>
 #include <ATen/core/interned_strings_class.h>
 
 namespace c10 {
diff --git a/aten/src/ATen/core/tensor_type.cpp b/aten/src/ATen/core/tensor_type.cpp
index d428aceb3d0..debd5e92bbc 100644
--- a/aten/src/ATen/core/tensor_type.cpp
+++ b/aten/src/ATen/core/tensor_type.cpp
@@ -1,4 +1,3 @@
-#include <ATen/core/Tensor.h>
 #include <ATen/core/jit_type.h>
 #include <c10/core/GradMode.h>
 
diff --git a/aten/src/ATen/core/type.cpp b/aten/src/ATen/core/type.cpp
index 35a729ccc9f..215f91eed68 100644
--- a/aten/src/ATen/core/type.cpp
+++ b/aten/src/ATen/core/type.cpp
@@ -1,10 +1,8 @@
 #include <ATen/core/Dict.h>
-#include <ATen/core/Tensor.h>
 #include <ATen/core/dynamic_type.h>
 #include <ATen/core/enum_type.h>
 #include <ATen/core/function.h>
 #include <ATen/core/function_schema.h>
-#include <ATen/core/grad_mode.h>
 #include <ATen/core/jit_type.h>
 #include <c10/macros/Macros.h>
 #include <c10/util/env.h>
diff --git a/aten/src/ATen/core/union_type.cpp b/aten/src/ATen/core/union_type.cpp
index 8731c2cbc49..6113041f154 100644
--- a/aten/src/ATen/core/union_type.cpp
+++ b/aten/src/ATen/core/union_type.cpp
@@ -1,10 +1,5 @@
 #include <ATen/core/Dict.h>
-#include <ATen/core/Tensor.h>
-#include <ATen/core/function.h>
-#include <ATen/core/function_schema.h>
-#include <ATen/core/grad_mode.h>
 #include <ATen/core/jit_type.h>
-#include <ATen/core/type_factory.h>
 #include <c10/macros/Macros.h>
 #include <c10/util/irange.h>
 #include <ostream>
diff --git a/c10/cuda/CUDADeviceAssertionHost.cpp b/c10/cuda/CUDADeviceAssertionHost.cpp
index 08e657a4116..43dbb92531c 100644
--- a/c10/cuda/CUDADeviceAssertionHost.cpp
+++ b/c10/cuda/CUDADeviceAssertionHost.cpp
@@ -3,7 +3,6 @@
 #include <c10/cuda/CUDAFunctions.h>
 #include <c10/util/env.h>
 #include <c10/util/irange.h>
-#include <cuda_runtime.h>
 
 #include <memory>
 #include <string>
diff --git a/c10/cuda/CUDAException.cpp b/c10/cuda/CUDAException.cpp
index 4e4419b4369..f0dbe49d2ea 100644
--- a/c10/cuda/CUDAException.cpp
+++ b/c10/cuda/CUDAException.cpp
@@ -2,7 +2,6 @@
 
 #include <c10/cuda/CUDADeviceAssertionHost.h>
 #include <c10/util/Exception.h>
-#include <cuda_runtime.h>
 
 #include <string>
 
diff --git a/c10/cuda/CUDAMiscFunctions.cpp b/c10/cuda/CUDAMiscFunctions.cpp
index 49bad41dda8..70bb0f841b3 100644
--- a/c10/cuda/CUDAMiscFunctions.cpp
+++ b/c10/cuda/CUDAMiscFunctions.cpp
@@ -1,6 +1,5 @@
 #include <c10/cuda/CUDAMiscFunctions.h>
 #include <c10/util/env.h>
-#include <cuda_runtime.h>
 #include <string>
 
 namespace c10::cuda {
diff --git a/torch/csrc/Device.cpp b/torch/csrc/Device.cpp
index da7b287369d..b3acb4e4bb4 100644
--- a/torch/csrc/Device.cpp
+++ b/torch/csrc/Device.cpp
@@ -2,15 +2,12 @@
 
 #include <torch/csrc/Exceptions.h>
 #include <torch/csrc/utils/object_ptr.h>
-#include <torch/csrc/utils/pybind.h>
 #include <torch/csrc/utils/python_arg_parser.h>
 #include <torch/csrc/utils/python_numbers.h>
 #include <torch/csrc/utils/python_strings.h>
 
-#include <ATen/Device.h>
 #include <c10/util/Exception.h>
 
-#include <structmember.h>
 #include <limits>
 #include <sstream>
 
diff --git a/torch/csrc/Dtype.cpp b/torch/csrc/Dtype.cpp
index c302378de81..bff17ca0cbc 100644
--- a/torch/csrc/Dtype.cpp
+++ b/torch/csrc/Dtype.cpp
@@ -1,15 +1,11 @@
 #include <torch/csrc/Dtype.h>
 
 #include <c10/core/ScalarType.h>
-#include <structmember.h>
 #include <torch/csrc/DynamicTypes.h>
 #include <torch/csrc/Exceptions.h>
 #include <torch/csrc/utils/object_ptr.h>
 #include <torch/csrc/utils/python_numbers.h>
 #include <torch/csrc/utils/python_strings.h>
-#include <torch/csrc/utils/pythoncapi_compat.h>
-#include <torch/csrc/utils/tensor_dtypes.h>
-#include <torch/csrc/utils/tensor_types.h>
 #include <cstring>
 
 PyObject* THPDtype_New(at::ScalarType scalar_type, const std::string& name) {
diff --git a/torch/csrc/DynamicTypes.cpp b/torch/csrc/DynamicTypes.cpp
index d5621146fef..9db1903eec3 100644
--- a/torch/csrc/DynamicTypes.cpp
+++ b/torch/csrc/DynamicTypes.cpp
@@ -1,18 +1,9 @@
-#include <torch/csrc/python_headers.h>
 
-#include <torch/csrc/Device.h>
 #include <torch/csrc/Dtype.h>
 #include <torch/csrc/DynamicTypes.h>
 #include <torch/csrc/Exceptions.h>
 #include <torch/csrc/Layout.h>
 #include <torch/csrc/Storage.h>
-#include <torch/csrc/autograd/generated/VariableType.h>
-#include <torch/csrc/utils/cuda_enabled.h>
-#include <torch/csrc/utils/device_lazy_init.h>
-#include <torch/csrc/utils/object_ptr.h>
-
-#include <ATen/ATen.h>
-#include <ATen/FunctionalStorageImpl.h>
 
 #include <array>
 #include <stdexcept>
diff --git a/torch/csrc/Event.cpp b/torch/csrc/Event.cpp
index fd7d72228fc..f5bb1b60eac 100644
--- a/torch/csrc/Event.cpp
+++ b/torch/csrc/Event.cpp
@@ -1,9 +1,6 @@
-#include <pybind11/pybind11.h>
 #include <torch/csrc/Device.h>
 #include <torch/csrc/Event.h>
 #include <torch/csrc/Stream.h>
-#include <torch/csrc/THP.h>
-#include <torch/csrc/utils/pybind.h>
 #include <torch/csrc/utils/pycfunction_helpers.h>
 #include <torch/csrc/utils/python_arg_parser.h>
 
@@ -12,7 +9,6 @@
 
 #include <c10/core/DeviceType.h>
 #include <c10/core/impl/DeviceGuardImplInterface.h>
-#include <structmember.h>
 #include <string>
 
 PyTypeObject* THPEventClass = nullptr;
diff --git a/torch/csrc/Layout.cpp b/torch/csrc/Layout.cpp
index 06b49d56f64..af7dfc74379 100644
--- a/torch/csrc/Layout.cpp
+++ b/torch/csrc/Layout.cpp
@@ -4,9 +4,6 @@
 #include <torch/csrc/utils/object_ptr.h>
 #include <torch/csrc/utils/python_strings.h>
 
-#include <ATen/Layout.h>
-
-#include <structmember.h>
 #include <cstring>
 #include <string>
 
diff --git a/torch/csrc/MemoryFormat.cpp b/torch/csrc/MemoryFormat.cpp
index 5bd3f9eed42..0a8e212500c 100644
--- a/torch/csrc/MemoryFormat.cpp
+++ b/torch/csrc/MemoryFormat.cpp
@@ -6,7 +6,6 @@
 
 #include <c10/core/MemoryFormat.h>
 
-#include <structmember.h>
 #include <cstring>
 #include <string>
 
diff --git a/torch/csrc/QScheme.cpp b/torch/csrc/QScheme.cpp
index 3fbabc1026f..e178ec9247e 100644
--- a/torch/csrc/QScheme.cpp
+++ b/torch/csrc/QScheme.cpp
@@ -4,9 +4,6 @@
 #include <torch/csrc/utils/object_ptr.h>
 #include <torch/csrc/utils/python_strings.h>
 
-#include <c10/core/QScheme.h>
-
-#include <structmember.h>
 #include <cstring>
 #include <string>
 
diff --git a/torch/csrc/Size.cpp b/torch/csrc/Size.cpp
index ea39424cf8e..7a136420d79 100644
--- a/torch/csrc/Size.cpp
+++ b/torch/csrc/Size.cpp
@@ -1,12 +1,12 @@
 #include <c10/util/irange.h>
 #include <pybind11/pytypes.h>
 #include <torch/csrc/Size.h>
-#include <torch/csrc/utils/pybind.h>
+// #include <torch/csrc/utils/pybind.h>
 
 #include <torch/csrc/utils/object_ptr.h>
-#include <torch/csrc/utils/python_arg_parser.h>
 #include <torch/csrc/utils/python_numbers.h>
 #include <torch/csrc/utils/python_strings.h>
+#include <torch/csrc/utils/python_symnode.h>
 #include <torch/csrc/utils/python_tuples.h>
 #include <string>
 
diff --git a/torch/csrc/Stream.cpp b/torch/csrc/Stream.cpp
index 6993f726597..3b290b0cfbe 100644
--- a/torch/csrc/Stream.cpp
+++ b/torch/csrc/Stream.cpp
@@ -1,9 +1,6 @@
-#include <pybind11/pybind11.h>
 #include <torch/csrc/Device.h>
 #include <torch/csrc/Event.h>
 #include <torch/csrc/Stream.h>
-#include <torch/csrc/THP.h>
-#include <torch/csrc/utils/pybind.h>
 #include <torch/csrc/utils/pycfunction_helpers.h>
 #include <torch/csrc/utils/python_arg_parser.h>
 
diff --git a/torch/csrc/TypeInfo.cpp b/torch/csrc/TypeInfo.cpp
index de23b795360..355202c7e40 100644
--- a/torch/csrc/TypeInfo.cpp
+++ b/torch/csrc/TypeInfo.cpp
@@ -2,18 +2,14 @@
 
 #include <torch/csrc/Exceptions.h>
 #include <torch/csrc/utils/object_ptr.h>
-#include <torch/csrc/utils/pybind.h>
 #include <torch/csrc/utils/python_arg_parser.h>
 #include <torch/csrc/utils/python_numbers.h>
 #include <torch/csrc/utils/python_strings.h>
-#include <torch/csrc/utils/tensor_dtypes.h>
 
 #include <ATen/Dispatch_v2.h>
 
 #include <c10/util/Exception.h>
 
-#include <structmember.h>
-#include <cstring>
 #include <limits>
 #include <sstream>
 
diff --git a/torch/csrc/distributed/autograd/functions/recvrpc_backward.cpp b/torch/csrc/distributed/autograd/functions/recvrpc_backward.cpp
index c2d4630bdd0..7bd98144439 100644
--- a/torch/csrc/distributed/autograd/functions/recvrpc_backward.cpp
+++ b/torch/csrc/distributed/autograd/functions/recvrpc_backward.cpp
@@ -1,4 +1,3 @@
-#include <ATen/core/functional.h>
 #include <c10/util/irange.h>
 #include <torch/csrc/distributed/autograd/functions/recvrpc_backward.h>
 #include <torch/csrc/distributed/autograd/rpc_messages/propagate_gradients_req.h>
diff --git a/torch/csrc/distributed/autograd/init.cpp b/torch/csrc/distributed/autograd/init.cpp
index 115d371524d..1d4bacc0943 100644
--- a/torch/csrc/distributed/autograd/init.cpp
+++ b/torch/csrc/distributed/autograd/init.cpp
@@ -2,10 +2,7 @@
 #include <torch/csrc/distributed/autograd/autograd.h>
 #include <torch/csrc/distributed/autograd/python_autograd.h>
 #include <torch/csrc/jit/python/pybind_utils.h>
-#include <torch/csrc/python_headers.h>
 #include <torch/csrc/utils/object_ptr.h>
-#include <torch/csrc/utils/pybind.h>
-#include <torch/types.h>
 
 namespace torch::distributed::autograd {
 
diff --git a/torch/csrc/distributed/autograd/rpc_messages/rpc_with_autograd.cpp b/torch/csrc/distributed/autograd/rpc_messages/rpc_with_autograd.cpp
index fd5ab54e58c..4e9af6d1240 100644
--- a/torch/csrc/distributed/autograd/rpc_messages/rpc_with_autograd.cpp
+++ b/torch/csrc/distributed/autograd/rpc_messages/rpc_with_autograd.cpp
@@ -2,7 +2,6 @@
 #include <torch/csrc/distributed/rpc/rpc_agent.h>
 #include <torch/csrc/distributed/rpc/utils.h>
 #include <torch/csrc/jit/serialization/pickle.h>
-#include <torch/csrc/utils/byte_order.h>
 
 namespace torch::distributed::autograd {
 
diff --git a/torch/csrc/distributed/autograd/utils.cpp b/torch/csrc/distributed/autograd/utils.cpp
index 84ddaa1a5ce..ec1bbf13375 100644
--- a/torch/csrc/distributed/autograd/utils.cpp
+++ b/torch/csrc/distributed/autograd/utils.cpp
@@ -1,7 +1,4 @@
-#include <ATen/ThreadLocalState.h>
-#include <c10/util/ThreadLocalDebugInfo.h>
 #include <torch/csrc/autograd/functions/utils.h>
-#include <torch/csrc/autograd/profiler.h>
 #include <torch/csrc/distributed/autograd/context/container.h>
 #include <torch/csrc/distributed/autograd/functions/recvrpc_backward.h>
 #include <torch/csrc/distributed/autograd/functions/sendrpc_backward.h>
diff --git a/torch/csrc/jit/api/function_impl.cpp b/torch/csrc/jit/api/function_impl.cpp
index 0c911970347..5fb7fc1f017 100644
--- a/torch/csrc/jit/api/function_impl.cpp
+++ b/torch/csrc/jit/api/function_impl.cpp
@@ -3,7 +3,6 @@
 #include <torch/csrc/jit/api/function_impl.h>
 #include <torch/csrc/jit/passes/inliner.h>
 
-#include <torch/csrc/jit/frontend/error_report.h>
 #include <torch/csrc/jit/passes/constant_pooling.h>
 #include <torch/csrc/jit/passes/constant_propagation.h>
 #include <torch/csrc/jit/passes/peephole.h>
diff --git a/torch/csrc/jit/api/module.cpp b/torch/csrc/jit/api/module.cpp
index 61c32680c7c..88d235b8a27 100644
--- a/torch/csrc/jit/api/module.cpp
+++ b/torch/csrc/jit/api/module.cpp
@@ -1,22 +1,15 @@
-#include <ATen/core/symbol.h>
 #include <ATen/record_function.h>
 #include <c10/util/Exception.h>
 #include <c10/util/StringUtil.h>
 #include <c10/util/irange.h>
-#include <torch/csrc/autograd/generated/variable_factories.h>
 #include <torch/csrc/jit/api/function_impl.h>
 #include <torch/csrc/jit/api/module.h>
-#include <torch/csrc/jit/frontend/error_report.h>
-#include <torch/csrc/jit/frontend/ir_emitter.h>
-#include <torch/csrc/jit/frontend/schema_matching.h>
 #include <torch/csrc/jit/jit_log.h>
-#include <torch/csrc/jit/passes/dead_code_elimination.h>
 #include <torch/csrc/jit/passes/freeze_module.h>
 #include <torch/csrc/jit/passes/frozen_conv_add_relu_fusion.h>
 #include <torch/csrc/jit/passes/frozen_graph_optimizations.h>
 #include <torch/csrc/jit/passes/frozen_linear_transpose.h>
 #include <torch/csrc/jit/passes/frozen_ops_to_mkldnn.h>
-#include <torch/csrc/jit/passes/inliner.h>
 #include <torch/csrc/jit/runtime/operator.h>
 
 #include <iostream>
diff --git a/torch/csrc/jit/backends/backend_debug_handler.cpp b/torch/csrc/jit/backends/backend_debug_handler.cpp
index 0d410341303..ec9f2e4fa56 100644
--- a/torch/csrc/jit/backends/backend_debug_handler.cpp
+++ b/torch/csrc/jit/backends/backend_debug_handler.cpp
@@ -1,7 +1,5 @@
 #include <torch/csrc/jit/backends/backend_debug_handler.h>
 
-#include <stack>
-
 namespace torch::jit {
 
 std::atomic<DebugHandleType> BackendDebugInfoRecorder::unique_debug_handle_{0};
diff --git a/torch/csrc/jit/backends/backend_init.cpp b/torch/csrc/jit/backends/backend_init.cpp
index b10aba884c7..ea71203412e 100644
--- a/torch/csrc/jit/backends/backend_init.cpp
+++ b/torch/csrc/jit/backends/backend_init.cpp
@@ -2,10 +2,8 @@
 
