Update on "Misc changes from compiled autograd branch"

This PR pulls out some standalone changes from #103822


cc voznesenskym penguinwu anijain2305 EikanWang jgong5 Guobing-Chen XiaobingSuper zhuhaozhe blzheng Xia-Weiwen wenzhe-nrv jiayisunx ipiszy chenyang78

[ghstack-poisoned]

Author: jansel

.ci/docker/common/install_base.sh

@@ -71,9 +71,7 @@ install_ubuntu() {
     libtool \
     vim \
     unzip \
-    gdb \
-    libxml2-dev \
-    libxslt-dev
+    gdb
 
   # Should resolve issues related to various apt package repository cert issues
   # see: https://github.com/pytorch/pytorch/issues/65931

.ci/docker/common/install_onnx.sh

@@ -24,7 +24,7 @@ pip_install \
   transformers==4.25.1
 
 # TODO: change this when onnx-script is on testPypi
-pip_install "onnxscript@git+https://github.com/microsoft/onnxscript@7e131c578f290ffad1f26bacda11a83daf5476ba"
+pip_install "onnxscript@git+https://github.com/microsoft/onnxscript@2bb3e9f2d094912f81cb63cecb412efb14c65738"
 
 # Cache the transformers model to be used later by ONNX tests. We need to run the transformers
 # package to download the model. By default, the model is cached at ~/.cache/huggingface/hub/

.ci/docker/requirements-ci.txt

@@ -124,7 +124,8 @@ opt-einsum==3.3
 #Pinned versions: 3.3
 #test that import: test_linalg.py
 
-#pillow
+pillow==9.2.0 ; python_version <= "3.8"
+pillow==9.5.0 ; python_version > "3.8"
 #Description:  Python Imaging Library fork
 #Pinned versions:
 #test that import:

.github/ci_commit_pins/huggingface.txt

@@ -1 +1 @@
-4.27.4
+4.30.2

aten/src/ATen/CPUGeneratorImpl.h

@@ -21,7 +21,7 @@ struct TORCH_API CPUGeneratorImpl : public c10::GeneratorImpl {
   uint64_t seed() override;
   void set_state(const c10::TensorImpl& new_state) override;
   c10::intrusive_ptr<c10::TensorImpl> get_state() const override;
-  static DeviceType device_type();
+  static c10::DeviceType device_type();
   uint32_t random();
   uint64_t random64();
   c10::optional<float> next_float_normal_sample();

aten/src/ATen/Context.h

@@ -34,7 +34,7 @@ class TORCH_API Context {
   Context();
 
   const Generator& defaultGenerator(Device device) {
-    DeviceType device_type = device.type();
+    c10::DeviceType device_type = device.type();
     initCUDAIfNeeded(device_type);
     initHIPIfNeeded(device_type);
     if (device_type == at::kCPU) {
@@ -44,18 +44,18 @@ class TORCH_API Context {
     } else if (device_type == at::kMPS) {
       return at::detail::getMPSHooks().getDefaultMPSGenerator();
     } else {
-      AT_ERROR(DeviceTypeName(device_type), " device type not enabled.");
+      AT_ERROR(c10::DeviceTypeName(device_type), " device type not enabled.");
     }
   }
-  Device getDeviceFromPtr(void* data, DeviceType device_type) {
+  Device getDeviceFromPtr(void* data, c10::DeviceType device_type) {
     initCUDAIfNeeded(device_type);
     initHIPIfNeeded(device_type);
     if (device_type == at::kCPU) {
-      return DeviceType::CPU;
+      return c10::DeviceType::CPU;
     } else if (device_type == at::kCUDA) {
       return at::detail::getCUDAHooks().getDeviceFromPtr(data);
     } else {
-      AT_ERROR(DeviceTypeName(device_type), " device type not enabled.");
+      AT_ERROR(c10::DeviceTypeName(device_type), " device type not enabled.");
     }
   }
   static bool isPinnedPtr(const void* data) {
@@ -96,19 +96,19 @@ class TORCH_API Context {
     return detail::getMPSHooks().hasMPS();
   }
   static bool hasIPU() {
-    return c10::impl::hasDeviceGuardImpl(at::DeviceType::IPU);
+    return c10::impl::hasDeviceGuardImpl(c10::DeviceType::IPU);
   }
   static bool hasXLA() {
-    return c10::impl::hasDeviceGuardImpl(at::DeviceType::XLA);
+    return c10::impl::hasDeviceGuardImpl(c10::DeviceType::XLA);
   }
   static bool hasXPU() {
     return detail::getXPUHooks().hasXPU();
   }
   static bool hasLazy() {
-    return c10::impl::hasDeviceGuardImpl(at::DeviceType::Lazy);
+    return c10::impl::hasDeviceGuardImpl(c10::DeviceType::Lazy);
   }
   static bool hasORT() {
-    return c10::impl::hasDeviceGuardImpl(at::DeviceType::ORT);
+    return c10::impl::hasDeviceGuardImpl(c10::DeviceType::ORT);
   }
   // defined in header so that getNonVariableType has ability to inline
   // call_once check. getNonVariableType is called fairly frequently
@@ -273,13 +273,13 @@ class TORCH_API Context {
   void unsetDefaultMobileCPUAllocator();
 
  private:
-  void initCUDAIfNeeded(DeviceType p) {
-    if (p == DeviceType::CUDA) {
+  void initCUDAIfNeeded(c10::DeviceType p) {
+    if (p == c10::DeviceType::CUDA) {
       lazyInitCUDA();
     }
   }
-  void initHIPIfNeeded(DeviceType p) {
-    if (p == DeviceType::HIP) {
+  void initHIPIfNeeded(c10::DeviceType p) {
+    if (p == c10::DeviceType::HIP) {
       lazyInitHIP();
     }
   }
@@ -428,7 +428,7 @@ static inline bool hasMKLDNN() {
 }
 
 static inline void manual_seed(uint64_t seed) {
-  auto gen = globalContext().defaultGenerator(DeviceType::CPU);
+  auto gen = globalContext().defaultGenerator(c10::DeviceType::CPU);
   {
     // See Note [Acquire lock when using random generators]
     std::lock_guard<std::mutex> lock(gen.mutex());
@@ -450,7 +450,7 @@ static inline void manual_seed(uint64_t seed) {
   }
 
   if (hasMPS()) {
-    auto mps_gen = globalContext().defaultGenerator(DeviceType::MPS);
+    auto mps_gen = globalContext().defaultGenerator(c10::DeviceType::MPS);
     // See Note [Acquire lock when using random generators]
     std::lock_guard<std::mutex> lock(mps_gen.mutex());
     mps_gen.set_current_seed(seed);

aten/src/ATen/TensorIterator.h

@@ -309,7 +309,7 @@ struct TORCH_API TensorIteratorBase : public impl::MetaBase {
   Device device(int arg = 0) const {
     return operands_[arg].device.value();
   }
-  DeviceType device_type(int arg = 0) const {
+  c10::DeviceType device_type(int arg = 0) const {
     return device(arg).type();
   }
   int64_t element_size(int arg) const {

aten/src/ATen/Utils.h

@@ -33,7 +33,7 @@ static inline std::vector<TensorImpl*> checked_dense_tensor_list_unwrap(
     ArrayRef<Tensor> tensors,
     const char* name,
     int pos,
-    DeviceType device_type,
+    c10::DeviceType device_type,
     ScalarType scalar_type) {
   std::vector<TensorImpl*> unwrapped;
   unwrapped.reserve(tensors.size());