Add files using upload-large-folder tool

.venv/Lib/site-packages/torch/include/ATen/ATen.h

@@ -0,0 +1,37 @@
+#pragma once
+
+#if !defined(_MSC_VER) && __cplusplus < 201703L
+#error C++17 or later compatible compiler is required to use ATen.
+#endif
+
+#include <ATen/Context.h>
+#include <ATen/Device.h>
+#include <ATen/DeviceGuard.h>
+#include <ATen/DimVector.h>
+#include <ATen/Dispatch.h>
+#include <ATen/Formatting.h>
+#include <ATen/Functions.h>
+#include <ATen/NamedTensor.h>
+#include <ATen/ScalarOps.h>
+#include <ATen/Tensor.h>
+#include <ATen/TensorGeometry.h>
+#include <ATen/TensorIndexing.h>
+#include <ATen/TensorOperators.h>
+#include <ATen/Version.h>
+#include <ATen/core/ATenGeneral.h>
+#include <ATen/core/Generator.h>
+#include <ATen/core/Reduction.h>
+#include <ATen/core/Scalar.h>
+#include <ATen/core/UnsafeFromTH.h>
+#include <ATen/core/ivalue.h>
+#include <ATen/core/jit_type.h>
+#include <c10/core/Allocator.h>
+#include <c10/core/InferenceMode.h>
+#include <c10/core/Layout.h>
+#include <c10/core/Storage.h>
+#include <c10/core/TensorOptions.h>
+#include <c10/util/Exception.h>
+
+// TODO: try to remove this
+// There is some back story, see https://github.com/pytorch/pytorch/issues/48684
+#include <ATen/NativeFunctions.h>

.venv/Lib/site-packages/torch/include/ATen/AccumulateType.h

@@ -0,0 +1,173 @@
+#pragma once
+#include <ATen/Config.h>
+#include <c10/core/DeviceType.h>
+#include <c10/core/ScalarType.h>
+#include <c10/util/BFloat16.h>
+#include <c10/util/Float8_e4m3fn.h>
+#include <c10/util/Float8_e4m3fnuz.h>
+#include <c10/util/Float8_e5m2.h>
+#include <c10/util/Float8_e5m2fnuz.h>
+#include <c10/util/Half.h>
+
+// Defines the accumulation type for a scalar type.
+// Example:
+//   using accscalar_t = acc_type<scalar_t, /*is_cuda*/true>;
+//
+// Accumulation types are an important concept in numeric computing
+// because you frequently want to perform intermediate computations
+// at a higher precision than the input and output precision, to avoid
+// compounding internal rounding errors.  Accumulation is the most
+// well-known intermediate computation (it is of great importance for
+// sum reduction and matrix multiply, for example), but in PyTorch
+// acc_type ends up getting used for all sorts of other intermediate
+// computations, so it perhaps would be more accurately (ahem) called an
+// "accurate" type.  acc_type is especially important for reduced
+// precision operations like float16 and bfloat16, where relatively
+// benign looking inputs can easily end up overflowing/underflowing.
+//
+// acc_type is parametrized by whether or not you are running on CUDA
+// or not, because on CUDA double precision operations are expensive
+// and so by default, we don't actually want to use double as an
+// acc_type on CUDA.  A lot of things are typed out below, but
+// basically, the table is generated by a few rules:
+//
+//  If bool:
+//      Use 'bool' as acc_type.
+//  If floating point:
+//      If CUDA, use 'float' as acc_type (unless scalar_t is double),
+//      otherwise (CPU) use 'double'
+//  If integral:
+//      Use 'int64_t' as acc_type
+//
+// You're not forced to use this template; if you happen to know
+// something specific about your use case, you can specify your own
+// desired behavior.  This template, however, will give you a reasonable
+// default that will work for all dtypes supported in PyTorch.
+
+#if defined(__CUDACC__)
+#include <cuda.h>
+#include <cuda_fp16.h>
+#elif defined(__HIPCC__)
+#include <hip/hip_fp16.h>
+#include <hip/hip_runtime.h>
+#endif
+
+namespace at {
+
+template <typename T, c10::DeviceType D>
+struct AccumulateTypeDevice {};
+
+template <typename T, bool>
+struct AccumulateType {};
+
+template <typename T>
+struct AccumulateType<T, false> {
+  using type = typename AccumulateTypeDevice<T, c10::DeviceType::CPU>::type;
+};
+
+template <typename T>
+struct AccumulateType<T, true> {
+  using type = typename AccumulateTypeDevice<T, c10::DeviceType::CUDA>::type;
+};
+
+template <typename T, c10::DeviceType device>
+using acc_type_device = typename AccumulateTypeDevice<T, device>::type;
+
+template <typename T, bool is_cuda>
+using acc_type = typename AccumulateType<T, is_cuda>::type;
+
+#define ACC_TYPE(t, acc_t, device_type)         \
+  template <>                                   \
+  struct AccumulateTypeDevice<t, device_type> { \
+    using type = acc_t;                         \
+  };
+#define MPS_ACC_TYPE(t, acc_t) ACC_TYPE(t, acc_t, c10::DeviceType::MPS)
+#define XPU_ACC_TYPE(t, acc_t) ACC_TYPE(t, acc_t, c10::DeviceType::XPU)
+#define CUDA_ACC_TYPE(t, acc_t) ACC_TYPE(t, acc_t, c10::DeviceType::CUDA)
+#define CPU_ACC_TYPE(t, acc_t) ACC_TYPE(t, acc_t, c10::DeviceType::CPU)
+
+MPS_ACC_TYPE(BFloat16, float);
+MPS_ACC_TYPE(Half, float);
+MPS_ACC_TYPE(Float8_e5m2, float);
+MPS_ACC_TYPE(Float8_e4m3fn, float);
+MPS_ACC_TYPE(Float8_e5m2fnuz, float);
+MPS_ACC_TYPE(Float8_e4m3fnuz, float);
+MPS_ACC_TYPE(float, float);
+MPS_ACC_TYPE(double, float);
+MPS_ACC_TYPE(int8_t, int64_t);
+MPS_ACC_TYPE(uint8_t, int64_t);
+MPS_ACC_TYPE(char, int64_t);
+MPS_ACC_TYPE(int16_t, int64_t);
+MPS_ACC_TYPE(int32_t, int64_t);
+MPS_ACC_TYPE(int64_t, int64_t);
+MPS_ACC_TYPE(bool, bool);
+MPS_ACC_TYPE(c10::complex<Half>, c10::complex<float>);
+MPS_ACC_TYPE(c10::complex<float>, c10::complex<float>);
+MPS_ACC_TYPE(c10::complex<double>, c10::complex<float>);
+
+XPU_ACC_TYPE(BFloat16, float);
+XPU_ACC_TYPE(Half, float);
+XPU_ACC_TYPE(Float8_e5m2, float);
+XPU_ACC_TYPE(Float8_e4m3fn, float);
+XPU_ACC_TYPE(Float8_e5m2fnuz, float);
+XPU_ACC_TYPE(Float8_e4m3fnuz, float);
+XPU_ACC_TYPE(float, float);
+XPU_ACC_TYPE(double, double);
+XPU_ACC_TYPE(int8_t, int64_t);
+XPU_ACC_TYPE(uint8_t, int64_t);
+XPU_ACC_TYPE(char, int64_t);
+XPU_ACC_TYPE(int16_t, int64_t);
+XPU_ACC_TYPE(int32_t, int64_t);
+XPU_ACC_TYPE(int64_t, int64_t);
+XPU_ACC_TYPE(bool, bool);
+XPU_ACC_TYPE(c10::complex<Half>, c10::complex<float>);
+XPU_ACC_TYPE(c10::complex<float>, c10::complex<float>);
+XPU_ACC_TYPE(c10::complex<double>, c10::complex<double>);
+
+#if defined(__CUDACC__) || defined(__HIPCC__)
+CUDA_ACC_TYPE(half, float);
+#endif
+CUDA_ACC_TYPE(BFloat16, float);
+CUDA_ACC_TYPE(Half, float);
+CUDA_ACC_TYPE(Float8_e5m2, float);
+CUDA_ACC_TYPE(Float8_e4m3fn, float);
+CUDA_ACC_TYPE(Float8_e5m2fnuz, float);
+CUDA_ACC_TYPE(Float8_e4m3fnuz, float);
+CUDA_ACC_TYPE(float, float);
+CUDA_ACC_TYPE(double, double);
+CUDA_ACC_TYPE(int8_t, int64_t);
+CUDA_ACC_TYPE(uint8_t, int64_t);
+CUDA_ACC_TYPE(char, int64_t);
+CUDA_ACC_TYPE(int16_t, int64_t);
+CUDA_ACC_TYPE(int32_t, int64_t);
+CUDA_ACC_TYPE(int64_t, int64_t);
+CUDA_ACC_TYPE(bool, bool);
+CUDA_ACC_TYPE(c10::complex<Half>, c10::complex<float>);
+CUDA_ACC_TYPE(c10::complex<float>, c10::complex<float>);
+CUDA_ACC_TYPE(c10::complex<double>, c10::complex<double>);
+
+CPU_ACC_TYPE(BFloat16, float);
+CPU_ACC_TYPE(Half, float);
+CPU_ACC_TYPE(Float8_e5m2, float);
+CPU_ACC_TYPE(Float8_e4m3fn, float);
+CPU_ACC_TYPE(Float8_e5m2fnuz, float);
+CPU_ACC_TYPE(Float8_e4m3fnuz, float);
+CPU_ACC_TYPE(float, double);
+CPU_ACC_TYPE(double, double);
+CPU_ACC_TYPE(int8_t, int64_t);
+CPU_ACC_TYPE(uint8_t, int64_t);
+CPU_ACC_TYPE(char, int64_t);
+CPU_ACC_TYPE(int16_t, int64_t);
+CPU_ACC_TYPE(int32_t, int64_t);
+CPU_ACC_TYPE(int64_t, int64_t);
+CPU_ACC_TYPE(bool, bool);
+CPU_ACC_TYPE(c10::complex<Half>, c10::complex<float>);
+CPU_ACC_TYPE(c10::complex<float>, c10::complex<double>);
+CPU_ACC_TYPE(c10::complex<double>, c10::complex<double>);
+
+TORCH_API c10::ScalarType toAccumulateType(
+    c10::ScalarType type,
+    c10::DeviceType device);
+TORCH_API c10::ScalarType toAccumulateType(c10::ScalarType type, bool is_cuda);
+
+} // namespace at

.venv/Lib/site-packages/torch/include/ATen/ArrayRef.h

@@ -0,0 +1,2 @@
+#pragma once
+#include <c10/util/ArrayRef.h>

.venv/Lib/site-packages/torch/include/ATen/Backend.h

@@ -0,0 +1,2 @@
+#pragma once
+#include <c10/core/Backend.h>

.venv/Lib/site-packages/torch/include/ATen/CPUApplyUtils.h

@@ -0,0 +1,343 @@
+#pragma once
+
+#include <ATen/CollapseDims.h>
+#include <ATen/Parallel.h>
+#include <ATen/TensorUtils.h>
+#include <c10/util/irange.h>
+#include <cstring>
+#include <limits>
+
+namespace at {
+
+/*
+ * The basic strategy for apply is as follows:
+ *
+ * 1. Starting with the outermost index, loop until we reach a dimension where
+ * the data is no longer contiguous, i.e. the stride at that dimension is not
+ * equal to the size of the tensor defined by the outer dimensions. Let's call
+ * this outer (contiguous) tensor A. Note that if the Tensor is contiguous, then
+ * A is equal to the entire Tensor. Let's call the inner tensor B.
+ *
+ * 2. We loop through the indices in B, starting at its outermost dimension. For
+ * example, if B is a 2x2 matrix, then we do:
+ *
+ * B[0][0]
+ * B[0][1]
+ * B[1][0]
+ * B[1][1]
+ *
+ * We set the offset into the underlying storage as (storageOffset + stride_B *
+ * index_B), i.e. basically we compute the offset into the storage as we would
+ * normally for a Tensor. But because we are guaranteed the subsequent data is
+ * contiguous in memory, we can simply loop for sizeof(A) iterations and perform
+ * the operation, without having to follow the order described by the strides of
+ * A.
+ *
+ * 3. As an optimization, we merge dimensions of A that are contiguous in
+ * memory. For example, if A is a 3x3x3x3 tensor narrowed from a 3x3x4x3 tensor,
+ * then the first two dimensions can be merged for the purposes of APPLY,
+ * reducing the number of nested loops.
+ */
+
+inline Tensor sort_strides(Tensor& tensor_) {
+  IntArrayRef strides = tensor_.strides();
+  std::vector<int64_t> indices;
+  indices.reserve(tensor_.ndimension());
+  for (const auto i : c10::irange(tensor_.ndimension())) {
+    indices.push_back(i);
+  }
+  std::sort(indices.begin(), indices.end(), [&strides](int64_t i1, int64_t i2) {
+    return strides[i1] > strides[i2];
+  });
+  Tensor tensor = tensor_.permute(indices);
+  return tensor;
+}
+
+template <typename T, int N>
+struct strided_tensor_iter_fixed {
+ public:
+  T* data_ = NULL;
+  int64_t dim_ = 0;
+
+  int64_t counter_[N] = {0};
+  int64_t sizes_[N] = {0};
+  int64_t strides_[N] = {0};
+
+  strided_tensor_iter_fixed(strided_tensor_iter_fixed const&) = delete;
+  void operator=(strided_tensor_iter_fixed const& x) = delete;
+  strided_tensor_iter_fixed(strided_tensor_iter_fixed&&) = default;
+  strided_tensor_iter_fixed(
+      Tensor& tensor,
+      C10_UNUSED bool sort_strides = false)
+      : data_(tensor.data_ptr<T>()) {
+    std::memset(counter_, 0, sizeof(int64_t) * N);
+    if (tensor.dim() > 0) {
+      std::memcpy(
+          sizes_, tensor.sizes().data(), tensor.dim() * sizeof(int64_t));
+      std::memcpy(
+          strides_, tensor.strides().data(), tensor.dim() * sizeof(int64_t));
+    }
+    dim_ = std::get<1>(collapse_dims(sizes_, strides_, tensor.ndimension()));
+  }
+};
+
+template <typename T>
+struct strided_tensor_iter {
+ private:
+ public:
+  T* data_ = NULL;
+  int64_t dim_;
+
+  std::vector<int64_t> counter_;
+  std::vector<int64_t> sizes_;
+  std::vector<int64_t> strides_;
+
+  strided_tensor_iter(strided_tensor_iter const&) = delete;
+  void operator=(strided_tensor_iter const& x) = delete;
+  strided_tensor_iter(strided_tensor_iter&&) = default;
+  strided_tensor_iter(Tensor& tensor)
+      : data_(tensor.data_ptr<T>()),
+        dim_(tensor.ndimension()),
+        counter_(dim_, 0),
+        sizes_(tensor.sizes().vec()),
+        strides_(tensor.strides().vec()) {
+    dim_ = std::get<1>(collapse_dims(sizes_.data(), strides_.data(), dim_));
+  }
+};
+
+inline bool _all_equal_numel(at::ArrayRef<Tensor> tensors) {
+  if (tensors.empty())
+    return true;
+  int64_t all_numel = tensors[0].numel();
+  for (const auto i : c10::irange(1, tensors.size())) {
+    if (tensors[i].numel() != all_numel)
+      return false;
+  }
+  return true;
+}
+
+inline std::string _all_equal_numel_error(at::ArrayRef<Tensor> tensors) {
+  std::ostringstream oss;
+  oss << "inconsistent tensor size, expected ";
+  for (size_t i = 0; i < tensors.size() - 1; i++) {
+    oss << tensors[i].sizes() << ", ";
+  }
+  oss << "and " << tensors[tensors.size() - 1].sizes()
+      << " to have the same number of elements, but got ";
+  for (size_t i = 0; i < tensors.size() - 1; i++) {
+    oss << tensors[i].numel() << ", ";
+  }
+  oss << "and " << tensors[tensors.size() - 1].numel()
+      << " elements respectively";
+  return oss.str();
+}
+
+inline bool _apply_preamble(ArrayRef<Tensor> tensors) {
+  checkDeviceType("CPU_tensor_apply", tensors, kCPU);
+  checkLayout("CPU_tensor_apply", tensors, kStrided);
+  if (!_all_equal_numel(tensors))
+    AT_ERROR(_all_equal_numel_error(tensors));
+  // An empty tensor has no elements
+  for (auto& t : tensors)
+    if (t.numel() == 0)
+      return false;
+  return true;
+}
+
+inline int64_t _max_dim_tensors(ArrayRef<Tensor> tensors) {
+  int64_t dim = 0;
+  for (auto& t : tensors)
+    dim = std::max(dim, t.ndimension());
+  return dim;
+}
+
+inline void iterate(int64_t /*size*/){};
+
+template <typename Arg, typename... Args>
+inline void iterate(int64_t size, Arg& iter, Args&... iter_tail) {
+  iter.counter_[iter.dim_ - 1] += size;
+  iter.data_ = iter.data_ + size * iter.strides_[iter.dim_ - 1];
+  iterate(size, iter_tail...);
+}
+
+inline bool iterate_continue() {
+  return true;
+};
+
+template <typename Arg, typename... Args>
+inline bool iterate_continue(Arg& iter, Args&... iter_tail) {
+  return iter.counter_[iter.dim_ - 1] < iter.sizes_[iter.dim_ - 1] &&
+      iterate_continue(iter_tail...);
+}
+
+inline int64_t max_iterate_size() {
+  return std::numeric_limits<int64_t>::max();
+};
+
+template <typename Arg, typename... Args>
+inline int64_t max_iterate_size(Arg& iter, Args&... iter_tail) {
+  return std::min(
+      (iter.sizes_[iter.dim_ - 1] - iter.counter_[iter.dim_ - 1]),
+      max_iterate_size(iter_tail...));
+}
+
+inline void iterate_overflow(){};
+
+template <typename Arg, typename... Args>
+inline void iterate_overflow(Arg& iter, Args&... iter_tail) {
+  if (iter.counter_[iter.dim_ - 1] == iter.sizes_[iter.dim_ - 1]) {
+    for (int64_t i = iter.dim_ - 1; i > 0; i--) {
+      if (iter.counter_[i] == iter.sizes_[i]) {
+        iter.counter_[i] = 0;
+        iter.counter_[i - 1]++;
+        iter.data_ = iter.data_ - (iter.sizes_[i] * iter.strides_[i]) +
+            iter.strides_[i - 1];
+      }
+    }
+  }
+  iterate_overflow(iter_tail...);
+}
+
+inline void forward(int64_t /*offset*/){};
+
+template <typename Arg, typename... Args>
+inline void forward(int64_t offset, Arg& iter, Args&... iter_tail) {
+  int64_t multi = offset;
+  for (int64_t i = iter.dim_ - 1; i >= 0; i--) {
+    int64_t inc = multi % iter.sizes_[i];
+    multi = multi / iter.sizes_[i];
+    iter.data_ = iter.data_ + inc * iter.strides_[i];
+    iter.counter_[i] += inc;
+  }
+  forward(offset, iter_tail...);
+}
+
+inline int64_t max_dim() {
+  return 0;
+}
+
+template <typename Arg, typename... Args>
+inline int64_t max_dim(Arg& iter, Args&... iter_tail) {
+  return std::max(iter.dim_, max_dim(iter_tail...));
+}
+
+inline void apply_op(){};
+
+template <typename Op, typename... Args>
+inline void apply_op(
+    int64_t numel,
+    int64_t offset,
+    const Op& op,
+    Args... iters) {
+  // For 0-dim tensors
+  if (numel == 1 && max_dim(iters...) == 0) {
+    op(*iters.data_...);
+    return;
+  }
+  if (offset > 0)
+    forward(offset, iters...);
+  // Splitting this into chunks helps the compiler create faster assembly
+  for (int64_t i = 0; i < numel;) {
+    for (; iterate_continue(iters...) && i < numel;) {
+      op(*iters.data_...);
+      iterate(1, iters...);
+      i++;
+    }
+    iterate_overflow(iters...);
+  }
+}
+
+/*
+  Apply a pointwise operator to sequence of tensors
+
+  The calling convention for op is a function/functor that takes the same
+  number of pointers of type scalar as the number of given tensors. For example,
+  to compute a = b * c, op would be of the form:
+  [](scalar* a_val, const scalar* b_val, const scalar* c_val) { a_val[0] =
+  b_val[0] * c_val[0]; };
+*/
+
+template <typename scalar1, typename scalar2, typename Op>
+inline void CPU_tensor_apply2(Tensor tensor1, Tensor tensor2, const Op op) {
+  if (!_apply_preamble({tensor1, tensor2}))
+    return;
+  if (_max_dim_tensors({tensor1, tensor2}) <= 8) {
+    apply_op(
+        tensor1.numel(),
+        0,
+        op,
+        strided_tensor_iter_fixed<scalar1, 8>(tensor1),
+        strided_tensor_iter_fixed<scalar2, 8>(tensor2));
+  } else {
+    apply_op(
+        tensor1.numel(),
+        0,
+        op,
+        strided_tensor_iter<scalar1>(tensor1),
+        strided_tensor_iter<scalar2>(tensor2));
+  }
+}
+
+template <typename scalar1, typename scalar2, typename scalar3, typename Op>
+inline void CPU_tensor_apply3(
+    Tensor tensor1,
+    Tensor tensor2,
+    Tensor tensor3,
+    const Op op) {
+  if (!_apply_preamble({tensor1, tensor2, tensor3}))
+    return;
+  if (_max_dim_tensors({tensor1, tensor2, tensor3}) <= 8) {
+    apply_op(
+        tensor1.numel(),
+        0,
+        op,
+        strided_tensor_iter_fixed<scalar1, 8>(tensor1),
+        strided_tensor_iter_fixed<scalar2, 8>(tensor2),
+        strided_tensor_iter_fixed<scalar3, 8>(tensor3));
+  } else {
+    apply_op(
+        tensor1.numel(),
+        0,
+        op,
+        strided_tensor_iter<scalar1>(tensor1),
+        strided_tensor_iter<scalar2>(tensor2),
+        strided_tensor_iter<scalar3>(tensor3));
+  }
+}
+
+template <
+    typename scalar1,
+    typename scalar2,
+    typename scalar3,
+    typename scalar4,
+    typename Op>
+inline void CPU_tensor_apply4(
+    Tensor tensor1,
+    Tensor tensor2,
+    Tensor tensor3,
+    Tensor tensor4,
+    const Op op) {
+  if (!_apply_preamble({tensor1, tensor2, tensor3, tensor4}))
+    return;
+  if (_max_dim_tensors({tensor1, tensor2, tensor3, tensor4}) <= 8) {
+    apply_op(
+        tensor1.numel(),
+        0,
+        op,
+        strided_tensor_iter_fixed<scalar1, 8>(tensor1),
+        strided_tensor_iter_fixed<scalar2, 8>(tensor2),
+        strided_tensor_iter_fixed<scalar3, 8>(tensor3),
+        strided_tensor_iter_fixed<scalar4, 8>(tensor4));
+  } else {
+    apply_op(
+        tensor1.numel(),
+        0,
+        op,
+        strided_tensor_iter<scalar1>(tensor1),
+        strided_tensor_iter<scalar2>(tensor2),
+        strided_tensor_iter<scalar3>(tensor3),
+        strided_tensor_iter<scalar4>(tensor4));
+  }
+}
+
+} // namespace at

.venv/Lib/site-packages/torch/include/ATen/CPUFixedAllocator.h

@@ -0,0 +1,33 @@
+#pragma once
+
+#include <c10/core/Allocator.h>
+#include <c10/util/Exception.h>
+
+// This file creates a fake allocator that just throws exceptions if
+// it is actually used.
+
+// state passed to the allocator is the std::function<void(void*)> called
+// when the blob is release by ATen
+
+namespace at {
+
+static cpu_fixed_malloc(void*, ptrdiff_t) {
+  AT_ERROR("attempting to resize a tensor view of an external blob");
+}
+
+static cpu_fixed_realloc(void*, void*, ptrdiff_t) {
+  AT_ERROR("attempting to resize a tensor view of an external blob");
+}
+
+static cpu_fixed_free(void* state, void* allocation) {
+  auto on_release = static_cast<std::function<void(void*)>*>(state);
+  (*on_release)(allocation);
+  delete on_release;
+}
+
+static Allocator CPU_fixed_allocator = {
+    cpu_fixed_malloc,
+    cpu_fixed_realloc,
+    cpu_fixed_free};
+
+} // namespace at

.venv/Lib/site-packages/torch/include/ATen/CPUFunctions.h

@@ -0,0 +1,29 @@
+#include <ATen/core/TensorBody.h>
+
+// TODO Undo all logic introduced for Note [Avoiding Include Cycles In Static Dispatch]
+// Code introduced to avoid cyclic dependency in static dispatch is no longer
+// needed as static dispatch logic is moved from TensorBody.h, which caused cycles in the first place,
+// to Operators.cpp for supporting multiple backends with multiple kernels.
+//
+// Note [Avoiding Include Cycles In Static Dispatch]
+// In order to avoid #include cycles in the static dispatch build, we've carefully split out
+// the static function definition files into {DispatchKey}Functions.h and {DispatchKey}Functions_inl.h.
+//
+// Without this split, the include cycle looks like TensorBody.h -> CPUFunctions.h -> TensorBody.h.
+// - TensorBody.h #includes CPUFunctions.h in the static dispatch build, because the tensor methods
+//   all need to call into the fastpath C++ API defined in CPUFunctions.h. The methods are also all
+//   directly inlined into TensorBody.h.
+// - CPUFunctions.h #includes TensorBody.h because it contains function declarations for the entire C++ API,
+//   which include functions that have defaultable std::optional<Tensor> arguments.
+//   That requires knowing the full Tensor class definition.
+//
+// We break the cycle by doing the following:
+// - Split out CPUFunction.h into two files: CPUFunctions.h and CPUFunctions_inl.h
+// - CPUFunction.h is a dummy file that just includes the Tensor class and includes CPUFunctions_inl.,
+// - CPUFunctions_inl.h includes everything else
+// - (only in the static dispatch build) TensorBody.h makes sure to finish defining the Tensor class,
+//   and then it includes CPUFunctions_inl.h.
+// - All other files that want the cpu fastpath functions can include CPUFunctions.h directly.
+// - This also means that static dispatch build, CPUFunctions.h only needs to
+//   #include TensorBody.h, and it will automatically bring in CPUFunctions_inl.h.
+#include <ATen/CPUFunctions_inl.h>