 #include <pybind11/iostream.h>
 #include <torch/csrc/jit/backends/backend_detail.h>
-#include <torch/csrc/jit/backends/backend_resolver.h>
 #include <torch/csrc/jit/python/module_python.h>
 #include <torch/csrc/jit/python/pybind_utils.h>
-#include <torch/csrc/utils/pybind.h>
 
 namespace torch::jit {
 
diff --git a/torch/csrc/jit/backends/nnapi/nnapi_backend_lib.cpp b/torch/csrc/jit/backends/nnapi/nnapi_backend_lib.cpp
index 18c1bc62b8c..b0e368d6a30 100644
--- a/torch/csrc/jit/backends/nnapi/nnapi_backend_lib.cpp
+++ b/torch/csrc/jit/backends/nnapi/nnapi_backend_lib.cpp
@@ -2,7 +2,6 @@
 
 #include <ATen/nnapi/nnapi_bind.h>
 #include <torch/csrc/jit/backends/backend.h>
-#include <torch/csrc/jit/backends/backend_exception.h>
 #include <torch/csrc/jit/mobile/import.h>
 #include <torch/csrc/jit/mobile/module.h>
 
diff --git a/torch/csrc/jit/backends/nnapi/nnapi_backend_preprocess.cpp b/torch/csrc/jit/backends/nnapi/nnapi_backend_preprocess.cpp
index 070e96c4f18..af6e9909dea 100644
--- a/torch/csrc/jit/backends/nnapi/nnapi_backend_preprocess.cpp
+++ b/torch/csrc/jit/backends/nnapi/nnapi_backend_preprocess.cpp
@@ -1,7 +1,5 @@
 #include <pybind11/pybind11.h>
-#include <torch/csrc/jit/backends/backend.h>
 #include <torch/csrc/jit/backends/backend_preprocess.h>
-#include <torch/csrc/jit/python/pybind_utils.h>
 #include <torch/csrc/utils/pybind.h>
 
 namespace py = pybind11;
diff --git a/torch/csrc/jit/codegen/cuda/interface.cpp b/torch/csrc/jit/codegen/cuda/interface.cpp
index 8dfa2bcc09c..d47c9f654d2 100644
--- a/torch/csrc/jit/codegen/cuda/interface.cpp
+++ b/torch/csrc/jit/codegen/cuda/interface.cpp
@@ -1,13 +1,5 @@
 #include <torch/csrc/jit/codegen/cuda/interface.h>
 
-#include <ATen/DynamicLibrary.h>
-#include <ATen/core/dispatch/OperatorOptions.h>
-#include <ATen/native/NonSymbolicBC.h>
-#include <ATen/native/TensorShape.h>
-#include <c10/util/irange.h>
-#include <torch/csrc/jit/runtime/custom_operator.h>
-#include <torch/csrc/jit/runtime/register_ops_utils.h>
-
 namespace torch::jit::fuser::cuda {
 
 static std::atomic<bool> cuda_fusion_guard_mode{true};
diff --git a/torch/csrc/jit/codegen/fuser/codegen.cpp b/torch/csrc/jit/codegen/fuser/codegen.cpp
index a5cd6f4e3a4..cb787cc2b58 100644
--- a/torch/csrc/jit/codegen/fuser/codegen.cpp
+++ b/torch/csrc/jit/codegen/fuser/codegen.cpp
@@ -1,11 +1,8 @@
 #include <torch/csrc/jit/codegen/fuser/codegen.h>
 
-#include <ATen/ATen.h>
 #include <ATen/code_template.h>
 #include <c10/util/Exception.h>
 #include <torch/csrc/jit/codegen/fuser/compiler.h>
-#include <torch/csrc/jit/codegen/fuser/interface.h>
-#include <torch/csrc/jit/codegen/fuser/tensor_info.h>
 #include <torch/csrc/jit/ir/ir.h>
 
 #include <torch/csrc/jit/codegen/fuser/cpu/resource_strings.h>
@@ -15,7 +12,6 @@
 #include <cstdint>
 #include <iostream>
 #include <sstream>
-#include <tuple>
 #include <vector>
 
 namespace torch::jit::fuser {
diff --git a/torch/csrc/jit/codegen/fuser/compiler.cpp b/torch/csrc/jit/codegen/fuser/compiler.cpp
index a1ff6cb613e..21d1a4734f7 100644
--- a/torch/csrc/jit/codegen/fuser/compiler.cpp
+++ b/torch/csrc/jit/codegen/fuser/compiler.cpp
@@ -1,25 +1,18 @@
 #include <torch/csrc/jit/codegen/fuser/compiler.h>
 
-#include <ATen/ATen.h>
 #include <ATen/core/jit_type.h>
 #include <c10/util/Exception.h>
 #include <c10/util/irange.h>
 #include <torch/csrc/jit/codegen/fuser/codegen.h>
-#include <torch/csrc/jit/codegen/fuser/interface.h>
 #include <torch/csrc/jit/codegen/fuser/kernel_cache.h>
 #include <torch/csrc/jit/codegen/fuser/tensor_desc.h>
 #include <torch/csrc/jit/ir/ir.h>
-#include <torch/csrc/jit/passes/canonicalize.h>
 #include <torch/csrc/jit/passes/shape_analysis.h>
 #include <torch/csrc/jit/runtime/operator.h>
 
 #include <atomic>
-#include <iostream>
 #include <memory>
-#include <sstream>
-#include <stdexcept>
 #include <string>
-#include <tuple>
 #include <unordered_set>
 #include <utility>
 
diff --git a/torch/csrc/jit/codegen/fuser/executor.cpp b/torch/csrc/jit/codegen/fuser/executor.cpp
index 67c4501dc27..d66c8f94db4 100644
--- a/torch/csrc/jit/codegen/fuser/executor.cpp
+++ b/torch/csrc/jit/codegen/fuser/executor.cpp
@@ -1,8 +1,6 @@
 #include <torch/csrc/jit/codegen/fuser/executor.h>
 
-#include <ATen/ATen.h>
 #include <ATen/ExpandUtils.h>
-#include <ATen/core/functional.h>
 #include <ATen/core/stack.h>
 #include <c10/util/irange.h>
 #include <torch/csrc/jit/codegen/fuser/compiler.h>
@@ -13,7 +11,6 @@
 #include <torch/csrc/jit/passes/graph_fuser.h>
 #include <optional>
 
-#include <algorithm>
 #include <vector>
 
 namespace torch::jit::fuser {
diff --git a/torch/csrc/jit/codegen/fuser/fallback.cpp b/torch/csrc/jit/codegen/fuser/fallback.cpp
index 698e2882d6a..a3655b63824 100644
--- a/torch/csrc/jit/codegen/fuser/fallback.cpp
+++ b/torch/csrc/jit/codegen/fuser/fallback.cpp
@@ -1,13 +1,11 @@
 #include <torch/csrc/jit/codegen/fuser/fallback.h>
 
-#include <ATen/core/functional.h> //fmap
 #include <ATen/core/stack.h>
 #include <c10/util/Exception.h>
 #include <torch/csrc/jit/codegen/fuser/kernel_cache.h>
 #include <torch/csrc/jit/ir/ir.h>
 #include <torch/csrc/jit/runtime/custom_operator.h>
 #include <torch/csrc/jit/runtime/interpreter.h>
-#include <stdexcept>
 
 namespace torch::jit::fuser {
 
diff --git a/torch/csrc/jit/codegen/fuser/interface.cpp b/torch/csrc/jit/codegen/fuser/interface.cpp
index 41efa23e2b4..90537815be4 100644
--- a/torch/csrc/jit/codegen/fuser/interface.cpp
+++ b/torch/csrc/jit/codegen/fuser/interface.cpp
@@ -3,11 +3,8 @@
 #include <torch/csrc/jit/codegen/fuser/compiler.h>
 #include <torch/csrc/jit/codegen/fuser/executor.h>
 #include <torch/csrc/jit/codegen/fuser/fallback.h>
-#include <torch/csrc/jit/codegen/fuser/kernel_cache.h>
 
 #include <c10/util/Exception.h>
-#include <c10/util/Flags.h>
-#include <stdexcept>
 
 namespace torch::jit {
 
diff --git a/torch/csrc/jit/codegen/onednn/decompose_silu.cpp b/torch/csrc/jit/codegen/onednn/decompose_silu.cpp
index 8a9e36c2973..0a03cf6c871 100644
--- a/torch/csrc/jit/codegen/onednn/decompose_silu.cpp
+++ b/torch/csrc/jit/codegen/onednn/decompose_silu.cpp
@@ -1,9 +1,7 @@
 #include <torch/csrc/jit/codegen/onednn/decompose_silu.h>
 #include <torch/csrc/jit/codegen/onednn/operator.h>
 
-#include <ATen/code_template.h>
 #include <torch/csrc/jit/passes/dead_code_elimination.h>
-#include <torch/csrc/jit/passes/subgraph_rewrite.h>
 
 namespace torch::jit::fuser::onednn {
 
diff --git a/torch/csrc/jit/codegen/onednn/graph_fuser.cpp b/torch/csrc/jit/codegen/onednn/graph_fuser.cpp
index 1c68edca761..2c6c96e6ede 100644
--- a/torch/csrc/jit/codegen/onednn/graph_fuser.cpp
+++ b/torch/csrc/jit/codegen/onednn/graph_fuser.cpp
@@ -1,9 +1,7 @@
 #include <torch/csrc/jit/codegen/onednn/graph_fuser.h>
-#include <torch/csrc/jit/codegen/onednn/graph_helper.h>
 #include <torch/csrc/jit/ir/alias_analysis.h>
 #include <torch/csrc/jit/passes/common_subexpression_elimination.h>
 #include <torch/csrc/jit/passes/dead_code_elimination.h>
-#include <torch/csrc/jit/passes/utils/subgraph_utils.h>
 
 namespace torch::jit::fuser::onednn {
 
diff --git a/torch/csrc/jit/codegen/onednn/graph_helper.cpp b/torch/csrc/jit/codegen/onednn/graph_helper.cpp
index 2ef9f3cfa95..46e65cac23d 100644
--- a/torch/csrc/jit/codegen/onednn/graph_helper.cpp
+++ b/torch/csrc/jit/codegen/onednn/graph_helper.cpp
@@ -1,7 +1,5 @@
-#include <torch/csrc/jit/codegen/onednn/LlgaTensorImpl.h>
 #include <torch/csrc/jit/codegen/onednn/graph_helper.h>
 
-#include <ATen/core/functional.h>
 #include <torch/csrc/jit/jit_log.h>
 #include <torch/csrc/jit/passes/utils/subgraph_utils.h>
 
diff --git a/torch/csrc/jit/codegen/onednn/graph_rewriter.cpp b/torch/csrc/jit/codegen/onednn/graph_rewriter.cpp
index c8d7617fe86..6780fffac01 100644
--- a/torch/csrc/jit/codegen/onednn/graph_rewriter.cpp
+++ b/torch/csrc/jit/codegen/onednn/graph_rewriter.cpp
@@ -1,9 +1,6 @@
 #include <torch/csrc/jit/codegen/onednn/graph_fuser.h>
 #include <torch/csrc/jit/ir/alias_analysis.h>
 #include <torch/csrc/jit/jit_log.h>
-#include <torch/csrc/jit/passes/common_subexpression_elimination.h>
-#include <torch/csrc/jit/passes/dead_code_elimination.h>
-#include <torch/csrc/jit/passes/utils/subgraph_utils.h>
 
 namespace torch::jit::fuser::onednn {
 
diff --git a/torch/csrc/jit/codegen/onednn/guard_shape.cpp b/torch/csrc/jit/codegen/onednn/guard_shape.cpp
index a71f980d631..f7f1dc3776e 100644
--- a/torch/csrc/jit/codegen/onednn/guard_shape.cpp
+++ b/torch/csrc/jit/codegen/onednn/guard_shape.cpp
@@ -2,8 +2,6 @@
 
 #include <torch/csrc/jit/jit_log.h>
 #include <torch/csrc/jit/passes/tensorexpr_fuser.h>
-#include <torch/csrc/jit/passes/utils/subgraph_utils.h>
-#include <torch/csrc/jit/runtime/graph_executor.h>
 
 namespace torch::jit::fuser::onednn {
 
diff --git a/torch/csrc/jit/codegen/onednn/interface.cpp b/torch/csrc/jit/codegen/onednn/interface.cpp
index 2d29c8fa0f7..459fd9684c4 100644
--- a/torch/csrc/jit/codegen/onednn/interface.cpp
+++ b/torch/csrc/jit/codegen/onednn/interface.cpp
@@ -8,8 +8,6 @@
 #include <torch/csrc/jit/codegen/onednn/layout_propagation.h>
 #include <torch/csrc/jit/codegen/onednn/prepare_binary.h>
 #include <torch/csrc/jit/jit_log.h>
-#include <torch/csrc/jit/passes/decompose_ops.h>
-#include <torch/csrc/jit/passes/pass_manager.h>
 #include <torch/csrc/jit/passes/remove_mutation.h>
 #include <torch/csrc/jit/passes/tensorexpr_fuser.h>
 #include <torch/csrc/jit/runtime/custom_operator.h>
diff --git a/torch/csrc/jit/codegen/onednn/kernel.cpp b/torch/csrc/jit/codegen/onednn/kernel.cpp
index 85afc5fa8dc..2d6d4892184 100644
--- a/torch/csrc/jit/codegen/onednn/kernel.cpp
+++ b/torch/csrc/jit/codegen/onednn/kernel.cpp
@@ -1,7 +1,6 @@
 #include <torch/csrc/jit/codegen/onednn/graph_helper.h>
 #include <torch/csrc/jit/codegen/onednn/kernel.h>
 
-#include <ATen/core/functional.h>
 #include <torch/csrc/jit/jit_log.h>
 
 namespace torch::jit::fuser::onednn {
diff --git a/torch/csrc/jit/frontend/builtin_functions.cpp b/torch/csrc/jit/frontend/builtin_functions.cpp
index 2225f58e54e..38f142fb0ee 100644
--- a/torch/csrc/jit/frontend/builtin_functions.cpp
+++ b/torch/csrc/jit/frontend/builtin_functions.cpp
@@ -1,8 +1,6 @@
 #include <torch/csrc/jit/frontend/builtin_functions.h>
 
 #include <ATen/code_template.h>
-#include <caffe2/serialize/versions.h>
-#include <torch/csrc/api/include/torch/jit.h>
 #include <torch/csrc/jit/frontend/resolver.h>
 
 namespace torch::jit {
diff --git a/torch/csrc/jit/frontend/canonicalize_modified_loop.cpp b/torch/csrc/jit/frontend/canonicalize_modified_loop.cpp
index f2ef8b0e953..63369535a9e 100644
--- a/torch/csrc/jit/frontend/canonicalize_modified_loop.cpp
+++ b/torch/csrc/jit/frontend/canonicalize_modified_loop.cpp
@@ -1,8 +1,6 @@
-#include <functional>
 #include <memory>
 #include <string>
 
-#include <torch/csrc/Export.h>
 #include <torch/csrc/jit/frontend/canonicalize_modified_loop.h>
 #include <torch/csrc/jit/ir/ir.h>
 #include <torch/csrc/jit/ir/ir_views.h>
diff --git a/torch/csrc/jit/frontend/error_report.cpp b/torch/csrc/jit/frontend/error_report.cpp
index 47a9343c538..6942c1bfb59 100644
--- a/torch/csrc/jit/frontend/error_report.cpp
+++ b/torch/csrc/jit/frontend/error_report.cpp
@@ -1,7 +1,5 @@
 #include <torch/csrc/jit/frontend/error_report.h>
 
-#include <torch/csrc/jit/frontend/tree.h>
-
 namespace torch::jit {
 
 // Avoid storing objects with destructor in thread_local for mobile build.
diff --git a/torch/csrc/jit/frontend/inline_loop_condition.cpp b/torch/csrc/jit/frontend/inline_loop_condition.cpp
index da23769f402..6d3129c31a1 100644
--- a/torch/csrc/jit/frontend/inline_loop_condition.cpp
+++ b/torch/csrc/jit/frontend/inline_loop_condition.cpp
@@ -1,8 +1,6 @@
 #include <functional>
 #include <memory>
-#include <string>
 
-#include <torch/csrc/Export.h>
 #include <torch/csrc/jit/frontend/inline_loop_condition.h>
 #include <torch/csrc/jit/ir/ir.h>
 
diff --git a/torch/csrc/jit/frontend/ir_emitter.cpp b/torch/csrc/jit/frontend/ir_emitter.cpp
index fba613b5ea8..f1941215fcb 100644
--- a/torch/csrc/jit/frontend/ir_emitter.cpp
+++ b/torch/csrc/jit/frontend/ir_emitter.cpp
@@ -2,10 +2,8 @@
 #include <torch/csrc/jit/frontend/tree_views.h>
 
 #include <c10/util/Exception.h>
-#include <c10/util/StringUtil.h>
 #include <c10/util/env.h>
 #include <c10/util/irange.h>
-#include <caffe2/serialize/versions.h>
 #include <torch/csrc/jit/api/function_impl.h>
 #include <torch/csrc/jit/frontend/canonicalize_modified_loop.h>
 #include <torch/csrc/jit/frontend/convert_to_ssa.h>
@@ -18,7 +16,6 @@
 #include <torch/csrc/jit/passes/canonicalize.h>
 #include <torch/csrc/jit/passes/constant_pooling.h>
 #include <torch/csrc/jit/passes/constant_propagation.h>
-#include <torch/csrc/jit/passes/dead_code_elimination.h>
 #include <torch/csrc/jit/passes/inline_forked_closures.h>
 #include <torch/csrc/jit/passes/inliner.h>
 #include <torch/csrc/jit/passes/lift_closures.h>
@@ -29,7 +26,6 @@
 #include <torch/csrc/jit/runtime/interpreter.h>
 #include <torch/csrc/jit/runtime/operator.h>
 #include <torch/csrc/jit/runtime/slice_indices_adjust.h>
-#include <torch/csrc/jit/testing/hooks_for_testing.h>
 
