diff --git a/aten/src/ATen/core/ATen_pch.h b/aten/src/ATen/core/ATen_pch.h
index 779aa8fac31..d2cbaf76240 100644
--- a/aten/src/ATen/core/ATen_pch.h
+++ b/aten/src/ATen/core/ATen_pch.h
@@ -24,20 +24,12 @@
 //    Where the threshold can be tweaked until c10 and some of ATen
 //    core are included but TORCH_ASSERT_NO_OPERATORS still passes.
 
-#include <cassert>
-#include <cctype>
 #include <cerrno>
-#include <climits>
-#include <clocale>
 #include <cmath>
 #include <cstddef>
 #include <cstdint>
-#include <cstdio>
 #include <cstdlib>
 #include <cstring>
-#include <ctime>
-#include <cwchar>
-#include <cwctype>
 
 #include <algorithm>
 #include <array>
@@ -46,31 +38,25 @@
 #include <complex>
 #include <deque>
 #include <exception>
-#include <fstream>
 #include <functional>
 #include <initializer_list>
 #include <iomanip>
-#include <ios>
 #include <iosfwd>
-#include <istream>
 #include <iterator>
 #include <limits>
-#include <locale>
+#include <list>
 #include <map>
 #include <memory>
 #include <mutex>
 #include <new>
 #include <numeric>
 #include <ostream>
-#include <ratio>
-#include <set>
 #include <sstream>
 #include <stdexcept>
-#include <streambuf>
 #include <string>
-#include <system_error>
 #include <tuple>
 #include <type_traits>
+#include <typeindex>
 #include <typeinfo>
 #include <unordered_map>
 #include <unordered_set>
@@ -80,36 +66,34 @@
 #include <c10/core/Allocator.h>
 #include <c10/core/AutogradState.h>
 #include <c10/core/Backend.h>
-#include <c10/core/CopyBytes.h>
 #include <c10/core/DefaultDtype.h>
 #include <c10/core/Device.h>
-#include <c10/core/DeviceGuard.h>
 #include <c10/core/DeviceType.h>
 #include <c10/core/DispatchKey.h>
 #include <c10/core/DispatchKeySet.h>
 #include <c10/core/GeneratorImpl.h>
-#include <c10/core/GradMode.h>
 #include <c10/core/InferenceMode.h>
 #include <c10/core/Layout.h>
 #include <c10/core/MemoryFormat.h>
+#include <c10/core/OptionalRef.h>
 #include <c10/core/QScheme.h>
+#include <c10/core/Scalar.h>
 #include <c10/core/ScalarType.h>
 #include <c10/core/ScalarTypeToTypeMeta.h>
 #include <c10/core/Storage.h>
 #include <c10/core/StorageImpl.h>
-#include <c10/core/Stream.h>
+#include <c10/core/SymBool.h>
+#include <c10/core/SymFloat.h>
 #include <c10/core/SymInt.h>
 #include <c10/core/SymIntArrayRef.h>
+#include <c10/core/SymNodeImpl.h>
 #include <c10/core/TensorImpl.h>
 #include <c10/core/TensorOptions.h>
 #include <c10/core/UndefinedTensorImpl.h>
 #include <c10/core/WrapDimMinimal.h>
-#include <c10/core/impl/DeviceGuardImplInterface.h>
-#include <c10/core/impl/InlineDeviceGuard.h>
 #include <c10/core/impl/LocalDispatchKeySet.h>
 #include <c10/core/impl/PyInterpreter.h>
 #include <c10/core/impl/SizesAndStrides.h>
-#include <c10/core/impl/VirtualGuardImpl.h>
 
 #include <c10/macros/Export.h>
 #include <c10/macros/Macros.h>
@@ -117,13 +101,16 @@
 #include <c10/util/AlignOf.h>
 #include <c10/util/ArrayRef.h>
 #include <c10/util/BFloat16.h>
-#include <c10/util/Backtrace.h>
 #include <c10/util/C++17.h>
 #include <c10/util/ConstexprCrc.h>
 #include <c10/util/Deprecated.h>
+#include <c10/util/DimVector.h>
 #include <c10/util/Exception.h>
 #include <c10/util/ExclusivelyOwned.h>
 #include <c10/util/Flags.h>
+#include <c10/util/Float8_e4m3fn.h>
+#include <c10/util/Float8_e5m2.h>
+#include <c10/util/FunctionRef.h>
 #include <c10/util/Half.h>
 #include <c10/util/IdWrapper.h>
 #include <c10/util/Logging.h>
@@ -142,8 +129,10 @@
 #include <c10/util/TypeTraits.h>
 #include <c10/util/UniqueVoidPtr.h>
 #include <c10/util/accumulate.h>
+#include <c10/util/bit_cast.h>
+#include <c10/util/bits.h>
 #include <c10/util/complex.h>
-#include <c10/util/flat_hash_map.h>
+#include <c10/util/floating_point_utils.h>
 #include <c10/util/in_place.h>
 #include <c10/util/intrusive_ptr.h>
 #include <c10/util/irange.h>
@@ -158,7 +147,10 @@
 #include <c10/util/string_utils.h>
 #include <c10/util/string_view.h>
 #include <c10/util/typeid.h>
+#include <c10/util/variant.h>
 
+#include <ATen/StorageUtils.h>
+#include <ATen/core/ATen_fwd.h>
 #include <ATen/core/DeprecatedTypeProperties.h>
 #include <ATen/core/DeprecatedTypePropertiesRegistry.h>
 #include <ATen/core/DimVector.h>
diff --git a/aten/src/ATen/cpu/vec/vec_base.h b/aten/src/ATen/cpu/vec/vec_base.h
index d0995ca8cf5..f68e9c6ebb3 100644
--- a/aten/src/ATen/cpu/vec/vec_base.h
+++ b/aten/src/ATen/cpu/vec/vec_base.h
@@ -21,7 +21,6 @@
 #include <functional>
 #include <cmath>
 #include <type_traits>
-#include <bitset>
 #include <climits>
 
 #include <ATen/cpu/vec/intrinsics.h>
diff --git a/c10/util/CallOnce.h b/c10/util/CallOnce.h
index a31600ef2e7..50467df5fa8 100644
--- a/c10/util/CallOnce.h
+++ b/c10/util/CallOnce.h
@@ -2,7 +2,6 @@
 
 #include <atomic>
 #include <mutex>
-#include <thread>
 #include <utility>
 
 #include <c10/macros/Macros.h>