.venv/Lib/site-packages/torch/include/ATen/CPUFunctions_inl.h

@@ -0,0 +1,540 @@
+#pragma once
+// @generated by torchgen/gen.py from DispatchKeyFunctions_inl.h
+
+// NB: The implementing C++ file is RegisterDispatchKey.cpp
+
+// The only #includes we need are for custom classes that have defaults in the C++ API
+#include <c10/core/MemoryFormat.h>
+#include <c10/core/Scalar.h>
+#include <ATen/core/Reduction.h>
+
+#if defined(AT_PER_OPERATOR_HEADERS) && defined(TORCH_ASSERT_ONLY_METHOD_OPERATORS)
+#error This change adds a dependency on all pytorch operators, meaning the     \
+  file will need to be re-compiled every time an operator is changed or added. \
+  Consider including a specific operator from                                  \
+  <ATen/ops/{my_operator}_cpu_dispatch.h>.                   \
+  See NOTE [TORCH_ASSERT_ONLY_METHOD_OPERATORS].
+#endif
+
+#include <ATen/ops/_adaptive_avg_pool2d_cpu_dispatch.h>
+#include <ATen/ops/_adaptive_avg_pool2d_backward_cpu_dispatch.h>
+#include <ATen/ops/_adaptive_avg_pool3d_cpu_dispatch.h>
+#include <ATen/ops/_adaptive_avg_pool3d_backward_cpu_dispatch.h>
+#include <ATen/ops/_add_relu_cpu_dispatch.h>
+#include <ATen/ops/_addmm_activation_cpu_dispatch.h>
+#include <ATen/ops/_aminmax_cpu_dispatch.h>
+#include <ATen/ops/_amp_foreach_non_finite_check_and_unscale_cpu_dispatch.h>
+#include <ATen/ops/_amp_update_scale_cpu_dispatch.h>
+#include <ATen/ops/_assert_async_cpu_dispatch.h>
+#include <ATen/ops/_batch_norm_with_update_cpu_dispatch.h>
+#include <ATen/ops/_cdist_backward_cpu_dispatch.h>
+#include <ATen/ops/_cdist_forward_cpu_dispatch.h>
+#include <ATen/ops/_cholesky_solve_helper_cpu_dispatch.h>
+#include <ATen/ops/_compute_linear_combination_cpu_dispatch.h>
+#include <ATen/ops/_convert_indices_from_coo_to_csr_cpu_dispatch.h>
+#include <ATen/ops/_convert_indices_from_csr_to_coo_cpu_dispatch.h>
+#include <ATen/ops/_convert_weight_to_int4pack_cpu_dispatch.h>
+#include <ATen/ops/_ctc_loss_cpu_dispatch.h>
+#include <ATen/ops/_ctc_loss_backward_cpu_dispatch.h>
+#include <ATen/ops/_cummax_helper_cpu_dispatch.h>
+#include <ATen/ops/_cummin_helper_cpu_dispatch.h>
+#include <ATen/ops/_dirichlet_grad_cpu_dispatch.h>
+#include <ATen/ops/_efficientzerotensor_cpu_dispatch.h>
+#include <ATen/ops/_embedding_bag_cpu_dispatch.h>
+#include <ATen/ops/_embedding_bag_backward_cpu_dispatch.h>
+#include <ATen/ops/_embedding_bag_dense_backward_cpu_dispatch.h>
+#include <ATen/ops/_embedding_bag_forward_only_cpu_dispatch.h>
+#include <ATen/ops/_embedding_bag_per_sample_weights_backward_cpu_dispatch.h>
+#include <ATen/ops/_empty_affine_quantized_cpu_dispatch.h>
+#include <ATen/ops/_empty_per_channel_affine_quantized_cpu_dispatch.h>
+#include <ATen/ops/_fake_quantize_learnable_per_channel_affine_cpu_dispatch.h>
+#include <ATen/ops/_fake_quantize_learnable_per_channel_affine_backward_cpu_dispatch.h>
+#include <ATen/ops/_fake_quantize_learnable_per_tensor_affine_cpu_dispatch.h>
+#include <ATen/ops/_fake_quantize_learnable_per_tensor_affine_backward_cpu_dispatch.h>
+#include <ATen/ops/_fake_quantize_per_tensor_affine_cachemask_tensor_qparams_cpu_dispatch.h>
+#include <ATen/ops/_fft_c2c_cpu_dispatch.h>
+#include <ATen/ops/_fft_c2r_cpu_dispatch.h>
+#include <ATen/ops/_fft_r2c_cpu_dispatch.h>
+#include <ATen/ops/_foobar_cpu_dispatch.h>
+#include <ATen/ops/_functional_assert_async_cpu_dispatch.h>
+#include <ATen/ops/_fused_adagrad_cpu_dispatch.h>
+#include <ATen/ops/_fused_adam_cpu_dispatch.h>
+#include <ATen/ops/_fused_adamw_cpu_dispatch.h>
+#include <ATen/ops/_fused_moving_avg_obs_fq_helper_cpu_dispatch.h>
+#include <ATen/ops/_fused_sdp_choice_cpu_dispatch.h>
+#include <ATen/ops/_fused_sgd_cpu_dispatch.h>
+#include <ATen/ops/_histogramdd_bin_edges_cpu_dispatch.h>
+#include <ATen/ops/_histogramdd_from_bin_cts_cpu_dispatch.h>
+#include <ATen/ops/_histogramdd_from_bin_tensors_cpu_dispatch.h>
+#include <ATen/ops/_index_put_impl_cpu_dispatch.h>
+#include <ATen/ops/_int_mm_cpu_dispatch.h>
+#include <ATen/ops/_jagged_to_padded_dense_forward_cpu_dispatch.h>
+#include <ATen/ops/_linalg_det_cpu_dispatch.h>
+#include <ATen/ops/_linalg_eigh_cpu_dispatch.h>
+#include <ATen/ops/_linalg_eigvals_cpu_dispatch.h>
+#include <ATen/ops/_linalg_slogdet_cpu_dispatch.h>
+#include <ATen/ops/_linalg_solve_ex_cpu_dispatch.h>
+#include <ATen/ops/_linalg_svd_cpu_dispatch.h>
+#include <ATen/ops/_local_scalar_dense_cpu_dispatch.h>
+#include <ATen/ops/_log_softmax_cpu_dispatch.h>
+#include <ATen/ops/_log_softmax_backward_data_cpu_dispatch.h>
+#include <ATen/ops/_logcumsumexp_cpu_dispatch.h>
+#include <ATen/ops/_make_dep_token_cpu_dispatch.h>
+#include <ATen/ops/_make_per_channel_quantized_tensor_cpu_dispatch.h>
+#include <ATen/ops/_make_per_tensor_quantized_tensor_cpu_dispatch.h>
+#include <ATen/ops/_masked_softmax_cpu_dispatch.h>
+#include <ATen/ops/_masked_softmax_backward_cpu_dispatch.h>
+#include <ATen/ops/_native_batch_norm_legit_cpu_dispatch.h>
+#include <ATen/ops/_native_multi_head_attention_cpu_dispatch.h>
+#include <ATen/ops/_nested_compute_contiguous_strides_offsets_cpu_dispatch.h>
+#include <ATen/ops/_nested_from_padded_cpu_dispatch.h>
+#include <ATen/ops/_nested_tensor_from_mask_cpu_dispatch.h>
+#include <ATen/ops/_nested_tensor_from_mask_left_aligned_cpu_dispatch.h>
+#include <ATen/ops/_nested_view_from_buffer_cpu_dispatch.h>
+#include <ATen/ops/_padded_dense_to_jagged_forward_cpu_dispatch.h>
+#include <ATen/ops/_pdist_backward_cpu_dispatch.h>
+#include <ATen/ops/_pdist_forward_cpu_dispatch.h>
+#include <ATen/ops/_prelu_kernel_cpu_dispatch.h>
+#include <ATen/ops/_prelu_kernel_backward_cpu_dispatch.h>
+#include <ATen/ops/_reshape_alias_cpu_dispatch.h>
+#include <ATen/ops/_sample_dirichlet_cpu_dispatch.h>
+#include <ATen/ops/_scaled_dot_product_flash_attention_for_cpu_cpu_dispatch.h>
+#include <ATen/ops/_scaled_dot_product_flash_attention_for_cpu_backward_cpu_dispatch.h>
+#include <ATen/ops/_segment_reduce_backward_cpu_dispatch.h>
+#include <ATen/ops/_slow_conv2d_backward_cpu_dispatch.h>
+#include <ATen/ops/_slow_conv2d_forward_cpu_dispatch.h>
+#include <ATen/ops/_softmax_cpu_dispatch.h>
+#include <ATen/ops/_softmax_backward_data_cpu_dispatch.h>
+#include <ATen/ops/_spdiags_cpu_dispatch.h>
+#include <ATen/ops/_stack_cpu_dispatch.h>
+#include <ATen/ops/_standard_gamma_cpu_dispatch.h>
+#include <ATen/ops/_standard_gamma_grad_cpu_dispatch.h>
+#include <ATen/ops/_test_functorch_fallback_cpu_dispatch.h>
+#include <ATen/ops/_test_optional_filled_intlist_cpu_dispatch.h>
+#include <ATen/ops/_test_optional_floatlist_cpu_dispatch.h>
+#include <ATen/ops/_test_optional_intlist_cpu_dispatch.h>
+#include <ATen/ops/_to_sparse_cpu_dispatch.h>
+#include <ATen/ops/_to_sparse_bsc_cpu_dispatch.h>
+#include <ATen/ops/_to_sparse_bsr_cpu_dispatch.h>
+#include <ATen/ops/_to_sparse_csc_cpu_dispatch.h>
+#include <ATen/ops/_to_sparse_csr_cpu_dispatch.h>
+#include <ATen/ops/_transform_bias_rescale_qkv_cpu_dispatch.h>
+#include <ATen/ops/_transformer_encoder_layer_fwd_cpu_dispatch.h>
+#include <ATen/ops/_unique_cpu_dispatch.h>
+#include <ATen/ops/_unique2_cpu_dispatch.h>
+#include <ATen/ops/_upsample_bicubic2d_aa_cpu_dispatch.h>
+#include <ATen/ops/_upsample_bicubic2d_aa_backward_cpu_dispatch.h>
+#include <ATen/ops/_upsample_bilinear2d_aa_cpu_dispatch.h>
+#include <ATen/ops/_upsample_bilinear2d_aa_backward_cpu_dispatch.h>
+#include <ATen/ops/_upsample_nearest_exact1d_cpu_dispatch.h>
+#include <ATen/ops/_upsample_nearest_exact1d_backward_cpu_dispatch.h>
+#include <ATen/ops/_upsample_nearest_exact2d_cpu_dispatch.h>
+#include <ATen/ops/_upsample_nearest_exact2d_backward_cpu_dispatch.h>
+#include <ATen/ops/_upsample_nearest_exact3d_cpu_dispatch.h>
+#include <ATen/ops/_upsample_nearest_exact3d_backward_cpu_dispatch.h>
+#include <ATen/ops/_validate_compressed_sparse_indices_cpu_dispatch.h>
+#include <ATen/ops/_weight_int4pack_mm_cpu_dispatch.h>
+#include <ATen/ops/_weight_int8pack_mm_cpu_dispatch.h>
+#include <ATen/ops/_weight_norm_interface_cpu_dispatch.h>
+#include <ATen/ops/_weight_norm_interface_backward_cpu_dispatch.h>
+#include <ATen/ops/abs_cpu_dispatch.h>
+#include <ATen/ops/acos_cpu_dispatch.h>
+#include <ATen/ops/acosh_cpu_dispatch.h>
+#include <ATen/ops/adaptive_avg_pool2d_cpu_dispatch.h>
+#include <ATen/ops/adaptive_avg_pool3d_cpu_dispatch.h>
+#include <ATen/ops/adaptive_avg_pool3d_backward_cpu_dispatch.h>
+#include <ATen/ops/adaptive_max_pool2d_cpu_dispatch.h>
+#include <ATen/ops/adaptive_max_pool2d_backward_cpu_dispatch.h>
+#include <ATen/ops/adaptive_max_pool3d_cpu_dispatch.h>
+#include <ATen/ops/adaptive_max_pool3d_backward_cpu_dispatch.h>
+#include <ATen/ops/add_cpu_dispatch.h>
+#include <ATen/ops/addbmm_cpu_dispatch.h>
+#include <ATen/ops/addcdiv_cpu_dispatch.h>
+#include <ATen/ops/addcmul_cpu_dispatch.h>
+#include <ATen/ops/addmm_cpu_dispatch.h>
+#include <ATen/ops/addmv_cpu_dispatch.h>
+#include <ATen/ops/addr_cpu_dispatch.h>
+#include <ATen/ops/all_cpu_dispatch.h>
+#include <ATen/ops/amax_cpu_dispatch.h>
+#include <ATen/ops/amin_cpu_dispatch.h>
+#include <ATen/ops/aminmax_cpu_dispatch.h>
+#include <ATen/ops/angle_cpu_dispatch.h>
+#include <ATen/ops/any_cpu_dispatch.h>
+#include <ATen/ops/arange_cpu_dispatch.h>
+#include <ATen/ops/argmax_cpu_dispatch.h>
+#include <ATen/ops/argmin_cpu_dispatch.h>
+#include <ATen/ops/as_strided_cpu_dispatch.h>
+#include <ATen/ops/asin_cpu_dispatch.h>
+#include <ATen/ops/asinh_cpu_dispatch.h>
+#include <ATen/ops/atan_cpu_dispatch.h>
+#include <ATen/ops/atan2_cpu_dispatch.h>
+#include <ATen/ops/atanh_cpu_dispatch.h>
+#include <ATen/ops/avg_pool2d_cpu_dispatch.h>
+#include <ATen/ops/avg_pool2d_backward_cpu_dispatch.h>
+#include <ATen/ops/avg_pool3d_cpu_dispatch.h>
+#include <ATen/ops/avg_pool3d_backward_cpu_dispatch.h>
+#include <ATen/ops/baddbmm_cpu_dispatch.h>
+#include <ATen/ops/batch_norm_backward_cpu_dispatch.h>
+#include <ATen/ops/batch_norm_update_stats_cpu_dispatch.h>
+#include <ATen/ops/bernoulli_cpu_dispatch.h>
+#include <ATen/ops/binary_cross_entropy_cpu_dispatch.h>
+#include <ATen/ops/binary_cross_entropy_backward_cpu_dispatch.h>
+#include <ATen/ops/bincount_cpu_dispatch.h>
+#include <ATen/ops/binomial_cpu_dispatch.h>
+#include <ATen/ops/bitwise_and_cpu_dispatch.h>
+#include <ATen/ops/bitwise_left_shift_cpu_dispatch.h>
+#include <ATen/ops/bitwise_not_cpu_dispatch.h>
+#include <ATen/ops/bitwise_or_cpu_dispatch.h>
+#include <ATen/ops/bitwise_right_shift_cpu_dispatch.h>
+#include <ATen/ops/bitwise_xor_cpu_dispatch.h>
+#include <ATen/ops/bmm_cpu_dispatch.h>
+#include <ATen/ops/bucketize_cpu_dispatch.h>
+#include <ATen/ops/cat_cpu_dispatch.h>
+#include <ATen/ops/cauchy_cpu_dispatch.h>
+#include <ATen/ops/ceil_cpu_dispatch.h>
+#include <ATen/ops/channel_shuffle_cpu_dispatch.h>
+#include <ATen/ops/cholesky_cpu_dispatch.h>
+#include <ATen/ops/cholesky_inverse_cpu_dispatch.h>
+#include <ATen/ops/clamp_cpu_dispatch.h>
+#include <ATen/ops/clamp_max_cpu_dispatch.h>
+#include <ATen/ops/clamp_min_cpu_dispatch.h>
+#include <ATen/ops/col2im_cpu_dispatch.h>
+#include <ATen/ops/complex_cpu_dispatch.h>
+#include <ATen/ops/conj_physical_cpu_dispatch.h>
+#include <ATen/ops/copysign_cpu_dispatch.h>
+#include <ATen/ops/cos_cpu_dispatch.h>
+#include <ATen/ops/cosh_cpu_dispatch.h>
+#include <ATen/ops/count_nonzero_cpu_dispatch.h>
+#include <ATen/ops/cumprod_cpu_dispatch.h>
+#include <ATen/ops/cumsum_cpu_dispatch.h>
+#include <ATen/ops/dequantize_cpu_dispatch.h>
+#include <ATen/ops/digamma_cpu_dispatch.h>
+#include <ATen/ops/div_cpu_dispatch.h>
+#include <ATen/ops/dot_cpu_dispatch.h>
+#include <ATen/ops/elu_cpu_dispatch.h>
+#include <ATen/ops/elu_backward_cpu_dispatch.h>
+#include <ATen/ops/embedding_dense_backward_cpu_dispatch.h>
+#include <ATen/ops/embedding_renorm_cpu_dispatch.h>
+#include <ATen/ops/empty_cpu_dispatch.h>
+#include <ATen/ops/empty_strided_cpu_dispatch.h>
+#include <ATen/ops/eq_cpu_dispatch.h>
+#include <ATen/ops/equal_cpu_dispatch.h>
+#include <ATen/ops/erf_cpu_dispatch.h>
+#include <ATen/ops/erfc_cpu_dispatch.h>
+#include <ATen/ops/erfinv_cpu_dispatch.h>
+#include <ATen/ops/exp_cpu_dispatch.h>
+#include <ATen/ops/exp2_cpu_dispatch.h>
+#include <ATen/ops/expm1_cpu_dispatch.h>
+#include <ATen/ops/exponential_cpu_dispatch.h>
+#include <ATen/ops/eye_cpu_dispatch.h>
+#include <ATen/ops/fake_quantize_per_channel_affine_cachemask_cpu_dispatch.h>
+#include <ATen/ops/fake_quantize_per_tensor_affine_cachemask_cpu_dispatch.h>
+#include <ATen/ops/fill_cpu_dispatch.h>
+#include <ATen/ops/flip_cpu_dispatch.h>
+#include <ATen/ops/floor_cpu_dispatch.h>
+#include <ATen/ops/floor_divide_cpu_dispatch.h>
+#include <ATen/ops/fmax_cpu_dispatch.h>
+#include <ATen/ops/fmin_cpu_dispatch.h>
+#include <ATen/ops/fmod_cpu_dispatch.h>
+#include <ATen/ops/frac_cpu_dispatch.h>
+#include <ATen/ops/fractional_max_pool2d_cpu_dispatch.h>
+#include <ATen/ops/fractional_max_pool2d_backward_cpu_dispatch.h>
+#include <ATen/ops/fractional_max_pool3d_cpu_dispatch.h>
+#include <ATen/ops/fractional_max_pool3d_backward_cpu_dispatch.h>
+#include <ATen/ops/frexp_cpu_dispatch.h>
+#include <ATen/ops/from_file_cpu_dispatch.h>
+#include <ATen/ops/gather_cpu_dispatch.h>
+#include <ATen/ops/gcd_cpu_dispatch.h>
+#include <ATen/ops/ge_cpu_dispatch.h>
+#include <ATen/ops/gelu_cpu_dispatch.h>
+#include <ATen/ops/gelu_backward_cpu_dispatch.h>
+#include <ATen/ops/geometric_cpu_dispatch.h>
+#include <ATen/ops/geqrf_cpu_dispatch.h>
+#include <ATen/ops/glu_cpu_dispatch.h>
+#include <ATen/ops/glu_backward_cpu_dispatch.h>
+#include <ATen/ops/glu_backward_jvp_cpu_dispatch.h>
+#include <ATen/ops/glu_jvp_cpu_dispatch.h>
+#include <ATen/ops/grid_sampler_2d_cpu_dispatch.h>
+#include <ATen/ops/grid_sampler_2d_backward_cpu_dispatch.h>
+#include <ATen/ops/grid_sampler_3d_cpu_dispatch.h>
+#include <ATen/ops/grid_sampler_3d_backward_cpu_dispatch.h>
+#include <ATen/ops/gt_cpu_dispatch.h>
+#include <ATen/ops/hardshrink_cpu_dispatch.h>
+#include <ATen/ops/hardshrink_backward_cpu_dispatch.h>
+#include <ATen/ops/hardsigmoid_cpu_dispatch.h>
+#include <ATen/ops/hardsigmoid_backward_cpu_dispatch.h>
+#include <ATen/ops/hardswish_cpu_dispatch.h>
+#include <ATen/ops/hardswish_backward_cpu_dispatch.h>
+#include <ATen/ops/hardtanh_cpu_dispatch.h>
+#include <ATen/ops/hardtanh_backward_cpu_dispatch.h>
+#include <ATen/ops/heaviside_cpu_dispatch.h>
+#include <ATen/ops/histc_cpu_dispatch.h>
+#include <ATen/ops/histogram_cpu_dispatch.h>
+#include <ATen/ops/huber_loss_cpu_dispatch.h>
+#include <ATen/ops/huber_loss_backward_cpu_dispatch.h>
+#include <ATen/ops/hypot_cpu_dispatch.h>
+#include <ATen/ops/i0_cpu_dispatch.h>
+#include <ATen/ops/igamma_cpu_dispatch.h>
+#include <ATen/ops/igammac_cpu_dispatch.h>
+#include <ATen/ops/im2col_cpu_dispatch.h>
+#include <ATen/ops/index_cpu_dispatch.h>
+#include <ATen/ops/index_add_cpu_dispatch.h>
+#include <ATen/ops/index_copy_cpu_dispatch.h>
+#include <ATen/ops/index_fill_cpu_dispatch.h>
+#include <ATen/ops/index_reduce_cpu_dispatch.h>
+#include <ATen/ops/index_select_cpu_dispatch.h>
+#include <ATen/ops/is_set_to_cpu_dispatch.h>
+#include <ATen/ops/isin_cpu_dispatch.h>
+#include <ATen/ops/isnan_cpu_dispatch.h>
+#include <ATen/ops/isneginf_cpu_dispatch.h>
+#include <ATen/ops/isposinf_cpu_dispatch.h>
+#include <ATen/ops/kthvalue_cpu_dispatch.h>
+#include <ATen/ops/lcm_cpu_dispatch.h>
+#include <ATen/ops/le_cpu_dispatch.h>
+#include <ATen/ops/leaky_relu_cpu_dispatch.h>
+#include <ATen/ops/leaky_relu_backward_cpu_dispatch.h>
+#include <ATen/ops/lerp_cpu_dispatch.h>
+#include <ATen/ops/lgamma_cpu_dispatch.h>
+#include <ATen/ops/linalg_cholesky_ex_cpu_dispatch.h>
+#include <ATen/ops/linalg_cross_cpu_dispatch.h>
+#include <ATen/ops/linalg_eig_cpu_dispatch.h>
+#include <ATen/ops/linalg_eigvals_cpu_dispatch.h>
+#include <ATen/ops/linalg_householder_product_cpu_dispatch.h>
+#include <ATen/ops/linalg_inv_ex_cpu_dispatch.h>
+#include <ATen/ops/linalg_ldl_factor_ex_cpu_dispatch.h>
+#include <ATen/ops/linalg_ldl_solve_cpu_dispatch.h>
+#include <ATen/ops/linalg_lstsq_cpu_dispatch.h>
+#include <ATen/ops/linalg_lu_cpu_dispatch.h>
+#include <ATen/ops/linalg_lu_factor_ex_cpu_dispatch.h>
+#include <ATen/ops/linalg_lu_solve_cpu_dispatch.h>
+#include <ATen/ops/linalg_matrix_exp_cpu_dispatch.h>
+#include <ATen/ops/linalg_qr_cpu_dispatch.h>
+#include <ATen/ops/linalg_solve_triangular_cpu_dispatch.h>
+#include <ATen/ops/linalg_vector_norm_cpu_dispatch.h>
+#include <ATen/ops/linspace_cpu_dispatch.h>
+#include <ATen/ops/log_cpu_dispatch.h>
+#include <ATen/ops/log10_cpu_dispatch.h>
+#include <ATen/ops/log1p_cpu_dispatch.h>
+#include <ATen/ops/log2_cpu_dispatch.h>
+#include <ATen/ops/log_normal_cpu_dispatch.h>
+#include <ATen/ops/log_sigmoid_backward_cpu_dispatch.h>
+#include <ATen/ops/log_sigmoid_forward_cpu_dispatch.h>
+#include <ATen/ops/logaddexp_cpu_dispatch.h>
+#include <ATen/ops/logaddexp2_cpu_dispatch.h>
+#include <ATen/ops/logical_and_cpu_dispatch.h>
+#include <ATen/ops/logical_not_cpu_dispatch.h>
+#include <ATen/ops/logical_or_cpu_dispatch.h>
+#include <ATen/ops/logical_xor_cpu_dispatch.h>
+#include <ATen/ops/logit_cpu_dispatch.h>
+#include <ATen/ops/logit_backward_cpu_dispatch.h>
+#include <ATen/ops/logspace_cpu_dispatch.h>
+#include <ATen/ops/lshift_cpu_dispatch.h>
+#include <ATen/ops/lt_cpu_dispatch.h>
+#include <ATen/ops/lu_unpack_cpu_dispatch.h>
+#include <ATen/ops/masked_fill_cpu_dispatch.h>
+#include <ATen/ops/masked_scatter_cpu_dispatch.h>
+#include <ATen/ops/masked_select_cpu_dispatch.h>
+#include <ATen/ops/max_cpu_dispatch.h>
+#include <ATen/ops/max_pool2d_with_indices_cpu_dispatch.h>
+#include <ATen/ops/max_pool2d_with_indices_backward_cpu_dispatch.h>
+#include <ATen/ops/max_pool3d_with_indices_cpu_dispatch.h>
+#include <ATen/ops/max_pool3d_with_indices_backward_cpu_dispatch.h>
+#include <ATen/ops/max_unpool2d_cpu_dispatch.h>
+#include <ATen/ops/max_unpool3d_cpu_dispatch.h>
+#include <ATen/ops/maximum_cpu_dispatch.h>
+#include <ATen/ops/mean_cpu_dispatch.h>
+#include <ATen/ops/median_cpu_dispatch.h>
+#include <ATen/ops/min_cpu_dispatch.h>
+#include <ATen/ops/minimum_cpu_dispatch.h>
+#include <ATen/ops/mish_cpu_dispatch.h>
+#include <ATen/ops/mish_backward_cpu_dispatch.h>
+#include <ATen/ops/mkldnn_rnn_layer_cpu_dispatch.h>
+#include <ATen/ops/mkldnn_rnn_layer_backward_cpu_dispatch.h>
+#include <ATen/ops/mm_cpu_dispatch.h>
+#include <ATen/ops/mode_cpu_dispatch.h>
+#include <ATen/ops/mse_loss_cpu_dispatch.h>
+#include <ATen/ops/mse_loss_backward_cpu_dispatch.h>
+#include <ATen/ops/mul_cpu_dispatch.h>
+#include <ATen/ops/multi_margin_loss_cpu_dispatch.h>
+#include <ATen/ops/multi_margin_loss_backward_cpu_dispatch.h>
+#include <ATen/ops/multilabel_margin_loss_backward_cpu_dispatch.h>
+#include <ATen/ops/multilabel_margin_loss_forward_cpu_dispatch.h>
+#include <ATen/ops/multinomial_cpu_dispatch.h>
+#include <ATen/ops/mvlgamma_cpu_dispatch.h>
+#include <ATen/ops/nan_to_num_cpu_dispatch.h>
+#include <ATen/ops/nanmedian_cpu_dispatch.h>
+#include <ATen/ops/nansum_cpu_dispatch.h>
+#include <ATen/ops/narrow_copy_cpu_dispatch.h>
+#include <ATen/ops/native_batch_norm_cpu_dispatch.h>
+#include <ATen/ops/native_batch_norm_backward_cpu_dispatch.h>
+#include <ATen/ops/native_channel_shuffle_cpu_dispatch.h>
+#include <ATen/ops/native_dropout_cpu_dispatch.h>
+#include <ATen/ops/native_dropout_backward_cpu_dispatch.h>
+#include <ATen/ops/native_group_norm_cpu_dispatch.h>
+#include <ATen/ops/native_group_norm_backward_cpu_dispatch.h>
+#include <ATen/ops/native_layer_norm_cpu_dispatch.h>
+#include <ATen/ops/native_layer_norm_backward_cpu_dispatch.h>
+#include <ATen/ops/ne_cpu_dispatch.h>
+#include <ATen/ops/neg_cpu_dispatch.h>
+#include <ATen/ops/nextafter_cpu_dispatch.h>
+#include <ATen/ops/nll_loss2d_backward_cpu_dispatch.h>
+#include <ATen/ops/nll_loss2d_forward_cpu_dispatch.h>
+#include <ATen/ops/nll_loss_backward_cpu_dispatch.h>
+#include <ATen/ops/nll_loss_forward_cpu_dispatch.h>
+#include <ATen/ops/nonzero_cpu_dispatch.h>
+#include <ATen/ops/nonzero_static_cpu_dispatch.h>
+#include <ATen/ops/norm_cpu_dispatch.h>
+#include <ATen/ops/normal_cpu_dispatch.h>
+#include <ATen/ops/ormqr_cpu_dispatch.h>
+#include <ATen/ops/pixel_shuffle_cpu_dispatch.h>
+#include <ATen/ops/pixel_unshuffle_cpu_dispatch.h>
+#include <ATen/ops/poisson_cpu_dispatch.h>
+#include <ATen/ops/polar_cpu_dispatch.h>
+#include <ATen/ops/polygamma_cpu_dispatch.h>
+#include <ATen/ops/pow_cpu_dispatch.h>
+#include <ATen/ops/prod_cpu_dispatch.h>
+#include <ATen/ops/put_cpu_dispatch.h>
+#include <ATen/ops/quantize_per_channel_cpu_dispatch.h>
+#include <ATen/ops/quantize_per_tensor_cpu_dispatch.h>
+#include <ATen/ops/quantize_per_tensor_dynamic_cpu_dispatch.h>
+#include <ATen/ops/random_cpu_dispatch.h>
+#include <ATen/ops/randperm_cpu_dispatch.h>
+#include <ATen/ops/range_cpu_dispatch.h>
+#include <ATen/ops/reciprocal_cpu_dispatch.h>
+#include <ATen/ops/reflection_pad1d_cpu_dispatch.h>
+#include <ATen/ops/reflection_pad1d_backward_cpu_dispatch.h>
+#include <ATen/ops/reflection_pad2d_cpu_dispatch.h>
+#include <ATen/ops/reflection_pad2d_backward_cpu_dispatch.h>
+#include <ATen/ops/reflection_pad3d_cpu_dispatch.h>
+#include <ATen/ops/reflection_pad3d_backward_cpu_dispatch.h>
+#include <ATen/ops/relu_cpu_dispatch.h>
+#include <ATen/ops/remainder_cpu_dispatch.h>
+#include <ATen/ops/renorm_cpu_dispatch.h>
+#include <ATen/ops/repeat_interleave_cpu_dispatch.h>
+#include <ATen/ops/replication_pad1d_cpu_dispatch.h>
+#include <ATen/ops/replication_pad1d_backward_cpu_dispatch.h>
+#include <ATen/ops/replication_pad2d_cpu_dispatch.h>
+#include <ATen/ops/replication_pad2d_backward_cpu_dispatch.h>
+#include <ATen/ops/replication_pad3d_cpu_dispatch.h>
+#include <ATen/ops/replication_pad3d_backward_cpu_dispatch.h>
+#include <ATen/ops/resize_cpu_dispatch.h>
+#include <ATen/ops/roll_cpu_dispatch.h>
+#include <ATen/ops/round_cpu_dispatch.h>
+#include <ATen/ops/rrelu_with_noise_cpu_dispatch.h>
+#include <ATen/ops/rshift_cpu_dispatch.h>
+#include <ATen/ops/rsqrt_cpu_dispatch.h>
+#include <ATen/ops/rsub_cpu_dispatch.h>
+#include <ATen/ops/scatter_cpu_dispatch.h>
+#include <ATen/ops/scatter_add_cpu_dispatch.h>
+#include <ATen/ops/scatter_reduce_cpu_dispatch.h>
+#include <ATen/ops/searchsorted_cpu_dispatch.h>
+#include <ATen/ops/segment_reduce_cpu_dispatch.h>
+#include <ATen/ops/set_cpu_dispatch.h>
+#include <ATen/ops/sgn_cpu_dispatch.h>
+#include <ATen/ops/sigmoid_cpu_dispatch.h>
+#include <ATen/ops/sigmoid_backward_cpu_dispatch.h>
+#include <ATen/ops/sign_cpu_dispatch.h>
+#include <ATen/ops/signbit_cpu_dispatch.h>
+#include <ATen/ops/silu_cpu_dispatch.h>
+#include <ATen/ops/silu_backward_cpu_dispatch.h>
+#include <ATen/ops/sin_cpu_dispatch.h>
+#include <ATen/ops/sinc_cpu_dispatch.h>
+#include <ATen/ops/sinh_cpu_dispatch.h>
+#include <ATen/ops/slow_conv3d_forward_cpu_dispatch.h>
+#include <ATen/ops/slow_conv_dilated2d_cpu_dispatch.h>
+#include <ATen/ops/slow_conv_dilated3d_cpu_dispatch.h>
+#include <ATen/ops/slow_conv_transpose2d_cpu_dispatch.h>
+#include <ATen/ops/slow_conv_transpose3d_cpu_dispatch.h>
+#include <ATen/ops/smooth_l1_loss_cpu_dispatch.h>
+#include <ATen/ops/smooth_l1_loss_backward_cpu_dispatch.h>
+#include <ATen/ops/softplus_cpu_dispatch.h>
+#include <ATen/ops/softplus_backward_cpu_dispatch.h>
+#include <ATen/ops/softshrink_cpu_dispatch.h>
+#include <ATen/ops/softshrink_backward_cpu_dispatch.h>
+#include <ATen/ops/sort_cpu_dispatch.h>
+#include <ATen/ops/special_airy_ai_cpu_dispatch.h>
+#include <ATen/ops/special_bessel_j0_cpu_dispatch.h>
+#include <ATen/ops/special_bessel_j1_cpu_dispatch.h>
+#include <ATen/ops/special_bessel_y0_cpu_dispatch.h>
+#include <ATen/ops/special_bessel_y1_cpu_dispatch.h>
+#include <ATen/ops/special_chebyshev_polynomial_t_cpu_dispatch.h>
+#include <ATen/ops/special_chebyshev_polynomial_u_cpu_dispatch.h>
+#include <ATen/ops/special_chebyshev_polynomial_v_cpu_dispatch.h>
+#include <ATen/ops/special_chebyshev_polynomial_w_cpu_dispatch.h>
+#include <ATen/ops/special_entr_cpu_dispatch.h>
+#include <ATen/ops/special_erfcx_cpu_dispatch.h>
+#include <ATen/ops/special_hermite_polynomial_h_cpu_dispatch.h>
+#include <ATen/ops/special_hermite_polynomial_he_cpu_dispatch.h>
+#include <ATen/ops/special_i0e_cpu_dispatch.h>
+#include <ATen/ops/special_i1_cpu_dispatch.h>
+#include <ATen/ops/special_i1e_cpu_dispatch.h>
+#include <ATen/ops/special_laguerre_polynomial_l_cpu_dispatch.h>
+#include <ATen/ops/special_legendre_polynomial_p_cpu_dispatch.h>
+#include <ATen/ops/special_log_ndtr_cpu_dispatch.h>
+#include <ATen/ops/special_modified_bessel_i0_cpu_dispatch.h>
+#include <ATen/ops/special_modified_bessel_i1_cpu_dispatch.h>
+#include <ATen/ops/special_modified_bessel_k0_cpu_dispatch.h>
+#include <ATen/ops/special_modified_bessel_k1_cpu_dispatch.h>
+#include <ATen/ops/special_ndtri_cpu_dispatch.h>
+#include <ATen/ops/special_scaled_modified_bessel_k0_cpu_dispatch.h>
+#include <ATen/ops/special_scaled_modified_bessel_k1_cpu_dispatch.h>
+#include <ATen/ops/special_shifted_chebyshev_polynomial_t_cpu_dispatch.h>
+#include <ATen/ops/special_shifted_chebyshev_polynomial_u_cpu_dispatch.h>
+#include <ATen/ops/special_shifted_chebyshev_polynomial_v_cpu_dispatch.h>
+#include <ATen/ops/special_shifted_chebyshev_polynomial_w_cpu_dispatch.h>
+#include <ATen/ops/special_spherical_bessel_j0_cpu_dispatch.h>
+#include <ATen/ops/special_xlog1py_cpu_dispatch.h>
+#include <ATen/ops/special_zeta_cpu_dispatch.h>
+#include <ATen/ops/sqrt_cpu_dispatch.h>
+#include <ATen/ops/sspaddmm_cpu_dispatch.h>
+#include <ATen/ops/std_cpu_dispatch.h>
+#include <ATen/ops/std_mean_cpu_dispatch.h>
+#include <ATen/ops/sub_cpu_dispatch.h>
+#include <ATen/ops/sum_cpu_dispatch.h>
+#include <ATen/ops/take_cpu_dispatch.h>
+#include <ATen/ops/tan_cpu_dispatch.h>
+#include <ATen/ops/tanh_cpu_dispatch.h>
+#include <ATen/ops/tanh_backward_cpu_dispatch.h>
+#include <ATen/ops/threshold_cpu_dispatch.h>
+#include <ATen/ops/threshold_backward_cpu_dispatch.h>
+#include <ATen/ops/to_mkldnn_cpu_dispatch.h>
+#include <ATen/ops/topk_cpu_dispatch.h>
+#include <ATen/ops/trace_cpu_dispatch.h>
+#include <ATen/ops/triangular_solve_cpu_dispatch.h>
+#include <ATen/ops/tril_cpu_dispatch.h>
+#include <ATen/ops/tril_indices_cpu_dispatch.h>
+#include <ATen/ops/triu_cpu_dispatch.h>
+#include <ATen/ops/triu_indices_cpu_dispatch.h>
+#include <ATen/ops/trunc_cpu_dispatch.h>
+#include <ATen/ops/unfold_cpu_dispatch.h>
+#include <ATen/ops/unfold_backward_cpu_dispatch.h>
+#include <ATen/ops/uniform_cpu_dispatch.h>
+#include <ATen/ops/unique_consecutive_cpu_dispatch.h>
+#include <ATen/ops/unique_dim_cpu_dispatch.h>
+#include <ATen/ops/unique_dim_consecutive_cpu_dispatch.h>
+#include <ATen/ops/upsample_bicubic2d_cpu_dispatch.h>
+#include <ATen/ops/upsample_bicubic2d_backward_cpu_dispatch.h>
+#include <ATen/ops/upsample_bilinear2d_cpu_dispatch.h>
+#include <ATen/ops/upsample_bilinear2d_backward_cpu_dispatch.h>
+#include <ATen/ops/upsample_linear1d_cpu_dispatch.h>
+#include <ATen/ops/upsample_linear1d_backward_cpu_dispatch.h>
+#include <ATen/ops/upsample_nearest1d_cpu_dispatch.h>
+#include <ATen/ops/upsample_nearest1d_backward_cpu_dispatch.h>
+#include <ATen/ops/upsample_nearest2d_cpu_dispatch.h>
+#include <ATen/ops/upsample_nearest2d_backward_cpu_dispatch.h>
+#include <ATen/ops/upsample_nearest3d_cpu_dispatch.h>
+#include <ATen/ops/upsample_nearest3d_backward_cpu_dispatch.h>
+#include <ATen/ops/upsample_trilinear3d_cpu_dispatch.h>
+#include <ATen/ops/upsample_trilinear3d_backward_cpu_dispatch.h>
+#include <ATen/ops/var_cpu_dispatch.h>
+#include <ATen/ops/var_mean_cpu_dispatch.h>
+#include <ATen/ops/vdot_cpu_dispatch.h>
+#include <ATen/ops/view_cpu_dispatch.h>
+#include <ATen/ops/view_as_complex_cpu_dispatch.h>
+#include <ATen/ops/view_as_real_cpu_dispatch.h>
+#include <ATen/ops/where_cpu_dispatch.h>
+#include <ATen/ops/xlogy_cpu_dispatch.h>
+#include <ATen/ops/zero_cpu_dispatch.h>
+
+
+