 #include <torch/csrc/jit/ir/constants.h>
 
@@ -39,7 +35,6 @@
 #include <ATen/core/interned_strings.h>
 #include <ATen/core/jit_type.h>
 #include <torch/csrc/jit/frontend/error_report.h>
-#include <climits>
 #include <set>
 #include <stack>
 
diff --git a/torch/csrc/jit/frontend/lexer.cpp b/torch/csrc/jit/frontend/lexer.cpp
index 187721671e6..7fd0b66bba5 100644
--- a/torch/csrc/jit/frontend/lexer.cpp
+++ b/torch/csrc/jit/frontend/lexer.cpp
@@ -1,7 +1,5 @@
 #include <torch/csrc/jit/frontend/lexer.h>
 
-#include <c10/util/Exception.h>
-
 #include <cstring>
 #include <string>
 #include <unordered_map>
diff --git a/torch/csrc/jit/frontend/schema_matching.cpp b/torch/csrc/jit/frontend/schema_matching.cpp
index c3525ac9c8a..83742b40ae9 100644
--- a/torch/csrc/jit/frontend/schema_matching.cpp
+++ b/torch/csrc/jit/frontend/schema_matching.cpp
@@ -4,7 +4,6 @@
 #include <ATen/core/jit_type.h>
 #include <c10/util/Exception.h>
 #include <c10/util/irange.h>
-#include <caffe2/serialize/versions.h>
 #include <torch/csrc/jit/frontend/builtin_functions.h>
 #include <torch/csrc/jit/frontend/error_report.h>
 #include <torch/csrc/jit/frontend/function_schema_parser.h>
diff --git a/torch/csrc/jit/frontend/sugared_value.cpp b/torch/csrc/jit/frontend/sugared_value.cpp
index f9a80cf4da5..9ebf9a7e06d 100644
--- a/torch/csrc/jit/frontend/sugared_value.cpp
+++ b/torch/csrc/jit/frontend/sugared_value.cpp
@@ -2,9 +2,7 @@
 
 #include <c10/util/irange.h>
 #include <torch/csrc/jit/frontend/schema_matching.h>
-#include <torch/csrc/jit/frontend/tree_views.h>
 #include <torch/csrc/jit/ir/ir.h>
-#include <torch/csrc/jit/passes/constant_propagation.h>
 
 namespace torch::jit {
 
diff --git a/torch/csrc/jit/frontend/tracer.cpp b/torch/csrc/jit/frontend/tracer.cpp
index 3ccbd5257ae..0a0709fdf50 100644
--- a/torch/csrc/jit/frontend/tracer.cpp
+++ b/torch/csrc/jit/frontend/tracer.cpp
@@ -1,25 +1,17 @@
 #include <torch/csrc/jit/frontend/tracer.h>
 
-#include <ATen/Backtrace.h>
 #include <ATen/ScalarOps.h>
 #include <ATen/TracerMode.h>
 #include <ATen/core/Dict.h>
-#include <ATen/core/functional.h>
 #include <c10/util/Exception.h>
 #include <c10/util/irange.h>
-#include <torch/csrc/autograd/engine.h>
-#include <torch/csrc/autograd/function.h>
 #include <torch/csrc/autograd/variable.h>
 #include <torch/csrc/jit/api/module.h>
 #include <torch/csrc/jit/ir/constants.h>
 #include <torch/csrc/jit/ir/ir.h>
-#include <torch/csrc/jit/passes/dead_code_elimination.h>
 #include <torch/csrc/jit/passes/fixup_trace_scope_blocks.h>
 #include <torch/csrc/jit/passes/inliner.h>
-#include <torch/csrc/jit/passes/lower_tuples.h>
 #include <torch/csrc/jit/passes/normalize_ops.h>
-#include <torch/csrc/jit/passes/remove_expands.h>
-#include <torch/csrc/utils/variadic.h>
 #include <torch/custom_class.h>
 
 #include <memory>
diff --git a/torch/csrc/jit/frontend/versioned_symbols.cpp b/torch/csrc/jit/frontend/versioned_symbols.cpp
index 0a468d12d02..6808804ba5f 100644
--- a/torch/csrc/jit/frontend/versioned_symbols.cpp
+++ b/torch/csrc/jit/frontend/versioned_symbols.cpp
@@ -1,8 +1,5 @@
 #include <torch/csrc/jit/frontend/versioned_symbols.h>
 
-#include <caffe2/serialize/versions.h>
-#include <torch/csrc/api/include/torch/jit.h>
-
 #include <unordered_map>
 
 namespace torch::jit {
diff --git a/torch/csrc/jit/ir/alias_analysis.cpp b/torch/csrc/jit/ir/alias_analysis.cpp
index 513258236ac..51dbb09db9e 100644
--- a/torch/csrc/jit/ir/alias_analysis.cpp
+++ b/torch/csrc/jit/ir/alias_analysis.cpp
@@ -6,7 +6,6 @@
 #include <torch/csrc/jit/api/function_impl.h>
 #include <torch/csrc/jit/jit_log.h>
 #include <torch/csrc/jit/passes/inliner.h>
-#include <torch/csrc/jit/passes/utils/subgraph_utils.h>
 #include <torch/csrc/jit/runtime/operator.h>
 #include <fstream>
 #include <iostream>
diff --git a/torch/csrc/jit/ir/constants.cpp b/torch/csrc/jit/ir/constants.cpp
index e17c981a746..d3524f1ac10 100644
--- a/torch/csrc/jit/ir/constants.cpp
+++ b/torch/csrc/jit/ir/constants.cpp
@@ -1,11 +1,7 @@
-#include <ATen/core/functional.h>
 #include <c10/util/Exception.h>
-#include <torch/csrc/autograd/variable.h>
 #include <torch/csrc/jit/ir/constants.h>
 #include <torch/csrc/jit/ir/ir.h>
-#include <torch/csrc/jit/runtime/custom_operator.h>
 #include <torch/csrc/jit/runtime/operator.h>
-#include <torch/csrc/jit/runtime/register_ops_utils.h>
 
 namespace torch::jit {
 
diff --git a/torch/csrc/jit/ir/ir.cpp b/torch/csrc/jit/ir/ir.cpp
index 9b00a703e35..c5dfa56b48a 100644
--- a/torch/csrc/jit/ir/ir.cpp
+++ b/torch/csrc/jit/ir/ir.cpp
@@ -1,16 +1,13 @@
 #include <torch/csrc/jit/ir/ir.h>
 
-#include <ATen/core/builtin_function.h>
 #include <ATen/core/function.h>
 #include <c10/util/Exception.h>
-#include <c10/util/StringUtil.h>
 #include <c10/util/irange.h>
 #include <torch/csrc/jit/api/function_impl.h>
 #include <torch/csrc/jit/frontend/error_report.h>
 #include <torch/csrc/jit/frontend/schema_matching.h>
 #include <torch/csrc/jit/ir/constants.h>
 #include <torch/csrc/jit/runtime/operator.h>
-#include <torch/csrc/jit/serialization/python_print.h>
 
 #include <algorithm>
 #include <iostream>
diff --git a/torch/csrc/jit/ir/irparser.cpp b/torch/csrc/jit/ir/irparser.cpp
index 2fadc7d573e..0fbf660da3b 100644
--- a/torch/csrc/jit/ir/irparser.cpp
+++ b/torch/csrc/jit/ir/irparser.cpp
@@ -1,6 +1,5 @@
 #include <torch/csrc/jit/ir/irparser.h>
 
-#include <ATen/EmptyTensor.h>
 #include <torch/csrc/jit/frontend/lexer.h>
 #include <torch/csrc/jit/frontend/parse_string_literal.h>
 #include <torch/csrc/jit/frontend/schema_type_parser.h>
@@ -9,7 +8,6 @@
 #ifndef AT_PER_OPERATOR_HEADERS
 #include <ATen/Functions.h>
 #else
-#include <ATen/ops/empty.h>
 #include <ATen/ops/empty_strided.h>
 #endif
 
diff --git a/torch/csrc/jit/ir/node_hashing.cpp b/torch/csrc/jit/ir/node_hashing.cpp
index 1551e610c3d..5e1e3c5aab1 100644
--- a/torch/csrc/jit/ir/node_hashing.cpp
+++ b/torch/csrc/jit/ir/node_hashing.cpp
@@ -1,15 +1,12 @@
 #include <torch/csrc/jit/ir/ir.h>
 
 #include <algorithm>
-#include <unordered_map>
 
-#include <ATen/core/functional.h>
 #include <ATen/core/symbol.h>
 #include <c10/util/Exception.h>
 #include <c10/util/hash.h>
 #include <c10/util/irange.h>
 #include <torch/csrc/jit/ir/node_hashing.h>
-#include <torch/csrc/jit/passes/common_subexpression_elimination.h>
 
 namespace torch::jit {
 
diff --git a/torch/csrc/jit/ir/type_hashing.cpp b/torch/csrc/jit/ir/type_hashing.cpp
index 5d1c03cb493..2929f5aabd6 100644
--- a/torch/csrc/jit/ir/type_hashing.cpp
+++ b/torch/csrc/jit/ir/type_hashing.cpp
@@ -1,8 +1,5 @@
 #include <torch/csrc/jit/ir/type_hashing.h>
 
-#include <ATen/core/functional.h>
-#include <ATen/core/jit_type.h>
-#include <ATen/core/qualified_name.h>
 #include <c10/util/hash.h>
 #include <torch/csrc/jit/ir/ir.h>
 
diff --git a/torch/csrc/jit/jit_log.cpp b/torch/csrc/jit/jit_log.cpp
index 83f0e158d31..9ae31ab11d1 100644
--- a/torch/csrc/jit/jit_log.cpp
+++ b/torch/csrc/jit/jit_log.cpp
@@ -11,7 +11,6 @@
 #include <c10/util/StringUtil.h>
 #include <c10/util/env.h>
 #include <torch/csrc/jit/api/function_impl.h>
-#include <torch/csrc/jit/frontend/error_report.h>
 #include <torch/csrc/jit/ir/ir.h>
 #include <torch/csrc/jit/jit_log.h>
 #include <torch/csrc/jit/serialization/python_print.h>
diff --git a/torch/csrc/jit/jit_opt_limit.cpp b/torch/csrc/jit/jit_opt_limit.cpp
index c4c1a230765..d182a70fda4 100644
--- a/torch/csrc/jit/jit_opt_limit.cpp
+++ b/torch/csrc/jit/jit_opt_limit.cpp
@@ -1,13 +1,9 @@
-#include <cstdlib>
 #include <sstream>
 #include <string>
 #include <utility>
 
-#include <ATen/core/function.h>
-#include <c10/util/Exception.h>
 #include <c10/util/StringUtil.h>
 #include <c10/util/env.h>
-#include <torch/csrc/jit/api/function_impl.h>
 #include <torch/csrc/jit/jit_opt_limit.h>
 
 // NOTE: Don't try to migrate jit to C++17 yet
diff --git a/torch/csrc/jit/mobile/compatibility/backport.cpp b/torch/csrc/jit/mobile/compatibility/backport.cpp
index e8d13b19557..0b264a650da 100644
--- a/torch/csrc/jit/mobile/compatibility/backport.cpp
+++ b/torch/csrc/jit/mobile/compatibility/backport.cpp
@@ -1,5 +1,3 @@
-#include <ATen/core/ivalue.h>
-#include <caffe2/serialize/file_adapter.h>
 #include <caffe2/serialize/inline_container.h>
 #include <torch/csrc/jit/mobile/compatibility/backport.h>
 #include <torch/csrc/jit/mobile/compatibility/backport_manager.h>
diff --git a/torch/csrc/jit/mobile/compatibility/backport_manager.cpp b/torch/csrc/jit/mobile/compatibility/backport_manager.cpp
index 4422608423e..c84e05f8a3f 100644
--- a/torch/csrc/jit/mobile/compatibility/backport_manager.cpp
+++ b/torch/csrc/jit/mobile/compatibility/backport_manager.cpp
@@ -1,11 +1,9 @@
 #include <ATen/core/ivalue.h>
 #include <c10/util/Exception.h>
-#include <caffe2/serialize/file_adapter.h>
 #include <caffe2/serialize/inline_container.h>
 #include <torch/csrc/jit/mobile/compatibility/backport_manager.h>
 #include <torch/csrc/jit/mobile/compatibility/model_compatibility.h>
 #include <torch/csrc/jit/mobile/import.h>
-#include <torch/csrc/jit/mobile/module.h>
 #include <torch/csrc/jit/serialization/export.h>
 #include <torch/csrc/jit/serialization/import.h>
 #include <torch/csrc/jit/serialization/pickler.h>
diff --git a/torch/csrc/jit/mobile/function.cpp b/torch/csrc/jit/mobile/function.cpp
index 87128a180a6..3dc960040c8 100644
--- a/torch/csrc/jit/mobile/function.cpp
+++ b/torch/csrc/jit/mobile/function.cpp
@@ -2,9 +2,7 @@
 #include <torch/csrc/jit/mobile/function.h>
 #include <torch/csrc/jit/mobile/interpreter.h>
 #include <torch/csrc/jit/mobile/parse_bytecode.h>
-#include <torch/csrc/jit/mobile/parse_operators.h>
 #include <torch/csrc/jit/mobile/prim_ops_registery.h>
-#include <torch/csrc/jit/mobile/type_parser.h>
 #include <torch/csrc/jit/runtime/instruction.h>
 #include <torch/csrc/jit/runtime/operator.h>
 
diff --git a/torch/csrc/jit/mobile/import.cpp b/torch/csrc/jit/mobile/import.cpp
index ab05e48143e..16b1cf29b4e 100644
--- a/torch/csrc/jit/mobile/import.cpp
+++ b/torch/csrc/jit/mobile/import.cpp
@@ -17,12 +17,10 @@
 #include <torch/csrc/jit/mobile/observer.h>
 #include <torch/csrc/jit/mobile/type_parser.h>
 #include <torch/csrc/jit/mobile/upgrader_mobile.h>
-#include <torch/csrc/jit/runtime/instruction.h>
 #include <torch/csrc/jit/serialization/import_export_constants.h>
 #include <torch/csrc/jit/serialization/import_export_functions.h>
 #include <torch/csrc/jit/serialization/import_read.h>
 #include <torch/custom_class.h>
-#include <torch/library.h>
 #include <optional>
 #include <string>
 #include <vector>
diff --git a/torch/csrc/jit/mobile/interpreter.cpp b/torch/csrc/jit/mobile/interpreter.cpp
index 41fc8d49efb..a0e0959d603 100644
--- a/torch/csrc/jit/mobile/interpreter.cpp
+++ b/torch/csrc/jit/mobile/interpreter.cpp
@@ -4,7 +4,6 @@
 #include <ATen/core/dynamic_type.h>
 #include <ATen/core/function.h>
 #include <ATen/core/jit_type.h>
-#include <ATen/core/operator_name.h>
 #include <ATen/record_function.h>
 #include <c10/util/Exception.h>
 #include <c10/util/irange.h>
@@ -13,7 +12,6 @@
 #include <torch/csrc/jit/mobile/observer.h>
 #include <torch/csrc/jit/mobile/promoted_prim_ops.h>
 #include <torch/csrc/jit/runtime/instruction.h>
-#include <torch/csrc/jit/runtime/jit_exception.h>
 #include <torch/csrc/jit/runtime/vararg_functions.h>
 
 namespace torch::jit {
diff --git a/torch/csrc/jit/mobile/module.cpp b/torch/csrc/jit/mobile/module.cpp
index fb38d70d6f3..fbe262b2fbc 100644
--- a/torch/csrc/jit/mobile/module.cpp
+++ b/torch/csrc/jit/mobile/module.cpp
@@ -1,12 +1,9 @@
 #include <torch/csrc/jit/mobile/module.h>
 
 #include <torch/csrc/jit/backends/backend_exception.h>
-#include <torch/csrc/jit/mobile/interpreter.h>
 #include <torch/csrc/jit/mobile/observer.h>
 #include <torch/csrc/jit/mobile/type_parser.h>
-#include <torch/csrc/jit/runtime/jit_exception.h>
 
-#include <ATen/record_function.h>
 #include <c10/util/ScopeExit.h>
 #include <c10/util/irange.h>
 
diff --git a/torch/csrc/jit/mobile/parse_bytecode.cpp b/torch/csrc/jit/mobile/parse_bytecode.cpp
index 1a1e278e371..1cb16613962 100644
--- a/torch/csrc/jit/mobile/parse_bytecode.cpp
+++ b/torch/csrc/jit/mobile/parse_bytecode.cpp
@@ -6,7 +6,6 @@
 #include <torch/csrc/jit/runtime/instruction.h>
 #include <torch/csrc/jit/serialization/import_export_constants.h>
 #include <torch/csrc/jit/serialization/import_export_functions.h>
-#include <torch/custom_class_detail.h>
 
 namespace torch::jit {
 
diff --git a/torch/csrc/jit/mobile/register_ops_common_utils.cpp b/torch/csrc/jit/mobile/register_ops_common_utils.cpp
index 11e1481a8de..147bd7cbd56 100644
--- a/torch/csrc/jit/mobile/register_ops_common_utils.cpp
+++ b/torch/csrc/jit/mobile/register_ops_common_utils.cpp
@@ -1,5 +1,4 @@
 #include <ATen/core/dynamic_type.h>
-#include <ATen/core/type_factory.h>
 #include <torch/csrc/jit/mobile/register_ops_common_utils.h>
 
 namespace torch::jit {
diff --git a/torch/csrc/jit/mobile/train/export_data.cpp b/torch/csrc/jit/mobile/train/export_data.cpp
index 2d0a91096a0..867a2be6b3d 100644
--- a/torch/csrc/jit/mobile/train/export_data.cpp
+++ b/torch/csrc/jit/mobile/train/export_data.cpp
@@ -2,7 +2,6 @@
 
