diff --git a/c10/core/DispatchKey.h b/c10/core/DispatchKey.h
index 5e6441e7d6a..40be3f9dc5d 100644
--- a/c10/core/DispatchKey.h
+++ b/c10/core/DispatchKey.h
@@ -2,6 +2,7 @@
 
 #include <c10/core/DeviceType.h>
 #include <c10/macros/Export.h>
+#include <cstdint>
 #include <ostream>
 #include <string>
 
diff --git a/c10/core/StorageImpl.cpp b/c10/core/StorageImpl.cpp
index 2651300a5fd..eb7312527f2 100644
--- a/c10/core/StorageImpl.cpp
+++ b/c10/core/StorageImpl.cpp
@@ -1,4 +1,5 @@
 #include <c10/core/StorageImpl.h>
+#include <c10/util/flat_hash_map.h>
 
 namespace c10 {
 
diff --git a/c10/core/StorageImpl.h b/c10/core/StorageImpl.h
index 04affff38c5..a0d912160f6 100644
--- a/c10/core/StorageImpl.h
+++ b/c10/core/StorageImpl.h
@@ -4,8 +4,6 @@
 #include <c10/core/SymInt.h>
 #include <c10/core/impl/PyObjectSlot.h>
 
-#include <c10/util/Optional.h>
-#include <c10/util/flat_hash_map.h>
 #include <c10/util/intrusive_ptr.h>
 
 namespace c10 {
diff --git a/c10/core/impl/LocalDispatchKeySet.cpp b/c10/core/impl/LocalDispatchKeySet.cpp
index 4a1a39c2316..0b6cf6fa66f 100644
--- a/c10/core/impl/LocalDispatchKeySet.cpp
+++ b/c10/core/impl/LocalDispatchKeySet.cpp
@@ -1,7 +1,5 @@
 #include <c10/core/impl/LocalDispatchKeySet.h>
 
-#include <iostream>
-
 namespace c10 {
 namespace impl {
 
diff --git a/c10/core/impl/LocalDispatchKeySet.h b/c10/core/impl/LocalDispatchKeySet.h
index 391b8cff493..9eb60237f35 100644
--- a/c10/core/impl/LocalDispatchKeySet.h
+++ b/c10/core/impl/LocalDispatchKeySet.h
@@ -1,8 +1,7 @@
 #pragma once
 
 #include <c10/core/DispatchKeySet.h>
-#include <c10/macros/Macros.h>
-#include <c10/util/Flags.h>
+#include <c10/macros/Export.h>
 
 // TLS management for DispatchKeySet (the "local" DispatchKeySet(s))
 //
diff --git a/c10/core/impl/cow/context.cpp b/c10/core/impl/cow/context.cpp
index 1392b480bc9..5ea072a94ee 100644
--- a/c10/core/impl/cow/context.cpp
+++ b/c10/core/impl/cow/context.cpp
@@ -2,6 +2,7 @@
 
 #include <c10/core/impl/cow/deleter.h>
 #include <c10/util/Exception.h>
+#include <mutex>
 
 namespace c10::impl {
 
diff --git a/c10/core/impl/cow/context.h b/c10/core/impl/cow/context.h
index ee57e3aabfc..f22e02f25d6 100644
--- a/c10/core/impl/cow/context.h
+++ b/c10/core/impl/cow/context.h
@@ -1,12 +1,11 @@
 #pragma once
 
-#include <c10/macros/Macros.h>
+#include <c10/macros/Export.h>
 #include <c10/util/UniqueVoidPtr.h>
 
 #include <atomic>
 #include <cstdint>
 #include <memory>
-#include <mutex>
 #include <shared_mutex>
 #include <variant>
 
diff --git a/c10/core/impl/cow/deleter.h b/c10/core/impl/cow/deleter.h
index fcdc287abaf..b9b4f33e1d2 100644
--- a/c10/core/impl/cow/deleter.h
+++ b/c10/core/impl/cow/deleter.h
@@ -5,7 +5,7 @@
 
 #pragma once
 
-#include <c10/macros/Macros.h>
+#include <c10/macros/Export.h>
 
 namespace c10 {
 namespace impl {
diff --git a/c10/mobile/CPUCachingAllocator.cpp b/c10/mobile/CPUCachingAllocator.cpp
index e59de377586..c1b7ee36662 100644
--- a/c10/mobile/CPUCachingAllocator.cpp
+++ b/c10/mobile/CPUCachingAllocator.cpp
@@ -1,6 +1,6 @@
-#include <c10/mobile/CPUCachingAllocator.h>
-
 #include <c10/core/impl/alloc_cpu.h>
+#include <c10/mobile/CPUCachingAllocator.h>
+#include <c10/util/Exception.h>
 
 namespace c10 {
 
diff --git a/c10/mobile/CPUCachingAllocator.h b/c10/mobile/CPUCachingAllocator.h
index 3f78c63a3fe..94e1305d80c 100644
--- a/c10/mobile/CPUCachingAllocator.h
+++ b/c10/mobile/CPUCachingAllocator.h
@@ -1,11 +1,7 @@
 #pragma once
 
-#include <algorithm>
-#include <deque>
-#include <memory>
 #include <mutex>
 
-#include <c10/util/Exception.h>
 #include <c10/util/SmallVector.h>
 #include <c10/util/flat_hash_map.h>
 
diff --git a/c10/util/flags_use_gflags.cpp b/c10/util/flags_use_gflags.cpp
index c63d6fd437f..e0d5bb44781 100644
--- a/c10/util/flags_use_gflags.cpp
+++ b/c10/util/flags_use_gflags.cpp
@@ -1,11 +1,10 @@
-#include <c10/util/Flags.h>
-
-#include <string>
-
 #include <c10/macros/Macros.h>
 
 #ifdef C10_USE_GFLAGS
 
+#include <c10/util/Flags.h>
+#include <string>
+
 namespace c10 {
 
 using std::string;
diff --git a/c10/util/signal_handler.cpp b/c10/util/signal_handler.cpp
index b60314d26c6..41def36b90a 100644
--- a/c10/util/signal_handler.cpp
+++ b/c10/util/signal_handler.cpp
@@ -1,18 +1,16 @@
 #include <c10/util/Backtrace.h>
+#include <c10/util/Logging.h>
 #include <c10/util/signal_handler.h>
 
 #if defined(C10_SUPPORTS_SIGNAL_HANDLER)
 
 // Normal signal handler implementation.
-#include <cxxabi.h>
 #include <dirent.h>
-#include <dlfcn.h>
-#include <fmt/format.h>
+#include <fmt/core.h>
 #include <sys/syscall.h>
-#include <sys/types.h>
 #include <unistd.h>
-#include <unwind.h>
 
+#include <cstdint>
 #include <cstdio>
 #include <cstdlib>
 #include <iostream>
diff --git a/c10/util/signal_handler.h b/c10/util/signal_handler.h
index 70295874844..c14f69a0491 100644
--- a/c10/util/signal_handler.h
+++ b/c10/util/signal_handler.h
@@ -5,7 +5,6 @@
 #include <mutex>
 
 #include <c10/macros/Export.h>
-#include <c10/util/Logging.h>
 
 #if defined(__APPLE__)
 #define C10_SUPPORTS_SIGNAL_HANDLER