.venv/Lib/site-packages/torch/include/ATen/CPUGeneratorImpl.h

@@ -0,0 +1,49 @@
+#pragma once
+
+#include <ATen/core/Generator.h>
+#include <ATen/core/MT19937RNGEngine.h>
+#include <c10/core/GeneratorImpl.h>
+#include <optional>
+
+namespace at {
+
+struct TORCH_API CPUGeneratorImpl : public c10::GeneratorImpl {
+  // Constructors
+  CPUGeneratorImpl(uint64_t seed_in = default_rng_seed_val);
+  ~CPUGeneratorImpl() override = default;
+
+  // CPUGeneratorImpl methods
+  std::shared_ptr<CPUGeneratorImpl> clone() const;
+  void set_current_seed(uint64_t seed) override;
+  void set_offset(uint64_t offset) override;
+  uint64_t get_offset() const override;
+  uint64_t current_seed() const override;
+  uint64_t seed() override;
+  void set_state(const c10::TensorImpl& new_state) override;
+  c10::intrusive_ptr<c10::TensorImpl> get_state() const override;
+  static c10::DeviceType device_type();
+  uint32_t random();
+  uint64_t random64();
+  std::optional<float> next_float_normal_sample();
+  std::optional<double> next_double_normal_sample();
+  void set_next_float_normal_sample(std::optional<float> randn);
+  void set_next_double_normal_sample(std::optional<double> randn);
+  at::mt19937 engine();
+  void set_engine(at::mt19937 engine);
+
+ private:
+  CPUGeneratorImpl* clone_impl() const override;
+  at::mt19937 engine_;
+  std::optional<float> next_float_normal_sample_;
+  std::optional<double> next_double_normal_sample_;
+};
+
+namespace detail {
+
+TORCH_API const Generator& getDefaultCPUGenerator();
+TORCH_API Generator
+createCPUGenerator(uint64_t seed_val = default_rng_seed_val);
+
+} // namespace detail
+
+} // namespace at

.venv/Lib/site-packages/torch/include/ATen/CUDAFunctions.h

@@ -0,0 +1,29 @@
+#include <ATen/core/TensorBody.h>
+
+// TODO Undo all logic introduced for Note [Avoiding Include Cycles In Static Dispatch]
+// Code introduced to avoid cyclic dependency in static dispatch is no longer
+// needed as static dispatch logic is moved from TensorBody.h, which caused cycles in the first place,
+// to Operators.cpp for supporting multiple backends with multiple kernels.
+//
+// Note [Avoiding Include Cycles In Static Dispatch]
+// In order to avoid #include cycles in the static dispatch build, we've carefully split out
+// the static function definition files into {DispatchKey}Functions.h and {DispatchKey}Functions_inl.h.
+//
+// Without this split, the include cycle looks like TensorBody.h -> CPUFunctions.h -> TensorBody.h.
+// - TensorBody.h #includes CPUFunctions.h in the static dispatch build, because the tensor methods
+//   all need to call into the fastpath C++ API defined in CPUFunctions.h. The methods are also all
+//   directly inlined into TensorBody.h.
+// - CPUFunctions.h #includes TensorBody.h because it contains function declarations for the entire C++ API,
+//   which include functions that have defaultable std::optional<Tensor> arguments.
+//   That requires knowing the full Tensor class definition.
+//
+// We break the cycle by doing the following:
+// - Split out CPUFunction.h into two files: CPUFunctions.h and CPUFunctions_inl.h
+// - CPUFunction.h is a dummy file that just includes the Tensor class and includes CPUFunctions_inl.,
+// - CPUFunctions_inl.h includes everything else
+// - (only in the static dispatch build) TensorBody.h makes sure to finish defining the Tensor class,
+//   and then it includes CPUFunctions_inl.h.
+// - All other files that want the cpu fastpath functions can include CPUFunctions.h directly.
+// - This also means that static dispatch build, CPUFunctions.h only needs to
+//   #include TensorBody.h, and it will automatically bring in CPUFunctions_inl.h.
+#include <ATen/CUDAFunctions_inl.h>