 #include <torch/csrc/jit/mobile/import_export_common.h>
 #include <torch/csrc/jit/mobile/module.h>
-#include <torch/csrc/jit/runtime/instruction.h>
 #include <torch/csrc/jit/serialization/flatbuffer_serializer.h>
 #include <torch/csrc/jit/serialization/pickler.h>
 #include <torch/csrc/jit/serialization/type_name_uniquer.h>
diff --git a/torch/csrc/jit/mobile/train/optim/sgd.cpp b/torch/csrc/jit/mobile/train/optim/sgd.cpp
index 1523c5629a9..1fedb07e3b4 100644
--- a/torch/csrc/jit/mobile/train/optim/sgd.cpp
+++ b/torch/csrc/jit/mobile/train/optim/sgd.cpp
@@ -1,10 +1,7 @@
 #include <torch/csrc/jit/mobile/train/optim/sgd.h>
 
-#include <torch/types.h>
 #include <torch/utils.h>
 
-#include <ATen/ATen.h>
-
 namespace torch::jit::mobile {
 
 bool SGDParamGroup::has_options() const {
diff --git a/torch/csrc/jit/mobile/train/sequential.cpp b/torch/csrc/jit/mobile/train/sequential.cpp
index 3b76db5e8d0..e249b2e340f 100644
--- a/torch/csrc/jit/mobile/train/sequential.cpp
+++ b/torch/csrc/jit/mobile/train/sequential.cpp
@@ -1,5 +1,4 @@
 #include <torch/csrc/jit/mobile/train/sequential.h>
-#include <torch/types.h>
 
 #include <algorithm>
 #include <cstddef>
diff --git a/torch/csrc/jit/mobile/upgrader_mobile.cpp b/torch/csrc/jit/mobile/upgrader_mobile.cpp
index 04bc12f1d10..c78fc439721 100644
--- a/torch/csrc/jit/mobile/upgrader_mobile.cpp
+++ b/torch/csrc/jit/mobile/upgrader_mobile.cpp
@@ -5,7 +5,6 @@
  * cd ~/pytorch && python torchgen/operator_versions/gen_mobile_upgraders.py
  */
 
-#include <caffe2/serialize/versions.h>
 #include <torch/csrc/jit/mobile/upgrader_mobile.h>
 
 namespace c10 {
diff --git a/torch/csrc/jit/operator_upgraders/utils.cpp b/torch/csrc/jit/operator_upgraders/utils.cpp
index 98819b08d64..fe110b5d570 100644
--- a/torch/csrc/jit/operator_upgraders/utils.cpp
+++ b/torch/csrc/jit/operator_upgraders/utils.cpp
@@ -2,9 +2,8 @@
 
 #include <caffe2/serialize/versions.h>
 #include <torch/csrc/jit/operator_upgraders/version_map.h>
-#include <iostream>
+#include <algorithm>
 #include <optional>
-#include <regex>
 #include <string>
 #include <vector>
 
diff --git a/torch/csrc/jit/passes/autocast.cpp b/torch/csrc/jit/passes/autocast.cpp
index 4699cceec5b..79ead2c5ee6 100644
--- a/torch/csrc/jit/passes/autocast.cpp
+++ b/torch/csrc/jit/passes/autocast.cpp
@@ -2,7 +2,6 @@
 #include <torch/csrc/jit/passes/autocast.h>
 
 #include <ATen/autocast_mode.h>
-#include <c10/core/ScalarType.h>
 #include <c10/util/Exception.h>
 #include <torch/csrc/jit/ir/ir.h>
 #include <torch/csrc/jit/jit_log.h>
diff --git a/torch/csrc/jit/passes/bailout_graph.cpp b/torch/csrc/jit/passes/bailout_graph.cpp
index 5bea5e42c0d..4fb339d0d53 100644
--- a/torch/csrc/jit/passes/bailout_graph.cpp
+++ b/torch/csrc/jit/passes/bailout_graph.cpp
@@ -2,14 +2,11 @@
 
 #include <ATen/core/function.h>
 #include <c10/util/irange.h>
-#include <torch/csrc/jit/ir/alias_analysis.h>
 #include <torch/csrc/jit/ir/ir_views.h>
 #include <torch/csrc/jit/jit_log.h>
 #include <torch/csrc/jit/passes/clear_profiling.h>
-#include <torch/csrc/jit/passes/constant_pooling.h>
 #include <torch/csrc/jit/passes/liveness.h>
 #include <memory>
-#include <unordered_set>
 #include <utility>
 
 namespace torch::jit {
diff --git a/torch/csrc/jit/passes/check_strict_fusion.cpp b/torch/csrc/jit/passes/check_strict_fusion.cpp
index 731382c3163..2a5ed995c10 100644
--- a/torch/csrc/jit/passes/check_strict_fusion.cpp
+++ b/torch/csrc/jit/passes/check_strict_fusion.cpp
@@ -1,7 +1,6 @@
 
 #include <torch/csrc/jit/passes/check_strict_fusion.h>
 
-#include <c10/util/Exception.h>
 #include <torch/csrc/jit/frontend/error_report.h>
 #include <torch/csrc/jit/ir/ir.h>
 #include <torch/csrc/jit/jit_log.h>
diff --git a/torch/csrc/jit/passes/common_subexpression_elimination.cpp b/torch/csrc/jit/passes/common_subexpression_elimination.cpp
index cfa0ee49788..e5d214762e2 100644
--- a/torch/csrc/jit/passes/common_subexpression_elimination.cpp
+++ b/torch/csrc/jit/passes/common_subexpression_elimination.cpp
@@ -5,8 +5,6 @@
 #include <torch/csrc/jit/ir/node_hashing.h>
 #include <torch/csrc/jit/jit_log.h>
 
-#include <unordered_map>
-
 namespace torch::jit {
 namespace {
 
diff --git a/torch/csrc/jit/passes/concat_opt.cpp b/torch/csrc/jit/passes/concat_opt.cpp
index a651458eb5e..b21a65ea98d 100644
--- a/torch/csrc/jit/passes/concat_opt.cpp
+++ b/torch/csrc/jit/passes/concat_opt.cpp
@@ -11,9 +11,7 @@
 #include <torch/csrc/jit/ir/ir.h>
 #include <torch/csrc/jit/ir/named_value.h>
 #include <torch/csrc/jit/jit_log.h>
-#include <torch/csrc/jit/passes/constant_pooling.h>
 #include <torch/csrc/jit/passes/dead_code_elimination.h>
-#include <torch/csrc/jit/passes/remove_mutation.h>
 #include <torch/csrc/jit/runtime/graph_iterator.h>
 
 namespace torch::jit {
diff --git a/torch/csrc/jit/passes/constant_propagation.cpp b/torch/csrc/jit/passes/constant_propagation.cpp
index 97d4c6262ed..5e95f1eae39 100644
--- a/torch/csrc/jit/passes/constant_propagation.cpp
+++ b/torch/csrc/jit/passes/constant_propagation.cpp
@@ -1,10 +1,8 @@
 #include <torch/csrc/jit/passes/constant_propagation.h>
 
-#include <ATen/core/functional.h>
 #include <ATen/core/ivalue.h>
 #include <c10/util/Exception.h>
 #include <c10/util/irange.h>
-#include <torch/csrc/autograd/variable.h>
 #include <torch/csrc/jit/ir/alias_analysis.h>
 #include <torch/csrc/jit/ir/constants.h>
 #include <torch/csrc/jit/ir/ir.h>
diff --git a/torch/csrc/jit/passes/create_autodiff_subgraphs.cpp b/torch/csrc/jit/passes/create_autodiff_subgraphs.cpp
index cac257125b0..d0c836df9ff 100644
--- a/torch/csrc/jit/passes/create_autodiff_subgraphs.cpp
+++ b/torch/csrc/jit/passes/create_autodiff_subgraphs.cpp
@@ -4,9 +4,7 @@
 #include <torch/csrc/jit/ir/alias_analysis.h>
 #include <torch/csrc/jit/ir/ir.h>
 #include <torch/csrc/jit/jit_log.h>
-#include <torch/csrc/jit/passes/canonicalize.h>
 #include <torch/csrc/jit/passes/common_subexpression_elimination.h>
-#include <torch/csrc/jit/passes/remove_redundant_profiles.h>
 #include <torch/csrc/jit/passes/utils/subgraph_utils.h>
 #include <torch/csrc/jit/runtime/autodiff.h>
 
diff --git a/torch/csrc/jit/passes/create_functional_graphs.cpp b/torch/csrc/jit/passes/create_functional_graphs.cpp
index 86e9fa13893..562659788d7 100644
--- a/torch/csrc/jit/passes/create_functional_graphs.cpp
+++ b/torch/csrc/jit/passes/create_functional_graphs.cpp
@@ -6,7 +6,6 @@
 #include <torch/csrc/jit/passes/utils/subgraph_utils.h>
 
 #include <cstddef>
-#include <limits>
 
 namespace torch::jit {
 
diff --git a/torch/csrc/jit/passes/dbr_quantization/remove_redundant_aliases.cpp b/torch/csrc/jit/passes/dbr_quantization/remove_redundant_aliases.cpp
index 1d35b30c050..6fcf40a4ded 100644
--- a/torch/csrc/jit/passes/dbr_quantization/remove_redundant_aliases.cpp
+++ b/torch/csrc/jit/passes/dbr_quantization/remove_redundant_aliases.cpp
@@ -2,7 +2,6 @@
 
 #include <torch/csrc/jit/ir/alias_analysis.h>
 #include <torch/csrc/jit/jit_log.h>
-#include <torch/csrc/jit/passes/quantization/helper.h>
 #include <torch/csrc/jit/runtime/graph_iterator.h>
 
 namespace torch::jit {
diff --git a/torch/csrc/jit/passes/dtype_analysis.cpp b/torch/csrc/jit/passes/dtype_analysis.cpp
index 9cbe6a93623..64f3165dbce 100644
--- a/torch/csrc/jit/passes/dtype_analysis.cpp
+++ b/torch/csrc/jit/passes/dtype_analysis.cpp
@@ -1,14 +1,9 @@
-#include <ATen/core/function_schema.h>
 #include <ATen/core/jit_type.h>
-#include <ATen/core/symbol.h>
-#include <c10/core/ScalarType.h>
 #include <c10/util/ArrayRef.h>
-#include <torch/csrc/jit/ir/alias_analysis.h>
 #include <torch/csrc/jit/ir/ir.h>
 #include <torch/csrc/jit/jit_log.h>
 #include <torch/csrc/jit/passes/dtype_analysis.h>
 #include <torch/csrc/jit/passes/utils/op_registry.h>
-#include <torch/library.h>
 #include <optional>
 
 #ifndef AT_PER_OPERATOR_HEADERS
diff --git a/torch/csrc/jit/passes/erase_number_types.cpp b/torch/csrc/jit/passes/erase_number_types.cpp
index 03b370576d5..fd5fbdbaf3d 100644
--- a/torch/csrc/jit/passes/erase_number_types.cpp
+++ b/torch/csrc/jit/passes/erase_number_types.cpp
@@ -2,7 +2,6 @@
 
 #include <torch/csrc/jit/ir/constants.h>
 #include <torch/csrc/jit/jit_log.h>
-#include <torch/csrc/jit/passes/dead_code_elimination.h>
 
 #include <ATen/ScalarOps.h>
 
diff --git a/torch/csrc/jit/passes/fixup_trace_scope_blocks.cpp b/torch/csrc/jit/passes/fixup_trace_scope_blocks.cpp
index 1bfa045d2d3..5320f88e12c 100644
--- a/torch/csrc/jit/passes/fixup_trace_scope_blocks.cpp
+++ b/torch/csrc/jit/passes/fixup_trace_scope_blocks.cpp
@@ -2,13 +2,10 @@
 
 #include <c10/util/irange.h>
 #include <torch/csrc/jit/frontend/schema_matching.h>
-#include <torch/csrc/jit/passes/canonicalize.h>
 #include <torch/csrc/jit/passes/dead_code_elimination.h>
 #include <torch/csrc/jit/passes/inliner.h>
 #include <torch/csrc/jit/passes/lower_tuples.h>
 
-#include <algorithm>
-
 namespace torch::jit {
 
 namespace {
diff --git a/torch/csrc/jit/passes/fold_conv_bn.cpp b/torch/csrc/jit/passes/fold_conv_bn.cpp
index c2cb24f3287..b1c6c52e99e 100644
--- a/torch/csrc/jit/passes/fold_conv_bn.cpp
+++ b/torch/csrc/jit/passes/fold_conv_bn.cpp
@@ -10,7 +10,6 @@
 #ifndef AT_PER_OPERATOR_HEADERS
 #include <ATen/Functions.h>
 #else
-#include <ATen/ops/empty_like.h>
 #include <ATen/ops/ones_like.h>
 #include <ATen/ops/rsqrt.h>
 #include <ATen/ops/zeros_like.h>
diff --git a/torch/csrc/jit/passes/frozen_concat_linear.cpp b/torch/csrc/jit/passes/frozen_concat_linear.cpp
index e2270aa8bd7..fc864a63469 100644
--- a/torch/csrc/jit/passes/frozen_concat_linear.cpp
+++ b/torch/csrc/jit/passes/frozen_concat_linear.cpp
@@ -1,14 +1,8 @@
-#include <c10/util/irange.h>
 #include <torch/csrc/jit/ir/alias_analysis.h>
 #include <torch/csrc/jit/ir/ir.h>
-#include <torch/csrc/jit/ir/ir_views.h>
 #include <torch/csrc/jit/jit_log.h>
 #include <torch/csrc/jit/passes/frozen_concat_linear.h>
-#include <torch/csrc/jit/passes/frozen_conv_folding.h>
-#include <torch/csrc/jit/passes/frozen_graph_optimizations.h>
-#include <torch/csrc/jit/passes/remove_dropout.h>
 #include <torch/csrc/jit/passes/utils/optimization_utils.h>
-#include <torch/csrc/jit/runtime/graph_executor.h>
 
 #ifndef AT_PER_OPERATOR_HEADERS
 #include <ATen/Functions.h>
diff --git a/torch/csrc/jit/passes/frozen_conv_add_relu_fusion.cpp b/torch/csrc/jit/passes/frozen_conv_add_relu_fusion.cpp
index 20edcdd9618..3434b35760c 100644
--- a/torch/csrc/jit/passes/frozen_conv_add_relu_fusion.cpp
+++ b/torch/csrc/jit/passes/frozen_conv_add_relu_fusion.cpp
@@ -1,14 +1,8 @@
-#include <ATen/Utils.h>
 
 #include <torch/csrc/jit/ir/constants.h>
 #include <torch/csrc/jit/ir/ir.h>
-#include <torch/csrc/jit/ir/subgraph_matcher.h>
 #include <torch/csrc/jit/passes/frozen_conv_add_relu_fusion.h>
-#include <torch/csrc/jit/passes/graph_rewrite_helper.h>
-#include <torch/csrc/jit/passes/remove_mutation.h>
-#include <torch/csrc/jit/passes/subgraph_rewrite.h>
 #ifdef USE_CUDA
-#include <ATen/cuda/CUDAConfig.h>
 #endif
 
 namespace torch::jit {
diff --git a/torch/csrc/jit/passes/frozen_conv_add_relu_fusion_cuda.cpp b/torch/csrc/jit/passes/frozen_conv_add_relu_fusion_cuda.cpp
index af0c0a6a788..be03a2750e6 100644
--- a/torch/csrc/jit/passes/frozen_conv_add_relu_fusion_cuda.cpp
+++ b/torch/csrc/jit/passes/frozen_conv_add_relu_fusion_cuda.cpp
@@ -1,4 +1,3 @@
-#include <ATen/Utils.h>
 
 #include <ATen/code_template.h>
 #include <ATen/cuda/CUDAConfig.h>
@@ -8,7 +7,6 @@
 #include <torch/csrc/jit/jit_log.h>
 #include <torch/csrc/jit/passes/frozen_conv_add_relu_fusion.h>
 #include <torch/csrc/jit/passes/graph_rewrite_helper.h>
-#include <torch/csrc/jit/passes/remove_mutation.h>
 #include <torch/csrc/jit/passes/subgraph_rewrite.h>
 
 namespace torch::jit {
diff --git a/torch/csrc/jit/passes/frozen_conv_folding.cpp b/torch/csrc/jit/passes/frozen_conv_folding.cpp
index 6bc75bfcc8c..e210f09ef32 100644
--- a/torch/csrc/jit/passes/frozen_conv_folding.cpp
+++ b/torch/csrc/jit/passes/frozen_conv_folding.cpp
@@ -1,4 +1,3 @@
-#include <ATen/Utils.h>
 #include <c10/core/ScalarType.h>
 #include <c10/util/Exception.h>
 #include <c10/util/accumulate.h>
@@ -11,7 +10,6 @@
 #include <torch/csrc/jit/passes/fold_conv_bn.h>
 #include <torch/csrc/jit/passes/frozen_conv_folding.h>
 #include <torch/csrc/jit/passes/utils/optimization_utils.h>
-#include <torch/csrc/jit/tensorexpr/types.h>
 