.venv/Lib/site-packages/torch/include/ATen/CUDAFunctions_inl.h

@@ -0,0 +1,623 @@
+#pragma once
+// @generated by torchgen/gen.py from DispatchKeyFunctions_inl.h
+
+// NB: The implementing C++ file is RegisterDispatchKey.cpp
+
+// The only #includes we need are for custom classes that have defaults in the C++ API
+#include <c10/core/MemoryFormat.h>
+#include <c10/core/Scalar.h>
+#include <ATen/core/Reduction.h>
+
+#if defined(AT_PER_OPERATOR_HEADERS) && defined(TORCH_ASSERT_ONLY_METHOD_OPERATORS)
+#error This change adds a dependency on all pytorch operators, meaning the     \
+  file will need to be re-compiled every time an operator is changed or added. \
+  Consider including a specific operator from                                  \
+  <ATen/ops/{my_operator}_cuda_dispatch.h>.                   \
+  See NOTE [TORCH_ASSERT_ONLY_METHOD_OPERATORS].
+#endif
+
+#include <ATen/ops/_adaptive_avg_pool2d_cuda_dispatch.h>
+#include <ATen/ops/_adaptive_avg_pool2d_backward_cuda_dispatch.h>
+#include <ATen/ops/_adaptive_avg_pool3d_cuda_dispatch.h>
+#include <ATen/ops/_adaptive_avg_pool3d_backward_cuda_dispatch.h>
+#include <ATen/ops/_addmm_activation_cuda_dispatch.h>
+#include <ATen/ops/_aminmax_cuda_dispatch.h>
+#include <ATen/ops/_amp_foreach_non_finite_check_and_unscale_cuda_dispatch.h>
+#include <ATen/ops/_amp_update_scale_cuda_dispatch.h>
+#include <ATen/ops/_assert_async_cuda_dispatch.h>
+#include <ATen/ops/_batch_norm_with_update_cuda_dispatch.h>
+#include <ATen/ops/_cdist_backward_cuda_dispatch.h>
+#include <ATen/ops/_cdist_forward_cuda_dispatch.h>
+#include <ATen/ops/_cholesky_solve_helper_cuda_dispatch.h>
+#include <ATen/ops/_chunk_cat_cuda_dispatch.h>
+#include <ATen/ops/_compute_linear_combination_cuda_dispatch.h>
+#include <ATen/ops/_conv_depthwise2d_cuda_dispatch.h>
+#include <ATen/ops/_convert_indices_from_coo_to_csr_cuda_dispatch.h>
+#include <ATen/ops/_convert_indices_from_csr_to_coo_cuda_dispatch.h>
+#include <ATen/ops/_convert_weight_to_int4pack_cuda_dispatch.h>
+#include <ATen/ops/_cslt_compress_cuda_dispatch.h>
+#include <ATen/ops/_cslt_sparse_mm_cuda_dispatch.h>
+#include <ATen/ops/_cslt_sparse_mm_search_cuda_dispatch.h>
+#include <ATen/ops/_ctc_loss_cuda_dispatch.h>
+#include <ATen/ops/_ctc_loss_backward_cuda_dispatch.h>
+#include <ATen/ops/_cudnn_ctc_loss_cuda_dispatch.h>
+#include <ATen/ops/_cudnn_init_dropout_state_cuda_dispatch.h>
+#include <ATen/ops/_cudnn_rnn_cuda_dispatch.h>
+#include <ATen/ops/_cudnn_rnn_backward_cuda_dispatch.h>
+#include <ATen/ops/_cudnn_rnn_flatten_weight_cuda_dispatch.h>
+#include <ATen/ops/_cummax_helper_cuda_dispatch.h>
+#include <ATen/ops/_cummin_helper_cuda_dispatch.h>
+#include <ATen/ops/_dirichlet_grad_cuda_dispatch.h>
+#include <ATen/ops/_efficient_attention_backward_cuda_dispatch.h>
+#include <ATen/ops/_efficient_attention_forward_cuda_dispatch.h>
+#include <ATen/ops/_efficientzerotensor_cuda_dispatch.h>
+#include <ATen/ops/_embedding_bag_cuda_dispatch.h>
+#include <ATen/ops/_embedding_bag_backward_cuda_dispatch.h>
+#include <ATen/ops/_embedding_bag_dense_backward_cuda_dispatch.h>
+#include <ATen/ops/_embedding_bag_forward_only_cuda_dispatch.h>
+#include <ATen/ops/_embedding_bag_per_sample_weights_backward_cuda_dispatch.h>
+#include <ATen/ops/_fake_quantize_learnable_per_channel_affine_cuda_dispatch.h>
+#include <ATen/ops/_fake_quantize_learnable_per_channel_affine_backward_cuda_dispatch.h>
+#include <ATen/ops/_fake_quantize_learnable_per_tensor_affine_cuda_dispatch.h>
+#include <ATen/ops/_fake_quantize_learnable_per_tensor_affine_backward_cuda_dispatch.h>
+#include <ATen/ops/_fake_quantize_per_tensor_affine_cachemask_tensor_qparams_cuda_dispatch.h>
+#include <ATen/ops/_fft_c2c_cuda_dispatch.h>
+#include <ATen/ops/_fft_c2r_cuda_dispatch.h>
+#include <ATen/ops/_fft_r2c_cuda_dispatch.h>
+#include <ATen/ops/_fill_mem_eff_dropout_mask_cuda_dispatch.h>
+#include <ATen/ops/_flash_attention_backward_cuda_dispatch.h>
+#include <ATen/ops/_flash_attention_forward_cuda_dispatch.h>
+#include <ATen/ops/_foreach_abs_cuda_dispatch.h>
+#include <ATen/ops/_foreach_acos_cuda_dispatch.h>
+#include <ATen/ops/_foreach_add_cuda_dispatch.h>
+#include <ATen/ops/_foreach_addcdiv_cuda_dispatch.h>
+#include <ATen/ops/_foreach_addcmul_cuda_dispatch.h>
+#include <ATen/ops/_foreach_asin_cuda_dispatch.h>
+#include <ATen/ops/_foreach_atan_cuda_dispatch.h>
+#include <ATen/ops/_foreach_ceil_cuda_dispatch.h>
+#include <ATen/ops/_foreach_clamp_max_cuda_dispatch.h>
+#include <ATen/ops/_foreach_clamp_min_cuda_dispatch.h>
+#include <ATen/ops/_foreach_copy_cuda_dispatch.h>
+#include <ATen/ops/_foreach_cos_cuda_dispatch.h>
+#include <ATen/ops/_foreach_cosh_cuda_dispatch.h>
+#include <ATen/ops/_foreach_div_cuda_dispatch.h>
+#include <ATen/ops/_foreach_erf_cuda_dispatch.h>
+#include <ATen/ops/_foreach_erfc_cuda_dispatch.h>
+#include <ATen/ops/_foreach_exp_cuda_dispatch.h>
+#include <ATen/ops/_foreach_expm1_cuda_dispatch.h>
+#include <ATen/ops/_foreach_floor_cuda_dispatch.h>
+#include <ATen/ops/_foreach_frac_cuda_dispatch.h>
+#include <ATen/ops/_foreach_lerp_cuda_dispatch.h>
+#include <ATen/ops/_foreach_lgamma_cuda_dispatch.h>
+#include <ATen/ops/_foreach_log_cuda_dispatch.h>
+#include <ATen/ops/_foreach_log10_cuda_dispatch.h>
+#include <ATen/ops/_foreach_log1p_cuda_dispatch.h>
+#include <ATen/ops/_foreach_log2_cuda_dispatch.h>
+#include <ATen/ops/_foreach_max_cuda_dispatch.h>
+#include <ATen/ops/_foreach_maximum_cuda_dispatch.h>
+#include <ATen/ops/_foreach_minimum_cuda_dispatch.h>
+#include <ATen/ops/_foreach_mul_cuda_dispatch.h>
+#include <ATen/ops/_foreach_neg_cuda_dispatch.h>
+#include <ATen/ops/_foreach_norm_cuda_dispatch.h>
+#include <ATen/ops/_foreach_pow_cuda_dispatch.h>
+#include <ATen/ops/_foreach_reciprocal_cuda_dispatch.h>
+#include <ATen/ops/_foreach_round_cuda_dispatch.h>
+#include <ATen/ops/_foreach_sigmoid_cuda_dispatch.h>
+#include <ATen/ops/_foreach_sign_cuda_dispatch.h>
+#include <ATen/ops/_foreach_sin_cuda_dispatch.h>
+#include <ATen/ops/_foreach_sinh_cuda_dispatch.h>
+#include <ATen/ops/_foreach_sqrt_cuda_dispatch.h>
+#include <ATen/ops/_foreach_sub_cuda_dispatch.h>
+#include <ATen/ops/_foreach_tan_cuda_dispatch.h>
+#include <ATen/ops/_foreach_tanh_cuda_dispatch.h>
+#include <ATen/ops/_foreach_trunc_cuda_dispatch.h>
+#include <ATen/ops/_foreach_zero_cuda_dispatch.h>
+#include <ATen/ops/_fused_adam_cuda_dispatch.h>
+#include <ATen/ops/_fused_adamw_cuda_dispatch.h>
+#include <ATen/ops/_fused_dropout_cuda_dispatch.h>
+#include <ATen/ops/_fused_moving_avg_obs_fq_helper_cuda_dispatch.h>
+#include <ATen/ops/_fused_sdp_choice_cuda_dispatch.h>
+#include <ATen/ops/_fused_sgd_cuda_dispatch.h>
+#include <ATen/ops/_index_put_impl_cuda_dispatch.h>
+#include <ATen/ops/_int_mm_cuda_dispatch.h>
+#include <ATen/ops/_jagged_to_padded_dense_forward_cuda_dispatch.h>
+#include <ATen/ops/_linalg_det_cuda_dispatch.h>
+#include <ATen/ops/_linalg_eigh_cuda_dispatch.h>
+#include <ATen/ops/_linalg_eigvals_cuda_dispatch.h>
+#include <ATen/ops/_linalg_slogdet_cuda_dispatch.h>
+#include <ATen/ops/_linalg_solve_ex_cuda_dispatch.h>
+#include <ATen/ops/_linalg_svd_cuda_dispatch.h>
+#include <ATen/ops/_local_scalar_dense_cuda_dispatch.h>
+#include <ATen/ops/_log_softmax_cuda_dispatch.h>
+#include <ATen/ops/_log_softmax_backward_data_cuda_dispatch.h>
+#include <ATen/ops/_logcumsumexp_cuda_dispatch.h>
+#include <ATen/ops/_make_per_channel_quantized_tensor_cuda_dispatch.h>
+#include <ATen/ops/_make_per_tensor_quantized_tensor_cuda_dispatch.h>
+#include <ATen/ops/_masked_scale_cuda_dispatch.h>
+#include <ATen/ops/_masked_softmax_cuda_dispatch.h>
+#include <ATen/ops/_masked_softmax_backward_cuda_dispatch.h>
+#include <ATen/ops/_mixed_dtypes_linear_cuda_dispatch.h>
+#include <ATen/ops/_native_batch_norm_legit_cuda_dispatch.h>
+#include <ATen/ops/_native_multi_head_attention_cuda_dispatch.h>
+#include <ATen/ops/_nested_compute_contiguous_strides_offsets_cuda_dispatch.h>
+#include <ATen/ops/_nested_from_padded_cuda_dispatch.h>
+#include <ATen/ops/_nested_tensor_from_mask_cuda_dispatch.h>
+#include <ATen/ops/_nested_tensor_from_mask_left_aligned_cuda_dispatch.h>
+#include <ATen/ops/_nested_view_from_buffer_cuda_dispatch.h>
+#include <ATen/ops/_padded_dense_to_jagged_forward_cuda_dispatch.h>
+#include <ATen/ops/_pdist_backward_cuda_dispatch.h>
+#include <ATen/ops/_pdist_forward_cuda_dispatch.h>
+#include <ATen/ops/_prelu_kernel_cuda_dispatch.h>
+#include <ATen/ops/_prelu_kernel_backward_cuda_dispatch.h>
+#include <ATen/ops/_reshape_alias_cuda_dispatch.h>
+#include <ATen/ops/_sample_dirichlet_cuda_dispatch.h>
+#include <ATen/ops/_scaled_dot_product_cudnn_attention_cuda_dispatch.h>
+#include <ATen/ops/_scaled_dot_product_cudnn_attention_backward_cuda_dispatch.h>
+#include <ATen/ops/_scaled_dot_product_efficient_attention_cuda_dispatch.h>
+#include <ATen/ops/_scaled_dot_product_efficient_attention_backward_cuda_dispatch.h>
+#include <ATen/ops/_scaled_dot_product_flash_attention_cuda_dispatch.h>
+#include <ATen/ops/_scaled_dot_product_flash_attention_backward_cuda_dispatch.h>
+#include <ATen/ops/_scaled_mm_cuda_dispatch.h>
+#include <ATen/ops/_segment_reduce_backward_cuda_dispatch.h>
+#include <ATen/ops/_slow_conv2d_backward_cuda_dispatch.h>
+#include <ATen/ops/_slow_conv2d_forward_cuda_dispatch.h>
+#include <ATen/ops/_softmax_cuda_dispatch.h>
+#include <ATen/ops/_softmax_backward_data_cuda_dispatch.h>
+#include <ATen/ops/_sparse_semi_structured_addmm_cuda_dispatch.h>
+#include <ATen/ops/_sparse_semi_structured_apply_cuda_dispatch.h>
+#include <ATen/ops/_sparse_semi_structured_apply_dense_cuda_dispatch.h>
+#include <ATen/ops/_sparse_semi_structured_linear_cuda_dispatch.h>
+#include <ATen/ops/_sparse_semi_structured_mm_cuda_dispatch.h>
+#include <ATen/ops/_sparse_semi_structured_tile_cuda_dispatch.h>
+#include <ATen/ops/_standard_gamma_cuda_dispatch.h>
+#include <ATen/ops/_standard_gamma_grad_cuda_dispatch.h>
+#include <ATen/ops/_thnn_fused_gru_cell_cuda_dispatch.h>
+#include <ATen/ops/_thnn_fused_gru_cell_backward_cuda_dispatch.h>
+#include <ATen/ops/_thnn_fused_lstm_cell_cuda_dispatch.h>
+#include <ATen/ops/_thnn_fused_lstm_cell_backward_impl_cuda_dispatch.h>
+#include <ATen/ops/_to_sparse_cuda_dispatch.h>
+#include <ATen/ops/_to_sparse_bsc_cuda_dispatch.h>
+#include <ATen/ops/_to_sparse_bsr_cuda_dispatch.h>
+#include <ATen/ops/_to_sparse_csc_cuda_dispatch.h>
+#include <ATen/ops/_to_sparse_csr_cuda_dispatch.h>
+#include <ATen/ops/_to_sparse_semi_structured_cuda_dispatch.h>
+#include <ATen/ops/_transform_bias_rescale_qkv_cuda_dispatch.h>
+#include <ATen/ops/_transformer_encoder_layer_fwd_cuda_dispatch.h>
+#include <ATen/ops/_triton_multi_head_attention_cuda_dispatch.h>
+#include <ATen/ops/_triton_scaled_dot_attention_cuda_dispatch.h>
+#include <ATen/ops/_unique_cuda_dispatch.h>
+#include <ATen/ops/_unique2_cuda_dispatch.h>
+#include <ATen/ops/_upsample_bicubic2d_aa_cuda_dispatch.h>
+#include <ATen/ops/_upsample_bicubic2d_aa_backward_cuda_dispatch.h>
+#include <ATen/ops/_upsample_bilinear2d_aa_cuda_dispatch.h>
+#include <ATen/ops/_upsample_bilinear2d_aa_backward_cuda_dispatch.h>
+#include <ATen/ops/_upsample_nearest_exact1d_cuda_dispatch.h>
+#include <ATen/ops/_upsample_nearest_exact1d_backward_cuda_dispatch.h>
+#include <ATen/ops/_upsample_nearest_exact2d_cuda_dispatch.h>
+#include <ATen/ops/_upsample_nearest_exact2d_backward_cuda_dispatch.h>
+#include <ATen/ops/_upsample_nearest_exact3d_cuda_dispatch.h>
+#include <ATen/ops/_upsample_nearest_exact3d_backward_cuda_dispatch.h>
+#include <ATen/ops/_use_cudnn_ctc_loss_cuda_dispatch.h>
+#include <ATen/ops/_validate_compressed_sparse_indices_cuda_dispatch.h>
+#include <ATen/ops/_weight_int4pack_mm_cuda_dispatch.h>
+#include <ATen/ops/_weight_norm_interface_cuda_dispatch.h>
+#include <ATen/ops/_weight_norm_interface_backward_cuda_dispatch.h>
+#include <ATen/ops/abs_cuda_dispatch.h>
+#include <ATen/ops/acos_cuda_dispatch.h>
+#include <ATen/ops/acosh_cuda_dispatch.h>
+#include <ATen/ops/adaptive_avg_pool2d_cuda_dispatch.h>
+#include <ATen/ops/adaptive_avg_pool3d_cuda_dispatch.h>
+#include <ATen/ops/adaptive_avg_pool3d_backward_cuda_dispatch.h>
+#include <ATen/ops/adaptive_max_pool2d_cuda_dispatch.h>
+#include <ATen/ops/adaptive_max_pool2d_backward_cuda_dispatch.h>
+#include <ATen/ops/adaptive_max_pool3d_cuda_dispatch.h>
+#include <ATen/ops/adaptive_max_pool3d_backward_cuda_dispatch.h>
+#include <ATen/ops/add_cuda_dispatch.h>
+#include <ATen/ops/addbmm_cuda_dispatch.h>
+#include <ATen/ops/addcdiv_cuda_dispatch.h>
+#include <ATen/ops/addcmul_cuda_dispatch.h>
+#include <ATen/ops/addmm_cuda_dispatch.h>
+#include <ATen/ops/addmv_cuda_dispatch.h>
+#include <ATen/ops/addr_cuda_dispatch.h>
+#include <ATen/ops/all_cuda_dispatch.h>
+#include <ATen/ops/amax_cuda_dispatch.h>
+#include <ATen/ops/amin_cuda_dispatch.h>
+#include <ATen/ops/aminmax_cuda_dispatch.h>
+#include <ATen/ops/angle_cuda_dispatch.h>
+#include <ATen/ops/any_cuda_dispatch.h>
+#include <ATen/ops/arange_cuda_dispatch.h>
+#include <ATen/ops/argmax_cuda_dispatch.h>
+#include <ATen/ops/argmin_cuda_dispatch.h>
+#include <ATen/ops/as_strided_cuda_dispatch.h>
+#include <ATen/ops/asin_cuda_dispatch.h>
+#include <ATen/ops/asinh_cuda_dispatch.h>
+#include <ATen/ops/atan_cuda_dispatch.h>
+#include <ATen/ops/atan2_cuda_dispatch.h>
+#include <ATen/ops/atanh_cuda_dispatch.h>
+#include <ATen/ops/avg_pool2d_cuda_dispatch.h>
+#include <ATen/ops/avg_pool2d_backward_cuda_dispatch.h>
+#include <ATen/ops/avg_pool3d_cuda_dispatch.h>
+#include <ATen/ops/avg_pool3d_backward_cuda_dispatch.h>
+#include <ATen/ops/baddbmm_cuda_dispatch.h>
+#include <ATen/ops/batch_norm_backward_cuda_dispatch.h>
+#include <ATen/ops/batch_norm_backward_elemt_cuda_dispatch.h>
+#include <ATen/ops/batch_norm_backward_reduce_cuda_dispatch.h>
+#include <ATen/ops/batch_norm_elemt_cuda_dispatch.h>
+#include <ATen/ops/batch_norm_gather_stats_cuda_dispatch.h>
+#include <ATen/ops/batch_norm_gather_stats_with_counts_cuda_dispatch.h>
+#include <ATen/ops/batch_norm_stats_cuda_dispatch.h>
+#include <ATen/ops/batch_norm_update_stats_cuda_dispatch.h>
+#include <ATen/ops/bernoulli_cuda_dispatch.h>
+#include <ATen/ops/binary_cross_entropy_cuda_dispatch.h>
+#include <ATen/ops/binary_cross_entropy_backward_cuda_dispatch.h>
+#include <ATen/ops/bincount_cuda_dispatch.h>
+#include <ATen/ops/binomial_cuda_dispatch.h>
+#include <ATen/ops/bitwise_and_cuda_dispatch.h>
+#include <ATen/ops/bitwise_left_shift_cuda_dispatch.h>
+#include <ATen/ops/bitwise_not_cuda_dispatch.h>
+#include <ATen/ops/bitwise_or_cuda_dispatch.h>
+#include <ATen/ops/bitwise_right_shift_cuda_dispatch.h>
+#include <ATen/ops/bitwise_xor_cuda_dispatch.h>
+#include <ATen/ops/bmm_cuda_dispatch.h>
+#include <ATen/ops/bucketize_cuda_dispatch.h>
+#include <ATen/ops/cat_cuda_dispatch.h>
+#include <ATen/ops/cauchy_cuda_dispatch.h>
+#include <ATen/ops/ceil_cuda_dispatch.h>
+#include <ATen/ops/channel_shuffle_cuda_dispatch.h>
+#include <ATen/ops/cholesky_cuda_dispatch.h>
+#include <ATen/ops/cholesky_inverse_cuda_dispatch.h>
+#include <ATen/ops/clamp_cuda_dispatch.h>
+#include <ATen/ops/clamp_max_cuda_dispatch.h>
+#include <ATen/ops/clamp_min_cuda_dispatch.h>
+#include <ATen/ops/col2im_cuda_dispatch.h>
+#include <ATen/ops/complex_cuda_dispatch.h>
+#include <ATen/ops/conj_physical_cuda_dispatch.h>
+#include <ATen/ops/conv_depthwise3d_cuda_dispatch.h>
+#include <ATen/ops/convolution_backward_cuda_dispatch.h>
+#include <ATen/ops/copysign_cuda_dispatch.h>
+#include <ATen/ops/cos_cuda_dispatch.h>
+#include <ATen/ops/cosh_cuda_dispatch.h>
+#include <ATen/ops/count_nonzero_cuda_dispatch.h>
+#include <ATen/ops/cudnn_affine_grid_generator_cuda_dispatch.h>
+#include <ATen/ops/cudnn_affine_grid_generator_backward_cuda_dispatch.h>
+#include <ATen/ops/cudnn_batch_norm_cuda_dispatch.h>
+#include <ATen/ops/cudnn_batch_norm_backward_cuda_dispatch.h>
+#include <ATen/ops/cudnn_convolution_cuda_dispatch.h>
+#include <ATen/ops/cudnn_convolution_add_relu_cuda_dispatch.h>
+#include <ATen/ops/cudnn_convolution_relu_cuda_dispatch.h>
+#include <ATen/ops/cudnn_convolution_transpose_cuda_dispatch.h>
+#include <ATen/ops/cudnn_grid_sampler_cuda_dispatch.h>
+#include <ATen/ops/cudnn_grid_sampler_backward_cuda_dispatch.h>
+#include <ATen/ops/cumprod_cuda_dispatch.h>
+#include <ATen/ops/cumsum_cuda_dispatch.h>
+#include <ATen/ops/dequantize_cuda_dispatch.h>
+#include <ATen/ops/digamma_cuda_dispatch.h>
+#include <ATen/ops/div_cuda_dispatch.h>
+#include <ATen/ops/dot_cuda_dispatch.h>
+#include <ATen/ops/elu_cuda_dispatch.h>
+#include <ATen/ops/elu_backward_cuda_dispatch.h>
+#include <ATen/ops/embedding_dense_backward_cuda_dispatch.h>
+#include <ATen/ops/embedding_renorm_cuda_dispatch.h>
+#include <ATen/ops/empty_cuda_dispatch.h>
+#include <ATen/ops/empty_strided_cuda_dispatch.h>
+#include <ATen/ops/eq_cuda_dispatch.h>
+#include <ATen/ops/equal_cuda_dispatch.h>
+#include <ATen/ops/erf_cuda_dispatch.h>
+#include <ATen/ops/erfc_cuda_dispatch.h>
+#include <ATen/ops/erfinv_cuda_dispatch.h>
+#include <ATen/ops/exp_cuda_dispatch.h>
+#include <ATen/ops/exp2_cuda_dispatch.h>
+#include <ATen/ops/expm1_cuda_dispatch.h>
+#include <ATen/ops/exponential_cuda_dispatch.h>
+#include <ATen/ops/eye_cuda_dispatch.h>
+#include <ATen/ops/fake_quantize_per_channel_affine_cachemask_cuda_dispatch.h>
+#include <ATen/ops/fake_quantize_per_tensor_affine_cachemask_cuda_dispatch.h>
+#include <ATen/ops/fill_cuda_dispatch.h>
+#include <ATen/ops/flip_cuda_dispatch.h>
+#include <ATen/ops/floor_cuda_dispatch.h>
+#include <ATen/ops/floor_divide_cuda_dispatch.h>
+#include <ATen/ops/fmax_cuda_dispatch.h>
+#include <ATen/ops/fmin_cuda_dispatch.h>
+#include <ATen/ops/fmod_cuda_dispatch.h>
+#include <ATen/ops/frac_cuda_dispatch.h>
+#include <ATen/ops/fractional_max_pool2d_cuda_dispatch.h>
+#include <ATen/ops/fractional_max_pool2d_backward_cuda_dispatch.h>
+#include <ATen/ops/fractional_max_pool3d_cuda_dispatch.h>
+#include <ATen/ops/fractional_max_pool3d_backward_cuda_dispatch.h>
+#include <ATen/ops/frexp_cuda_dispatch.h>
+#include <ATen/ops/gather_cuda_dispatch.h>
+#include <ATen/ops/gcd_cuda_dispatch.h>
+#include <ATen/ops/ge_cuda_dispatch.h>
+#include <ATen/ops/gelu_cuda_dispatch.h>
+#include <ATen/ops/gelu_backward_cuda_dispatch.h>
+#include <ATen/ops/geometric_cuda_dispatch.h>
+#include <ATen/ops/geqrf_cuda_dispatch.h>
+#include <ATen/ops/glu_cuda_dispatch.h>
+#include <ATen/ops/glu_backward_cuda_dispatch.h>
+#include <ATen/ops/glu_backward_jvp_cuda_dispatch.h>
+#include <ATen/ops/glu_jvp_cuda_dispatch.h>
+#include <ATen/ops/grid_sampler_2d_cuda_dispatch.h>
+#include <ATen/ops/grid_sampler_2d_backward_cuda_dispatch.h>
+#include <ATen/ops/grid_sampler_3d_cuda_dispatch.h>
+#include <ATen/ops/grid_sampler_3d_backward_cuda_dispatch.h>
+#include <ATen/ops/gt_cuda_dispatch.h>
+#include <ATen/ops/hardshrink_cuda_dispatch.h>
+#include <ATen/ops/hardshrink_backward_cuda_dispatch.h>
+#include <ATen/ops/hardsigmoid_cuda_dispatch.h>
+#include <ATen/ops/hardsigmoid_backward_cuda_dispatch.h>
+#include <ATen/ops/hardswish_cuda_dispatch.h>
+#include <ATen/ops/hardswish_backward_cuda_dispatch.h>
+#include <ATen/ops/hardtanh_cuda_dispatch.h>
+#include <ATen/ops/hardtanh_backward_cuda_dispatch.h>
+#include <ATen/ops/heaviside_cuda_dispatch.h>
+#include <ATen/ops/histc_cuda_dispatch.h>
+#include <ATen/ops/huber_loss_cuda_dispatch.h>
+#include <ATen/ops/huber_loss_backward_cuda_dispatch.h>
+#include <ATen/ops/hypot_cuda_dispatch.h>
+#include <ATen/ops/i0_cuda_dispatch.h>
+#include <ATen/ops/igamma_cuda_dispatch.h>
+#include <ATen/ops/igammac_cuda_dispatch.h>
+#include <ATen/ops/im2col_cuda_dispatch.h>
+#include <ATen/ops/index_cuda_dispatch.h>
+#include <ATen/ops/index_add_cuda_dispatch.h>
+#include <ATen/ops/index_copy_cuda_dispatch.h>
+#include <ATen/ops/index_fill_cuda_dispatch.h>
+#include <ATen/ops/index_reduce_cuda_dispatch.h>
+#include <ATen/ops/index_select_cuda_dispatch.h>
+#include <ATen/ops/is_set_to_cuda_dispatch.h>
+#include <ATen/ops/isin_cuda_dispatch.h>
+#include <ATen/ops/isnan_cuda_dispatch.h>
+#include <ATen/ops/isneginf_cuda_dispatch.h>
+#include <ATen/ops/isposinf_cuda_dispatch.h>
+#include <ATen/ops/kthvalue_cuda_dispatch.h>
+#include <ATen/ops/lcm_cuda_dispatch.h>
+#include <ATen/ops/le_cuda_dispatch.h>
+#include <ATen/ops/leaky_relu_cuda_dispatch.h>
+#include <ATen/ops/leaky_relu_backward_cuda_dispatch.h>
+#include <ATen/ops/lerp_cuda_dispatch.h>
+#include <ATen/ops/lgamma_cuda_dispatch.h>
+#include <ATen/ops/linalg_cholesky_ex_cuda_dispatch.h>
+#include <ATen/ops/linalg_cross_cuda_dispatch.h>
+#include <ATen/ops/linalg_eig_cuda_dispatch.h>
+#include <ATen/ops/linalg_eigvals_cuda_dispatch.h>
+#include <ATen/ops/linalg_householder_product_cuda_dispatch.h>
+#include <ATen/ops/linalg_inv_ex_cuda_dispatch.h>
+#include <ATen/ops/linalg_ldl_factor_ex_cuda_dispatch.h>
+#include <ATen/ops/linalg_ldl_solve_cuda_dispatch.h>
+#include <ATen/ops/linalg_lstsq_cuda_dispatch.h>
+#include <ATen/ops/linalg_lu_cuda_dispatch.h>
+#include <ATen/ops/linalg_lu_factor_ex_cuda_dispatch.h>
+#include <ATen/ops/linalg_lu_solve_cuda_dispatch.h>
+#include <ATen/ops/linalg_matrix_exp_cuda_dispatch.h>
+#include <ATen/ops/linalg_qr_cuda_dispatch.h>
+#include <ATen/ops/linalg_solve_triangular_cuda_dispatch.h>
+#include <ATen/ops/linalg_vector_norm_cuda_dispatch.h>
+#include <ATen/ops/linspace_cuda_dispatch.h>
+#include <ATen/ops/log_cuda_dispatch.h>
+#include <ATen/ops/log10_cuda_dispatch.h>
+#include <ATen/ops/log1p_cuda_dispatch.h>
+#include <ATen/ops/log2_cuda_dispatch.h>
+#include <ATen/ops/log_normal_cuda_dispatch.h>
+#include <ATen/ops/log_sigmoid_backward_cuda_dispatch.h>
+#include <ATen/ops/log_sigmoid_forward_cuda_dispatch.h>
+#include <ATen/ops/logaddexp_cuda_dispatch.h>
+#include <ATen/ops/logaddexp2_cuda_dispatch.h>
+#include <ATen/ops/logical_and_cuda_dispatch.h>
+#include <ATen/ops/logical_not_cuda_dispatch.h>
+#include <ATen/ops/logical_or_cuda_dispatch.h>
+#include <ATen/ops/logical_xor_cuda_dispatch.h>
+#include <ATen/ops/logit_cuda_dispatch.h>
+#include <ATen/ops/logit_backward_cuda_dispatch.h>
+#include <ATen/ops/logspace_cuda_dispatch.h>
+#include <ATen/ops/lshift_cuda_dispatch.h>
+#include <ATen/ops/lt_cuda_dispatch.h>
+#include <ATen/ops/lu_unpack_cuda_dispatch.h>
+#include <ATen/ops/masked_fill_cuda_dispatch.h>
+#include <ATen/ops/masked_scatter_cuda_dispatch.h>
+#include <ATen/ops/masked_select_cuda_dispatch.h>
+#include <ATen/ops/max_cuda_dispatch.h>
+#include <ATen/ops/max_pool2d_with_indices_cuda_dispatch.h>
+#include <ATen/ops/max_pool2d_with_indices_backward_cuda_dispatch.h>
+#include <ATen/ops/max_pool3d_with_indices_cuda_dispatch.h>
+#include <ATen/ops/max_pool3d_with_indices_backward_cuda_dispatch.h>
+#include <ATen/ops/max_unpool2d_cuda_dispatch.h>
+#include <ATen/ops/max_unpool3d_cuda_dispatch.h>
+#include <ATen/ops/maximum_cuda_dispatch.h>
+#include <ATen/ops/mean_cuda_dispatch.h>
+#include <ATen/ops/median_cuda_dispatch.h>
+#include <ATen/ops/min_cuda_dispatch.h>
+#include <ATen/ops/minimum_cuda_dispatch.h>
+#include <ATen/ops/miopen_batch_norm_cuda_dispatch.h>
+#include <ATen/ops/miopen_batch_norm_backward_cuda_dispatch.h>
+#include <ATen/ops/miopen_convolution_cuda_dispatch.h>
+#include <ATen/ops/miopen_convolution_add_relu_cuda_dispatch.h>
+#include <ATen/ops/miopen_convolution_relu_cuda_dispatch.h>
+#include <ATen/ops/miopen_convolution_transpose_cuda_dispatch.h>
+#include <ATen/ops/miopen_depthwise_convolution_cuda_dispatch.h>
+#include <ATen/ops/miopen_rnn_cuda_dispatch.h>
+#include <ATen/ops/miopen_rnn_backward_cuda_dispatch.h>
+#include <ATen/ops/mish_cuda_dispatch.h>
+#include <ATen/ops/mish_backward_cuda_dispatch.h>
+#include <ATen/ops/mm_cuda_dispatch.h>
+#include <ATen/ops/mode_cuda_dispatch.h>
+#include <ATen/ops/mse_loss_cuda_dispatch.h>
+#include <ATen/ops/mse_loss_backward_cuda_dispatch.h>
+#include <ATen/ops/mul_cuda_dispatch.h>
+#include <ATen/ops/multi_margin_loss_cuda_dispatch.h>
+#include <ATen/ops/multi_margin_loss_backward_cuda_dispatch.h>
+#include <ATen/ops/multilabel_margin_loss_backward_cuda_dispatch.h>
+#include <ATen/ops/multilabel_margin_loss_forward_cuda_dispatch.h>
+#include <ATen/ops/multinomial_cuda_dispatch.h>
+#include <ATen/ops/mvlgamma_cuda_dispatch.h>
+#include <ATen/ops/nan_to_num_cuda_dispatch.h>
+#include <ATen/ops/nanmedian_cuda_dispatch.h>
+#include <ATen/ops/nansum_cuda_dispatch.h>
+#include <ATen/ops/native_batch_norm_cuda_dispatch.h>
+#include <ATen/ops/native_batch_norm_backward_cuda_dispatch.h>
+#include <ATen/ops/native_dropout_cuda_dispatch.h>
+#include <ATen/ops/native_dropout_backward_cuda_dispatch.h>
+#include <ATen/ops/native_group_norm_cuda_dispatch.h>
+#include <ATen/ops/native_group_norm_backward_cuda_dispatch.h>
+#include <ATen/ops/native_layer_norm_cuda_dispatch.h>
+#include <ATen/ops/native_layer_norm_backward_cuda_dispatch.h>
+#include <ATen/ops/ne_cuda_dispatch.h>
+#include <ATen/ops/neg_cuda_dispatch.h>
+#include <ATen/ops/nextafter_cuda_dispatch.h>
+#include <ATen/ops/nll_loss2d_backward_cuda_dispatch.h>
+#include <ATen/ops/nll_loss2d_forward_cuda_dispatch.h>
+#include <ATen/ops/nll_loss_backward_cuda_dispatch.h>
+#include <ATen/ops/nll_loss_forward_cuda_dispatch.h>
+#include <ATen/ops/nonzero_cuda_dispatch.h>
+#include <ATen/ops/norm_cuda_dispatch.h>
+#include <ATen/ops/normal_cuda_dispatch.h>
+#include <ATen/ops/ormqr_cuda_dispatch.h>
+#include <ATen/ops/poisson_cuda_dispatch.h>
+#include <ATen/ops/polar_cuda_dispatch.h>
+#include <ATen/ops/polygamma_cuda_dispatch.h>
+#include <ATen/ops/pow_cuda_dispatch.h>
+#include <ATen/ops/prod_cuda_dispatch.h>
+#include <ATen/ops/put_cuda_dispatch.h>
+#include <ATen/ops/quantize_per_channel_cuda_dispatch.h>
+#include <ATen/ops/quantize_per_tensor_cuda_dispatch.h>
+#include <ATen/ops/quantize_per_tensor_dynamic_cuda_dispatch.h>
+#include <ATen/ops/random_cuda_dispatch.h>
+#include <ATen/ops/randperm_cuda_dispatch.h>
+#include <ATen/ops/range_cuda_dispatch.h>
+#include <ATen/ops/reciprocal_cuda_dispatch.h>
+#include <ATen/ops/record_stream_cuda_dispatch.h>
+#include <ATen/ops/reflection_pad1d_cuda_dispatch.h>
+#include <ATen/ops/reflection_pad1d_backward_cuda_dispatch.h>
+#include <ATen/ops/reflection_pad2d_cuda_dispatch.h>
+#include <ATen/ops/reflection_pad2d_backward_cuda_dispatch.h>
+#include <ATen/ops/reflection_pad3d_cuda_dispatch.h>
+#include <ATen/ops/reflection_pad3d_backward_cuda_dispatch.h>
+#include <ATen/ops/relu_cuda_dispatch.h>
+#include <ATen/ops/remainder_cuda_dispatch.h>
+#include <ATen/ops/renorm_cuda_dispatch.h>
+#include <ATen/ops/repeat_interleave_cuda_dispatch.h>
+#include <ATen/ops/replication_pad1d_cuda_dispatch.h>
+#include <ATen/ops/replication_pad1d_backward_cuda_dispatch.h>
+#include <ATen/ops/replication_pad2d_cuda_dispatch.h>
+#include <ATen/ops/replication_pad2d_backward_cuda_dispatch.h>
+#include <ATen/ops/replication_pad3d_cuda_dispatch.h>
+#include <ATen/ops/replication_pad3d_backward_cuda_dispatch.h>
+#include <ATen/ops/resize_cuda_dispatch.h>
+#include <ATen/ops/roll_cuda_dispatch.h>
+#include <ATen/ops/round_cuda_dispatch.h>
+#include <ATen/ops/rrelu_with_noise_cuda_dispatch.h>
+#include <ATen/ops/rshift_cuda_dispatch.h>
+#include <ATen/ops/rsqrt_cuda_dispatch.h>
+#include <ATen/ops/rsub_cuda_dispatch.h>
+#include <ATen/ops/scatter_cuda_dispatch.h>
+#include <ATen/ops/scatter_add_cuda_dispatch.h>
+#include <ATen/ops/scatter_reduce_cuda_dispatch.h>
+#include <ATen/ops/searchsorted_cuda_dispatch.h>
+#include <ATen/ops/segment_reduce_cuda_dispatch.h>
+#include <ATen/ops/set_cuda_dispatch.h>
+#include <ATen/ops/sgn_cuda_dispatch.h>
+#include <ATen/ops/sigmoid_cuda_dispatch.h>
+#include <ATen/ops/sigmoid_backward_cuda_dispatch.h>
+#include <ATen/ops/sign_cuda_dispatch.h>
+#include <ATen/ops/signbit_cuda_dispatch.h>
+#include <ATen/ops/silu_cuda_dispatch.h>
+#include <ATen/ops/silu_backward_cuda_dispatch.h>
+#include <ATen/ops/sin_cuda_dispatch.h>
+#include <ATen/ops/sinc_cuda_dispatch.h>
+#include <ATen/ops/sinh_cuda_dispatch.h>
+#include <ATen/ops/slow_conv_dilated2d_cuda_dispatch.h>
+#include <ATen/ops/slow_conv_dilated3d_cuda_dispatch.h>
+#include <ATen/ops/slow_conv_transpose2d_cuda_dispatch.h>
+#include <ATen/ops/slow_conv_transpose3d_cuda_dispatch.h>
+#include <ATen/ops/smooth_l1_loss_cuda_dispatch.h>
+#include <ATen/ops/smooth_l1_loss_backward_cuda_dispatch.h>
+#include <ATen/ops/softplus_cuda_dispatch.h>
+#include <ATen/ops/softplus_backward_cuda_dispatch.h>
+#include <ATen/ops/softshrink_cuda_dispatch.h>
+#include <ATen/ops/softshrink_backward_cuda_dispatch.h>
+#include <ATen/ops/sort_cuda_dispatch.h>
+#include <ATen/ops/special_airy_ai_cuda_dispatch.h>
+#include <ATen/ops/special_bessel_j0_cuda_dispatch.h>
+#include <ATen/ops/special_bessel_j1_cuda_dispatch.h>
+#include <ATen/ops/special_bessel_y0_cuda_dispatch.h>
+#include <ATen/ops/special_bessel_y1_cuda_dispatch.h>
+#include <ATen/ops/special_chebyshev_polynomial_t_cuda_dispatch.h>
+#include <ATen/ops/special_chebyshev_polynomial_u_cuda_dispatch.h>
+#include <ATen/ops/special_chebyshev_polynomial_v_cuda_dispatch.h>
+#include <ATen/ops/special_chebyshev_polynomial_w_cuda_dispatch.h>
+#include <ATen/ops/special_entr_cuda_dispatch.h>
+#include <ATen/ops/special_erfcx_cuda_dispatch.h>
+#include <ATen/ops/special_hermite_polynomial_h_cuda_dispatch.h>
+#include <ATen/ops/special_hermite_polynomial_he_cuda_dispatch.h>
+#include <ATen/ops/special_i0e_cuda_dispatch.h>
+#include <ATen/ops/special_i1_cuda_dispatch.h>
+#include <ATen/ops/special_i1e_cuda_dispatch.h>
+#include <ATen/ops/special_laguerre_polynomial_l_cuda_dispatch.h>
+#include <ATen/ops/special_legendre_polynomial_p_cuda_dispatch.h>
+#include <ATen/ops/special_log_ndtr_cuda_dispatch.h>
+#include <ATen/ops/special_modified_bessel_i0_cuda_dispatch.h>
+#include <ATen/ops/special_modified_bessel_i1_cuda_dispatch.h>
+#include <ATen/ops/special_modified_bessel_k0_cuda_dispatch.h>
+#include <ATen/ops/special_modified_bessel_k1_cuda_dispatch.h>
+#include <ATen/ops/special_ndtri_cuda_dispatch.h>
+#include <ATen/ops/special_scaled_modified_bessel_k0_cuda_dispatch.h>
+#include <ATen/ops/special_scaled_modified_bessel_k1_cuda_dispatch.h>
+#include <ATen/ops/special_shifted_chebyshev_polynomial_t_cuda_dispatch.h>
+#include <ATen/ops/special_shifted_chebyshev_polynomial_u_cuda_dispatch.h>
+#include <ATen/ops/special_shifted_chebyshev_polynomial_v_cuda_dispatch.h>
+#include <ATen/ops/special_shifted_chebyshev_polynomial_w_cuda_dispatch.h>
+#include <ATen/ops/special_spherical_bessel_j0_cuda_dispatch.h>
+#include <ATen/ops/special_xlog1py_cuda_dispatch.h>
+#include <ATen/ops/special_zeta_cuda_dispatch.h>
+#include <ATen/ops/split_with_sizes_copy_cuda_dispatch.h>
+#include <ATen/ops/sqrt_cuda_dispatch.h>
+#include <ATen/ops/sspaddmm_cuda_dispatch.h>
+#include <ATen/ops/std_cuda_dispatch.h>
+#include <ATen/ops/std_mean_cuda_dispatch.h>
+#include <ATen/ops/sub_cuda_dispatch.h>
+#include <ATen/ops/sum_cuda_dispatch.h>
+#include <ATen/ops/take_cuda_dispatch.h>
+#include <ATen/ops/tan_cuda_dispatch.h>
+#include <ATen/ops/tanh_cuda_dispatch.h>
+#include <ATen/ops/tanh_backward_cuda_dispatch.h>
+#include <ATen/ops/threshold_cuda_dispatch.h>
+#include <ATen/ops/threshold_backward_cuda_dispatch.h>
+#include <ATen/ops/topk_cuda_dispatch.h>
+#include <ATen/ops/trace_cuda_dispatch.h>
+#include <ATen/ops/triangular_solve_cuda_dispatch.h>
+#include <ATen/ops/tril_cuda_dispatch.h>
+#include <ATen/ops/tril_indices_cuda_dispatch.h>
+#include <ATen/ops/triu_cuda_dispatch.h>
+#include <ATen/ops/triu_indices_cuda_dispatch.h>
+#include <ATen/ops/trunc_cuda_dispatch.h>
+#include <ATen/ops/unfold_cuda_dispatch.h>
+#include <ATen/ops/unfold_backward_cuda_dispatch.h>
+#include <ATen/ops/uniform_cuda_dispatch.h>
+#include <ATen/ops/unique_consecutive_cuda_dispatch.h>
+#include <ATen/ops/unique_dim_cuda_dispatch.h>
+#include <ATen/ops/unique_dim_consecutive_cuda_dispatch.h>
+#include <ATen/ops/upsample_bicubic2d_cuda_dispatch.h>
+#include <ATen/ops/upsample_bicubic2d_backward_cuda_dispatch.h>
+#include <ATen/ops/upsample_bilinear2d_cuda_dispatch.h>
+#include <ATen/ops/upsample_bilinear2d_backward_cuda_dispatch.h>
+#include <ATen/ops/upsample_linear1d_cuda_dispatch.h>
+#include <ATen/ops/upsample_linear1d_backward_cuda_dispatch.h>
+#include <ATen/ops/upsample_nearest1d_cuda_dispatch.h>
+#include <ATen/ops/upsample_nearest1d_backward_cuda_dispatch.h>
+#include <ATen/ops/upsample_nearest2d_cuda_dispatch.h>
+#include <ATen/ops/upsample_nearest2d_backward_cuda_dispatch.h>
+#include <ATen/ops/upsample_nearest3d_cuda_dispatch.h>
+#include <ATen/ops/upsample_nearest3d_backward_cuda_dispatch.h>
+#include <ATen/ops/upsample_trilinear3d_cuda_dispatch.h>
+#include <ATen/ops/upsample_trilinear3d_backward_cuda_dispatch.h>
+#include <ATen/ops/var_cuda_dispatch.h>
+#include <ATen/ops/var_mean_cuda_dispatch.h>
+#include <ATen/ops/vdot_cuda_dispatch.h>
+#include <ATen/ops/view_cuda_dispatch.h>
+#include <ATen/ops/view_as_complex_cuda_dispatch.h>
+#include <ATen/ops/view_as_real_cuda_dispatch.h>
+#include <ATen/ops/where_cuda_dispatch.h>
+#include <ATen/ops/xlogy_cuda_dispatch.h>
+#include <ATen/ops/zero_cuda_dispatch.h>
+
+
+

.venv/Lib/site-packages/torch/include/ATen/CompositeExplicitAutogradNonFunctionalFunctions.h

@@ -0,0 +1,29 @@
+#include <ATen/core/TensorBody.h>
+
+// TODO Undo all logic introduced for Note [Avoiding Include Cycles In Static Dispatch]
+// Code introduced to avoid cyclic dependency in static dispatch is no longer
+// needed as static dispatch logic is moved from TensorBody.h, which caused cycles in the first place,
+// to Operators.cpp for supporting multiple backends with multiple kernels.
+//
+// Note [Avoiding Include Cycles In Static Dispatch]
+// In order to avoid #include cycles in the static dispatch build, we've carefully split out
+// the static function definition files into {DispatchKey}Functions.h and {DispatchKey}Functions_inl.h.
+//
+// Without this split, the include cycle looks like TensorBody.h -> CPUFunctions.h -> TensorBody.h.
+// - TensorBody.h #includes CPUFunctions.h in the static dispatch build, because the tensor methods
+//   all need to call into the fastpath C++ API defined in CPUFunctions.h. The methods are also all
+//   directly inlined into TensorBody.h.
+// - CPUFunctions.h #includes TensorBody.h because it contains function declarations for the entire C++ API,
+//   which include functions that have defaultable std::optional<Tensor> arguments.
+//   That requires knowing the full Tensor class definition.
+//
+// We break the cycle by doing the following:
+// - Split out CPUFunction.h into two files: CPUFunctions.h and CPUFunctions_inl.h
+// - CPUFunction.h is a dummy file that just includes the Tensor class and includes CPUFunctions_inl.,
+// - CPUFunctions_inl.h includes everything else
+// - (only in the static dispatch build) TensorBody.h makes sure to finish defining the Tensor class,
+//   and then it includes CPUFunctions_inl.h.
+// - All other files that want the cpu fastpath functions can include CPUFunctions.h directly.
+// - This also means that static dispatch build, CPUFunctions.h only needs to
+//   #include TensorBody.h, and it will automatically bring in CPUFunctions_inl.h.
+#include <ATen/CompositeExplicitAutogradNonFunctionalFunctions_inl.h>

.venv/Lib/site-packages/torch/include/ATen/CompositeExplicitAutogradNonFunctionalFunctions_inl.h

@@ -0,0 +1,323 @@
+#pragma once
+// @generated by torchgen/gen.py from DispatchKeyFunctions_inl.h
+
+// NB: The implementing C++ file is RegisterDispatchKey.cpp
+
+// The only #includes we need are for custom classes that have defaults in the C++ API
+#include <c10/core/MemoryFormat.h>
+#include <c10/core/Scalar.h>
+#include <ATen/core/Reduction.h>
+
+#if defined(AT_PER_OPERATOR_HEADERS) && defined(TORCH_ASSERT_ONLY_METHOD_OPERATORS)
+#error This change adds a dependency on all pytorch operators, meaning the     \
+  file will need to be re-compiled every time an operator is changed or added. \
+  Consider including a specific operator from                                  \
+  <ATen/ops/{my_operator}_compositeexplicitautogradnonfunctional_dispatch.h>.                   \
+  See NOTE [TORCH_ASSERT_ONLY_METHOD_OPERATORS].
+#endif
+
+#include <ATen/ops/_addmm_activation_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/_conj_copy_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/_convert_indices_from_coo_to_csr_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/_convert_indices_from_csr_to_coo_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/_fw_primal_copy_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/_indices_copy_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/_linalg_det_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/_linalg_eigh_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/_linalg_slogdet_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/_linalg_solve_ex_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/_linalg_svd_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/_log_softmax_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/_log_softmax_backward_data_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/_make_dual_copy_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/_neg_view_copy_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/_nested_get_values_copy_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/_nested_view_from_buffer_copy_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/_nested_view_from_jagged_copy_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/_reshape_alias_copy_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/_softmax_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/_softmax_backward_data_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/_sparse_broadcast_to_copy_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/_test_autograd_multiple_dispatch_view_copy_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/_trilinear_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/_upsample_bicubic2d_aa_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/_upsample_bicubic2d_aa_backward_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/_upsample_bilinear2d_aa_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/_upsample_bilinear2d_aa_backward_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/_upsample_nearest_exact1d_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/_upsample_nearest_exact1d_backward_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/_upsample_nearest_exact2d_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/_upsample_nearest_exact2d_backward_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/_upsample_nearest_exact3d_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/_upsample_nearest_exact3d_backward_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/_values_copy_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/acos_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/acosh_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/adaptive_max_pool2d_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/adaptive_max_pool2d_backward_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/adaptive_max_pool3d_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/adaptive_max_pool3d_backward_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/add_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/addcdiv_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/addcmul_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/addmm_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/addmv_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/alias_copy_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/all_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/amax_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/amin_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/aminmax_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/any_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/argmax_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/argmin_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/as_strided_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/as_strided_copy_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/as_strided_scatter_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/asin_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/asinh_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/atan_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/atan2_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/atanh_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/avg_pool2d_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/avg_pool2d_backward_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/avg_pool3d_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/avg_pool3d_backward_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/baddbmm_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/bernoulli_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/bitwise_and_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/bitwise_left_shift_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/bitwise_not_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/bitwise_or_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/bitwise_right_shift_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/bitwise_xor_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/bmm_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/cat_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/ccol_indices_copy_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/ceil_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/clamp_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/clamp_max_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/clamp_min_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/col_indices_copy_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/copy_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/copysign_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/cos_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/cosh_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/crow_indices_copy_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/cumprod_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/cumsum_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/detach_copy_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/diag_embed_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/diagonal_copy_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/diagonal_scatter_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/digamma_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/div_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/elu_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/elu_backward_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/eq_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/erf_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/erfc_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/erfinv_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/exp_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/exp2_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/expand_copy_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/expm1_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/floor_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/fmax_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/fmin_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/fmod_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/frac_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/fractional_max_pool2d_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/fractional_max_pool2d_backward_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/fractional_max_pool3d_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/gather_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/gcd_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/ge_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/gelu_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/gelu_backward_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/glu_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/gt_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/hardshrink_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/hardshrink_backward_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/hardsigmoid_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/hardsigmoid_backward_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/heaviside_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/hypot_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/i0_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/igamma_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/igammac_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/index_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/index_add_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/index_copy_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/index_reduce_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/indices_copy_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/isin_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/isneginf_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/isposinf_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/lcm_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/le_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/leaky_relu_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/leaky_relu_backward_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/lerp_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/lgamma_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/lift_fresh_copy_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/linalg_cholesky_ex_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/linalg_cross_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/linalg_inv_ex_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/linalg_ldl_factor_ex_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/linalg_ldl_solve_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/linalg_lu_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/linalg_lu_factor_ex_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/linalg_lu_solve_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/linalg_pinv_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/linalg_qr_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/linalg_vector_norm_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/log_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/log10_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/log1p_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/log2_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/logaddexp_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/logaddexp2_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/logit_backward_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/logsumexp_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/lt_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/lu_unpack_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/max_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/max_pool2d_with_indices_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/max_pool2d_with_indices_backward_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/maximum_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/mean_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/min_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/minimum_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/mish_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/mm_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/mse_loss_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/mul_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/narrow_copy_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/ne_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/neg_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/new_empty_strided_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/nextafter_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/nll_loss_backward_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/nll_loss_forward_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/norm_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/permute_copy_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/pixel_shuffle_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/pixel_unshuffle_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/polygamma_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/pow_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/prod_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/reciprocal_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/reflection_pad1d_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/reflection_pad1d_backward_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/reflection_pad3d_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/reflection_pad3d_backward_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/remainder_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/renorm_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/replication_pad1d_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/replication_pad1d_backward_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/replication_pad2d_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/replication_pad3d_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/round_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/row_indices_copy_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/rsqrt_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/scatter_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/scatter_add_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/scatter_reduce_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/select_backward_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/select_copy_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/select_scatter_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/sgn_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/sigmoid_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/sigmoid_backward_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/sign_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/signbit_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/silu_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/silu_backward_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/sin_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/sinc_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/sinh_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/slice_copy_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/slice_scatter_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/slow_conv_transpose2d_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/smooth_l1_loss_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/softplus_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/softplus_backward_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/softshrink_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/softshrink_backward_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/sort_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/special_airy_ai_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/special_bessel_j0_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/special_bessel_j1_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/special_bessel_y0_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/special_bessel_y1_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/special_chebyshev_polynomial_t_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/special_chebyshev_polynomial_u_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/special_chebyshev_polynomial_v_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/special_chebyshev_polynomial_w_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/special_entr_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/special_erfcx_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/special_hermite_polynomial_h_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/special_hermite_polynomial_he_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/special_i0e_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/special_i1_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/special_i1e_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/special_laguerre_polynomial_l_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/special_legendre_polynomial_p_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/special_log_ndtr_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/special_modified_bessel_i0_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/special_modified_bessel_i1_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/special_modified_bessel_k0_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/special_modified_bessel_k1_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/special_ndtri_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/special_scaled_modified_bessel_k0_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/special_scaled_modified_bessel_k1_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/special_shifted_chebyshev_polynomial_t_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/special_shifted_chebyshev_polynomial_u_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/special_shifted_chebyshev_polynomial_v_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/special_shifted_chebyshev_polynomial_w_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/special_spherical_bessel_j0_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/special_xlog1py_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/special_zeta_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/split_copy_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/split_with_sizes_copy_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/sqrt_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/squeeze_copy_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/sub_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/sum_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/t_copy_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/tan_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/tanh_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/tanh_backward_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/threshold_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/threshold_backward_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/topk_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/transpose_copy_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/triangular_solve_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/tril_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/triu_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/trunc_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/unbind_copy_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/unfold_copy_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/unsqueeze_copy_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/upsample_bicubic2d_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/upsample_bicubic2d_backward_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/upsample_bilinear2d_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/upsample_bilinear2d_backward_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/upsample_linear1d_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/upsample_linear1d_backward_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/upsample_nearest1d_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/upsample_nearest1d_backward_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/upsample_nearest2d_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/upsample_nearest2d_backward_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/upsample_nearest3d_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/upsample_nearest3d_backward_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/upsample_trilinear3d_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/upsample_trilinear3d_backward_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/values_copy_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/view_as_complex_copy_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/view_as_real_copy_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/view_copy_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/xlogy_compositeexplicitautogradnonfunctional_dispatch.h>
+
+
+