 #ifndef AT_PER_OPERATOR_HEADERS
 #include <ATen/Functions.h>
diff --git a/torch/csrc/jit/passes/frozen_graph_optimizations.cpp b/torch/csrc/jit/passes/frozen_graph_optimizations.cpp
index e76575a2370..f086906a19f 100644
--- a/torch/csrc/jit/passes/frozen_graph_optimizations.cpp
+++ b/torch/csrc/jit/passes/frozen_graph_optimizations.cpp
@@ -1,12 +1,8 @@
-#include <c10/util/irange.h>
-#include <torch/csrc/jit/ir/alias_analysis.h>
-#include <torch/csrc/jit/ir/ir_views.h>
 #include <torch/csrc/jit/passes/frozen_concat_linear.h>
 #include <torch/csrc/jit/passes/frozen_conv_folding.h>
 #include <torch/csrc/jit/passes/frozen_graph_optimizations.h>
 #include <torch/csrc/jit/passes/frozen_linear_folding.h>
 #include <torch/csrc/jit/passes/remove_dropout.h>
-#include <torch/csrc/jit/runtime/graph_executor.h>
 
 namespace torch::jit {
 
diff --git a/torch/csrc/jit/passes/frozen_linear_transpose.cpp b/torch/csrc/jit/passes/frozen_linear_transpose.cpp
index 9595227d258..ccd97942f2c 100644
--- a/torch/csrc/jit/passes/frozen_linear_transpose.cpp
+++ b/torch/csrc/jit/passes/frozen_linear_transpose.cpp
@@ -1,9 +1,7 @@
 #include <torch/csrc/jit/ir/ir.h>
-#include <torch/csrc/jit/ir/ir_views.h>
 #include <torch/csrc/jit/jit_log.h>
 #include <torch/csrc/jit/passes/frozen_linear_transpose.h>
 #include <torch/csrc/jit/passes/utils/optimization_utils.h>
-#include <torch/csrc/jit/runtime/graph_executor.h>
 #include <torch/csrc/jit/runtime/graph_iterator.h>
 
 #ifndef AT_PER_OPERATOR_HEADERS
@@ -12,7 +10,6 @@
 #include <ATen/ops/transpose.h>
 #endif
 
-#include <iostream>
 #include <utility>
 
 namespace torch::jit {
diff --git a/torch/csrc/jit/passes/fuse_relu.cpp b/torch/csrc/jit/passes/fuse_relu.cpp
index 1a8ee88b3da..953dc8fe2c3 100644
--- a/torch/csrc/jit/passes/fuse_relu.cpp
+++ b/torch/csrc/jit/passes/fuse_relu.cpp
@@ -1,7 +1,6 @@
 #include <torch/csrc/jit/passes/fuse_relu.h>
 
 #include <torch/csrc/jit/ir/ir.h>
-#include <torch/csrc/jit/ir/subgraph_matcher.h>
 #include <torch/csrc/jit/passes/subgraph_rewrite.h>
 
 namespace torch::jit {
diff --git a/torch/csrc/jit/passes/graph_fuser.cpp b/torch/csrc/jit/passes/graph_fuser.cpp
index 8dfa836f87b..03c418260e2 100644
--- a/torch/csrc/jit/passes/graph_fuser.cpp
+++ b/torch/csrc/jit/passes/graph_fuser.cpp
@@ -3,18 +3,14 @@
 #include <c10/util/Exception.h>
 #include <c10/util/irange.h>
 #include <torch/csrc/jit/codegen/fuser/interface.h>
-#include <torch/csrc/jit/frontend/ir_emitter.h>
 #include <torch/csrc/jit/ir/alias_analysis.h>
 #include <torch/csrc/jit/passes/common_subexpression_elimination.h>
 #include <torch/csrc/jit/passes/constant_pooling.h>
 #include <torch/csrc/jit/passes/dead_code_elimination.h>
 #include <torch/csrc/jit/passes/tensorexpr_fuser.h>
-#include <torch/csrc/jit/passes/utils/subgraph_utils.h>
 #include <torch/csrc/jit/runtime/autodiff.h>
-#include <torch/csrc/jit/runtime/custom_operator.h>
 #include <torch/csrc/jit/runtime/operator.h>
 
-#include <queue>
 #include <unordered_map>
 #include <utility>
 
diff --git a/torch/csrc/jit/passes/guard_elimination.cpp b/torch/csrc/jit/passes/guard_elimination.cpp
index 7b0fed5dc15..5f76a0ce0cf 100644
--- a/torch/csrc/jit/passes/guard_elimination.cpp
+++ b/torch/csrc/jit/passes/guard_elimination.cpp
@@ -2,9 +2,6 @@
 
 #include <torch/csrc/jit/ir/alias_analysis.h>
 #include <torch/csrc/jit/jit_log.h>
-#include <torch/csrc/jit/passes/constant_propagation.h>
-#include <torch/csrc/jit/passes/peephole.h>
-#include <torch/csrc/jit/runtime/graph_executor.h>
 #include <memory>
 #include <unordered_set>
 
diff --git a/torch/csrc/jit/passes/hoist_conv_packed_params.cpp b/torch/csrc/jit/passes/hoist_conv_packed_params.cpp
index 5ef4e5d576c..1222b7cb39b 100644
--- a/torch/csrc/jit/passes/hoist_conv_packed_params.cpp
+++ b/torch/csrc/jit/passes/hoist_conv_packed_params.cpp
@@ -2,8 +2,6 @@
 
 #include <torch/csrc/jit/api/module.h>
 #include <torch/csrc/jit/jit_log.h>
-#include <torch/csrc/jit/passes/constant_pooling.h>
-#include <torch/csrc/jit/passes/constant_propagation.h>
 #include <torch/csrc/jit/passes/hoist_conv_packed_params.h>
 #include <torch/csrc/jit/passes/quantization/helper.h>
 
diff --git a/torch/csrc/jit/passes/inliner.cpp b/torch/csrc/jit/passes/inliner.cpp
index 1ddbb02f927..9c06f748e43 100644
--- a/torch/csrc/jit/passes/inliner.cpp
+++ b/torch/csrc/jit/passes/inliner.cpp
@@ -2,8 +2,6 @@
 
 #include <ATen/core/interned_strings.h>
 #include <torch/csrc/jit/api/function_impl.h>
-#include <torch/csrc/jit/api/module.h>
-#include <torch/csrc/jit/frontend/error_report.h>
 #include <torch/csrc/jit/jit_log.h>
 
 namespace torch::jit {
diff --git a/torch/csrc/jit/passes/insert_guards.cpp b/torch/csrc/jit/passes/insert_guards.cpp
index 2bb810199e8..602a5086e73 100644
--- a/torch/csrc/jit/passes/insert_guards.cpp
+++ b/torch/csrc/jit/passes/insert_guards.cpp
@@ -1,7 +1,6 @@
 #include <torch/csrc/jit/passes/insert_guards.h>
 #include <torch/csrc/jit/runtime/profiling_record.h>
 #include <memory>
-#include <unordered_set>
 
 namespace torch::jit {
 
diff --git a/torch/csrc/jit/passes/integer_value_refinement.cpp b/torch/csrc/jit/passes/integer_value_refinement.cpp
index 7405608bb4c..c760c5cb137 100644
--- a/torch/csrc/jit/passes/integer_value_refinement.cpp
+++ b/torch/csrc/jit/passes/integer_value_refinement.cpp
@@ -1,4 +1,3 @@
-#include <ATen/core/jit_type.h>
 #include <torch/csrc/jit/ir/ir.h>
 #include <torch/csrc/jit/jit_log.h>
 #include <torch/csrc/jit/passes/integer_value_refinement.h>
diff --git a/torch/csrc/jit/passes/liveness.cpp b/torch/csrc/jit/passes/liveness.cpp
index 138c6fc78f7..5fc13b44f17 100644
--- a/torch/csrc/jit/passes/liveness.cpp
+++ b/torch/csrc/jit/passes/liveness.cpp
@@ -1,8 +1,6 @@
 #include <torch/csrc/jit/passes/liveness.h>
 
-#include <torch/csrc/jit/ir/alias_analysis.h>
 #include <torch/csrc/jit/ir/ir_views.h>
-#include <torch/csrc/jit/passes/constant_pooling.h>
 #include <iostream>
 #include <memory>
 
diff --git a/torch/csrc/jit/passes/lower_tuples.cpp b/torch/csrc/jit/passes/lower_tuples.cpp
index ff8c1642f62..cfeb04f5f19 100644
--- a/torch/csrc/jit/passes/lower_tuples.cpp
+++ b/torch/csrc/jit/passes/lower_tuples.cpp
@@ -1,6 +1,5 @@
 #include <torch/csrc/jit/passes/lower_tuples.h>
 
-#include <ATen/core/functional.h>
 #include <c10/util/Exception.h>
 #include <c10/util/irange.h>
 #include <torch/csrc/jit/ir/constants.h>
diff --git a/torch/csrc/jit/passes/metal_rewrite.cpp b/torch/csrc/jit/passes/metal_rewrite.cpp
index 630701cab6d..82400a1cdcb 100644
--- a/torch/csrc/jit/passes/metal_rewrite.cpp
+++ b/torch/csrc/jit/passes/metal_rewrite.cpp
@@ -1,9 +1,5 @@
-#include <ATen/core/jit_type.h>
-#include <c10/util/irange.h>
 
 #include <torch/csrc/jit/ir/ir.h>
-#include <torch/csrc/jit/ir/subgraph_matcher.h>
-#include <torch/csrc/jit/passes/constant_pooling.h>
 #include <torch/csrc/jit/passes/fold_conv_bn.h>
 #include <torch/csrc/jit/passes/freeze_module.h>
 #include <torch/csrc/jit/passes/fuse_linear.h>
diff --git a/torch/csrc/jit/passes/mkldnn_rewrite.cpp b/torch/csrc/jit/passes/mkldnn_rewrite.cpp
index 769d96eec21..934f44a9ccf 100644
--- a/torch/csrc/jit/passes/mkldnn_rewrite.cpp
+++ b/torch/csrc/jit/passes/mkldnn_rewrite.cpp
@@ -4,7 +4,6 @@
 #include <torch/csrc/jit/jit_log.h>
 #include <torch/csrc/jit/passes/constant_propagation.h>
 #include <torch/csrc/jit/passes/dead_code_elimination.h>
-#include <torch/csrc/jit/passes/graph_rewrite_helper.h>
 #include <torch/csrc/jit/passes/mkldnn_rewrite.h>
 #include <torch/csrc/jit/tensorexpr/kernel.h>
 
diff --git a/torch/csrc/jit/passes/normalize_ops.cpp b/torch/csrc/jit/passes/normalize_ops.cpp
index 1c0a453c28c..4ce0afa2d30 100644
--- a/torch/csrc/jit/passes/normalize_ops.cpp
+++ b/torch/csrc/jit/passes/normalize_ops.cpp
@@ -1,7 +1,5 @@
 #include <torch/csrc/jit/passes/normalize_ops.h>
 
-#include <c10/util/Exception.h>
-
 namespace torch::jit {
 
 namespace {
diff --git a/torch/csrc/jit/passes/onnx.cpp b/torch/csrc/jit/passes/onnx.cpp
index d3231222cb9..720688ccc76 100644
--- a/torch/csrc/jit/passes/onnx.cpp
+++ b/torch/csrc/jit/passes/onnx.cpp
@@ -1,9 +1,7 @@
 #include <torch/csrc/jit/passes/onnx.h>
 
-#include <ATen/core/functional.h>
 #include <c10/util/Exception.h>
 #include <c10/util/irange.h>
-#include <torch/csrc/autograd/function.h>
 #include <torch/csrc/autograd/symbolic.h>
 #include <torch/csrc/jit/ir/constants.h>
 #include <torch/csrc/jit/jit_log.h>
@@ -13,7 +11,6 @@
 #include <torch/csrc/jit/passes/onnx/onnx_log.h>
 #include <torch/csrc/jit/passes/onnx/shape_type_inference.h>
 #include <torch/csrc/jit/python/python_ir.h>
-#include <torch/csrc/utils/pybind.h>
 #include <sstream>
 
 namespace torch::jit {
diff --git a/torch/csrc/jit/passes/onnx/constant_map.cpp b/torch/csrc/jit/passes/onnx/constant_map.cpp
index 902dc5f8924..60699a1e75e 100644
--- a/torch/csrc/jit/passes/onnx/constant_map.cpp
+++ b/torch/csrc/jit/passes/onnx/constant_map.cpp
@@ -1,7 +1,5 @@
 #include <c10/util/irange.h>
-#include <torch/csrc/jit/jit_log.h>
 #include <torch/csrc/jit/passes/onnx/constant_map.h>
-#include <torch/csrc/jit/passes/onnx/helper.h>
 #include <iostream>
 #include <sstream>
 #include <string>
diff --git a/torch/csrc/jit/passes/onnx/eval_peephole.cpp b/torch/csrc/jit/passes/onnx/eval_peephole.cpp
index 0334d5706a6..72fd0cb9690 100644
--- a/torch/csrc/jit/passes/onnx/eval_peephole.cpp
+++ b/torch/csrc/jit/passes/onnx/eval_peephole.cpp
@@ -1,10 +1,8 @@
 #include <torch/csrc/jit/jit_log.h>
 #include <torch/csrc/jit/passes/onnx/eval_peephole.h>
 #include <torch/csrc/jit/passes/onnx/helper.h>
-#include <torch/torch.h>
 
 #include <c10/util/irange.h>
-#include <algorithm>
 
 namespace torch::jit {
 
diff --git a/torch/csrc/jit/passes/onnx/fixup_onnx_controlflow.cpp b/torch/csrc/jit/passes/onnx/fixup_onnx_controlflow.cpp
index 2687ee9fb07..2f18a6d8c99 100644
--- a/torch/csrc/jit/passes/onnx/fixup_onnx_controlflow.cpp
+++ b/torch/csrc/jit/passes/onnx/fixup_onnx_controlflow.cpp
@@ -3,9 +3,7 @@
 #include <ATen/InitialTensorOptions.h>
 #include <c10/util/irange.h>
 #include <torch/csrc/jit/jit_log.h>
-#include <torch/csrc/jit/passes/dead_code_elimination.h>
 #include <torch/csrc/jit/passes/onnx/helper.h>
-#include <torch/csrc/jit/passes/onnx/peephole.h>
 #include <torch/csrc/jit/passes/onnx/shape_type_inference.h>
 
 namespace torch::jit {
diff --git a/torch/csrc/jit/passes/onnx/helper.cpp b/torch/csrc/jit/passes/onnx/helper.cpp
index 8eab378c892..3897a8d5cae 100644
--- a/torch/csrc/jit/passes/onnx/helper.cpp
+++ b/torch/csrc/jit/passes/onnx/helper.cpp
@@ -10,8 +10,6 @@
 #include <ATen/ops/unsqueeze.h>
 #endif
 
-#include <onnx/onnx_pb.h>
-
 namespace torch::jit {
 namespace onnx {
 using namespace ::c10::onnx;
diff --git a/torch/csrc/jit/passes/onnx/list_model_parameters.cpp b/torch/csrc/jit/passes/onnx/list_model_parameters.cpp
index f9aa740c44a..491106f0cb2 100644
--- a/torch/csrc/jit/passes/onnx/list_model_parameters.cpp
+++ b/torch/csrc/jit/passes/onnx/list_model_parameters.cpp
@@ -1,7 +1,5 @@
 #include <torch/csrc/jit/frontend/error_report.h>
 #include <torch/csrc/jit/jit_log.h>
-#include <torch/csrc/jit/passes/dead_code_elimination.h>
-#include <torch/csrc/jit/passes/onnx/helper.h>
 #include <torch/csrc/jit/passes/onnx/list_model_parameters.h>
 
 namespace torch::jit {
diff --git a/torch/csrc/jit/passes/onnx/pattern_conversion/autograd_function_process.cpp b/torch/csrc/jit/passes/onnx/pattern_conversion/autograd_function_process.cpp
index 3283b82eb46..e7f228f29b2 100644
--- a/torch/csrc/jit/passes/onnx/pattern_conversion/autograd_function_process.cpp
+++ b/torch/csrc/jit/passes/onnx/pattern_conversion/autograd_function_process.cpp
@@ -1,8 +1,6 @@
 #include <torch/csrc/jit/passes/onnx/pattern_conversion/autograd_function_process.h>
 
 #include <torch/csrc/jit/jit_log.h>
-#include <torch/csrc/jit/passes/onnx/helper.h>
-#include <torch/csrc/jit/passes/utils/subgraph_utils.h>
 
 namespace torch::jit {
 
diff --git a/torch/csrc/jit/passes/onnx/pattern_conversion/pattern_encapsulation.cpp b/torch/csrc/jit/passes/onnx/pattern_conversion/pattern_encapsulation.cpp
index a51801ac836..186a25873ef 100644
--- a/torch/csrc/jit/passes/onnx/pattern_conversion/pattern_encapsulation.cpp
+++ b/torch/csrc/jit/passes/onnx/pattern_conversion/pattern_encapsulation.cpp
@@ -1,8 +1,5 @@
-#include <torch/csrc/jit/passes/dead_code_elimination.h>
-#include <torch/csrc/jit/passes/onnx.h>
 #include <torch/csrc/jit/passes/onnx/pattern_conversion/common.h>
 #include <torch/csrc/jit/passes/onnx/pattern_conversion/pattern_encapsulation.h>
-#include <torch/csrc/jit/passes/onnx/remove_inplace_ops_for_onnx.h>
 
 // EDITING THIS FILE? READ THIS FIRST!
 // see Note [Edit Pattern Encapsulation] in pattern_encapsulation.h
diff --git a/torch/csrc/jit/passes/onnx/preprocess_for_onnx.cpp b/torch/csrc/jit/passes/onnx/preprocess_for_onnx.cpp
index 5f35a85b2aa..a00e98708e2 100644
--- a/torch/csrc/jit/passes/onnx/preprocess_for_onnx.cpp
+++ b/torch/csrc/jit/passes/onnx/preprocess_for_onnx.cpp
@@ -4,7 +4,6 @@
 #include <c10/util/irange.h>
 
 #include <torch/csrc/jit/jit_log.h>
-#include <torch/csrc/jit/passes/onnx/helper.h>
 
 namespace torch::jit {
 
diff --git a/torch/csrc/jit/passes/peephole.cpp b/torch/csrc/jit/passes/peephole.cpp
index 92dfa86da4b..125a8f53b79 100644
--- a/torch/csrc/jit/passes/peephole.cpp
+++ b/torch/csrc/jit/passes/peephole.cpp
@@ -2,8 +2,6 @@
 
 #include <ATen/core/jit_type.h>
 #include <c10/util/irange.h>
-#include <torch/csrc/jit/ir/alias_analysis.h>
-#include <torch/csrc/jit/ir/ir_views.h>
 #include <torch/csrc/jit/jit_log.h>
 #include <torch/csrc/jit/passes/concat_opt.h>
 #include <torch/csrc/jit/passes/dead_code_elimination.h>
@@ -11,7 +9,6 @@
 #include <torch/csrc/jit/passes/peephole_dict_idioms.h>
 #include <torch/csrc/jit/passes/peephole_list_idioms.h>
 #include <torch/csrc/jit/passes/peephole_non_tensor.h>
-#include <torch/csrc/jit/runtime/graph_executor.h>
 
 namespace torch::jit {
 
diff --git a/torch/csrc/jit/passes/peephole_alias_sensitive.cpp b/torch/csrc/jit/passes/peephole_alias_sensitive.cpp
index e3fca5c215f..e6ec265fc98 100644
--- a/torch/csrc/jit/passes/peephole_alias_sensitive.cpp
+++ b/torch/csrc/jit/passes/peephole_alias_sensitive.cpp
@@ -1,12 +1,6 @@
-#include <ATen/core/jit_type.h>
 #include <torch/csrc/jit/ir/alias_analysis.h>
-#include <torch/csrc/jit/ir/ir_views.h>
 #include <torch/csrc/jit/jit_log.h>
-#include <torch/csrc/jit/passes/dead_code_elimination.h>
-#include <torch/csrc/jit/passes/peephole.h>
 #include <torch/csrc/jit/passes/peephole_alias_sensitive.h>
-#include <torch/csrc/jit/runtime/graph_executor.h>
-#include <unordered_set>
 
 namespace torch::jit {
 
diff --git a/torch/csrc/jit/passes/peephole_list_idioms.cpp b/torch/csrc/jit/passes/peephole_list_idioms.cpp
index e07496dee2e..71734d32bbf 100644
--- a/torch/csrc/jit/passes/peephole_list_idioms.cpp
+++ b/torch/csrc/jit/passes/peephole_list_idioms.cpp
@@ -2,11 +2,8 @@
 #include <torch/csrc/jit/ir/alias_analysis.h>
 #include <torch/csrc/jit/ir/ir_views.h>
 #include <torch/csrc/jit/jit_log.h>
-#include <torch/csrc/jit/passes/dead_code_elimination.h>
-#include <torch/csrc/jit/passes/peephole.h>
 #include <torch/csrc/jit/passes/peephole_list_idioms.h>
 #include <torch/csrc/jit/passes/value_refinement_utils.h>
-#include <torch/csrc/jit/runtime/graph_executor.h>
 #include <torch/csrc/jit/runtime/slice_indices_adjust.h>
 #include <limits>
 #include <utility>
diff --git a/torch/csrc/jit/passes/peephole_non_tensor.cpp b/torch/csrc/jit/passes/peephole_non_tensor.cpp
index dbc8fce10da..a6bd622fc5d 100644
--- a/torch/csrc/jit/passes/peephole_non_tensor.cpp
+++ b/torch/csrc/jit/passes/peephole_non_tensor.cpp
@@ -1,4 +1,3 @@
-#include <torch/csrc/jit/passes/peephole.h>
 #include <torch/csrc/jit/passes/peephole_non_tensor.h>
 
 #include <ATen/core/jit_type.h>
diff --git a/torch/csrc/jit/passes/prepack_folding.cpp b/torch/csrc/jit/passes/prepack_folding.cpp
index 608432602dd..6efd4427585 100644
--- a/torch/csrc/jit/passes/prepack_folding.cpp
+++ b/torch/csrc/jit/passes/prepack_folding.cpp
@@ -1,7 +1,6 @@
 #include <stack>
 
 #include <torch/csrc/jit/api/module.h>
-#include <torch/csrc/jit/passes/constant_pooling.h>
 #include <torch/csrc/jit/passes/constant_propagation.h>
 #include <torch/csrc/jit/passes/prepack_folding.h>
 
diff --git a/torch/csrc/jit/passes/quantization/insert_observers.cpp b/torch/csrc/jit/passes/quantization/insert_observers.cpp
index 5fab2350444..d1dc726faaa 100644
--- a/torch/csrc/jit/passes/quantization/insert_observers.cpp
+++ b/torch/csrc/jit/passes/quantization/insert_observers.cpp
@@ -4,8 +4,6 @@
 #include <torch/csrc/jit/frontend/schema_matching.h>
 #include <torch/csrc/jit/ir/subgraph_matcher.h>
 #include <torch/csrc/jit/jit_log.h>
-#include <torch/csrc/jit/passes/constant_pooling.h>
-#include <torch/csrc/jit/passes/constant_propagation.h>
 #include <torch/csrc/jit/passes/fuse_linear.h>
 #include <torch/csrc/jit/passes/graph_rewrite_helper.h>
 #include <torch/csrc/jit/passes/inline_fork_wait.h>
diff --git a/torch/csrc/jit/passes/refine_tuple_types.cpp b/torch/csrc/jit/passes/refine_tuple_types.cpp
index 08d91d43150..14349438eef 100644
--- a/torch/csrc/jit/passes/refine_tuple_types.cpp
+++ b/torch/csrc/jit/passes/refine_tuple_types.cpp
@@ -1,8 +1,6 @@
 #include <torch/csrc/jit/passes/refine_tuple_types.h>
 #include <torch/csrc/jit/runtime/graph_iterator.h>
 
-#include <ATen/core/type_factory.h>
-
 #include <utility>
 
 namespace torch::jit {
diff --git a/torch/csrc/jit/passes/remove_redundant_profiles.cpp b/torch/csrc/jit/passes/remove_redundant_profiles.cpp
index 1bfb6396eba..e636433cff8 100644
--- a/torch/csrc/jit/passes/remove_redundant_profiles.cpp
+++ b/torch/csrc/jit/passes/remove_redundant_profiles.cpp
@@ -1,8 +1,6 @@
-#include <torch/csrc/jit/passes/dead_code_elimination.h>
 #include <torch/csrc/jit/passes/remove_redundant_profiles.h>
 
 #include <torch/csrc/jit/ir/alias_analysis.h>
-#include <torch/csrc/jit/ir/ir_views.h>
 #include <torch/csrc/jit/jit_log.h>
 
 namespace torch::jit {
diff --git a/torch/csrc/jit/passes/replacement_of_old_operators.cpp b/torch/csrc/jit/passes/replacement_of_old_operators.cpp
index 090f4a46b14..4e9f123918b 100644
--- a/torch/csrc/jit/passes/replacement_of_old_operators.cpp
+++ b/torch/csrc/jit/passes/replacement_of_old_operators.cpp
@@ -8,7 +8,6 @@
 #include <torch/csrc/jit/operator_upgraders/utils.h>
 #include <torch/csrc/jit/operator_upgraders/version_map.h>
 #include <torch/csrc/jit/runtime/graph_iterator.h>
-#include <limits>
 #include <string>
 #include <unordered_map>
 #include <utility>
diff --git a/torch/csrc/jit/passes/requires_grad_analysis.cpp b/torch/csrc/jit/passes/requires_grad_analysis.cpp
index 88367b58a81..17a8289ba75 100644
--- a/torch/csrc/jit/passes/requires_grad_analysis.cpp
+++ b/torch/csrc/jit/passes/requires_grad_analysis.cpp
@@ -1,6 +1,5 @@
 #include <torch/csrc/jit/passes/requires_grad_analysis.h>
 
-#include <ATen/core/jit_type.h>
 #include <c10/util/irange.h>
 #include <torch/csrc/jit/ir/constants.h>
 #include <torch/csrc/jit/ir/ir.h>
diff --git a/torch/csrc/jit/passes/restore_mutation.cpp b/torch/csrc/jit/passes/restore_mutation.cpp
index 8e02f4f55e2..fbefcd7ed7a 100644
--- a/torch/csrc/jit/passes/restore_mutation.cpp
+++ b/torch/csrc/jit/passes/restore_mutation.cpp
@@ -1,5 +1,3 @@
-#include <ATen/core/jit_type.h>
-#include <ATen/core/symbol.h>
 #include <torch/csrc/jit/passes/remove_mutation.h>
 #include <torch/csrc/jit/passes/restore_mutation.h>
 
diff --git a/torch/csrc/jit/passes/shape_analysis.cpp b/torch/csrc/jit/passes/shape_analysis.cpp
index 57dc2552c66..7493667a2f0 100644
--- a/torch/csrc/jit/passes/shape_analysis.cpp
+++ b/torch/csrc/jit/passes/shape_analysis.cpp
@@ -11,9 +11,6 @@
 #include <torch/csrc/jit/runtime/exception_message.h>
 #include <torch/csrc/jit/runtime/operator.h>
 
-#include <torch/csrc/autograd/variable.h>
-
-#include <ATen/DeviceGuard.h>
 #include <ATen/ExpandUtils.h>
 #include <ATen/core/symbol.h>
 
diff --git a/torch/csrc/jit/passes/symbolic_shape_analysis.cpp b/torch/csrc/jit/passes/symbolic_shape_analysis.cpp
index 999f8247b7c..75ec0e12016 100644
--- a/torch/csrc/jit/passes/symbolic_shape_analysis.cpp
+++ b/torch/csrc/jit/passes/symbolic_shape_analysis.cpp
@@ -12,16 +12,12 @@
 #include <torch/csrc/jit/passes/dead_code_elimination.h>
 #include <torch/csrc/jit/passes/integer_value_refinement.h>
 #include <torch/csrc/jit/passes/loop_unrolling.h>
-#include <torch/csrc/jit/passes/lower_tuples.h>
-#include <torch/csrc/jit/passes/peephole.h>
 #include <torch/csrc/jit/passes/peephole_list_idioms.h>
 #include <torch/csrc/jit/passes/peephole_non_tensor.h>
 #include <torch/csrc/jit/passes/remove_mutation.h>
 #include <torch/csrc/jit/passes/shape_analysis.h>
 #include <torch/csrc/jit/passes/symbolic_shape_analysis.h>
 #include <torch/csrc/jit/passes/symbolic_shape_cache.h>
-#include <torch/csrc/jit/passes/tensorexpr_fuser.h>
-#include <torch/csrc/jit/runtime/exception_message.h>
 #include <torch/csrc/jit/runtime/symbolic_shape_registry.h>
 #include <algorithm>
 #include <memory>
diff --git a/torch/csrc/jit/passes/symbolic_shape_runtime_fusion.cpp b/torch/csrc/jit/passes/symbolic_shape_runtime_fusion.cpp
index 60363116571..c8c6953f344 100644
--- a/torch/csrc/jit/passes/symbolic_shape_runtime_fusion.cpp
+++ b/torch/csrc/jit/passes/symbolic_shape_runtime_fusion.cpp
@@ -1,7 +1,6 @@
 #include <ATen/core/functional.h>
 #include <ATen/core/interned_strings.h>
 #include <c10/core/MemoryFormat.h>
-#include <c10/core/ScalarType.h>
 #include <c10/util/Exception.h>
 #include <torch/csrc/jit/ir/ir.h>
 #include <torch/csrc/jit/ir/ir_views.h>
diff --git a/torch/csrc/jit/passes/tensorexpr_fuser.cpp b/torch/csrc/jit/passes/tensorexpr_fuser.cpp
index 672a9949c6b..e0f03324454 100644
--- a/torch/csrc/jit/passes/tensorexpr_fuser.cpp
+++ b/torch/csrc/jit/passes/tensorexpr_fuser.cpp
@@ -5,7 +5,6 @@
 #include <ATen/record_function.h>
 #include <c10/util/FunctionRef.h>
 #include <c10/util/irange.h>
-#include <torch/csrc/jit/codegen/cuda/interface.h>
 #include <torch/csrc/jit/codegen/fuser/interface.h>
 #include <torch/csrc/jit/ir/alias_analysis.h>
 #include <torch/csrc/jit/jit_log.h>
@@ -13,7 +12,6 @@
 #include <torch/csrc/jit/passes/common_subexpression_elimination.h>
 #include <torch/csrc/jit/passes/constant_pooling.h>
 #include <torch/csrc/jit/passes/dead_code_elimination.h>
-#include <torch/csrc/jit/passes/pass_manager.h>
 #include <torch/csrc/jit/passes/remove_redundant_profiles.h>
 #include <torch/csrc/jit/passes/symbolic_shape_runtime_fusion.h>
 #include <torch/csrc/jit/passes/utils/subgraph_utils.h>
diff --git a/torch/csrc/jit/passes/update_differentiable_graph_requires_grad.cpp b/torch/csrc/jit/passes/update_differentiable_graph_requires_grad.cpp
index 3333bfeefb1..fb46771cdbc 100644
--- a/torch/csrc/jit/passes/update_differentiable_graph_requires_grad.cpp
+++ b/torch/csrc/jit/passes/update_differentiable_graph_requires_grad.cpp
@@ -1,7 +1,6 @@
 #include <torch/csrc/jit/passes/update_differentiable_graph_requires_grad.h>
 
 #include <torch/csrc/jit/ir/ir.h>
-#include <torch/csrc/jit/passes/utils/subgraph_utils.h>
 
 namespace torch::jit {
 
diff --git a/torch/csrc/jit/passes/utils/memory_dag.cpp b/torch/csrc/jit/passes/utils/memory_dag.cpp
index 8ad213082f5..1b56cddf79d 100644
--- a/torch/csrc/jit/passes/utils/memory_dag.cpp
+++ b/torch/csrc/jit/passes/utils/memory_dag.cpp
@@ -2,7 +2,6 @@
 
 #include <c10/util/flat_hash_map.h>
 #include <algorithm>
-#include <queue>
 
 namespace torch::jit {
 namespace {
diff --git a/torch/csrc/jit/passes/utils/subgraph_utils.cpp b/torch/csrc/jit/passes/utils/subgraph_utils.cpp
index f54adbd7223..6f92e821e5b 100644
--- a/torch/csrc/jit/passes/utils/subgraph_utils.cpp
+++ b/torch/csrc/jit/passes/utils/subgraph_utils.cpp
@@ -3,7 +3,6 @@
 #include <torch/csrc/jit/passes/canonicalize.h>
 
 #include <ATen/core/symbol.h>
-#include <c10/util/StringUtil.h>
 #include <c10/util/irange.h>
 #include <torch/csrc/jit/jit_log.h>
 
diff --git a/torch/csrc/jit/passes/vulkan_rewrite.cpp b/torch/csrc/jit/passes/vulkan_rewrite.cpp
index 7d9b3b8210c..4914a2f8186 100644
--- a/torch/csrc/jit/passes/vulkan_rewrite.cpp
+++ b/torch/csrc/jit/passes/vulkan_rewrite.cpp
@@ -1,4 +1,3 @@
-#include <ATen/core/jit_type.h>
 #include <torch/csrc/jit/ir/ir.h>
 #include <torch/csrc/jit/ir/subgraph_matcher.h>
 #include <torch/csrc/jit/passes/dead_code_elimination.h>
diff --git a/torch/csrc/jit/python/pybind_utils.cpp b/torch/csrc/jit/python/pybind_utils.cpp
index 9f7c2756d0d..31f24bf1b4b 100644
--- a/torch/csrc/jit/python/pybind_utils.cpp
+++ b/torch/csrc/jit/python/pybind_utils.cpp
@@ -1,4 +1,3 @@
-#include <torch/csrc/jit/ir/graph_utils.h>
 #include <torch/csrc/jit/python/module_python.h>
 #include <torch/csrc/jit/python/pybind_utils.h>
 #include <torch/csrc/jit/python/python_dict.h>
@@ -8,7 +7,6 @@
 
 #include <ATen/ScalarOps.h>
 
-#include <c10/core/QScheme.h>
 #include <c10/util/irange.h>
 #include <torch/csrc/utils/python_arg_parser.h>
 
diff --git a/torch/csrc/jit/python/python_custom_class.cpp b/torch/csrc/jit/python/python_custom_class.cpp
index 32ba91df0ab..25f5088368e 100644
--- a/torch/csrc/jit/python/python_custom_class.cpp
+++ b/torch/csrc/jit/python/python_custom_class.cpp
@@ -1,8 +1,6 @@
 #include <torch/csrc/jit/python/pybind_utils.h>
 #include <torch/csrc/jit/python/python_custom_class.h>
 
-#include <torch/csrc/jit/frontend/sugared_value.h>
-
 #include <fmt/format.h>
 
 namespace torch::jit {
diff --git a/torch/csrc/jit/python/python_dict.cpp b/torch/csrc/jit/python/python_dict.cpp
index ea64f5a985d..82fd4449a4f 100644
--- a/torch/csrc/jit/python/python_dict.cpp
+++ b/torch/csrc/jit/python/python_dict.cpp
@@ -1,5 +1,4 @@
 #include <ATen/core/ivalue.h>
-#include <pybind11/cast.h>
 #include <pybind11/detail/common.h>
 #include <torch/csrc/jit/python/pybind_utils.h>
 #include <torch/csrc/jit/python/python_dict.h>
diff --git a/torch/csrc/jit/python/python_interpreter.cpp b/torch/csrc/jit/python/python_interpreter.cpp
index 7b29134cf0e..7e78cbd28f7 100644
--- a/torch/csrc/jit/python/python_interpreter.cpp
+++ b/torch/csrc/jit/python/python_interpreter.cpp
@@ -1,24 +1,11 @@
 #include <torch/csrc/jit/runtime/interpreter.h>
-#include <torch/csrc/python_headers.h>
 
-#include <torch/csrc/autograd/edge.h>
-#include <torch/csrc/autograd/function.h>
-#include <torch/csrc/autograd/profiler.h>
-#include <torch/csrc/autograd/variable.h>
 #include <torch/csrc/jit/ir/ir.h>
 #include <torch/csrc/jit/python/pybind_utils.h>
 #include <torch/csrc/jit/python/python_ir.h>
 #include <torch/csrc/jit/runtime/custom_operator.h>
-#include <torch/csrc/jit/runtime/graph_executor.h>
 #include <torch/csrc/jit/runtime/operator.h>
 