.venv/Lib/site-packages/torch/include/ATen/CompositeImplicitAutogradFunctions.h

@@ -0,0 +1,29 @@
+#include <ATen/core/TensorBody.h>
+
+// TODO Undo all logic introduced for Note [Avoiding Include Cycles In Static Dispatch]
+// Code introduced to avoid cyclic dependency in static dispatch is no longer
+// needed as static dispatch logic is moved from TensorBody.h, which caused cycles in the first place,
+// to Operators.cpp for supporting multiple backends with multiple kernels.
+//
+// Note [Avoiding Include Cycles In Static Dispatch]
+// In order to avoid #include cycles in the static dispatch build, we've carefully split out
+// the static function definition files into {DispatchKey}Functions.h and {DispatchKey}Functions_inl.h.
+//
+// Without this split, the include cycle looks like TensorBody.h -> CPUFunctions.h -> TensorBody.h.
+// - TensorBody.h #includes CPUFunctions.h in the static dispatch build, because the tensor methods
+//   all need to call into the fastpath C++ API defined in CPUFunctions.h. The methods are also all
+//   directly inlined into TensorBody.h.
+// - CPUFunctions.h #includes TensorBody.h because it contains function declarations for the entire C++ API,
+//   which include functions that have defaultable std::optional<Tensor> arguments.
+//   That requires knowing the full Tensor class definition.
+//
+// We break the cycle by doing the following:
+// - Split out CPUFunction.h into two files: CPUFunctions.h and CPUFunctions_inl.h
+// - CPUFunction.h is a dummy file that just includes the Tensor class and includes CPUFunctions_inl.,
+// - CPUFunctions_inl.h includes everything else
+// - (only in the static dispatch build) TensorBody.h makes sure to finish defining the Tensor class,
+//   and then it includes CPUFunctions_inl.h.
+// - All other files that want the cpu fastpath functions can include CPUFunctions.h directly.
+// - This also means that static dispatch build, CPUFunctions.h only needs to
+//   #include TensorBody.h, and it will automatically bring in CPUFunctions_inl.h.
+#include <ATen/CompositeImplicitAutogradFunctions_inl.h>

.venv/Lib/site-packages/torch/include/ATen/CompositeImplicitAutogradFunctions_inl.h

@@ -0,0 +1,502 @@
+#pragma once
+// @generated by torchgen/gen.py from DispatchKeyFunctions_inl.h
+
+// NB: The implementing C++ file is RegisterDispatchKey.cpp
+
+// The only #includes we need are for custom classes that have defaults in the C++ API
+#include <c10/core/MemoryFormat.h>
+#include <c10/core/Scalar.h>
+#include <ATen/core/Reduction.h>
+
+#if defined(AT_PER_OPERATOR_HEADERS) && defined(TORCH_ASSERT_ONLY_METHOD_OPERATORS)
+#error This change adds a dependency on all pytorch operators, meaning the     \
+  file will need to be re-compiled every time an operator is changed or added. \
+  Consider including a specific operator from                                  \
+  <ATen/ops/{my_operator}_compositeimplicitautograd_dispatch.h>.                   \
+  See NOTE [TORCH_ASSERT_ONLY_METHOD_OPERATORS].
+#endif
+
+#include <ATen/ops/_add_batch_dim_compositeimplicitautograd_dispatch.h>
+#include <ATen/ops/_assert_tensor_metadata_compositeimplicitautograd_dispatch.h>
+#include <ATen/ops/_autocast_to_full_precision_compositeimplicitautograd_dispatch.h>
+#include <ATen/ops/_autocast_to_reduced_precision_compositeimplicitautograd_dispatch.h>
+#include <ATen/ops/_backward_compositeimplicitautograd_dispatch.h>
+#include <ATen/ops/_batch_norm_impl_index_compositeimplicitautograd_dispatch.h>
+#include <ATen/ops/_batch_norm_impl_index_backward_compositeimplicitautograd_dispatch.h>
+#include <ATen/ops/_cast_Byte_compositeimplicitautograd_dispatch.h>
+#include <ATen/ops/_cast_Char_compositeimplicitautograd_dispatch.h>
+#include <ATen/ops/_cast_Double_compositeimplicitautograd_dispatch.h>
+#include <ATen/ops/_cast_Float_compositeimplicitautograd_dispatch.h>
+#include <ATen/ops/_cast_Half_compositeimplicitautograd_dispatch.h>
+#include <ATen/ops/_cast_Int_compositeimplicitautograd_dispatch.h>
+#include <ATen/ops/_cast_Long_compositeimplicitautograd_dispatch.h>
+#include <ATen/ops/_cast_Short_compositeimplicitautograd_dispatch.h>
+#include <ATen/ops/_choose_qparams_per_tensor_compositeimplicitautograd_dispatch.h>
+#include <ATen/ops/_convolution_compositeimplicitautograd_dispatch.h>
+#include <ATen/ops/_convolution_double_backward_compositeimplicitautograd_dispatch.h>
+#include <ATen/ops/_convolution_mode_compositeimplicitautograd_dispatch.h>
+#include <ATen/ops/_cufft_clear_plan_cache_compositeimplicitautograd_dispatch.h>
+#include <ATen/ops/_cufft_get_plan_cache_max_size_compositeimplicitautograd_dispatch.h>
+#include <ATen/ops/_cufft_get_plan_cache_size_compositeimplicitautograd_dispatch.h>
+#include <ATen/ops/_cufft_set_plan_cache_max_size_compositeimplicitautograd_dispatch.h>
+#include <ATen/ops/_debug_has_internal_overlap_compositeimplicitautograd_dispatch.h>
+#include <ATen/ops/_dim_arange_compositeimplicitautograd_dispatch.h>
+#include <ATen/ops/_embedding_bag_sparse_backward_compositeimplicitautograd_dispatch.h>
+#include <ATen/ops/_gather_sparse_backward_compositeimplicitautograd_dispatch.h>
+#include <ATen/ops/_grid_sampler_2d_cpu_fallback_backward_compositeimplicitautograd_dispatch.h>
+#include <ATen/ops/_has_compatible_shallow_copy_type_compositeimplicitautograd_dispatch.h>
+#include <ATen/ops/_is_zerotensor_compositeimplicitautograd_dispatch.h>
+#include <ATen/ops/_lu_with_info_compositeimplicitautograd_dispatch.h>
+#include <ATen/ops/_nnpack_available_compositeimplicitautograd_dispatch.h>
+#include <ATen/ops/_pack_padded_sequence_backward_compositeimplicitautograd_dispatch.h>
+#include <ATen/ops/_pad_circular_compositeimplicitautograd_dispatch.h>
+#include <ATen/ops/_pad_enum_compositeimplicitautograd_dispatch.h>
+#include <ATen/ops/_pad_packed_sequence_compositeimplicitautograd_dispatch.h>
+#include <ATen/ops/_propagate_xla_data_compositeimplicitautograd_dispatch.h>
+#include <ATen/ops/_remove_batch_dim_compositeimplicitautograd_dispatch.h>
+#include <ATen/ops/_reshape_from_tensor_compositeimplicitautograd_dispatch.h>
+#include <ATen/ops/_rowwise_prune_compositeimplicitautograd_dispatch.h>
+#include <ATen/ops/_saturate_weight_to_fp16_compositeimplicitautograd_dispatch.h>
+#include <ATen/ops/_scaled_dot_product_attention_math_compositeimplicitautograd_dispatch.h>
+#include <ATen/ops/_shape_as_tensor_compositeimplicitautograd_dispatch.h>
+#include <ATen/ops/_sobol_engine_draw_compositeimplicitautograd_dispatch.h>
+#include <ATen/ops/_sobol_engine_ff_compositeimplicitautograd_dispatch.h>
+#include <ATen/ops/_sobol_engine_initialize_state_compositeimplicitautograd_dispatch.h>
+#include <ATen/ops/_sobol_engine_scramble_compositeimplicitautograd_dispatch.h>
+#include <ATen/ops/_sparse_bsc_tensor_unsafe_compositeimplicitautograd_dispatch.h>
+#include <ATen/ops/_sparse_bsr_tensor_unsafe_compositeimplicitautograd_dispatch.h>
+#include <ATen/ops/_sparse_compressed_tensor_unsafe_compositeimplicitautograd_dispatch.h>
+#include <ATen/ops/_sparse_coo_tensor_unsafe_compositeimplicitautograd_dispatch.h>
+#include <ATen/ops/_sparse_csc_tensor_unsafe_compositeimplicitautograd_dispatch.h>
+#include <ATen/ops/_sparse_csr_tensor_unsafe_compositeimplicitautograd_dispatch.h>
+#include <ATen/ops/_sparse_log_softmax_compositeimplicitautograd_dispatch.h>
+#include <ATen/ops/_sparse_mm_compositeimplicitautograd_dispatch.h>
+#include <ATen/ops/_sparse_softmax_compositeimplicitautograd_dispatch.h>
+#include <ATen/ops/_sparse_sum_compositeimplicitautograd_dispatch.h>
+#include <ATen/ops/_test_ambiguous_defaults_compositeimplicitautograd_dispatch.h>
+#include <ATen/ops/_test_autograd_multiple_dispatch_compositeimplicitautograd_dispatch.h>
+#include <ATen/ops/_test_check_tensor_compositeimplicitautograd_dispatch.h>
+#include <ATen/ops/_test_serialization_subcmul_compositeimplicitautograd_dispatch.h>
+#include <ATen/ops/_test_string_default_compositeimplicitautograd_dispatch.h>
+#include <ATen/ops/_thnn_differentiable_gru_cell_backward_compositeimplicitautograd_dispatch.h>
+#include <ATen/ops/_thnn_differentiable_lstm_cell_backward_compositeimplicitautograd_dispatch.h>
+#include <ATen/ops/_thnn_fused_lstm_cell_backward_compositeimplicitautograd_dispatch.h>
+#include <ATen/ops/_to_cpu_compositeimplicitautograd_dispatch.h>
+#include <ATen/ops/_unpack_dual_compositeimplicitautograd_dispatch.h>
+#include <ATen/ops/_upsample_bicubic2d_aa_compositeimplicitautograd_dispatch.h>
+#include <ATen/ops/_upsample_bilinear2d_aa_compositeimplicitautograd_dispatch.h>
+#include <ATen/ops/_upsample_nearest_exact1d_compositeimplicitautograd_dispatch.h>
+#include <ATen/ops/_upsample_nearest_exact2d_compositeimplicitautograd_dispatch.h>
+#include <ATen/ops/_upsample_nearest_exact3d_compositeimplicitautograd_dispatch.h>
+#include <ATen/ops/_use_cudnn_rnn_flatten_weight_compositeimplicitautograd_dispatch.h>
+#include <ATen/ops/_validate_sparse_bsc_tensor_args_compositeimplicitautograd_dispatch.h>
+#include <ATen/ops/_validate_sparse_bsr_tensor_args_compositeimplicitautograd_dispatch.h>
+#include <ATen/ops/_validate_sparse_compressed_tensor_args_compositeimplicitautograd_dispatch.h>
+#include <ATen/ops/_validate_sparse_coo_tensor_args_compositeimplicitautograd_dispatch.h>
+#include <ATen/ops/_validate_sparse_csc_tensor_args_compositeimplicitautograd_dispatch.h>
+#include <ATen/ops/_validate_sparse_csr_tensor_args_compositeimplicitautograd_dispatch.h>
+#include <ATen/ops/_version_compositeimplicitautograd_dispatch.h>
+#include <ATen/ops/_weight_norm_compositeimplicitautograd_dispatch.h>
+#include <ATen/ops/_weight_norm_differentiable_backward_compositeimplicitautograd_dispatch.h>
+#include <ATen/ops/_wrapped_linear_prepack_compositeimplicitautograd_dispatch.h>
+#include <ATen/ops/_wrapped_quantized_linear_prepacked_compositeimplicitautograd_dispatch.h>
+#include <ATen/ops/absolute_compositeimplicitautograd_dispatch.h>
+#include <ATen/ops/adaptive_avg_pool1d_compositeimplicitautograd_dispatch.h>
+#include <ATen/ops/adaptive_avg_pool2d_compositeimplicitautograd_dispatch.h>
+#include <ATen/ops/adaptive_avg_pool3d_compositeimplicitautograd_dispatch.h>
+#include <ATen/ops/adaptive_max_pool1d_compositeimplicitautograd_dispatch.h>
+#include <ATen/ops/adjoint_compositeimplicitautograd_dispatch.h>
+#include <ATen/ops/affine_grid_generator_backward_compositeimplicitautograd_dispatch.h>
+#include <ATen/ops/align_as_compositeimplicitautograd_dispatch.h>
+#include <ATen/ops/align_tensors_compositeimplicitautograd_dispatch.h>
+#include <ATen/ops/align_to_compositeimplicitautograd_dispatch.h>
+#include <ATen/ops/all_compositeimplicitautograd_dispatch.h>
+#include <ATen/ops/alpha_dropout_compositeimplicitautograd_dispatch.h>
+#include <ATen/ops/and_compositeimplicitautograd_dispatch.h>
+#include <ATen/ops/any_compositeimplicitautograd_dispatch.h>
+#include <ATen/ops/arccos_compositeimplicitautograd_dispatch.h>
+#include <ATen/ops/arccosh_compositeimplicitautograd_dispatch.h>
+#include <ATen/ops/arcsin_compositeimplicitautograd_dispatch.h>
+#include <ATen/ops/arcsinh_compositeimplicitautograd_dispatch.h>
+#include <ATen/ops/arctan_compositeimplicitautograd_dispatch.h>
+#include <ATen/ops/arctan2_compositeimplicitautograd_dispatch.h>
+#include <ATen/ops/arctanh_compositeimplicitautograd_dispatch.h>
+#include <ATen/ops/argsort_compositeimplicitautograd_dispatch.h>
+#include <ATen/ops/argwhere_compositeimplicitautograd_dispatch.h>
+#include <ATen/ops/atleast_1d_compositeimplicitautograd_dispatch.h>
+#include <ATen/ops/atleast_2d_compositeimplicitautograd_dispatch.h>
+#include <ATen/ops/atleast_3d_compositeimplicitautograd_dispatch.h>
+#include <ATen/ops/avg_pool1d_compositeimplicitautograd_dispatch.h>
+#include <ATen/ops/batch_norm_compositeimplicitautograd_dispatch.h>
+#include <ATen/ops/bilinear_compositeimplicitautograd_dispatch.h>
+#include <ATen/ops/broadcast_tensors_compositeimplicitautograd_dispatch.h>
+#include <ATen/ops/broadcast_to_compositeimplicitautograd_dispatch.h>
+#include <ATen/ops/can_cast_compositeimplicitautograd_dispatch.h>
+#include <ATen/ops/cartesian_prod_compositeimplicitautograd_dispatch.h>
+#include <ATen/ops/cat_compositeimplicitautograd_dispatch.h>
+#include <ATen/ops/cdist_compositeimplicitautograd_dispatch.h>
+#include <ATen/ops/chain_matmul_compositeimplicitautograd_dispatch.h>
+#include <ATen/ops/chalf_compositeimplicitautograd_dispatch.h>
+#include <ATen/ops/choose_qparams_optimized_compositeimplicitautograd_dispatch.h>
+#include <ATen/ops/chunk_compositeimplicitautograd_dispatch.h>
+#include <ATen/ops/clip_compositeimplicitautograd_dispatch.h>
+#include <ATen/ops/coalesce_compositeimplicitautograd_dispatch.h>
+#include <ATen/ops/column_stack_compositeimplicitautograd_dispatch.h>
+#include <ATen/ops/combinations_compositeimplicitautograd_dispatch.h>
+#include <ATen/ops/concat_compositeimplicitautograd_dispatch.h>
+#include <ATen/ops/concatenate_compositeimplicitautograd_dispatch.h>
+#include <ATen/ops/conj_compositeimplicitautograd_dispatch.h>
+#include <ATen/ops/conj_physical_compositeimplicitautograd_dispatch.h>
+#include <ATen/ops/contiguous_compositeimplicitautograd_dispatch.h>
+#include <ATen/ops/conv1d_compositeimplicitautograd_dispatch.h>
+#include <ATen/ops/conv2d_compositeimplicitautograd_dispatch.h>
+#include <ATen/ops/conv3d_compositeimplicitautograd_dispatch.h>
+#include <ATen/ops/conv_tbc_backward_compositeimplicitautograd_dispatch.h>
+#include <ATen/ops/conv_transpose1d_compositeimplicitautograd_dispatch.h>
+#include <ATen/ops/conv_transpose2d_compositeimplicitautograd_dispatch.h>
+#include <ATen/ops/conv_transpose3d_compositeimplicitautograd_dispatch.h>
+#include <ATen/ops/corrcoef_compositeimplicitautograd_dispatch.h>
+#include <ATen/ops/cosine_embedding_loss_compositeimplicitautograd_dispatch.h>
+#include <ATen/ops/cosine_similarity_compositeimplicitautograd_dispatch.h>
+#include <ATen/ops/cov_compositeimplicitautograd_dispatch.h>
+#include <ATen/ops/cross_compositeimplicitautograd_dispatch.h>
+#include <ATen/ops/cross_entropy_loss_compositeimplicitautograd_dispatch.h>
+#include <ATen/ops/ctc_loss_compositeimplicitautograd_dispatch.h>
+#include <ATen/ops/cudnn_is_acceptable_compositeimplicitautograd_dispatch.h>
+#include <ATen/ops/cummax_compositeimplicitautograd_dispatch.h>
+#include <ATen/ops/cummaxmin_backward_compositeimplicitautograd_dispatch.h>
+#include <ATen/ops/cummin_compositeimplicitautograd_dispatch.h>
+#include <ATen/ops/cumprod_compositeimplicitautograd_dispatch.h>
+#include <ATen/ops/cumprod_backward_compositeimplicitautograd_dispatch.h>
+#include <ATen/ops/cumsum_compositeimplicitautograd_dispatch.h>
+#include <ATen/ops/cumulative_trapezoid_compositeimplicitautograd_dispatch.h>
+#include <ATen/ops/data_compositeimplicitautograd_dispatch.h>
+#include <ATen/ops/det_compositeimplicitautograd_dispatch.h>
+#include <ATen/ops/diag_compositeimplicitautograd_dispatch.h>
+#include <ATen/ops/diagflat_compositeimplicitautograd_dispatch.h>
+#include <ATen/ops/diagonal_compositeimplicitautograd_dispatch.h>
+#include <ATen/ops/diff_compositeimplicitautograd_dispatch.h>
+#include <ATen/ops/divide_compositeimplicitautograd_dispatch.h>
+#include <ATen/ops/dropout_compositeimplicitautograd_dispatch.h>
+#include <ATen/ops/dsplit_compositeimplicitautograd_dispatch.h>
+#include <ATen/ops/dstack_compositeimplicitautograd_dispatch.h>
+#include <ATen/ops/einsum_compositeimplicitautograd_dispatch.h>
+#include <ATen/ops/embedding_backward_compositeimplicitautograd_dispatch.h>
+#include <ATen/ops/embedding_bag_compositeimplicitautograd_dispatch.h>
+#include <ATen/ops/embedding_sparse_backward_compositeimplicitautograd_dispatch.h>
+#include <ATen/ops/empty_compositeimplicitautograd_dispatch.h>
+#include <ATen/ops/expand_as_compositeimplicitautograd_dispatch.h>
+#include <ATen/ops/fake_quantize_per_channel_affine_compositeimplicitautograd_dispatch.h>
+#include <ATen/ops/fake_quantize_per_channel_affine_cachemask_backward_compositeimplicitautograd_dispatch.h>
+#include <ATen/ops/fake_quantize_per_tensor_affine_compositeimplicitautograd_dispatch.h>
+#include <ATen/ops/fake_quantize_per_tensor_affine_cachemask_backward_compositeimplicitautograd_dispatch.h>
+#include <ATen/ops/fbgemm_linear_fp16_weight_compositeimplicitautograd_dispatch.h>
+#include <ATen/ops/fbgemm_linear_fp16_weight_fp32_activation_compositeimplicitautograd_dispatch.h>
+#include <ATen/ops/fbgemm_linear_int8_weight_compositeimplicitautograd_dispatch.h>
+#include <ATen/ops/fbgemm_linear_int8_weight_fp32_activation_compositeimplicitautograd_dispatch.h>
+#include <ATen/ops/fbgemm_linear_quantize_weight_compositeimplicitautograd_dispatch.h>
+#include <ATen/ops/fbgemm_pack_gemm_matrix_fp16_compositeimplicitautograd_dispatch.h>
+#include <ATen/ops/fbgemm_pack_quantized_matrix_compositeimplicitautograd_dispatch.h>
+#include <ATen/ops/feature_alpha_dropout_compositeimplicitautograd_dispatch.h>
+#include <ATen/ops/feature_dropout_compositeimplicitautograd_dispatch.h>
+#include <ATen/ops/fft_fft_compositeimplicitautograd_dispatch.h>
+#include <ATen/ops/fft_fft2_compositeimplicitautograd_dispatch.h>
+#include <ATen/ops/fft_fftn_compositeimplicitautograd_dispatch.h>
+#include <ATen/ops/fft_fftshift_compositeimplicitautograd_dispatch.h>
+#include <ATen/ops/fft_hfft_compositeimplicitautograd_dispatch.h>
+#include <ATen/ops/fft_hfft2_compositeimplicitautograd_dispatch.h>
+#include <ATen/ops/fft_hfftn_compositeimplicitautograd_dispatch.h>
+#include <ATen/ops/fft_ifft_compositeimplicitautograd_dispatch.h>
+#include <ATen/ops/fft_ifft2_compositeimplicitautograd_dispatch.h>
+#include <ATen/ops/fft_ifftn_compositeimplicitautograd_dispatch.h>
+#include <ATen/ops/fft_ifftshift_compositeimplicitautograd_dispatch.h>
+#include <ATen/ops/fft_ihfft_compositeimplicitautograd_dispatch.h>
+#include <ATen/ops/fft_ihfft2_compositeimplicitautograd_dispatch.h>
+#include <ATen/ops/fft_ihfftn_compositeimplicitautograd_dispatch.h>
+#include <ATen/ops/fft_irfft_compositeimplicitautograd_dispatch.h>
+#include <ATen/ops/fft_irfft2_compositeimplicitautograd_dispatch.h>
+#include <ATen/ops/fft_irfftn_compositeimplicitautograd_dispatch.h>
+#include <ATen/ops/fft_rfft_compositeimplicitautograd_dispatch.h>
+#include <ATen/ops/fft_rfft2_compositeimplicitautograd_dispatch.h>
+#include <ATen/ops/fft_rfftn_compositeimplicitautograd_dispatch.h>
+#include <ATen/ops/fill_diagonal_compositeimplicitautograd_dispatch.h>
+#include <ATen/ops/fix_compositeimplicitautograd_dispatch.h>
+#include <ATen/ops/flatten_compositeimplicitautograd_dispatch.h>
+#include <ATen/ops/flatten_dense_tensors_compositeimplicitautograd_dispatch.h>
+#include <ATen/ops/fliplr_compositeimplicitautograd_dispatch.h>
+#include <ATen/ops/flipud_compositeimplicitautograd_dispatch.h>
+#include <ATen/ops/float_power_compositeimplicitautograd_dispatch.h>
+#include <ATen/ops/frobenius_norm_compositeimplicitautograd_dispatch.h>
+#include <ATen/ops/fused_moving_avg_obs_fake_quant_compositeimplicitautograd_dispatch.h>
+#include <ATen/ops/gather_compositeimplicitautograd_dispatch.h>
+#include <ATen/ops/gather_backward_compositeimplicitautograd_dispatch.h>
+#include <ATen/ops/ger_compositeimplicitautograd_dispatch.h>
+#include <ATen/ops/gradient_compositeimplicitautograd_dispatch.h>
+#include <ATen/ops/greater_compositeimplicitautograd_dispatch.h>
+#include <ATen/ops/greater_equal_compositeimplicitautograd_dispatch.h>
+#include <ATen/ops/grid_sampler_compositeimplicitautograd_dispatch.h>
+#include <ATen/ops/group_norm_compositeimplicitautograd_dispatch.h>
+#include <ATen/ops/gru_compositeimplicitautograd_dispatch.h>
+#include <ATen/ops/gru_cell_compositeimplicitautograd_dispatch.h>
+#include <ATen/ops/hinge_embedding_loss_compositeimplicitautograd_dispatch.h>
+#include <ATen/ops/histogramdd_compositeimplicitautograd_dispatch.h>
+#include <ATen/ops/hsplit_compositeimplicitautograd_dispatch.h>
+#include <ATen/ops/hstack_compositeimplicitautograd_dispatch.h>
+#include <ATen/ops/imag_compositeimplicitautograd_dispatch.h>
+#include <ATen/ops/index_add_compositeimplicitautograd_dispatch.h>
+#include <ATen/ops/index_copy_compositeimplicitautograd_dispatch.h>
+#include <ATen/ops/index_fill_compositeimplicitautograd_dispatch.h>
+#include <ATen/ops/index_select_compositeimplicitautograd_dispatch.h>
+#include <ATen/ops/index_select_backward_compositeimplicitautograd_dispatch.h>
+#include <ATen/ops/infinitely_differentiable_gelu_backward_compositeimplicitautograd_dispatch.h>
+#include <ATen/ops/inner_compositeimplicitautograd_dispatch.h>
+#include <ATen/ops/instance_norm_compositeimplicitautograd_dispatch.h>
+#include <ATen/ops/inverse_compositeimplicitautograd_dispatch.h>
+#include <ATen/ops/is_complex_compositeimplicitautograd_dispatch.h>
+#include <ATen/ops/is_conj_compositeimplicitautograd_dispatch.h>
+#include <ATen/ops/is_distributed_compositeimplicitautograd_dispatch.h>
+#include <ATen/ops/is_floating_point_compositeimplicitautograd_dispatch.h>
+#include <ATen/ops/is_inference_compositeimplicitautograd_dispatch.h>
+#include <ATen/ops/is_leaf_compositeimplicitautograd_dispatch.h>
+#include <ATen/ops/is_neg_compositeimplicitautograd_dispatch.h>
+#include <ATen/ops/is_nonzero_compositeimplicitautograd_dispatch.h>
+#include <ATen/ops/is_signed_compositeimplicitautograd_dispatch.h>
+#include <ATen/ops/is_vulkan_available_compositeimplicitautograd_dispatch.h>
+#include <ATen/ops/isclose_compositeimplicitautograd_dispatch.h>
+#include <ATen/ops/isfinite_compositeimplicitautograd_dispatch.h>
+#include <ATen/ops/isreal_compositeimplicitautograd_dispatch.h>
+#include <ATen/ops/istft_compositeimplicitautograd_dispatch.h>
+#include <ATen/ops/item_compositeimplicitautograd_dispatch.h>
+#include <ATen/ops/kl_div_compositeimplicitautograd_dispatch.h>
+#include <ATen/ops/kron_compositeimplicitautograd_dispatch.h>
+#include <ATen/ops/kthvalue_compositeimplicitautograd_dispatch.h>
+#include <ATen/ops/l1_loss_compositeimplicitautograd_dispatch.h>
+#include <ATen/ops/layer_norm_compositeimplicitautograd_dispatch.h>
+#include <ATen/ops/ldexp_compositeimplicitautograd_dispatch.h>
+#include <ATen/ops/less_compositeimplicitautograd_dispatch.h>
+#include <ATen/ops/less_equal_compositeimplicitautograd_dispatch.h>
+#include <ATen/ops/linalg_cholesky_compositeimplicitautograd_dispatch.h>
+#include <ATen/ops/linalg_cond_compositeimplicitautograd_dispatch.h>
+#include <ATen/ops/linalg_det_compositeimplicitautograd_dispatch.h>
+#include <ATen/ops/linalg_diagonal_compositeimplicitautograd_dispatch.h>
+#include <ATen/ops/linalg_eigh_compositeimplicitautograd_dispatch.h>
+#include <ATen/ops/linalg_eigvals_compositeimplicitautograd_dispatch.h>
+#include <ATen/ops/linalg_eigvalsh_compositeimplicitautograd_dispatch.h>
+#include <ATen/ops/linalg_inv_compositeimplicitautograd_dispatch.h>
+#include <ATen/ops/linalg_ldl_factor_compositeimplicitautograd_dispatch.h>
+#include <ATen/ops/linalg_lu_factor_compositeimplicitautograd_dispatch.h>
+#include <ATen/ops/linalg_matmul_compositeimplicitautograd_dispatch.h>
+#include <ATen/ops/linalg_matrix_norm_compositeimplicitautograd_dispatch.h>
+#include <ATen/ops/linalg_matrix_power_compositeimplicitautograd_dispatch.h>
+#include <ATen/ops/linalg_matrix_rank_compositeimplicitautograd_dispatch.h>
+#include <ATen/ops/linalg_multi_dot_compositeimplicitautograd_dispatch.h>
+#include <ATen/ops/linalg_norm_compositeimplicitautograd_dispatch.h>
+#include <ATen/ops/linalg_pinv_compositeimplicitautograd_dispatch.h>
+#include <ATen/ops/linalg_slogdet_compositeimplicitautograd_dispatch.h>
+#include <ATen/ops/linalg_solve_compositeimplicitautograd_dispatch.h>
+#include <ATen/ops/linalg_solve_ex_compositeimplicitautograd_dispatch.h>
+#include <ATen/ops/linalg_svd_compositeimplicitautograd_dispatch.h>
+#include <ATen/ops/linalg_svdvals_compositeimplicitautograd_dispatch.h>
+#include <ATen/ops/linalg_tensorinv_compositeimplicitautograd_dispatch.h>
+#include <ATen/ops/linalg_tensorsolve_compositeimplicitautograd_dispatch.h>
+#include <ATen/ops/linalg_vander_compositeimplicitautograd_dispatch.h>
+#include <ATen/ops/linalg_vecdot_compositeimplicitautograd_dispatch.h>
+#include <ATen/ops/linear_compositeimplicitautograd_dispatch.h>
+#include <ATen/ops/log_sigmoid_compositeimplicitautograd_dispatch.h>
+#include <ATen/ops/log_softmax_compositeimplicitautograd_dispatch.h>
+#include <ATen/ops/logcumsumexp_compositeimplicitautograd_dispatch.h>
+#include <ATen/ops/logdet_compositeimplicitautograd_dispatch.h>
+#include <ATen/ops/logsumexp_compositeimplicitautograd_dispatch.h>
+#include <ATen/ops/lstm_compositeimplicitautograd_dispatch.h>
+#include <ATen/ops/lstm_cell_compositeimplicitautograd_dispatch.h>
+#include <ATen/ops/lu_solve_compositeimplicitautograd_dispatch.h>
+#include <ATen/ops/mH_compositeimplicitautograd_dispatch.h>
+#include <ATen/ops/mT_compositeimplicitautograd_dispatch.h>
+#include <ATen/ops/margin_ranking_loss_compositeimplicitautograd_dispatch.h>
+#include <ATen/ops/masked_select_backward_compositeimplicitautograd_dispatch.h>
+#include <ATen/ops/matmul_compositeimplicitautograd_dispatch.h>
+#include <ATen/ops/matrix_H_compositeimplicitautograd_dispatch.h>
+#include <ATen/ops/matrix_exp_compositeimplicitautograd_dispatch.h>
+#include <ATen/ops/matrix_exp_backward_compositeimplicitautograd_dispatch.h>
+#include <ATen/ops/matrix_power_compositeimplicitautograd_dispatch.h>
+#include <ATen/ops/max_compositeimplicitautograd_dispatch.h>
+#include <ATen/ops/max_pool1d_compositeimplicitautograd_dispatch.h>
+#include <ATen/ops/max_pool1d_with_indices_compositeimplicitautograd_dispatch.h>
+#include <ATen/ops/max_pool2d_compositeimplicitautograd_dispatch.h>
+#include <ATen/ops/max_pool3d_compositeimplicitautograd_dispatch.h>
+#include <ATen/ops/mean_compositeimplicitautograd_dispatch.h>
+#include <ATen/ops/median_compositeimplicitautograd_dispatch.h>
+#include <ATen/ops/meshgrid_compositeimplicitautograd_dispatch.h>
+#include <ATen/ops/min_compositeimplicitautograd_dispatch.h>
+#include <ATen/ops/mish_backward_compositeimplicitautograd_dispatch.h>
+#include <ATen/ops/mode_compositeimplicitautograd_dispatch.h>
+#include <ATen/ops/moveaxis_compositeimplicitautograd_dispatch.h>
+#include <ATen/ops/movedim_compositeimplicitautograd_dispatch.h>
+#include <ATen/ops/msort_compositeimplicitautograd_dispatch.h>
+#include <ATen/ops/multilabel_margin_loss_compositeimplicitautograd_dispatch.h>
+#include <ATen/ops/multiply_compositeimplicitautograd_dispatch.h>
+#include <ATen/ops/nanmean_compositeimplicitautograd_dispatch.h>
+#include <ATen/ops/nanmedian_compositeimplicitautograd_dispatch.h>
+#include <ATen/ops/nanquantile_compositeimplicitautograd_dispatch.h>
+#include <ATen/ops/narrow_compositeimplicitautograd_dispatch.h>
+#include <ATen/ops/native_channel_shuffle_compositeimplicitautograd_dispatch.h>
+#include <ATen/ops/negative_compositeimplicitautograd_dispatch.h>
+#include <ATen/ops/nested_to_padded_tensor_compositeimplicitautograd_dispatch.h>
+#include <ATen/ops/nll_loss_compositeimplicitautograd_dispatch.h>
+#include <ATen/ops/nll_loss2d_compositeimplicitautograd_dispatch.h>
+#include <ATen/ops/nll_loss_nd_compositeimplicitautograd_dispatch.h>
+#include <ATen/ops/nonzero_numpy_compositeimplicitautograd_dispatch.h>
+#include <ATen/ops/norm_compositeimplicitautograd_dispatch.h>
+#include <ATen/ops/norm_except_dim_compositeimplicitautograd_dispatch.h>
+#include <ATen/ops/not_equal_compositeimplicitautograd_dispatch.h>
+#include <ATen/ops/nuclear_norm_compositeimplicitautograd_dispatch.h>
+#include <ATen/ops/numpy_T_compositeimplicitautograd_dispatch.h>
+#include <ATen/ops/one_hot_compositeimplicitautograd_dispatch.h>
+#include <ATen/ops/or_compositeimplicitautograd_dispatch.h>
+#include <ATen/ops/orgqr_compositeimplicitautograd_dispatch.h>
+#include <ATen/ops/outer_compositeimplicitautograd_dispatch.h>
+#include <ATen/ops/output_nr_compositeimplicitautograd_dispatch.h>
+#include <ATen/ops/pad_compositeimplicitautograd_dispatch.h>
+#include <ATen/ops/pad_sequence_compositeimplicitautograd_dispatch.h>
+#include <ATen/ops/pairwise_distance_compositeimplicitautograd_dispatch.h>
+#include <ATen/ops/pdist_compositeimplicitautograd_dispatch.h>
+#include <ATen/ops/pin_memory_compositeimplicitautograd_dispatch.h>
+#include <ATen/ops/pinverse_compositeimplicitautograd_dispatch.h>
+#include <ATen/ops/poisson_nll_loss_compositeimplicitautograd_dispatch.h>
+#include <ATen/ops/positive_compositeimplicitautograd_dispatch.h>
+#include <ATen/ops/prelu_compositeimplicitautograd_dispatch.h>
+#include <ATen/ops/prod_compositeimplicitautograd_dispatch.h>
+#include <ATen/ops/promote_types_compositeimplicitautograd_dispatch.h>
+#include <ATen/ops/qr_compositeimplicitautograd_dispatch.h>
+#include <ATen/ops/quantile_compositeimplicitautograd_dispatch.h>
+#include <ATen/ops/quantized_gru_cell_compositeimplicitautograd_dispatch.h>
+#include <ATen/ops/quantized_lstm_cell_compositeimplicitautograd_dispatch.h>
+#include <ATen/ops/quantized_rnn_relu_cell_compositeimplicitautograd_dispatch.h>
+#include <ATen/ops/quantized_rnn_tanh_cell_compositeimplicitautograd_dispatch.h>
+#include <ATen/ops/rand_compositeimplicitautograd_dispatch.h>
+#include <ATen/ops/randn_compositeimplicitautograd_dispatch.h>
+#include <ATen/ops/ravel_compositeimplicitautograd_dispatch.h>
+#include <ATen/ops/real_compositeimplicitautograd_dispatch.h>
+#include <ATen/ops/refine_names_compositeimplicitautograd_dispatch.h>
+#include <ATen/ops/relu6_compositeimplicitautograd_dispatch.h>
+#include <ATen/ops/rename_compositeimplicitautograd_dispatch.h>
+#include <ATen/ops/repeat_interleave_compositeimplicitautograd_dispatch.h>
+#include <ATen/ops/requires_grad_compositeimplicitautograd_dispatch.h>
+#include <ATen/ops/reshape_compositeimplicitautograd_dispatch.h>
+#include <ATen/ops/reshape_as_compositeimplicitautograd_dispatch.h>
+#include <ATen/ops/resolve_conj_compositeimplicitautograd_dispatch.h>
+#include <ATen/ops/resolve_neg_compositeimplicitautograd_dispatch.h>
+#include <ATen/ops/result_type_compositeimplicitautograd_dispatch.h>
+#include <ATen/ops/retain_grad_compositeimplicitautograd_dispatch.h>
+#include <ATen/ops/retains_grad_compositeimplicitautograd_dispatch.h>
+#include <ATen/ops/rms_norm_compositeimplicitautograd_dispatch.h>
+#include <ATen/ops/rnn_relu_compositeimplicitautograd_dispatch.h>
+#include <ATen/ops/rnn_relu_cell_compositeimplicitautograd_dispatch.h>
+#include <ATen/ops/rnn_tanh_compositeimplicitautograd_dispatch.h>
+#include <ATen/ops/rnn_tanh_cell_compositeimplicitautograd_dispatch.h>
+#include <ATen/ops/row_stack_compositeimplicitautograd_dispatch.h>
+#include <ATen/ops/rrelu_compositeimplicitautograd_dispatch.h>
+#include <ATen/ops/scaled_dot_product_attention_compositeimplicitautograd_dispatch.h>
+#include <ATen/ops/scatter_compositeimplicitautograd_dispatch.h>
+#include <ATen/ops/scatter_add_compositeimplicitautograd_dispatch.h>
+#include <ATen/ops/select_compositeimplicitautograd_dispatch.h>
+#include <ATen/ops/selu_compositeimplicitautograd_dispatch.h>
+#include <ATen/ops/set_compositeimplicitautograd_dispatch.h>
+#include <ATen/ops/set_data_compositeimplicitautograd_dispatch.h>
+#include <ATen/ops/silu_backward_compositeimplicitautograd_dispatch.h>
+#include <ATen/ops/size_compositeimplicitautograd_dispatch.h>
+#include <ATen/ops/slogdet_compositeimplicitautograd_dispatch.h>
+#include <ATen/ops/slow_conv3d_compositeimplicitautograd_dispatch.h>
+#include <ATen/ops/smm_compositeimplicitautograd_dispatch.h>
+#include <ATen/ops/softmax_compositeimplicitautograd_dispatch.h>
+#include <ATen/ops/sort_compositeimplicitautograd_dispatch.h>
+#include <ATen/ops/sparse_bsc_tensor_compositeimplicitautograd_dispatch.h>
+#include <ATen/ops/sparse_bsr_tensor_compositeimplicitautograd_dispatch.h>
+#include <ATen/ops/sparse_coo_tensor_compositeimplicitautograd_dispatch.h>
+#include <ATen/ops/sparse_csc_tensor_compositeimplicitautograd_dispatch.h>
+#include <ATen/ops/sparse_csr_tensor_compositeimplicitautograd_dispatch.h>
+#include <ATen/ops/special_digamma_compositeimplicitautograd_dispatch.h>
+#include <ATen/ops/special_erf_compositeimplicitautograd_dispatch.h>
+#include <ATen/ops/special_erfc_compositeimplicitautograd_dispatch.h>
+#include <ATen/ops/special_erfinv_compositeimplicitautograd_dispatch.h>
+#include <ATen/ops/special_exp2_compositeimplicitautograd_dispatch.h>
+#include <ATen/ops/special_expit_compositeimplicitautograd_dispatch.h>
+#include <ATen/ops/special_expm1_compositeimplicitautograd_dispatch.h>
+#include <ATen/ops/special_gammainc_compositeimplicitautograd_dispatch.h>
+#include <ATen/ops/special_gammaincc_compositeimplicitautograd_dispatch.h>
+#include <ATen/ops/special_gammaln_compositeimplicitautograd_dispatch.h>
+#include <ATen/ops/special_i0_compositeimplicitautograd_dispatch.h>
+#include <ATen/ops/special_log1p_compositeimplicitautograd_dispatch.h>
+#include <ATen/ops/special_log_softmax_compositeimplicitautograd_dispatch.h>
+#include <ATen/ops/special_logit_compositeimplicitautograd_dispatch.h>
+#include <ATen/ops/special_logsumexp_compositeimplicitautograd_dispatch.h>
+#include <ATen/ops/special_multigammaln_compositeimplicitautograd_dispatch.h>
+#include <ATen/ops/special_ndtr_compositeimplicitautograd_dispatch.h>
+#include <ATen/ops/special_polygamma_compositeimplicitautograd_dispatch.h>
+#include <ATen/ops/special_psi_compositeimplicitautograd_dispatch.h>
+#include <ATen/ops/special_round_compositeimplicitautograd_dispatch.h>
+#include <ATen/ops/special_sinc_compositeimplicitautograd_dispatch.h>
+#include <ATen/ops/special_softmax_compositeimplicitautograd_dispatch.h>
+#include <ATen/ops/special_xlogy_compositeimplicitautograd_dispatch.h>
+#include <ATen/ops/split_compositeimplicitautograd_dispatch.h>
+#include <ATen/ops/square_compositeimplicitautograd_dispatch.h>
+#include <ATen/ops/squeeze_compositeimplicitautograd_dispatch.h>
+#include <ATen/ops/sspaddmm_compositeimplicitautograd_dispatch.h>
+#include <ATen/ops/std_compositeimplicitautograd_dispatch.h>
+#include <ATen/ops/std_mean_compositeimplicitautograd_dispatch.h>
+#include <ATen/ops/stft_compositeimplicitautograd_dispatch.h>
+#include <ATen/ops/stride_compositeimplicitautograd_dispatch.h>
+#include <ATen/ops/subtract_compositeimplicitautograd_dispatch.h>
+#include <ATen/ops/sum_compositeimplicitautograd_dispatch.h>
+#include <ATen/ops/sum_to_size_compositeimplicitautograd_dispatch.h>
+#include <ATen/ops/svd_compositeimplicitautograd_dispatch.h>
+#include <ATen/ops/swapaxes_compositeimplicitautograd_dispatch.h>
+#include <ATen/ops/swapdims_compositeimplicitautograd_dispatch.h>
+#include <ATen/ops/sym_numel_compositeimplicitautograd_dispatch.h>
+#include <ATen/ops/sym_size_compositeimplicitautograd_dispatch.h>
+#include <ATen/ops/sym_storage_offset_compositeimplicitautograd_dispatch.h>
+#include <ATen/ops/sym_stride_compositeimplicitautograd_dispatch.h>
+#include <ATen/ops/take_along_dim_compositeimplicitautograd_dispatch.h>
+#include <ATen/ops/tensor_split_compositeimplicitautograd_dispatch.h>
+#include <ATen/ops/tensordot_compositeimplicitautograd_dispatch.h>
+#include <ATen/ops/thnn_conv2d_compositeimplicitautograd_dispatch.h>
+#include <ATen/ops/tile_compositeimplicitautograd_dispatch.h>
+#include <ATen/ops/to_compositeimplicitautograd_dispatch.h>
+#include <ATen/ops/to_dense_compositeimplicitautograd_dispatch.h>
+#include <ATen/ops/to_dense_backward_compositeimplicitautograd_dispatch.h>
+#include <ATen/ops/to_mkldnn_backward_compositeimplicitautograd_dispatch.h>
+#include <ATen/ops/to_sparse_compositeimplicitautograd_dispatch.h>
+#include <ATen/ops/to_sparse_bsc_compositeimplicitautograd_dispatch.h>
+#include <ATen/ops/to_sparse_bsr_compositeimplicitautograd_dispatch.h>
+#include <ATen/ops/to_sparse_csc_compositeimplicitautograd_dispatch.h>
+#include <ATen/ops/to_sparse_csr_compositeimplicitautograd_dispatch.h>
+#include <ATen/ops/trace_backward_compositeimplicitautograd_dispatch.h>
+#include <ATen/ops/transpose_compositeimplicitautograd_dispatch.h>
+#include <ATen/ops/trapezoid_compositeimplicitautograd_dispatch.h>
+#include <ATen/ops/trapz_compositeimplicitautograd_dispatch.h>
+#include <ATen/ops/triplet_margin_loss_compositeimplicitautograd_dispatch.h>
+#include <ATen/ops/true_divide_compositeimplicitautograd_dispatch.h>
+#include <ATen/ops/type_as_compositeimplicitautograd_dispatch.h>
+#include <ATen/ops/unbind_compositeimplicitautograd_dispatch.h>
+#include <ATen/ops/unflatten_compositeimplicitautograd_dispatch.h>
+#include <ATen/ops/unflatten_dense_tensors_compositeimplicitautograd_dispatch.h>
+#include <ATen/ops/unsafe_chunk_compositeimplicitautograd_dispatch.h>
+#include <ATen/ops/upsample_bicubic2d_compositeimplicitautograd_dispatch.h>
+#include <ATen/ops/upsample_bilinear2d_compositeimplicitautograd_dispatch.h>
+#include <ATen/ops/upsample_linear1d_compositeimplicitautograd_dispatch.h>
+#include <ATen/ops/upsample_nearest1d_compositeimplicitautograd_dispatch.h>
+#include <ATen/ops/upsample_nearest2d_compositeimplicitautograd_dispatch.h>
+#include <ATen/ops/upsample_nearest3d_compositeimplicitautograd_dispatch.h>
+#include <ATen/ops/upsample_trilinear3d_compositeimplicitautograd_dispatch.h>
+#include <ATen/ops/value_selecting_reduction_backward_compositeimplicitautograd_dispatch.h>
+#include <ATen/ops/vander_compositeimplicitautograd_dispatch.h>
+#include <ATen/ops/var_compositeimplicitautograd_dispatch.h>
+#include <ATen/ops/var_mean_compositeimplicitautograd_dispatch.h>
+#include <ATen/ops/view_as_compositeimplicitautograd_dispatch.h>
+#include <ATen/ops/vsplit_compositeimplicitautograd_dispatch.h>
+#include <ATen/ops/vstack_compositeimplicitautograd_dispatch.h>
+#include <ATen/ops/where_compositeimplicitautograd_dispatch.h>
+#include <ATen/ops/xor_compositeimplicitautograd_dispatch.h>
+
+
+