-#include <pybind11/pybind11.h>
-#include <torch/csrc/Exceptions.h>
-#include <torch/csrc/autograd/python_engine.h>
-#include <torch/csrc/autograd/python_variable.h>
-#include <torch/csrc/jit/python/pybind.h>
-#include <torch/csrc/utils/pybind.h>
-
 namespace py = pybind11;
 
 namespace torch::jit {
diff --git a/torch/csrc/jit/python/python_ir.cpp b/torch/csrc/jit/python/python_ir.cpp
index 6e5dcde957d..bd1290cbdf9 100644
--- a/torch/csrc/jit/python/python_ir.cpp
+++ b/torch/csrc/jit/python/python_ir.cpp
@@ -12,11 +12,7 @@
 #include <torch/csrc/jit/passes/shape_analysis.h>
 #include <torch/csrc/jit/passes/symbolic_shape_analysis.h>
 #include <torch/csrc/jit/python/pybind.h>
-#include <torch/csrc/jit/python/python_tracer.h>
-#include <torch/csrc/jit/runtime/argument_spec.h>
 #include <torch/csrc/jit/serialization/export.h>
-#include <torch/csrc/jit/serialization/python_print.h>
-#include <torch/csrc/python_headers.h>
 #include <torch/csrc/utils/pybind.h>
 #include <torch/csrc/utils/python_strings.h>
 #include <iostream>
diff --git a/torch/csrc/jit/python/python_sugared_value.cpp b/torch/csrc/jit/python/python_sugared_value.cpp
index 8b16e089aa5..26c8fe067a6 100644
--- a/torch/csrc/jit/python/python_sugared_value.cpp
+++ b/torch/csrc/jit/python/python_sugared_value.cpp
@@ -1,7 +1,6 @@
 #include <torch/csrc/jit/python/python_sugared_value.h>
 
 #include <ATen/core/interned_strings.h>
-#include <c10/core/ScalarType.h>
 #include <pybind11/pytypes.h>
 #include <torch/csrc/Dtype.h>
 #include <torch/csrc/Layout.h>
@@ -9,15 +8,12 @@
 #include <torch/csrc/jit/frontend/schema_matching.h>
 #include <torch/csrc/jit/python/module_python.h>
 #include <torch/csrc/utils/pybind.h>
-#include <climits>
 #include <memory>
 #include <sstream>
 #include <string>
 #include <tuple>
 #include <vector>
 
-#include <Python.h>
-
 namespace torch::jit {
 
 std::string typeString(py::handle h) {
diff --git a/torch/csrc/jit/python/python_tracer.cpp b/torch/csrc/jit/python/python_tracer.cpp
index 92103119973..5cf3bd900f3 100644
--- a/torch/csrc/jit/python/python_tracer.cpp
+++ b/torch/csrc/jit/python/python_tracer.cpp
@@ -6,7 +6,6 @@
 #include <torch/csrc/jit/passes/lower_tuples.h>
 #include <torch/csrc/jit/python/pybind.h>
 #include <torch/csrc/jit/python/python_tracer.h>
-#include <torch/csrc/jit/serialization/export.h>
 #include <torch/csrc/utils/python_strings.h>
 
 #include <c10/util/Exception.h>
diff --git a/torch/csrc/jit/runtime/autodiff.cpp b/torch/csrc/jit/runtime/autodiff.cpp
index f1e58a9bd3e..214a07872d0 100644
--- a/torch/csrc/jit/runtime/autodiff.cpp
+++ b/torch/csrc/jit/runtime/autodiff.cpp
@@ -1,6 +1,5 @@
 #include <torch/csrc/jit/runtime/autodiff.h>
 
-#include <ATen/core/functional.h>
 #include <c10/util/Exception.h>
 #include <c10/util/irange.h>
 #include <torch/csrc/jit/jit_log.h>
diff --git a/torch/csrc/jit/runtime/decomposition_registry.cpp b/torch/csrc/jit/runtime/decomposition_registry.cpp
index 31ee76d1429..fbaa10ee32b 100644
--- a/torch/csrc/jit/runtime/decomposition_registry.cpp
+++ b/torch/csrc/jit/runtime/decomposition_registry.cpp
@@ -1,4 +1,3 @@
-#include <torch/csrc/jit/frontend/ir_emitter.h>
 #include <torch/csrc/jit/jit_log.h>
 #include <torch/csrc/jit/passes/constant_propagation.h>
 #include <torch/csrc/jit/passes/peephole.h>
diff --git a/torch/csrc/jit/runtime/decomposition_registry_util.cpp b/torch/csrc/jit/runtime/decomposition_registry_util.cpp
index d0a4fa3b04f..ad48d1cd893 100644
--- a/torch/csrc/jit/runtime/decomposition_registry_util.cpp
+++ b/torch/csrc/jit/runtime/decomposition_registry_util.cpp
@@ -5,10 +5,7 @@
  * To re-generate, please run:
  * cd ~/pytorch && python torchgen/decompositions/gen_jit_decompositions.py
  */
-#include <torch/csrc/jit/jit_log.h>
-#include <torch/csrc/jit/passes/inliner.h>
 #include <torch/csrc/jit/runtime/decomposition_registry_util.h>
-#include <torch/csrc/jit/runtime/operator.h>
 
 namespace torch::jit {
 
diff --git a/torch/csrc/jit/runtime/graph_executor.cpp b/torch/csrc/jit/runtime/graph_executor.cpp
index bb152df094f..4bdab8c5dcb 100644
--- a/torch/csrc/jit/runtime/graph_executor.cpp
+++ b/torch/csrc/jit/runtime/graph_executor.cpp
@@ -30,7 +30,6 @@
 #include <torch/csrc/jit/passes/shape_analysis.h>
 #include <torch/csrc/jit/passes/specialize_autogradzero.h>
 #include <torch/csrc/jit/passes/tensorexpr_fuser.h>
-#include <torch/csrc/jit/resource_guard.h>
 #include <torch/csrc/jit/runtime/argument_spec.h>
 #include <torch/csrc/jit/runtime/autodiff.h>
 #include <torch/csrc/jit/runtime/custom_operator.h>
diff --git a/torch/csrc/jit/runtime/interpreter.cpp b/torch/csrc/jit/runtime/interpreter.cpp
index 95b74376d2e..7fd16c08a9e 100644
--- a/torch/csrc/jit/runtime/interpreter.cpp
+++ b/torch/csrc/jit/runtime/interpreter.cpp
@@ -1,17 +1,10 @@
 #include <torch/csrc/jit/runtime/interpreter.h>
 
-#include <ATen/Parallel.h>
 #include <ATen/core/ivalue.h>
 #include <ATen/record_function.h>
-#include <c10/core/thread_pool.h>
 #include <c10/macros/Macros.h>
 #include <c10/util/Exception.h>
 #include <c10/util/irange.h>
-#include <torch/csrc/autograd/edge.h>
-#include <torch/csrc/autograd/grad_mode.h>
-#include <torch/csrc/autograd/profiler.h>
-#include <torch/csrc/autograd/variable.h>
-#include <torch/csrc/jit/api/compilation_unit.h>
 #include <torch/csrc/jit/api/function_impl.h>
 #include <torch/csrc/jit/ir/constants.h>
 #include <torch/csrc/jit/ir/ir.h>
@@ -40,7 +33,6 @@ using torch::distributed::autograd::DistAutogradContainer;
 #include <mutex>
 #include <ostream>
 #include <stdexcept>
-#include <typeinfo>
 #include <unordered_map>
 #include <unordered_set>
 #include <utility>
diff --git a/torch/csrc/jit/runtime/jit_trace.cpp b/torch/csrc/jit/runtime/jit_trace.cpp
index 45be4fe21bb..8a1daabf54c 100644
--- a/torch/csrc/jit/runtime/jit_trace.cpp
+++ b/torch/csrc/jit/runtime/jit_trace.cpp
@@ -1,16 +1,10 @@
 
-#include <ATen/ATen.h>
-#include <ATen/Parallel.h>
 #include <ATen/core/ivalue.h>
 #include <ATen/core/symbol.h>
 #include <torch/csrc/jit/ir/ir_views.h>
 #include <torch/csrc/jit/jit_log.h>
 #include <torch/csrc/jit/passes/dead_code_elimination.h>
-#include <torch/csrc/jit/passes/freeze_module.h>
-#include <torch/csrc/jit/passes/frozen_graph_optimizations.h>
 #include <torch/csrc/jit/passes/inliner.h>
-#include <torch/csrc/jit/passes/insert_guards.h>
-#include <torch/csrc/jit/passes/remove_mutation.h>
 #include <torch/csrc/jit/runtime/graph_executor.h>
 #include <torch/csrc/jit/runtime/interpreter.h>
 #include <torch/csrc/jit/runtime/jit_trace.h>
diff --git a/torch/csrc/jit/runtime/operator.cpp b/torch/csrc/jit/runtime/operator.cpp
index 6f9dec70cdd..30105754c5e 100644
--- a/torch/csrc/jit/runtime/operator.cpp
+++ b/torch/csrc/jit/runtime/operator.cpp
@@ -1,6 +1,5 @@
 #include <torch/csrc/jit/runtime/operator.h>
 
-#include <ATen/ATen.h>
 #include <ATen/core/interned_strings.h>
 #include <c10/util/irange.h>
 #include <torch/csrc/jit/frontend/edit_distance.h>
diff --git a/torch/csrc/jit/runtime/profiling_graph_executor_impl.cpp b/torch/csrc/jit/runtime/profiling_graph_executor_impl.cpp
index 98acf24dd1d..680244b363c 100644
--- a/torch/csrc/jit/runtime/profiling_graph_executor_impl.cpp
+++ b/torch/csrc/jit/runtime/profiling_graph_executor_impl.cpp
@@ -3,7 +3,6 @@
 #include <c10/util/irange.h>
 #include <torch/csrc/jit/jit_log.h>
 #include <torch/csrc/jit/passes/add_if_then_else.h>
-#include <torch/csrc/jit/passes/bailout_graph.h>
 #include <torch/csrc/jit/passes/batch_mm.h>
 #include <torch/csrc/jit/passes/canonicalize_graph_fuser_ops.h>
 #include <torch/csrc/jit/passes/check_strict_fusion.h>
@@ -16,11 +15,9 @@
 #include <torch/csrc/jit/passes/dead_code_elimination.h>
 #include <torch/csrc/jit/passes/decompose_ops.h>
 #include <torch/csrc/jit/passes/graph_fuser.h>
-#include <torch/csrc/jit/passes/guard_elimination.h>
 #include <torch/csrc/jit/passes/inline_autodiff_subgraphs.h>
 #include <torch/csrc/jit/passes/inliner.h>
 #include <torch/csrc/jit/passes/inplace_check.h>
-#include <torch/csrc/jit/passes/insert_guards.h>
 #include <torch/csrc/jit/passes/loop_unrolling.h>
 #include <torch/csrc/jit/passes/lower_grad_of.h>
 #include <torch/csrc/jit/passes/lower_tuples.h>
diff --git a/torch/csrc/jit/runtime/profiling_record.cpp b/torch/csrc/jit/runtime/profiling_record.cpp
index 30e8c58d65a..fb01aa2a255 100644
--- a/torch/csrc/jit/runtime/profiling_record.cpp
+++ b/torch/csrc/jit/runtime/profiling_record.cpp
@@ -2,7 +2,6 @@
 
 #include <ATen/core/symbol.h>
 #include <c10/util/irange.h>
-#include <torch/csrc/jit/codegen/cuda/interface.h>
 #include <torch/csrc/jit/jit_log.h>
 #include <torch/csrc/jit/passes/clear_profiling.h>
 #include <torch/csrc/jit/passes/tensorexpr_fuser.h>
diff --git a/torch/csrc/jit/runtime/register_c10_ops.cpp b/torch/csrc/jit/runtime/register_c10_ops.cpp
index 85e8c0a2b03..be7bfbd4acd 100644
--- a/torch/csrc/jit/runtime/register_c10_ops.cpp
+++ b/torch/csrc/jit/runtime/register_c10_ops.cpp
@@ -1,8 +1,5 @@
-#include <ATen/core/ATenOpList.h>
 #include <ATen/core/dispatch/Dispatcher.h>
 #include <ATen/record_function.h>
-#include <torch/csrc/jit/frontend/tracer.h>
-#include <torch/csrc/jit/ir/ir.h>
 #include <torch/csrc/jit/runtime/operator.h>
 
 namespace torch::jit {
diff --git a/torch/csrc/jit/runtime/register_cuda_ops.cpp b/torch/csrc/jit/runtime/register_cuda_ops.cpp
index 9ca5e01b0dd..dec70000d2c 100644
--- a/torch/csrc/jit/runtime/register_cuda_ops.cpp
+++ b/torch/csrc/jit/runtime/register_cuda_ops.cpp
@@ -1,6 +1,5 @@
 // This file registers special JIT operators used to implement the PyTorch CUDA
 // API in TorchScript.
-#include <torch/csrc/api/include/torch/utils.h>
 #include <torch/csrc/jit/cuda/cuda.h>
 #include <torch/csrc/jit/ir/ir.h>
 #include <torch/csrc/jit/runtime/custom_operator.h>
diff --git a/torch/csrc/jit/runtime/register_distributed_ops.cpp b/torch/csrc/jit/runtime/register_distributed_ops.cpp
index a09a0f99f25..8ce967aca0f 100644
--- a/torch/csrc/jit/runtime/register_distributed_ops.cpp
+++ b/torch/csrc/jit/runtime/register_distributed_ops.cpp
@@ -1,8 +1,5 @@
-#include <ATen/ATen.h>
-#include <ATen/core/op_registration/op_registration.h>
 #include <torch/csrc/distributed/autograd/autograd.h>
 #include <torch/csrc/distributed/autograd/context/container.h>
-#include <torch/csrc/distributed/autograd/engine/dist_engine.h>
 #include <torch/csrc/distributed/rpc/rpc_agent.h>
 #include <torch/csrc/distributed/rpc/rref_impl.h>
 #include <torch/csrc/distributed/rpc/torchscript_functions.h>
diff --git a/torch/csrc/jit/runtime/register_prim_ops_fulljit.cpp b/torch/csrc/jit/runtime/register_prim_ops_fulljit.cpp
index b09cc45ce33..b74fc4316c2 100644
--- a/torch/csrc/jit/runtime/register_prim_ops_fulljit.cpp
+++ b/torch/csrc/jit/runtime/register_prim_ops_fulljit.cpp
@@ -4,25 +4,14 @@
 #include <ATen/core/ivalue.h>
 #include <c10/util/ApproximateClock.h>
 #include <c10/util/irange.h>
-#include <torch/csrc/autograd/profiler.h>
 #include <torch/csrc/jit/frontend/tracer.h>
 
 #include <algorithm>
-#include <bitset>
-#include <cctype>
 #include <cmath>
-#include <exception>
 #include <fstream>
 #include <iostream>
-#include <limits>
-#include <memory>
-#include <mutex>
-#include <ostream>
 #include <stdexcept>
 #include <string>
-#include <typeinfo>
-#include <unordered_map>
-#include <unordered_set>
 #include <utility>
 #include <vector>
 
diff --git a/torch/csrc/jit/runtime/register_special_ops.cpp b/torch/csrc/jit/runtime/register_special_ops.cpp
index 0f2447e05a9..c7343914cb6 100644
--- a/torch/csrc/jit/runtime/register_special_ops.cpp
+++ b/torch/csrc/jit/runtime/register_special_ops.cpp
@@ -2,7 +2,6 @@
 #include <torch/library.h>
 
 #include <ATen/ExpandUtils.h>
-#include <ATen/NativeFunctions.h>
 #include <ATen/core/jit_type.h>
 #include <c10/core/DefaultDtype.h>
 #include <c10/util/irange.h>
@@ -10,11 +9,9 @@
 #include <torch/csrc/jit/ir/ir.h>
 #include <torch/csrc/jit/runtime/custom_operator.h>
 #include <torch/csrc/jit/runtime/operator.h>
-#include <torch/csrc/jit/runtime/vararg_functions.h>
 
 #include <ATen/InitialTensorOptions.h>
 #include <c10/core/ScalarType.h>
-#include <torch/csrc/jit/frontend/error_report.h>
 
 #include <sstream>
 
diff --git a/torch/csrc/jit/runtime/script_profile.cpp b/torch/csrc/jit/runtime/script_profile.cpp
index a1e1ad6972e..a9151d0e00f 100644
--- a/torch/csrc/jit/runtime/script_profile.cpp
+++ b/torch/csrc/jit/runtime/script_profile.cpp
@@ -7,7 +7,6 @@
 
 #include <c10/util/Exception.h>
 #include <c10/util/intrusive_ptr.h>
-#include <torch/csrc/jit/api/function_impl.h>
 
 namespace torch::jit {
 
diff --git a/torch/csrc/jit/runtime/serialized_shape_function_registry.cpp b/torch/csrc/jit/runtime/serialized_shape_function_registry.cpp
index d77e0b3a10d..89537c4b404 100644
--- a/torch/csrc/jit/runtime/serialized_shape_function_registry.cpp
+++ b/torch/csrc/jit/runtime/serialized_shape_function_registry.cpp
@@ -6,9 +6,6 @@
  * cd ~/pytorch && python
  * torchgen/shape_functions/gen_jit_shape_functions.py
  */
-#include <torch/csrc/jit/jit_log.h>
-#include <torch/csrc/jit/passes/inliner.h>
-#include <torch/csrc/jit/runtime/operator.h>
 #include <torch/csrc/jit/runtime/serialized_shape_function_registry.h>
 
 // clang-format off
diff --git a/torch/csrc/jit/runtime/static/fusion.cpp b/torch/csrc/jit/runtime/static/fusion.cpp
index 61f2e5614ef..1dc66c85f1d 100644
--- a/torch/csrc/jit/runtime/static/fusion.cpp
+++ b/torch/csrc/jit/runtime/static/fusion.cpp
@@ -6,7 +6,6 @@
 #include <torch/csrc/jit/passes/canonicalize.h>
 #include <torch/csrc/jit/passes/constant_pooling.h>
 #include <torch/csrc/jit/passes/dead_code_elimination.h>
-#include <torch/csrc/jit/passes/freeze_module.h>
 #include <torch/csrc/jit/passes/remove_mutation.h>
 #include <torch/csrc/jit/passes/tensorexpr_fuser.h>
 #include <torch/csrc/jit/passes/utils/subgraph_utils.h>
diff --git a/torch/csrc/jit/runtime/static/impl.cpp b/torch/csrc/jit/runtime/static/impl.cpp
index 8ad348bb162..4cd12cf19fb 100644
--- a/torch/csrc/jit/runtime/static/impl.cpp
+++ b/torch/csrc/jit/runtime/static/impl.cpp
@@ -3,7 +3,6 @@
 #include <ATen/MemoryOverlap.h>
 #include <ATen/core/symbol.h>
 #include <ATen/record_function.h>
-#include <c10/core/CPUAllocator.h>
 #include <c10/core/InferenceMode.h>
 #include <c10/macros/Macros.h>
 #include <c10/util/MaybeOwned.h>
@@ -17,14 +16,12 @@
 #include <torch/csrc/jit/passes/eliminate_no_ops.h>
 #include <torch/csrc/jit/passes/freeze_module.h>
 #include <torch/csrc/jit/passes/remove_mutation.h>
-#include <torch/csrc/jit/passes/subgraph_rewrite.h>
 #include <torch/csrc/jit/passes/variadic_ops.h>
 #include <torch/csrc/jit/runtime/graph_iterator.h>
 #include <torch/csrc/jit/runtime/static/fusion.h>
 #include <torch/csrc/jit/runtime/static/memory_planner.h>
 #include <torch/csrc/jit/runtime/static/ops.h>
 #include <torch/csrc/jit/runtime/static/passes.h>
-#include <torch/csrc/jit/runtime/vararg_functions.h>
 #include <algorithm>
 #include <cstdint>
 #include <iostream>
diff --git a/torch/csrc/jit/runtime/static/memory_planner.cpp b/torch/csrc/jit/runtime/static/memory_planner.cpp
index 8660183867e..d1051b94b63 100644
--- a/torch/csrc/jit/runtime/static/memory_planner.cpp
+++ b/torch/csrc/jit/runtime/static/memory_planner.cpp
@@ -1,11 +1,8 @@
 #include <c10/core/alignment.h>
 #include <torch/csrc/jit/runtime/static/memory_planner.h>
 
-#include <ATen/Tensor.h>
-#include <torch/csrc/jit/ir/alias_analysis.h>
 #include <torch/csrc/jit/jit_log.h>
 #include <torch/csrc/jit/runtime/static/impl.h>
-#include <iterator>
 
 namespace torch::jit {
 
diff --git a/torch/csrc/jit/runtime/static/native_ops.cpp b/torch/csrc/jit/runtime/static/native_ops.cpp
index 716202f4568..9478dd98fce 100644
--- a/torch/csrc/jit/runtime/static/native_ops.cpp
+++ b/torch/csrc/jit/runtime/static/native_ops.cpp
@@ -3,13 +3,8 @@
 #include <torch/csrc/jit/runtime/static/ops.h>
 
 #include <ATen/CPUFunctions.h>
-#include <ATen/NativeFunctions.h>
-#include <ATen/ScalarOps.h>
-#include <ATen/TensorUtils.h>
 #include <ATen/native/IndexingUtils.h>
 #include <ATen/native/NonSymbolicBC.h>
-#include <ATen/native/Resize.h>
-#include <ATen/native/TensorAdvancedIndexing.h>
 #include <c10/util/intrusive_ptr.h>
 #include <c10/util/irange.h>
 #include <c10/util/ssize.h>
diff --git a/torch/csrc/jit/runtime/static/passes.cpp b/torch/csrc/jit/runtime/static/passes.cpp
index fdb0919da45..1029dd7019f 100644
--- a/torch/csrc/jit/runtime/static/passes.cpp
+++ b/torch/csrc/jit/runtime/static/passes.cpp
@@ -2,8 +2,6 @@
 
 #include <torch/csrc/jit/ir/alias_analysis.h>
 #include <torch/csrc/jit/ir/subgraph_matcher.h>
-#include <torch/csrc/jit/passes/constant_pooling.h>
-#include <torch/csrc/jit/passes/constant_propagation.h>
 #include <torch/csrc/jit/passes/subgraph_rewrite.h>
 #include <torch/csrc/jit/passes/variadic_ops.h>
 #include <torch/csrc/jit/runtime/graph_iterator.h>
diff --git a/torch/csrc/jit/runtime/symbolic_shape_registry.cpp b/torch/csrc/jit/runtime/symbolic_shape_registry.cpp
index b1f0f410f14..6fb34bc2027 100644
--- a/torch/csrc/jit/runtime/symbolic_shape_registry.cpp
+++ b/torch/csrc/jit/runtime/symbolic_shape_registry.cpp
@@ -1,5 +1,4 @@
 #include <c10/util/Exception.h>
-#include <torch/csrc/jit/frontend/ir_emitter.h>
 #include <torch/csrc/jit/ir/ir_views.h>
 #include <torch/csrc/jit/jit_log.h>
 #include <torch/csrc/jit/passes/inliner.h>
diff --git a/torch/csrc/jit/runtime/symbolic_shape_registry_util.cpp b/torch/csrc/jit/runtime/symbolic_shape_registry_util.cpp
index ac0cd61fd2f..c14277aebeb 100644
--- a/torch/csrc/jit/runtime/symbolic_shape_registry_util.cpp
+++ b/torch/csrc/jit/runtime/symbolic_shape_registry_util.cpp
@@ -1,9 +1,4 @@
-#include <torch/csrc/jit/frontend/ir_emitter.h>
-#include <torch/csrc/jit/jit_log.h>
-#include <torch/csrc/jit/passes/inliner.h>
-#include <torch/csrc/jit/runtime/operator.h>
 #include <torch/csrc/jit/runtime/symbolic_shape_registry_util.h>
-#include <unordered_map>
 
 namespace torch::jit {
 
diff --git a/torch/csrc/jit/testing/file_check.cpp b/torch/csrc/jit/testing/file_check.cpp
index fb1280400a8..0e792934472 100644
--- a/torch/csrc/jit/testing/file_check.cpp
+++ b/torch/csrc/jit/testing/file_check.cpp
@@ -12,7 +12,6 @@
 #include <c10/util/Exception.h>
 #include <c10/util/StringUtil.h>
 #include <c10/util/irange.h>
-#include <torch/csrc/Export.h>
 #include <torch/csrc/jit/frontend/source_range.h>
 #include <torch/csrc/jit/ir/ir.h>
 #include <torch/csrc/jit/testing/file_check.h>
diff --git a/torch/csrc/python_dimname.cpp b/torch/csrc/python_dimname.cpp
index d7046552f80..07f604600b2 100644
--- a/torch/csrc/python_dimname.cpp
+++ b/torch/csrc/python_dimname.cpp
@@ -1,5 +1,4 @@
 #include <c10/util/flat_hash_map.h>
-#include <torch/csrc/Exceptions.h>
 #include <torch/csrc/python_dimname.h>
 #include <torch/csrc/utils/python_strings.h>
 
diff --git a/torch/csrc/utils/cpp_stacktraces.cpp b/torch/csrc/utils/cpp_stacktraces.cpp
index 641dffe08bc..79c8253b91a 100644
--- a/torch/csrc/utils/cpp_stacktraces.cpp
+++ b/torch/csrc/utils/cpp_stacktraces.cpp
@@ -1,8 +1,5 @@
 #include <torch/csrc/utils/cpp_stacktraces.h>
 
-#include <cstdlib>
-#include <cstring>
-
 #include <c10/util/Exception.h>
 #include <c10/util/env.h>
 
diff --git a/torch/csrc/utils/device_lazy_init.cpp b/torch/csrc/utils/device_lazy_init.cpp
index e531cca4fb2..6083b55064c 100644
--- a/torch/csrc/utils/device_lazy_init.cpp
+++ b/torch/csrc/utils/device_lazy_init.cpp
@@ -3,7 +3,6 @@
 #include <torch/csrc/utils/device_lazy_init.h>
 
 #include <torch/csrc/Exceptions.h>
-#include <torch/csrc/python_headers.h>
 #include <torch/csrc/utils/object_ptr.h>
 
 #ifndef WIN32
diff --git a/torch/csrc/utils/disable_torch_function.cpp b/torch/csrc/utils/disable_torch_function.cpp
index becbe1681f0..d75c0351fb6 100644
--- a/torch/csrc/utils/disable_torch_function.cpp
+++ b/torch/csrc/utils/disable_torch_function.cpp
@@ -1,7 +1,6 @@
 #include <torch/csrc/Exceptions.h>
 #include <torch/csrc/autograd/python_variable.h>
 #include <torch/csrc/utils/disable_torch_function.h>
-#include <torch/csrc/utils/pybind.h>
 #include <torch/csrc/utils/python_strings.h>
 
 #include <ATen/PythonTorchFunctionTLS.h>
diff --git a/torch/csrc/utils/init.cpp b/torch/csrc/utils/init.cpp
index 30e4082b033..986df49c571 100644
--- a/torch/csrc/utils/init.cpp
+++ b/torch/csrc/utils/init.cpp
@@ -2,9 +2,6 @@
 #include <torch/csrc/utils/init.h>
 #include <torch/csrc/utils/throughput_benchmark.h>
 
-#include <pybind11/functional.h>
-#include <torch/csrc/utils/pybind.h>
-
 namespace torch::throughput_benchmark {
 
 void initThroughputBenchmarkBindings(PyObject* module) {
diff --git a/torch/csrc/utils/object_ptr.cpp b/torch/csrc/utils/object_ptr.cpp
index ff314fdad14..c77797c0a48 100644
--- a/torch/csrc/utils/object_ptr.cpp
+++ b/torch/csrc/utils/object_ptr.cpp
@@ -1,8 +1,6 @@
 #include <c10/macros/Macros.h>
 #include <torch/csrc/utils/object_ptr.h>
 
-#include <torch/csrc/python_headers.h>
-
 template <>
 TORCH_PYTHON_API void THPPointer<PyObject>::free() {
   if (ptr && C10_LIKELY(Py_IsInitialized()))
diff --git a/torch/csrc/utils/python_dispatch.cpp b/torch/csrc/utils/python_dispatch.cpp
index 3380bb0a13e..69971fe0983 100644
--- a/torch/csrc/utils/python_dispatch.cpp
+++ b/torch/csrc/utils/python_dispatch.cpp
@@ -1,14 +1,11 @@
 #include <torch/csrc/jit/frontend/function_schema_parser.h>
 #include <torch/csrc/utils/python_dispatch.h>
 
-#include <ATen/ATen.h>
 #include <ATen/DTensorState.h>
-#include <ATen/FuncTorchTLS.h>
 #include <ATen/FunctionalTensorWrapper.h>
 #include <ATen/TensorSubclassLikeUtils.h>
 #include <ATen/autocast_mode.h>
 #include <ATen/core/NestedIntSymNodeImpl.h>
-#include <ATen/core/PythonOpRegistrationTrampoline.h>
 #include <ATen/core/dispatch/Dispatcher.h>
 
 #include <ATen/functorch/BatchedTensorImpl.h>
@@ -22,14 +19,9 @@
 #include <torch/csrc/utils/tensor_new.h>
 
 #include <c10/util/flat_hash_map.h>
-#include <pybind11/operators.h>
-#include <pybind11/stl.h>
 #include <torch/csrc/inductor/aoti_eager/kernel_holder.h>
-#include <torch/csrc/utils/pybind.h>
 #include <torch/csrc/utils/python_raii.h>
 
-#include <cstdlib>
-#include <cstring>
 #include <iostream>
 #include <utility>
 
diff --git a/torch/csrc/utils/tensor_apply.cpp b/torch/csrc/utils/tensor_apply.cpp
index c8a731d8d5f..efb0b2c6889 100644
--- a/torch/csrc/utils/tensor_apply.cpp
+++ b/torch/csrc/utils/tensor_apply.cpp
@@ -1,11 +1,9 @@
 #include <torch/csrc/utils/tensor_apply.h>
 
 #include <ATen/ExpandUtils.h>
-#include <ATen/TensorUtils.h>
 #include <c10/util/irange.h>
 
 #include <torch/csrc/Exceptions.h>
-#include <torch/csrc/utils/python_numbers.h>
 #include <torch/csrc/utils/python_scalars.h>
 
 using namespace at;
diff --git a/torch/csrc/utils/tensor_dtypes.cpp b/torch/csrc/utils/tensor_dtypes.cpp
index e7c58540d74..39df9be6886 100644
--- a/torch/csrc/utils/tensor_dtypes.cpp
+++ b/torch/csrc/utils/tensor_dtypes.cpp
@@ -1,7 +1,6 @@
 #include <torch/csrc/Dtype.h>
 #include <torch/csrc/DynamicTypes.h>
 #include <torch/csrc/Exceptions.h>
-#include <torch/csrc/python_headers.h>
 #include <torch/csrc/utils/object_ptr.h>
 #include <torch/csrc/utils/tensor_dtypes.h>
 
diff --git a/torch/csrc/utils/tensor_layouts.cpp b/torch/csrc/utils/tensor_layouts.cpp
index be8816c8a9a..d0bccbcf910 100644
--- a/torch/csrc/utils/tensor_layouts.cpp
+++ b/torch/csrc/utils/tensor_layouts.cpp
@@ -1,9 +1,6 @@
-#include <ATen/Layout.h>
-#include <c10/core/ScalarType.h>
 #include <torch/csrc/DynamicTypes.h>
 #include <torch/csrc/Exceptions.h>
 #include <torch/csrc/Layout.h>
-#include <torch/csrc/python_headers.h>
 #include <torch/csrc/utils/object_ptr.h>
 #include <torch/csrc/utils/tensor_layouts.h>
 
diff --git a/torch/csrc/utils/tensor_list.cpp b/torch/csrc/utils/tensor_list.cpp
index f25175af2dc..0a264e11e35 100644
--- a/torch/csrc/utils/tensor_list.cpp
+++ b/torch/csrc/utils/tensor_list.cpp
@@ -2,10 +2,8 @@
 #include <torch/csrc/utils/tensor_list.h>
 
 #include <c10/util/irange.h>
-#include <pybind11/pybind11.h>
 #include <torch/csrc/Exceptions.h>
 #include <torch/csrc/autograd/python_variable.h>
-#include <torch/csrc/utils/pybind.h>
 #include <torch/csrc/utils/python_scalars.h>
 
 using namespace at;
diff --git a/torch/csrc/utils/tensor_memoryformats.cpp b/torch/csrc/utils/tensor_memoryformats.cpp
index 28d56291bc9..c1a3ff32649 100644
--- a/torch/csrc/utils/tensor_memoryformats.cpp
+++ b/torch/csrc/utils/tensor_memoryformats.cpp
@@ -1,11 +1,9 @@
 #include <torch/csrc/utils/tensor_memoryformats.h>
 
 #include <c10/core/MemoryFormat.h>
-#include <torch/csrc/DynamicTypes.h>
 #include <torch/csrc/Exceptions.h>
 #include <torch/csrc/MemoryFormat.h>
 
-#include <torch/csrc/python_headers.h>
 #include <torch/csrc/utils/object_ptr.h>
 
 namespace torch::utils {
diff --git a/torch/csrc/utils/tensor_qschemes.cpp b/torch/csrc/utils/tensor_qschemes.cpp
index 4c2e6f20557..f85d091bd57 100644
--- a/torch/csrc/utils/tensor_qschemes.cpp
+++ b/torch/csrc/utils/tensor_qschemes.cpp
@@ -2,11 +2,9 @@
 
 #include <c10/core/QScheme.h>
 #include <c10/util/irange.h>
-#include <torch/csrc/DynamicTypes.h>
 #include <torch/csrc/Exceptions.h>
 #include <torch/csrc/QScheme.h>
 
-#include <torch/csrc/python_headers.h>
 #include <torch/csrc/utils/object_ptr.h>
 
 namespace torch::utils {
diff --git a/torch/csrc/utils/tensor_types.cpp b/torch/csrc/utils/tensor_types.cpp
index c46baea82a4..620086f9ad5 100644
--- a/torch/csrc/utils/tensor_types.cpp
+++ b/torch/csrc/utils/tensor_types.cpp
@@ -1,10 +1,8 @@
-#include <Python.h>
 
 #include <torch/csrc/utils/tensor_types.h>
 
 #include <ATen/Context.h>
 #include <ATen/Formatting.h>
-#include <torch/csrc/Exceptions.h>
 #include <torch/csrc/autograd/generated/VariableType.h>
 #include <torch/csrc/tensor/python_tensor.h>
 
diff --git a/torch/csrc/utils/throughput_benchmark.cpp b/torch/csrc/utils/throughput_benchmark.cpp
index 2f0ba77979a..8e8016567c7 100644
--- a/torch/csrc/utils/throughput_benchmark.cpp
+++ b/torch/csrc/utils/throughput_benchmark.cpp
@@ -1,8 +1,6 @@
 #include <torch/csrc/utils/throughput_benchmark.h>
 
-#include <pybind11/pybind11.h>
 #include <torch/csrc/jit/python/pybind_utils.h>
-#include <torch/csrc/utils/pybind.h>
 
 namespace torch::throughput_benchmark {