.venv/Lib/site-packages/torch/include/ATen/CompositeImplicitAutogradNestedTensorFunctions.h

@@ -0,0 +1,29 @@
+#include <ATen/core/TensorBody.h>
+
+// TODO Undo all logic introduced for Note [Avoiding Include Cycles In Static Dispatch]
+// Code introduced to avoid cyclic dependency in static dispatch is no longer
+// needed as static dispatch logic is moved from TensorBody.h, which caused cycles in the first place,
+// to Operators.cpp for supporting multiple backends with multiple kernels.
+//
+// Note [Avoiding Include Cycles In Static Dispatch]
+// In order to avoid #include cycles in the static dispatch build, we've carefully split out
+// the static function definition files into {DispatchKey}Functions.h and {DispatchKey}Functions_inl.h.
+//
+// Without this split, the include cycle looks like TensorBody.h -> CPUFunctions.h -> TensorBody.h.
+// - TensorBody.h #includes CPUFunctions.h in the static dispatch build, because the tensor methods
+//   all need to call into the fastpath C++ API defined in CPUFunctions.h. The methods are also all
+//   directly inlined into TensorBody.h.
+// - CPUFunctions.h #includes TensorBody.h because it contains function declarations for the entire C++ API,
+//   which include functions that have defaultable std::optional<Tensor> arguments.
+//   That requires knowing the full Tensor class definition.
+//
+// We break the cycle by doing the following:
+// - Split out CPUFunction.h into two files: CPUFunctions.h and CPUFunctions_inl.h
+// - CPUFunction.h is a dummy file that just includes the Tensor class and includes CPUFunctions_inl.,
+// - CPUFunctions_inl.h includes everything else
+// - (only in the static dispatch build) TensorBody.h makes sure to finish defining the Tensor class,
+//   and then it includes CPUFunctions_inl.h.
+// - All other files that want the cpu fastpath functions can include CPUFunctions.h directly.
+// - This also means that static dispatch build, CPUFunctions.h only needs to
+//   #include TensorBody.h, and it will automatically bring in CPUFunctions_inl.h.
+#include <ATen/CompositeImplicitAutogradNestedTensorFunctions_inl.h>

.venv/Lib/site-packages/torch/include/ATen/CompositeImplicitAutogradNestedTensorFunctions_inl.h

@@ -0,0 +1,25 @@
+#pragma once
+// @generated by torchgen/gen.py from DispatchKeyFunctions_inl.h
+
+// NB: The implementing C++ file is RegisterDispatchKey.cpp
+
+// The only #includes we need are for custom classes that have defaults in the C++ API
+#include <c10/core/MemoryFormat.h>
+#include <c10/core/Scalar.h>
+#include <ATen/core/Reduction.h>
+
+#if defined(AT_PER_OPERATOR_HEADERS) && defined(TORCH_ASSERT_ONLY_METHOD_OPERATORS)
+#error This change adds a dependency on all pytorch operators, meaning the     \
+  file will need to be re-compiled every time an operator is changed or added. \
+  Consider including a specific operator from                                  \
+  <ATen/ops/{my_operator}_compositeimplicitautogradnestedtensor_dispatch.h>.                   \
+  See NOTE [TORCH_ASSERT_ONLY_METHOD_OPERATORS].
+#endif
+
+#include <ATen/ops/randn_like_compositeimplicitautogradnestedtensor_dispatch.h>
+#include <ATen/ops/reshape_compositeimplicitautogradnestedtensor_dispatch.h>
+#include <ATen/ops/reshape_as_compositeimplicitautogradnestedtensor_dispatch.h>
+#include <ATen/ops/zeros_like_compositeimplicitautogradnestedtensor_dispatch.h>
+
+
+

.venv/Lib/site-packages/torch/include/ATen/Config.h

@@ -0,0 +1,21 @@
+#pragma once
+
+// Test these using #if AT_MKL_ENABLED(), not #ifdef, so that it's
+// obvious if you forgot to include Config.h
+//    c.f. https://stackoverflow.com/questions/33759787/generating-an-error-if-checked-boolean-macro-is-not-defined
+//
+// DO NOT put the macros for CUDA libraries in this file; they belong in cuda/CUDAConfig.h
+
+#define AT_MKLDNN_ENABLED() 1
+#define AT_MKLDNN_ACL_ENABLED() 0
+#define AT_MKL_ENABLED() 1
+#define AT_MKL_SEQUENTIAL() 0
+#define AT_POCKETFFT_ENABLED() 0
+#define AT_NNPACK_ENABLED() 0
+#define CAFFE2_STATIC_LINK_CUDA() 0
+#define AT_BUILD_WITH_BLAS() 1
+#define AT_BUILD_WITH_LAPACK() 1
+#define AT_PARALLEL_OPENMP 1
+#define AT_PARALLEL_NATIVE 0
+#define AT_BLAS_F2C() 0
+#define AT_BLAS_USE_CBLAS_DOT() 0

.venv/Lib/site-packages/torch/include/ATen/Context.h

@@ -0,0 +1,610 @@
+#pragma once
+
+#include <ATen/BlasBackend.h>
+#include <ATen/CPUGeneratorImpl.h>
+#include <ATen/DeviceAccelerator.h>
+#include <ATen/LinalgBackend.h>
+#include <ATen/core/ATenGeneral.h>
+#include <ATen/core/DeprecatedTypeProperties.h>
+#include <ATen/core/Generator.h>
+#include <ATen/core/LegacyTypeDispatch.h>
+#include <ATen/detail/AcceleratorHooksInterface.h>
+#include <ATen/detail/CUDAHooksInterface.h>
+#include <ATen/detail/HIPHooksInterface.h>
+#include <ATen/detail/IPUHooksInterface.h>
+#include <ATen/detail/MAIAHooksInterface.h>
+#include <ATen/detail/MPSHooksInterface.h>
+#include <ATen/detail/MTIAHooksInterface.h>
+#include <ATen/detail/PrivateUse1HooksInterface.h>
+#include <ATen/detail/XPUHooksInterface.h>
+#include <c10/core/QEngine.h>
+#include <c10/core/impl/DeviceGuardImplInterface.h>
+#include <c10/util/CallOnce.h>
+#include <c10/util/Exception.h>
+#include <c10/util/env.h>
+#include <c10/util/irange.h>
+
+#include <cstdint>
+#include <mutex>
+
+namespace at {
+
+class Tensor;
+
+enum class TORCH_API Float32MatmulPrecision { HIGHEST, HIGH, MEDIUM };
+
+class TORCH_API Context {
+ public:
+  Context();
+
+  const Generator& defaultGenerator(Device device) {
+    c10::DeviceType device_type = device.type();
+    initCUDAIfNeeded(device_type);
+    initHIPIfNeeded(device_type);
+    if (device_type == at::kCPU) {
+      return at::detail::getDefaultCPUGenerator();
+    } else if (device_type == at::kCUDA) {
+      return at::detail::getCUDAHooks().getDefaultCUDAGenerator(device.index());
+    } else if (device_type == at::kMPS) {
+      return at::detail::getMPSHooks().getDefaultMPSGenerator();
+    } else if (device_type == at::kXPU) {
+      return at::detail::getXPUHooks().getDefaultXPUGenerator(device.index());
+    } else if (device_type == at::kIPU) {
+      return at::detail::getIPUHooks().getDefaultIPUGenerator(device.index());
+    } else if (device_type == at::kPrivateUse1) {
+      return at::detail::getPrivateUse1Hooks().getDefaultGenerator(
+          device.index());
+    } else {
+      AT_ERROR(c10::DeviceTypeName(device_type), " device type not enabled.");
+    }
+  }
+  const AcceleratorHooksInterface& getAcceleratorHooksInterface(
+      std::optional<c10::DeviceType> opt_device_type = std::nullopt) {
+    c10::DeviceType device_type = opt_device_type.has_value()
+        ? opt_device_type.value()
+        : at::getAccelerator(true).value();
+    if (device_type == at::kCUDA) {
+      return at::detail::getCUDAHooks();
+    } else if (device_type == at::kXPU) {
+      return at::detail::getXPUHooks();
+    } else if (device_type == at::kMPS) {
+      return at::detail::getMPSHooks();
+    } else if (device_type == at::kPrivateUse1) {
+      return at::detail::getPrivateUse1Hooks();
+    } else if (device_type == at::kMTIA) {
+      return at::detail::getMTIAHooks();
+    } else if (device_type == at::kHIP) {
+      return at::detail::getHIPHooks();
+    } else {
+      AT_ERROR(
+          c10::DeviceTypeName(device_type), " device type not an accelerator.");
+    }
+  }
+  Device getDeviceFromPtr(void* data, c10::DeviceType device_type) {
+    initCUDAIfNeeded(device_type);
+    initHIPIfNeeded(device_type);
+    initXPUIfNeeded(device_type);
+    if (device_type == at::kCPU) {
+      return c10::DeviceType::CPU;
+    } else if (device_type == at::kCUDA) {
+      return at::detail::getCUDAHooks().getDeviceFromPtr(data);
+    } else if (device_type == at::kXPU) {
+      return at::detail::getXPUHooks().getDeviceFromPtr(data);
+    } else if (device_type == at::kPrivateUse1) {
+      return at::detail::getPrivateUse1Hooks().getDeviceFromPtr(data);
+    } else {
+      AT_ERROR(c10::DeviceTypeName(device_type), " device type not enabled.");
+    }
+  }
+  bool isPinnedPtr(
+      const void* data,
+      std::optional<c10::DeviceType> device_type = std::nullopt) {
+    auto opt_device_type =
+        device_type.has_value() ? device_type : at::getAccelerator();
+    if (!opt_device_type.has_value() || // there is no accelerator
+        !at::isAccelerator(
+            opt_device_type.value())) { // passed device not an accelerator
+      return false;
+    }
+    return getAcceleratorHooksInterface(opt_device_type.value())
+        .isPinnedPtr(data);
+  }
+  Allocator* getPinnedMemoryAllocator(
+      std::optional<c10::DeviceType> device_type = std::nullopt) {
+    return getAcceleratorHooksInterface(device_type).getPinnedMemoryAllocator();
+  }
+  static bool hasOpenMP();
+  static bool hasMKL();
+  static bool hasLAPACK();
+  static bool hasMKLDNN();
+  static bool hasMAGMA() {
+    return detail::getCUDAHooks().hasMAGMA();
+  }
+  static bool hasCUDA() {
+    return detail::getCUDAHooks().hasCUDA();
+  }
+  static bool hasMTIA() {
+    return detail::getMTIAHooks().hasMTIA();
+  }
+  static bool hasCUDART() {
+    return detail::getCUDAHooks().hasCUDART();
+  }
+  static long versionCUDART() {
+    return detail::getCUDAHooks().versionCUDART();
+  }
+  static bool hasCuDNN() {
+    return detail::getCUDAHooks().hasCuDNN();
+  }
+  static long versionCuDNN() {
+    return detail::getCUDAHooks().versionCuDNN();
+  }
+  static bool hasCuSOLVER() {
+    return detail::getCUDAHooks().hasCuSOLVER();
+  }
+  static bool hasCuBLASLt() {
+    return detail::getCUDAHooks().hasCuBLASLt();
+  }
+  static bool hasHIP() {
+    return detail::getHIPHooks().hasHIP();
+  }
+  static bool hasMPS() {
+    return detail::getMPSHooks().hasMPS();
+  }
+  static bool hasIPU() {
+    return c10::impl::hasDeviceGuardImpl(c10::DeviceType::IPU);
+  }
+  static bool hasXLA() {
+    return c10::impl::hasDeviceGuardImpl(c10::DeviceType::XLA);
+  }
+  static bool hasXPU() {
+    return detail::getXPUHooks().hasXPU();
+  }
+  static bool hasLazy() {
+    return c10::impl::hasDeviceGuardImpl(c10::DeviceType::Lazy);
+  }
+  static bool hasMAIA() {
+    return c10::impl::hasDeviceGuardImpl(c10::DeviceType::MAIA);
+  }
+  // defined in header so that getNonVariableType has ability to inline
+  // call_once check. getNonVariableType is called fairly frequently
+  void lazyInitCUDA() {
+    c10::call_once(thc_init, [&] { detail::getCUDAHooks().initCUDA(); });
+  }
+  void lazyInitHIP() {
+    c10::call_once(thh_init, [&] { detail::getHIPHooks().initHIP(); });
+  }
+  void lazyInitXPU() {
+    c10::call_once(thx_init, [&] { detail::getXPUHooks().initXPU(); });
+  }
+  void lazyInitMTIA() {
+    c10::call_once(th_mtia_init, [&] { detail::getMTIAHooks().initMTIA(); });
+  }
+  void lazyInitPrivateUse1() {
+    c10::call_once(thp_init, [&] {
+      if (isPrivateUse1HooksRegistered()) {
+        at::detail::getPrivateUse1Hooks().initPrivateUse1();
+      }
+    });
+  }
+  static const at::cuda::NVRTC& getNVRTC() {
+    return detail::getCUDAHooks().nvrtc();
+  }
+
+  static bool setFlushDenormal(bool on);
+
+  // NB: This method is *purely* whether or not a user requested
+  // that CuDNN was enabled, it doesn't actually say anything about
+  // whether or not CuDNN is actually usable.  Use cudnn_is_acceptable
+  // to test this instead
+  bool userEnabledCuDNN() const;
+  void setUserEnabledCuDNN(bool e);
+  bool userEnabledMkldnn() const;
+  void setUserEnabledMkldnn(bool e);
+  bool benchmarkCuDNN() const;
+  void setBenchmarkCuDNN(bool);
+  int benchmarkLimitCuDNN() const;
+  void setBenchmarkLimitCuDNN(int);
+  bool deterministicCuDNN() const;
+  void setDeterministicCuDNN(bool);
+  bool deterministicMkldnn() const;
+  void setDeterministicMkldnn(bool);
+  bool userEnabledNNPACK() const;
+  void setUserEnabledNNPACK(bool e);
+
+  // Note [Disabling Fused SDP Kernels]
+  // ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
+  // Flash and Memory Efficient SDP kernels are enabled by default.
+  // However, they can be disabled by setting
+  // at::globalContext().setUserEnabledFlashSDP(false) flag.
+  // This is useful for debugging purposes. For example, if you want to
+  // compare the performance of the flash SDP kernels with the unfused
+  // kernel, you can disable the flash SDP kernels. By disabling
+  // the math SDP kernel, you can force your code to use flash kernels.
+  // The math SDP kernel can be disabled by setting
+  // at::globalContext().setUserEnabledMathSDP(false) flag.
+  void setSDPUseFlash(bool);
+  bool userEnabledFlashSDP() const;
+
+  void setSDPUseMemEfficient(bool);
+  bool userEnabledMemEfficientSDP() const;
+
+  void setSDPUseMath(bool);
+  bool userEnabledMathSDP() const;
+
+  void setSDPUseCuDNN(bool);
+  bool userEnabledCuDNNSDP() const;
+
+  void setAllowFP16BF16ReductionMathSDP(bool);
+  bool allowFP16BF16ReductionMathSDP() const;
+
+  void setSDPUseOverrideable(bool);
+  bool userEnabledOverrideableSDP() const;
+
+  at::LinalgBackend linalgPreferredBackend() const;
+  void setLinalgPreferredBackend(at::LinalgBackend);
+
+  at::BlasBackend blasPreferredBackend();
+  void setBlasPreferredBackend(at::BlasBackend);
+
+  // Note [Enabling Deterministic Operations]
+  // ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
+  // Operations in PyTorch that normally act nondeterministically, but have an
+  // alternate deterministic implementation, should satisfy the following
+  // requirements:
+  //
+  // * Include this comment: "See Note [Enabling Deterministic Operations]"
+  //
+  // * Check the value of `at::globalContext().deterministicAlgorithms()` to
+  // toggle
+  //   between nondeterministic and deterministic implementations.
+  //
+  // * Have an entry in the list of PyTorch operations that toggle between
+  // nondeterministic
+  //   and deterministic implementations, in the docstring of
+  //   `use_deterministic_algorithms()` in torch/__init__.py
+  //
+  // `example_func()` below shows an example of toggling between
+  // nondeterministic and deterministic implementations:
+  //
+  //    void example_func() {
+  //      // See Note [Enabling Deterministic Operations]
+  //      if (at::globalContext().deterministicAlgorithms()) {
+  //        example_func_deterministic();
+  //      } else {
+  //        example_func_nondeterministic();
+  //      }
+  //    }
+
+  bool deterministicAlgorithms() const;
+  bool deterministicAlgorithmsWarnOnly() const;
+  void setDeterministicAlgorithms(bool, bool);
+  bool deterministicFillUninitializedMemory() const;
+  void setDeterministicFillUninitializedMemory(bool);
+
+  // Note [Writing Nondeterministic Operations]
+  // ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
+  // Operations in PyTorch that act nondeterministically and do not have an
+  // alternate deterministic implementation should satisfy the following
+  // requirements:
+  //
+  // * Include this comment: "See Note [Writing Nondeterministic Operations]"
+  //
+  // * Include a comment explaining why the operation is nondeterministic.
+  //
+  // * Throw an error when `Context::deterministicAlgorithms()` is true. Most
+  //   of the time, this should be accomplished by calling
+  //   `at::globalContext().alertNotDeterminstic()`.  However, if the
+  //   nondeterministic behavior is caused by the CuBLAS workspace
+  //   configuration in CUDA >= 10.2,
+  //   `at::globalContext().alertCuBLASConfigNotDeterministic()` should be
+  //   called instead (in this case, a comment explaining why the operation is
+  //   nondeterministic is not necessary). See below for details on these
+  //   methods.
+  //
+  // * Have an entry in the list of nondeterministic PyTorch operations in the
+  //   docstring of `use_deterministic_algorithms()` in torch/__init__.py
+  //
+  // * Have a test function in `test/test_torch.py` whose name begins with
+  //   `test_nondeterministic_alert_`. Alternatively, if CuBLAS workspace
+  //   configuration is the reason for nondeterminism, the operation should be
+  //   included in the `test_cublas_config_nondeterministic_alert` test. Any new
+  //   tests should ideally follow a pattern similar to the existing ones.
+  //
+  // `example_func()` below shows an example of the comments and error-throwing
+  // code for a nondeterministic operation:
+  //
+  //    void example_func() {
+  //      // See Note [Writing Nondeterministic Operations]
+  //      // Nondeterministic because <reason>
+  //      at::globalContext().alertNondeterministic("example_func");
+  //      ...
+  //    }
+
+  // Throws an error if `Context::deterministicAlgorithms()` is true
+  static void alertNotDeterministic(c10::string_view const& caller);
+
+  // Throws an error if `Context::deterministicAlgorithms()` is true, CUDA
+  // >= 10.2, and CUBLAS_WORKSPACE_CONFIG is not set to either ":16:8" or
+  // ":4096:8". For more details:
+  // https://docs.nvidia.com/cuda/cublas/index.html#results-reproducibility
+  void alertCuBLASConfigNotDeterministic() const;
+
+  void setFloat32MatmulPrecision(const std::string& s);
+  bool allowTF32CuDNN() const;
+  void setAllowTF32CuDNN(bool);
+  bool allowTF32CuBLAS() const;
+  void setAllowTF32CuBLAS(bool);
+  Float32MatmulPrecision float32MatmulPrecision() const;
+  void setFloat32MatmulPrecision(Float32MatmulPrecision p);
+  bool allowFP16ReductionCuBLAS() const;
+  void setAllowFP16ReductionCuBLAS(bool);
+  bool allowBF16ReductionCuBLAS() const;
+  void setAllowBF16ReductionCuBLAS(bool);
+  at::QEngine qEngine() const;
+  void setQEngine(at::QEngine e);
+  static const std::vector<at::QEngine>& supportedQEngines();
+  static bool isXNNPACKAvailable();
+  void setCheckSparseTensorInvariants(bool e);
+  bool checkSparseTensorInvariants() const;
+  // This method is used to release the original weight after pre-packing.
+  // It should be called once before loading/running the model.
+  // NB: By default it is set to true for mobile builds.
+  void setReleaseWeightsWhenPrepacking(bool e);
+  bool releaseWeightsWhenPrepacking() const;
+
+  void setDisplayVmapFallbackWarnings(bool enabled);
+  bool areVmapFallbackWarningsEnabled() const;
+
+  void setDefaultMobileCPUAllocator();
+  void unsetDefaultMobileCPUAllocator();
+  bool allowFP16ReductionCPU() const;
+  void setAllowFP16ReductionCPU(bool);
+
+ private:
+  void initCUDAIfNeeded(c10::DeviceType p) {
+    if (p == c10::DeviceType::CUDA) {
+      lazyInitCUDA();
+    }
+  }
+  void initHIPIfNeeded(c10::DeviceType p) {
+    if (p == c10::DeviceType::HIP) {
+      lazyInitHIP();
+    }
+  }
+  void initXPUIfNeeded(c10::DeviceType p) {
+    if (p == c10::DeviceType::XPU) {
+      lazyInitXPU();
+    }
+  }
+  static bool checkCuBLASConfigDeterministic();
+  c10::once_flag thc_init;
+  c10::once_flag thh_init;
+  c10::once_flag thx_init;
+  c10::once_flag th_mtia_init;
+  c10::once_flag thp_init;
+  bool enabled_cudnn = true;
+  bool deterministic_cudnn = false;
+  bool deterministic_mkldnn = false;
+  bool _deterministic_algorithms = false;
+  bool _deterministic_algorithms_warn_only = false;
+  bool _deterministic_fill_uninitialized_memory = true;
+  bool enabled_flashSDP = true;
+  bool enabled_mem_efficientSDP = true;
+  bool enabled_mathSDP = true;
+  bool enabled_cudnnSDP = true;
+  bool enabled_overrideable = true;
+  bool allow_fp16_bf16_reduction_mathSDP = false;
+#ifdef USE_ROCM
+  bool benchmark_cudnn = true;
+#else
+  bool benchmark_cudnn = false;
+#endif
+  Float32MatmulPrecision float32_matmul_precision =
+      c10::utils::check_env("TORCH_ALLOW_TF32_CUBLAS_OVERRIDE") == true
+      ? at::Float32MatmulPrecision::HIGH
+      : at::Float32MatmulPrecision::HIGHEST;
+  int benchmark_limit_cudnn = 10;
+  bool allow_tf32_cudnn = true;
+  bool allow_fp16_reduction_cublas = true;
+  bool allow_bf16_reduction_cublas = true;
+  bool enabled_mkldnn = true;
+  bool enabled_nnpack = true;
+  at::LinalgBackend linalg_preferred_backend =
+      c10::utils::check_env("TORCH_LINALG_PREFER_CUSOLVER") == true
+      ? at::LinalgBackend::Cusolver
+      : at::LinalgBackend::Default;
+  at::BlasBackend blas_preferred_backend =
+#ifdef USE_ROCM
+      (c10::utils::check_env("TORCH_BLAS_PREFER_HIPBLASLT") != false)
+#else
+      (c10::utils::check_env("TORCH_BLAS_PREFER_CUBLASLT") == true)
+#endif
+      ? at::BlasBackend::Cublaslt
+      : at::BlasBackend::Cublas;
+#ifdef C10_MOBILE
+  bool release_original_weights = true;
+#else
+  bool release_original_weights = false;
+#endif
+  bool display_vmap_fallback_warnings_ = false;
+  std::optional<at::QEngine> quantized_engine = std::nullopt;
+  bool enable_sparse_tensor_invariant_checks = false;
+  bool allow_fp16_reduction_cpu = false;
+
+  Allocator* prev_allocator_ptr_{nullptr};
+};
+
+TORCH_API Context& globalContext();
+
+inline void init() {
+  globalContext();
+}
+
+TORCH_API Allocator* getCPUAllocator();
+
+inline DeprecatedTypeProperties& getDeprecatedTypeProperties(
+    Backend p,
+    ScalarType s) {
+  return globalDeprecatedTypePropertiesRegistry().getDeprecatedTypeProperties(
+      p, s);
+}
+
+inline DeprecatedTypeProperties& CPU(ScalarType s) {
+  return globalDeprecatedTypePropertiesRegistry().getDeprecatedTypeProperties(
+      Backend::CPU, s);
+}
+
+inline DeprecatedTypeProperties& CUDA(ScalarType s) {
+  return globalDeprecatedTypePropertiesRegistry().getDeprecatedTypeProperties(
+      Backend::CUDA, s);
+}
+
+inline DeprecatedTypeProperties& HIP(ScalarType s) {
+  return globalDeprecatedTypePropertiesRegistry().getDeprecatedTypeProperties(
+      Backend::HIP, s);
+}
+
+inline DeprecatedTypeProperties& MPS(ScalarType s) {
+  return globalDeprecatedTypePropertiesRegistry().getDeprecatedTypeProperties(
+      Backend::MPS, s);
+}
+
+inline bool hasCUDA() {
+  return globalContext().hasCUDA();
+}
+
+inline bool hasMTIA() {
+  return globalContext().hasMTIA();
+}
+
+inline bool hasHIP() {
+  return globalContext().hasHIP();
+}
+
+inline bool hasIPU() {
+  return globalContext().hasIPU();
+}
+
+inline bool hasXLA() {
+  return globalContext().hasXLA();
+}
+
+inline bool hasMPS() {
+  return globalContext().hasMPS();
+}
+
+inline bool hasMAIA() {
+  return globalContext().hasMAIA();
+}
+
+inline bool hasXPU() {
+  return globalContext().hasXPU();
+}
+
+// Despite its name, this function returns the number of *CUDA* GPUs.
+inline size_t getNumGPUs() {
+  // WARNING: DO NOT ADD LOGIC TO HANDLE OTHER DEVICE TYPES TO THIS
+  // FUNCTION.  If you are interested in interrogating the number of
+  // devices for a specific device type, add that function to the
+  // relevant library (e.g., similar to at::cuda::device_count())
+  if (hasCUDA() && hasHIP()) {
+    throw std::runtime_error(
+        "Enabling both CUDA and HIP in ATen is not supported, as HIP masquerades "
+        "to be CUDA (e.g., when you say CUDA, on a HIP build of ATen, this actually "
+        "means HIP.  Rebuild PyTorch with one or the other disabled.");
+  } else if (hasCUDA()) {
+    return detail::getCUDAHooks().getNumGPUs();
+  } else if (hasHIP()) {
+    return detail::getHIPHooks().getNumGPUs();
+  } else {
+    return 0;
+  }
+}
+
+inline bool hasOpenMP() {
+  return globalContext().hasOpenMP();
+}
+
+inline bool hasMKL() {
+  return globalContext().hasMKL();
+}
+
+inline bool hasLAPACK() {
+  return globalContext().hasLAPACK();
+}
+
+inline bool hasMAGMA() {
+  return globalContext().hasMAGMA();
+}
+
+inline bool hasMKLDNN() {
+  return globalContext().hasMKLDNN();
+}
+
+inline void manual_seed(uint64_t seed) {
+  auto gen = globalContext().defaultGenerator(c10::DeviceType::CPU);
+  {
+    // See Note [Acquire lock when using random generators]
+    std::lock_guard<std::mutex> lock(gen.mutex());
+    gen.set_current_seed(seed);
+  }
+  // NB: Sometimes we build with CUDA, but we don't have any GPUs
+  // available. In that case, we must not seed CUDA; it will fail!
+  const auto cuda_num_gpus = detail::getCUDAHooks().getNumGPUs();
+  if (hasCUDA() && cuda_num_gpus > 0) {
+    for (const auto i : c10::irange(cuda_num_gpus)) {
+      auto cuda_gen = globalContext().defaultGenerator(
+          Device(at::kCUDA, static_cast<c10::DeviceIndex>(i)));
+      {
+        // See Note [Acquire lock when using random generators]
+        std::lock_guard<std::mutex> lock(cuda_gen.mutex());
+        cuda_gen.set_current_seed(seed);
+      }
+    }
+  }
+
+  const auto xpu_num_gpus = detail::getXPUHooks().getNumGPUs();
+  if (hasXPU() && xpu_num_gpus) {
+    for (const auto i : c10::irange(xpu_num_gpus)) {
+      auto xpu_gen = globalContext().defaultGenerator(
+          Device(at::kXPU, static_cast<c10::DeviceIndex>(i)));
+      {
+        // See Note [Acquire lock when using random generators]
+        std::lock_guard<std::mutex> lock(xpu_gen.mutex());
+        xpu_gen.set_current_seed(seed);
+      }
+    }
+  }
+
+  if (hasMPS()) {
+    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);
+  }
+}
+
+// When the global flag `allow_tf32` is set to true, cuBLAS handles are
+// automatically configured to use math mode CUBLAS_TF32_TENSOR_OP_MATH.
+// For some operators, such as addmv, TF32 offers no performance improvement
+// but causes precision loss. To help this case, this class implements
+// a RAII guard that can be used to quickly disable TF32 within its scope.
+//
+// Usage:
+//     NoTF32Guard disable_tf32;
+struct TORCH_API NoTF32Guard {
+  NoTF32Guard();
+  ~NoTF32Guard();
+  static bool should_disable_tf32();
+
+ private:
+  bool changed = false;
+};
+
+struct TORCH_API ROCmBackwardPassGuard {
+  ROCmBackwardPassGuard();
+  ~ROCmBackwardPassGuard();
+  static bool is_backward_pass();
+};
+
+} // namespace at

.venv/Lib/site-packages/torch/include/ATen/Device.h

@@ -0,0 +1,2 @@
+#pragma once
+#include <c10/core/Device.h>

.venv/Lib/site-packages/torch/include/ATen/DeviceAccelerator.h

@@ -0,0 +1,27 @@
+#pragma once
+
+#include <c10/core/DeviceType.h>
+#include <c10/macros/Macros.h>
+
+#include <ATen/detail/MTIAHooksInterface.h>
+#include <optional>
+
+// This file defines the top level Accelerator concept for PyTorch.
+// A device is an accelerator per the definition here if:
+// - It is mutually exclusive with all other accelerators
+// - It performs asynchronous compute via a Stream/Event system
+// - It provides a set of common APIs as defined by AcceleratorHooksInterface
+//
+// As of today, accelerator devices are (in no particular order):
+// CUDA, MTIA, XPU, HIP, MPS, PrivateUse1
+
+namespace at {
+
+// Ensures that only one accelerator is available (at
+// compile time if possible) and return it.
+// When checked is true, the returned optional always has a value.
+TORCH_API std::optional<c10::DeviceType> getAccelerator(bool checked = false);
+
+TORCH_API bool isAccelerator(c10::DeviceType d);
+
+} // namespace at

.venv/Lib/site-packages/torch/include/ATen/DeviceGuard.h

@@ -0,0 +1,41 @@
+#pragma once
+
+#include <ATen/core/IListRef.h>
+#include <ATen/core/Tensor.h>
+#include <c10/core/DeviceGuard.h>
+#include <c10/core/ScalarType.h> // TensorList whyyyyy
+
+namespace at {
+
+// Are you here because you're wondering why DeviceGuard(tensor) no
+// longer works?  For code organization reasons, we have temporarily(?)
+// removed this constructor from DeviceGuard.  The new way to
+// spell it is:
+//
+//    OptionalDeviceGuard guard(device_of(tensor));
+
+/// Return the Device of a Tensor, if the Tensor is defined.
+inline std::optional<Device> device_of(const Tensor& t) {
+  if (t.defined()) {
+    return std::make_optional(t.device());
+  } else {
+    return std::nullopt;
+  }
+}
+
+inline std::optional<Device> device_of(const std::optional<Tensor>& t) {
+  return t.has_value() ? device_of(t.value()) : std::nullopt;
+}
+
+/// Return the Device of a TensorList, if the list is non-empty and
+/// the first Tensor is defined.  (This function implicitly assumes
+/// that all tensors in the list have the same device.)
+inline std::optional<Device> device_of(ITensorListRef t) {
+  if (!t.empty()) {
+    return device_of(t.front());
+  } else {
+    return std::nullopt;
+  }
+}
+
+} // namespace at

.venv/Lib/site-packages/torch/include/ATen/DimVector.h

@@ -0,0 +1,2 @@
+#pragma once
+#include <ATen/core/DimVector.h>

.venv/Lib/site-packages/torch/include/ATen/Dimname.h

@@ -0,0 +1 @@
+#include <ATen/core/Dimname.h>

.venv/Lib/site-packages/torch/include/ATen/Dispatch.h

@@ -0,0 +1,808 @@
+#pragma once
+
+#include <ATen/core/DeprecatedTypeProperties.h>
+#include <c10/macros/Macros.h>
+#include <c10/util/Exception.h>
+#include <c10/util/Half.h>
+#include <c10/util/Metaprogramming.h>
+#include <c10/util/complex.h>
+#include <c10/util/string_view.h>
+
+#ifdef __CUDACC__
+#include <cuda.h> // For CUDA_VERSION
+#endif
+
+#ifdef TEMPLATE_SELECTIVE_BUILD
+#include <ATen/selected_mobile_ops.h>
+#else
+namespace at {
+/**
+ * The method should_include_kernel_dtype() returns true/false
+ * based on whether the switching code for a specific dtype should be
+ * included based on build time constants generated from tracing model
+ * execution. This method will be implemented via code-generation and
+ * included in this file when code-gen is ready.
+ */
+inline constexpr bool should_include_kernel_dtype(
+    const char* /*kernel_tag_str*/,
+    at::ScalarType /*scalar_type*/
+) {
+  return true;
+}
+} // namespace at
+#endif
+
+/**
+ * In the Facebook internal build (using BUCK), this macro is enabled by
+ * passing in -c pt.enable_record_kernel_dtype=1 when building the tracer
+ * binary.
+ */
+#if defined ENABLE_RECORD_KERNEL_FUNCTION_DTYPE
+namespace at {
+namespace detail {
+TORCH_API void record_kernel_function_dtype(std::string name);
+}
+} // namespace at
+
+#define RECORD_KERNEL_FUNCTION_DTYPE(NAME, enum_type) \
+  at::detail::record_kernel_function_dtype(           \
+      std::string(NAME) + "$" + toString(enum_type));
+#else
+#define RECORD_KERNEL_FUNCTION_DTYPE(NAME, enum_type)
+#endif
+
+#define AT_PRIVATE_CHECK_SELECTIVE_BUILD(enum_type)   \
+  do {                                                \
+    if constexpr (!at::should_include_kernel_dtype(   \
+                      at_dispatch_name, enum_type)) { \
+      AT_ERROR(                                       \
+          "dtype '",                                  \
+          toString(enum_type),                        \
+          "' not selected for kernel tag ",           \
+          at_dispatch_name);                          \
+    }                                                 \
+  } while (0)
+
+#define AT_PRIVATE_CASE_TYPE_USING_HINT(enum_type, HINT, ...)           \
+  case enum_type: {                                                     \
+    AT_PRIVATE_CHECK_SELECTIVE_BUILD(enum_type);                        \
+    using HINT C10_UNUSED = c10::impl::ScalarTypeToCPPTypeT<enum_type>; \
+    return __VA_ARGS__();                                               \
+  }
+
+#define AT_DISPATCH_CASE(enum_type, ...) \
+  AT_PRIVATE_CASE_TYPE_USING_HINT(enum_type, scalar_t, __VA_ARGS__)
+
+#define AT_DISPATCH_CASE_QINT(enum_type, scalar_type, ...)            \
+  case enum_type: {                                                   \
+    AT_PRIVATE_CHECK_SELECTIVE_BUILD(enum_type);                      \
+    using scalar_t = scalar_type;                                     \
+    using underlying_t C10_UNUSED = typename scalar_t::underlying;    \
+    const auto& SCALAR_TYPE C10_UNUSED = enum_type;                   \
+    const auto& UNDERLYING_TYPE C10_UNUSED = toUnderlying(enum_type); \
+    return __VA_ARGS__();                                             \
+  }
+
+#define AT_QINT_SUB_BYTE_PRIVATE_CASE_TYPE(                           \
+    enum_type, scalar_type, bitwidth, qmin, qmax, ...)                \
+  case enum_type: {                                                   \
+    AT_PRIVATE_CHECK_SELECTIVE_BUILD(enum_type);                      \
+    using scalar_t = scalar_type;                                     \
+    using underlying_t C10_UNUSED = typename scalar_t::underlying;    \
+    const auto& SCALAR_TYPE C10_UNUSED = enum_type;                   \
+    const auto& UNDERLYING_TYPE C10_UNUSED = toUnderlying(enum_type); \
+    C10_UNUSED int bit_width = bitwidth;                              \
+    C10_UNUSED int64_t quant_min = qmin;                              \
+    C10_UNUSED int64_t quant_max = qmax;                              \
+    return __VA_ARGS__();                                             \
+  }
+
+namespace detail {
+
+inline at::ScalarType scalar_type(at::ScalarType s) {
+  return s;
+}
+
+C10_DEPRECATED_MESSAGE(
+    "passing at::DeprecatedTypeProperties to an AT_DISPATCH macro is deprecated, "
+    "pass an at::ScalarType instead")
+inline at::ScalarType scalar_type(const at::DeprecatedTypeProperties& t) {
+  return t.scalarType();
+}
+
+C10_DEPRECATED_MESSAGE(
+    "AT_DISPATCH_ALL_TYPES_AND_HALF is deprecated, "
+    "use AT_DISPATCH_ALL_TYPES_AND(at::ScalarType::Half, ...) instead")
+inline void deprecated_AT_DISPATCH_ALL_TYPES_AND_HALF() {}
+
+C10_DEPRECATED_MESSAGE(
+    "AT_DISPATCH_ALL_TYPES_AND_HALF_AND_COMPLEX is deprecated, "
+    "use AT_DISPATCH_ALL_TYPES_AND_COMPLEX_AND(at::ScalarType::Half, ...) "
+    "instead")
+inline void deprecated_AT_DISPATCH_ALL_TYPES_AND_HALF_AND_COMPLEX() {}
+
+} // namespace detail
+
+// The AT_DISPATCH_* family of macros provides the ability to
+// conveniently generate specializations of a kernel over all of the
+// dtypes we care about in PyTorch.  We call it "dispatch" because
+// we are "dispatching" to the correct, dtype-specific kernel.
+//
+// A standard usage looks like:
+//
+//      AT_DISPATCH_ALL_TYPES(self.scalar_type(), "op_name", [&] {
+//          // Your code here, with 'scalar_t' now defined to
+//          // be the dtype in question
+//      });
+//
+// There are many variations of this macro, so it's important to
+// understand exactly /which/ dtypes you want to get instantiated, as
+// well as what the "default" set is.
+//
+// The default set of dtypes that are instantiated (e.g., by
+// AT_DISPATCH_ALL_TYPES) are floating point types (float, double),
+// and integral types (int32_t, int64_t, int16_t, int8_t, uint8_t),
+// but NOT booleans (bool), half-precision floats (Half) or
+// complex number (c10::complex<float>, c10::complex<double>).
+// This "cut" is somewhat historical (the default types are the
+// ones that TH historically supported), but it also reflects the
+// fact that the non-default types are "poorly" behaved (booleans
+// are NOT integers mod 2, half precision operations ~essentially
+// don't exist on CPU, complex numbers are an experimental application).
+//
+// Here are the questions you should generally ask to decide which
+// dispatch you want:
+//
+// 1. Is this an integral or floating point specific operation?
+//    (If so, you'll want one of the FLOATING or INTEGRAL macros.)
+//
+// 2. Should half be supported?  (If you're on CPU, the answer is almost
+//    definitely no.  If you do want support, use one of the AND_HALF
+//    macros)
+//
+// Much rarer situations:
+//
+// 3. Should bool be supported?  (You often have to write your kernel
+//    differently if arithmetic operations are involved.)  If so,
+//    Use AT_DISPATCH_ALL_TYPES_AND along with ScalarType::Bool
+//
+// 4. Should complex be supported?  The answer is almost always no,
+//    unless you are working on "generic" code that should work on
+//    all dtypes.
+//
+// Parameters:
+// -----------
+//
+// 1. The NAME argument is a "tag" that is used to trace and then
+//    conditionally compile fragments of the case statements such
+//    that the kernel functions are specialized only for the dtypes
+//    that are needed. The NAME parameter *must* be a build time
+//    const char* (can't be std::string, etc...)
+//
+// Please ensure that the NAME is unique for every implementation
+// or you run the risk of over-including code for the kernel
+// functions. There is no risk of missing out on any code, so
+// it's mostly a risk of a Type-2 error, and not a Type-1 error.
+//
+// Switch-like syntax:
+// -------------------
+// There is also a switch-case like syntax which is useful if a kernel
+// needs to be specialized for particular scalar types
+//
+//      AT_DISPATCH_SWITCH(self.scalar_type(), "op_name",
+//          AT_DISPATCH_CASE_INTEGRAL_TYPES([&] {
+//            op_integral<scalar_t>(iter);
+//          })
+//          AT_DISPATCH_CASE_FLOATING_TYPES([&] {
+//            op_floating<scalar_t>(iter);
+//          })
+//          AT_DISPATCH_CASE(kBool, [&] {
+//            op_bool(iter);
+//          })
+//      );
+//
+// For each AT_DISPATCH_FOO macro, there is a corresponding
+// AT_DISPATCH_CASE_FOO macro which can be used inside of an
+// AT_DISPATCH_SWITCH block.
+
+// NB: the the_type variable is not used, but we have kept it for
+// backwards compatibility.  It's probably not used by anyone though;
+// but we're just being safe (and it doesn't hurt.)  Note we must
+// use it to shut up warnings about unused store.
+
+#define AT_DISPATCH_SWITCH(TYPE, NAME, ...)                                 \
+  [&] {                                                                     \
+    const auto& the_type = TYPE;                                            \
+    constexpr const char* at_dispatch_name = NAME;                          \
+    /* don't use TYPE again in case it is an expensive or side-effect op */ \
+    at::ScalarType _st = ::detail::scalar_type(the_type);                   \
+    RECORD_KERNEL_FUNCTION_DTYPE(at_dispatch_name, _st);                    \
+    switch (_st) {                                                          \
+      __VA_ARGS__                                                           \
+      default:                                                              \
+        AT_ERROR(                                                           \
+            '"',                                                            \
+            at_dispatch_name,                                               \
+            "\" not implemented for '",                                     \
+            toString(_st),                                                  \
+            "'");                                                           \
+    }                                                                       \
+  }()
+
+#define AT_DISPATCH_CASE_FLOATING_TYPES(...)            \
+  AT_DISPATCH_CASE(at::ScalarType::Double, __VA_ARGS__) \
+  AT_DISPATCH_CASE(at::ScalarType::Float, __VA_ARGS__)
+
+#define AT_DISPATCH_FLOATING_TYPES(TYPE, NAME, ...) \
+  AT_DISPATCH_SWITCH(TYPE, NAME, AT_DISPATCH_CASE_FLOATING_TYPES(__VA_ARGS__))
+
+#define AT_DISPATCH_CASE_FLOATING_TYPES_AND_HALF(...)   \
+  AT_DISPATCH_CASE(at::ScalarType::Double, __VA_ARGS__) \
+  AT_DISPATCH_CASE(at::ScalarType::Float, __VA_ARGS__)  \
+  AT_DISPATCH_CASE(at::ScalarType::Half, __VA_ARGS__)
+
+#define AT_DISPATCH_FLOATING_TYPES_AND_HALF(TYPE, NAME, ...) \
+  AT_DISPATCH_SWITCH(                                        \
+      TYPE, NAME, AT_DISPATCH_CASE_FLOATING_TYPES_AND_HALF(__VA_ARGS__))
+
+#define AT_DISPATCH_CASE_REDUCED_FLOATING_TYPES(...)  \
+  AT_DISPATCH_CASE(at::ScalarType::Half, __VA_ARGS__) \
+  AT_DISPATCH_CASE(at::ScalarType::BFloat16, __VA_ARGS__)
+
+#define AT_DISPATCH_REDUCED_FLOATING_TYPES(TYPE, NAME, ...) \
+  AT_DISPATCH_SWITCH(                                       \
+      TYPE, NAME, AT_DISPATCH_CASE_REDUCED_FLOATING_TYPES(__VA_ARGS__))
+
+#define AT_DISPATCH_CASE_FLOATING_TYPES_AND(SCALARTYPE, ...) \
+  AT_DISPATCH_CASE_FLOATING_TYPES(__VA_ARGS__)               \
+  AT_DISPATCH_CASE(SCALARTYPE, __VA_ARGS__)
+
+#define AT_DISPATCH_FLOATING_TYPES_AND(SCALARTYPE, TYPE, NAME, ...) \
+  AT_DISPATCH_SWITCH(                                               \
+      TYPE,                                                         \
+      NAME,                                                         \
+      AT_DISPATCH_CASE_FLOATING_TYPES_AND(SCALARTYPE, __VA_ARGS__))
+
+#define AT_DISPATCH_CASE_FLOATING_TYPES_AND2(SCALARTYPE1, SCALARTYPE2, ...) \
+  AT_DISPATCH_CASE_FLOATING_TYPES(__VA_ARGS__)                              \
+  AT_DISPATCH_CASE(SCALARTYPE1, __VA_ARGS__)                                \
+  AT_DISPATCH_CASE(SCALARTYPE2, __VA_ARGS__)
+
+#define AT_DISPATCH_FLOATING_TYPES_AND2(       \
+    SCALARTYPE1, SCALARTYPE2, TYPE, NAME, ...) \
+  AT_DISPATCH_SWITCH(                          \
+      TYPE,                                    \
+      NAME,                                    \
+      AT_DISPATCH_CASE_FLOATING_TYPES_AND2(    \
+          SCALARTYPE1, SCALARTYPE2, __VA_ARGS__))
+
+#define AT_DISPATCH_CASE_FLOATING_TYPES_AND3(   \
+    SCALARTYPE1, SCALARTYPE2, SCALARTYPE3, ...) \
+  AT_DISPATCH_CASE_FLOATING_TYPES(__VA_ARGS__)  \
+  AT_DISPATCH_CASE(SCALARTYPE1, __VA_ARGS__)    \
+  AT_DISPATCH_CASE(SCALARTYPE2, __VA_ARGS__)    \
+  AT_DISPATCH_CASE(SCALARTYPE3, __VA_ARGS__)
+
+#define AT_DISPATCH_FLOATING_TYPES_AND3(                    \
+    SCALARTYPE1, SCALARTYPE2, SCALARTYPE3, TYPE, NAME, ...) \
+  AT_DISPATCH_SWITCH(                                       \
+      TYPE,                                                 \
+      NAME,                                                 \
+      AT_DISPATCH_CASE_FLOATING_TYPES_AND3(                 \
+          SCALARTYPE1, SCALARTYPE2, SCALARTYPE3, __VA_ARGS__))
+
+#define AT_DISPATCH_CASE_FLOATING_TYPES_AND4(                \
+    SCALARTYPE1, SCALARTYPE2, SCALARTYPE3, SCALARTYPE4, ...) \
+  AT_DISPATCH_CASE_FLOATING_TYPES(__VA_ARGS__)               \
+  AT_DISPATCH_CASE(SCALARTYPE1, __VA_ARGS__)                 \
+  AT_DISPATCH_CASE(SCALARTYPE2, __VA_ARGS__)                 \
+  AT_DISPATCH_CASE(SCALARTYPE3, __VA_ARGS__)                 \
+  AT_DISPATCH_CASE(SCALARTYPE4, __VA_ARGS__)
+
+#define AT_DISPATCH_FLOATING_TYPES_AND4(                                 \
+    SCALARTYPE1, SCALARTYPE2, SCALARTYPE3, SCALARTYPE4, TYPE, NAME, ...) \
+  AT_DISPATCH_SWITCH(                                                    \
+      TYPE,                                                              \
+      NAME,                                                              \
+      AT_DISPATCH_CASE_FLOATING_TYPES_AND4(                              \
+          SCALARTYPE1, SCALARTYPE2, SCALARTYPE3, SCALARTYPE4, __VA_ARGS__))
+
+#define AT_DISPATCH_CASE_COMPLEX_TYPES(...)                    \
+  AT_DISPATCH_CASE(at::ScalarType::ComplexDouble, __VA_ARGS__) \
+  AT_DISPATCH_CASE(at::ScalarType::ComplexFloat, __VA_ARGS__)
+
+#define AT_DISPATCH_COMPLEX_TYPES(TYPE, NAME, ...) \
+  AT_DISPATCH_SWITCH(TYPE, NAME, AT_DISPATCH_CASE_COMPLEX_TYPES(__VA_ARGS__))
+
+#define AT_DISPATCH_CASE_COMPLEX_TYPES_AND(SCALARTYPE, ...) \
+  AT_DISPATCH_CASE_COMPLEX_TYPES(__VA_ARGS__)               \
+  AT_DISPATCH_CASE(SCALARTYPE, __VA_ARGS__)
+
+#define AT_DISPATCH_COMPLEX_TYPES_AND(SCALARTYPE, TYPE, NAME, ...) \
+  AT_DISPATCH_SWITCH(                                              \
+      TYPE, NAME, AT_DISPATCH_CASE_COMPLEX_TYPES_AND(SCALARTYPE, __VA_ARGS__))
+
+#define AT_DISPATCH_CASE_FLOATING_AND_COMPLEX_TYPES(...) \
+  AT_DISPATCH_CASE_FLOATING_TYPES(__VA_ARGS__)           \
+  AT_DISPATCH_CASE_COMPLEX_TYPES(__VA_ARGS__)
+
+#define AT_DISPATCH_FLOATING_AND_COMPLEX_TYPES(TYPE, NAME, ...) \
+  AT_DISPATCH_SWITCH(                                           \
+      TYPE, NAME, AT_DISPATCH_CASE_FLOATING_AND_COMPLEX_TYPES(__VA_ARGS__))
+
+#define AT_DISPATCH_CASE_FLOATING_AND_COMPLEX_TYPES_AND1(SCALARTYPE, ...) \
+  AT_DISPATCH_CASE_FLOATING_AND_COMPLEX_TYPES(__VA_ARGS__)                \
+  AT_DISPATCH_CASE(SCALARTYPE, __VA_ARGS__)
+
+#define AT_DISPATCH_FLOATING_AND_COMPLEX_TYPES_AND1(    \
+    SCALARTYPE, TYPE, NAME, ...)                        \
+  AT_DISPATCH_SWITCH(                                   \
+      TYPE,                                             \
+      NAME,                                             \
+      AT_DISPATCH_CASE_FLOATING_AND_COMPLEX_TYPES_AND1( \
+          SCALARTYPE, __VA_ARGS__))
+
+#define AT_DISPATCH_CASE_FLOATING_AND_COMPLEX_TYPES_AND2(  \
+    SCALARTYPE1, SCALARTYPE2, ...)                         \
+  AT_DISPATCH_CASE_FLOATING_AND_COMPLEX_TYPES(__VA_ARGS__) \
+  AT_DISPATCH_CASE(SCALARTYPE1, __VA_ARGS__)               \
+  AT_DISPATCH_CASE(SCALARTYPE2, __VA_ARGS__)
+
+#define AT_DISPATCH_FLOATING_AND_COMPLEX_TYPES_AND2(    \
+    SCALARTYPE1, SCALARTYPE2, TYPE, NAME, ...)          \
+  AT_DISPATCH_SWITCH(                                   \
+      TYPE,                                             \
+      NAME,                                             \
+      AT_DISPATCH_CASE_FLOATING_AND_COMPLEX_TYPES_AND2( \
+          SCALARTYPE1, SCALARTYPE2, __VA_ARGS__))
+
+#define AT_DISPATCH_CASE_FLOATING_AND_COMPLEX_TYPES_AND3(  \
+    SCALARTYPE1, SCALARTYPE2, SCALARTYPE3, ...)            \
+  AT_DISPATCH_CASE_FLOATING_AND_COMPLEX_TYPES(__VA_ARGS__) \
+  AT_DISPATCH_CASE(SCALARTYPE1, __VA_ARGS__)               \
+  AT_DISPATCH_CASE(SCALARTYPE2, __VA_ARGS__)               \
+  AT_DISPATCH_CASE(SCALARTYPE3, __VA_ARGS__)
+
+#define AT_DISPATCH_FLOATING_AND_COMPLEX_TYPES_AND3(        \
+    SCALARTYPE1, SCALARTYPE2, SCALARTYPE3, TYPE, NAME, ...) \
+  AT_DISPATCH_SWITCH(                                       \
+      TYPE,                                                 \
+      NAME,                                                 \
+      AT_DISPATCH_CASE_FLOATING_AND_COMPLEX_TYPES_AND3(     \
+          SCALARTYPE1, SCALARTYPE2, SCALARTYPE3, __VA_ARGS__))
+
+#define AT_DISPATCH_CASE_FLOATING_AND_COMPLEX_TYPES_AND4(    \
+    SCALARTYPE1, SCALARTYPE2, SCALARTYPE3, SCALARTYPE4, ...) \
+  AT_DISPATCH_CASE_FLOATING_AND_COMPLEX_TYPES(__VA_ARGS__)   \
+  AT_DISPATCH_CASE(SCALARTYPE1, __VA_ARGS__)                 \
+  AT_DISPATCH_CASE(SCALARTYPE2, __VA_ARGS__)                 \
+  AT_DISPATCH_CASE(SCALARTYPE3, __VA_ARGS__)                 \
+  AT_DISPATCH_CASE(SCALARTYPE4, __VA_ARGS__)
+
+#define AT_DISPATCH_FLOATING_AND_COMPLEX_TYPES_AND4(                     \
+    SCALARTYPE1, SCALARTYPE2, SCALARTYPE3, SCALARTYPE4, TYPE, NAME, ...) \
+  AT_DISPATCH_SWITCH(                                                    \
+      TYPE,                                                              \
+      NAME,                                                              \
+      AT_DISPATCH_CASE_FLOATING_AND_COMPLEX_TYPES_AND4(                  \
+          SCALARTYPE1, SCALARTYPE2, SCALARTYPE3, SCALARTYPE4, __VA_ARGS__))
+
+#define AT_DISPATCH_CASE_FLOATING_AND_COMPLEX_TYPES_AND5(                 \
+    SCALARTYPE1, SCALARTYPE2, SCALARTYPE3, SCALARTYPE4, SCALARTYPE5, ...) \
+  AT_DISPATCH_CASE_FLOATING_AND_COMPLEX_TYPES(__VA_ARGS__)                \
+  AT_DISPATCH_CASE(SCALARTYPE1, __VA_ARGS__)                              \
+  AT_DISPATCH_CASE(SCALARTYPE2, __VA_ARGS__)                              \
+  AT_DISPATCH_CASE(SCALARTYPE3, __VA_ARGS__)                              \
+  AT_DISPATCH_CASE(SCALARTYPE4, __VA_ARGS__)                              \
+  AT_DISPATCH_CASE(SCALARTYPE5, __VA_ARGS__)
+
+#define AT_DISPATCH_FLOATING_AND_COMPLEX_TYPES_AND5(    \
+    SCALARTYPE1,                                        \
+    SCALARTYPE2,                                        \
+    SCALARTYPE3,                                        \
+    SCALARTYPE4,                                        \
+    SCALARTYPE5,                                        \
+    TYPE,                                               \
+    NAME,                                               \
+    ...)                                                \
+  AT_DISPATCH_SWITCH(                                   \
+      TYPE,                                             \
+      NAME,                                             \
+      AT_DISPATCH_CASE_FLOATING_AND_COMPLEX_TYPES_AND5( \
+          SCALARTYPE1,                                  \
+          SCALARTYPE2,                                  \
+          SCALARTYPE3,                                  \
+          SCALARTYPE4,                                  \
+          SCALARTYPE5,                                  \
+          __VA_ARGS__))
+
+#define AT_DISPATCH_CASE_FLOATING_AND_COMPLEX_TYPES_AND6(  \
+    SCALARTYPE1,                                           \
+    SCALARTYPE2,                                           \
+    SCALARTYPE3,                                           \
+    SCALARTYPE4,                                           \
+    SCALARTYPE5,                                           \
+    SCALARTYPE6,                                           \
+    ...)                                                   \
+  AT_DISPATCH_CASE_FLOATING_AND_COMPLEX_TYPES(__VA_ARGS__) \
+  AT_DISPATCH_CASE(SCALARTYPE1, __VA_ARGS__)               \
+  AT_DISPATCH_CASE(SCALARTYPE2, __VA_ARGS__)               \
+  AT_DISPATCH_CASE(SCALARTYPE3, __VA_ARGS__)               \
+  AT_DISPATCH_CASE(SCALARTYPE4, __VA_ARGS__)               \
+  AT_DISPATCH_CASE(SCALARTYPE5, __VA_ARGS__)               \
+  AT_DISPATCH_CASE(SCALARTYPE6, __VA_ARGS__)
+
+#define AT_DISPATCH_FLOATING_AND_COMPLEX_TYPES_AND6(    \
+    SCALARTYPE1,                                        \
+    SCALARTYPE2,                                        \
+    SCALARTYPE3,                                        \
+    SCALARTYPE4,                                        \
+    SCALARTYPE5,                                        \
+    SCALARTYPE6,                                        \
+    TYPE,                                               \
+    NAME,                                               \
+    ...)                                                \
+  AT_DISPATCH_SWITCH(                                   \
+      TYPE,                                             \
+      NAME,                                             \
+      AT_DISPATCH_CASE_FLOATING_AND_COMPLEX_TYPES_AND6( \
+          SCALARTYPE1,                                  \
+          SCALARTYPE2,                                  \
+          SCALARTYPE3,                                  \
+          SCALARTYPE4,                                  \
+          SCALARTYPE5,                                  \
+          SCALARTYPE6,                                  \
+          __VA_ARGS__))
+
+#define AT_DISPATCH_CASE_INTEGRAL_TYPES(...)          \
+  AT_DISPATCH_CASE(at::ScalarType::Byte, __VA_ARGS__) \
+  AT_DISPATCH_CASE(at::ScalarType::Char, __VA_ARGS__) \
+  AT_DISPATCH_CASE(at::ScalarType::Int, __VA_ARGS__)  \
+  AT_DISPATCH_CASE(at::ScalarType::Long, __VA_ARGS__) \
+  AT_DISPATCH_CASE(at::ScalarType::Short, __VA_ARGS__)
+
+#define AT_DISPATCH_INTEGRAL_TYPES(TYPE, NAME, ...) \
+  AT_DISPATCH_SWITCH(TYPE, NAME, AT_DISPATCH_CASE_INTEGRAL_TYPES(__VA_ARGS__))
+
+#define AT_DISPATCH_CASE_INTEGRAL_TYPES_AND(SCALARTYPE, ...) \
+  AT_DISPATCH_CASE_INTEGRAL_TYPES(__VA_ARGS__)               \
+  AT_DISPATCH_CASE(SCALARTYPE, __VA_ARGS__)
+
+#define AT_DISPATCH_INTEGRAL_TYPES_AND(SCALARTYPE, TYPE, NAME, ...) \
+  AT_DISPATCH_SWITCH(                                               \
+      TYPE,                                                         \
+      NAME,                                                         \
+      AT_DISPATCH_CASE_INTEGRAL_TYPES_AND(SCALARTYPE, __VA_ARGS__))
+
+#define AT_DISPATCH_CASE_ALL_TYPES(...)        \
+  AT_DISPATCH_CASE_INTEGRAL_TYPES(__VA_ARGS__) \
+  AT_DISPATCH_CASE_FLOATING_TYPES(__VA_ARGS__)
+
+#define AT_DISPATCH_ALL_TYPES(TYPE, NAME, ...) \
+  AT_DISPATCH_SWITCH(TYPE, NAME, AT_DISPATCH_CASE_ALL_TYPES(__VA_ARGS__))
+
+#define AT_DISPATCH_CASE_QINT_TYPES(...)                      \
+  AT_DISPATCH_CASE_QINT(at::kQInt8, at::qint8, __VA_ARGS__)   \
+  AT_DISPATCH_CASE_QINT(at::kQUInt8, at::quint8, __VA_ARGS__) \
+  AT_DISPATCH_CASE_QINT(at::kQInt32, at::qint32, __VA_ARGS__)
+
+#define AT_DISPATCH_QINT_TYPES(TYPE, NAME, ...) \
+  AT_DISPATCH_SWITCH(TYPE, NAME, AT_DISPATCH_CASE_QINT_TYPES(__VA_ARGS__))
+
+#define AT_DISPATCH_CASE_QINT_TYPES_AND(SCALARTYPE, ...) \
+  AT_DISPATCH_CASE_QINT_TYPES(__VA_ARGS__)               \
+  AT_DISPATCH_CASE(SCALARTYPE, __VA_ARGS__)
+
+#define AT_DISPATCH_QINT_TYPES_AND(SCALARTYPE, TYPE, NAME, ...) \
+  AT_DISPATCH_SWITCH(                                           \
+      TYPE, NAME, AT_DISPATCH_CASE_QINT_TYPES_AND(SCALARTYPE, __VA_ARGS__))
+
+#define AT_DISPATCH_CASE_QINT_BYTE_TYPES(...)               \
+  AT_DISPATCH_CASE_QINT(at::kQInt8, at::qint8, __VA_ARGS__) \
+  AT_DISPATCH_CASE_QINT(at::kQUInt8, at::quint8, __VA_ARGS__)
+
+#define AT_DISPATCH_QINT_BYTE_TYPES(TYPE, NAME, ...) \
+  AT_DISPATCH_SWITCH(TYPE, NAME, AT_DISPATCH_CASE_QINT_BYTE_TYPES(__VA_ARGS__))
+
+#define AT_DISPATCH_CASE_QINT_AND_SUB_BYTE_TYPES(...)                     \
+  AT_QINT_SUB_BYTE_PRIVATE_CASE_TYPE(                                     \
+      at::kQInt8, at::qint8, CHAR_BIT, SCHAR_MIN, SCHAR_MAX, __VA_ARGS__) \
+  AT_QINT_SUB_BYTE_PRIVATE_CASE_TYPE(                                     \
+      at::kQUInt8, at::quint8, CHAR_BIT, 0, UCHAR_MAX, __VA_ARGS__)       \
+  AT_QINT_SUB_BYTE_PRIVATE_CASE_TYPE(                                     \
+      at::kQInt32,                                                        \
+      at::qint32,                                                         \
+      CHAR_BIT * sizeof(int),                                             \
+      INT_MIN,                                                            \
+      INT_MAX,                                                            \
+      __VA_ARGS__)                                                        \
+  AT_QINT_SUB_BYTE_PRIVATE_CASE_TYPE(                                     \
+      at::kQUInt4x2, at::quint4x2, 4, 0, 15, __VA_ARGS__)                 \
+  AT_QINT_SUB_BYTE_PRIVATE_CASE_TYPE(                                     \
+      at::kQUInt2x4, at::quint2x4, 2, 0, 3, __VA_ARGS__)
+
+#define AT_DISPATCH_QINT_AND_SUB_BYTE_TYPES(TYPE, NAME, ...) \
+  AT_DISPATCH_SWITCH(                                        \
+      TYPE, NAME, AT_DISPATCH_CASE_QINT_AND_SUB_BYTE_TYPES(__VA_ARGS__))
+
+#define AT_DISPATCH_CASE_ALL_TYPES_AND_COMPLEX(...) \
+  AT_DISPATCH_CASE_ALL_TYPES(__VA_ARGS__)           \
+  AT_DISPATCH_CASE_COMPLEX_TYPES(__VA_ARGS__)
+
+#define AT_DISPATCH_ALL_TYPES_AND_COMPLEX(TYPE, NAME, ...) \
+  AT_DISPATCH_SWITCH(                                      \
+      TYPE, NAME, AT_DISPATCH_CASE_ALL_TYPES_AND_COMPLEX(__VA_ARGS__))
+
+#define AT_DISPATCH_CASE_ALL_TYPES_AND(SCALARTYPE, ...) \
+  AT_DISPATCH_CASE_ALL_TYPES(__VA_ARGS__)               \
+  AT_DISPATCH_CASE(SCALARTYPE, __VA_ARGS__)
+
+#define AT_DISPATCH_ALL_TYPES_AND(SCALARTYPE, TYPE, NAME, ...) \
+  AT_DISPATCH_SWITCH(                                          \
+      TYPE, NAME, AT_DISPATCH_CASE_ALL_TYPES_AND(SCALARTYPE, __VA_ARGS__))
+
+#define AT_DISPATCH_CASE_ALL_TYPES_AND_COMPLEX_AND(SCALARTYPE, ...) \
+  AT_DISPATCH_CASE_ALL_TYPES_AND_COMPLEX(__VA_ARGS__)               \
+  AT_DISPATCH_CASE(SCALARTYPE, __VA_ARGS__)
+
+#define AT_DISPATCH_ALL_TYPES_AND_COMPLEX_AND(SCALARTYPE, TYPE, NAME, ...) \
+  AT_DISPATCH_SWITCH(                                                      \
+      TYPE,                                                                \
+      NAME,                                                                \
+      AT_DISPATCH_CASE_ALL_TYPES_AND_COMPLEX_AND(SCALARTYPE, __VA_ARGS__))
+
+#define AT_DISPATCH_CASE_ALL_TYPES_AND2(SCALARTYPE1, SCALARTYPE2, ...) \
+  AT_DISPATCH_CASE_ALL_TYPES(__VA_ARGS__)                              \
+  AT_DISPATCH_CASE(SCALARTYPE1, __VA_ARGS__)                           \
+  AT_DISPATCH_CASE(SCALARTYPE2, __VA_ARGS__)
+
+#define AT_DISPATCH_ALL_TYPES_AND2(SCALARTYPE1, SCALARTYPE2, TYPE, NAME, ...) \
+  AT_DISPATCH_SWITCH(                                                         \
+      TYPE,                                                                   \
+      NAME,                                                                   \
+      AT_DISPATCH_CASE_ALL_TYPES_AND2(SCALARTYPE1, SCALARTYPE2, __VA_ARGS__))
+
+#define AT_DISPATCH_CASE_ALL_TYPES_AND_COMPLEX_AND2(  \
+    SCALARTYPE1, SCALARTYPE2, ...)                    \
+  AT_DISPATCH_CASE_ALL_TYPES_AND_COMPLEX(__VA_ARGS__) \
+  AT_DISPATCH_CASE(SCALARTYPE1, __VA_ARGS__)          \
+  AT_DISPATCH_CASE(SCALARTYPE2, __VA_ARGS__)
+
+#define AT_DISPATCH_ALL_TYPES_AND_COMPLEX_AND2(    \
+    SCALARTYPE1, SCALARTYPE2, TYPE, NAME, ...)     \
+  AT_DISPATCH_SWITCH(                              \
+      TYPE,                                        \
+      NAME,                                        \
+      AT_DISPATCH_CASE_ALL_TYPES_AND_COMPLEX_AND2( \
+          SCALARTYPE1, SCALARTYPE2, __VA_ARGS__))
+
+#define AT_DISPATCH_CASE_ALL_TYPES_AND3(        \
+    SCALARTYPE1, SCALARTYPE2, SCALARTYPE3, ...) \
+  AT_DISPATCH_CASE_ALL_TYPES(__VA_ARGS__)       \
+  AT_DISPATCH_CASE(SCALARTYPE1, __VA_ARGS__)    \
+  AT_DISPATCH_CASE(SCALARTYPE2, __VA_ARGS__)    \
+  AT_DISPATCH_CASE(SCALARTYPE3, __VA_ARGS__)
+
+#define AT_DISPATCH_ALL_TYPES_AND3(                         \
+    SCALARTYPE1, SCALARTYPE2, SCALARTYPE3, TYPE, NAME, ...) \
+  AT_DISPATCH_SWITCH(                                       \
+      TYPE,                                                 \
+      NAME,                                                 \
+      AT_DISPATCH_CASE_ALL_TYPES_AND3(                      \
+          SCALARTYPE1, SCALARTYPE2, SCALARTYPE3, __VA_ARGS__))
+
+#define AT_DISPATCH_CASE_ALL_TYPES_AND_COMPLEX_AND3(  \
+    SCALARTYPE1, SCALARTYPE2, SCALARTYPE3, ...)       \
+  AT_DISPATCH_CASE_ALL_TYPES_AND_COMPLEX(__VA_ARGS__) \
+  AT_DISPATCH_CASE(SCALARTYPE1, __VA_ARGS__)          \
+  AT_DISPATCH_CASE(SCALARTYPE2, __VA_ARGS__)          \
+  AT_DISPATCH_CASE(SCALARTYPE3, __VA_ARGS__)
+
+#define AT_DISPATCH_ALL_TYPES_AND_COMPLEX_AND3(             \
+    SCALARTYPE1, SCALARTYPE2, SCALARTYPE3, TYPE, NAME, ...) \
+  AT_DISPATCH_SWITCH(                                       \
+      TYPE,                                                 \
+      NAME,                                                 \
+      AT_DISPATCH_CASE_ALL_TYPES_AND_COMPLEX_AND3(          \
+          SCALARTYPE1, SCALARTYPE2, SCALARTYPE3, __VA_ARGS__))
+
+#define AT_DISPATCH_CASE_ALL_TYPES_AND_COMPLEX_AND4(         \
+    SCALARTYPE1, SCALARTYPE2, SCALARTYPE3, SCALARTYPE4, ...) \
+  AT_DISPATCH_CASE_ALL_TYPES_AND_COMPLEX(__VA_ARGS__)        \
+  AT_DISPATCH_CASE(SCALARTYPE1, __VA_ARGS__)                 \
+  AT_DISPATCH_CASE(SCALARTYPE2, __VA_ARGS__)                 \
+  AT_DISPATCH_CASE(SCALARTYPE3, __VA_ARGS__)                 \
+  AT_DISPATCH_CASE(SCALARTYPE4, __VA_ARGS__)
+
+#define AT_DISPATCH_ALL_TYPES_AND_COMPLEX_AND4(                          \
+    SCALARTYPE1, SCALARTYPE2, SCALARTYPE3, SCALARTYPE4, TYPE, NAME, ...) \
+  AT_DISPATCH_SWITCH(                                                    \
+      TYPE,                                                              \
+      NAME,                                                              \
+      AT_DISPATCH_CASE_ALL_TYPES_AND_COMPLEX_AND4(                       \
+          SCALARTYPE1, SCALARTYPE2, SCALARTYPE3, SCALARTYPE4, __VA_ARGS__))
+
+#define AT_DISPATCH_CASE_ALL_TYPES_AND_COMPLEX_AND5(                      \
+    SCALARTYPE1, SCALARTYPE2, SCALARTYPE3, SCALARTYPE4, SCALARTYPE5, ...) \
+  AT_DISPATCH_CASE_ALL_TYPES_AND_COMPLEX(__VA_ARGS__)                     \
+  AT_DISPATCH_CASE(SCALARTYPE1, __VA_ARGS__)                              \
+  AT_DISPATCH_CASE(SCALARTYPE2, __VA_ARGS__)                              \
+  AT_DISPATCH_CASE(SCALARTYPE3, __VA_ARGS__)                              \
+  AT_DISPATCH_CASE(SCALARTYPE4, __VA_ARGS__)                              \
+  AT_DISPATCH_CASE(SCALARTYPE5, __VA_ARGS__)
+
+#define AT_DISPATCH_ALL_TYPES_AND_COMPLEX_AND5(    \
+    SCALARTYPE1,                                   \
+    SCALARTYPE2,                                   \
+    SCALARTYPE3,                                   \
+    SCALARTYPE4,                                   \
+    SCALARTYPE5,                                   \
+    TYPE,                                          \
+    NAME,                                          \
+    ...)                                           \
+  AT_DISPATCH_SWITCH(                              \
+      TYPE,                                        \
+      NAME,                                        \
+      AT_DISPATCH_CASE_ALL_TYPES_AND_COMPLEX_AND5( \
+          SCALARTYPE1,                             \
+          SCALARTYPE2,                             \
+          SCALARTYPE3,                             \
+          SCALARTYPE4,                             \
+          SCALARTYPE5,                             \
+          __VA_ARGS__))
+
+#define AT_DISPATCH_CASE_ALL_TYPES_AND_COMPLEX_AND6(  \
+    SCALARTYPE1,                                      \
+    SCALARTYPE2,                                      \
+    SCALARTYPE3,                                      \
+    SCALARTYPE4,                                      \
+    SCALARTYPE5,                                      \
+    SCALARTYPE6,                                      \
+    ...)                                              \
+  AT_DISPATCH_CASE_ALL_TYPES_AND_COMPLEX(__VA_ARGS__) \
+  AT_DISPATCH_CASE(SCALARTYPE1, __VA_ARGS__)          \
+  AT_DISPATCH_CASE(SCALARTYPE2, __VA_ARGS__)          \
+  AT_DISPATCH_CASE(SCALARTYPE3, __VA_ARGS__)          \
+  AT_DISPATCH_CASE(SCALARTYPE4, __VA_ARGS__)          \
+  AT_DISPATCH_CASE(SCALARTYPE5, __VA_ARGS__)          \
+  AT_DISPATCH_CASE(SCALARTYPE6, __VA_ARGS__)
+
+#define AT_DISPATCH_ALL_TYPES_AND_COMPLEX_AND6(    \
+    SCALARTYPE1,                                   \
+    SCALARTYPE2,                                   \
+    SCALARTYPE3,                                   \
+    SCALARTYPE4,                                   \
+    SCALARTYPE5,                                   \
+    SCALARTYPE6,                                   \
+    TYPE,                                          \
+    NAME,                                          \
+    ...)                                           \
+  AT_DISPATCH_SWITCH(                              \
+      TYPE,                                        \
+      NAME,                                        \
+      AT_DISPATCH_CASE_ALL_TYPES_AND_COMPLEX_AND6( \
+          SCALARTYPE1,                             \
+          SCALARTYPE2,                             \
+          SCALARTYPE3,                             \
+          SCALARTYPE4,                             \
+          SCALARTYPE5,                             \
+          SCALARTYPE6,                             \
+          __VA_ARGS__))
+
+#define AT_DISPATCH_CASE_ALL_TYPES_AND_COMPLEX_AND7(  \
+    SCALARTYPE1,                                      \
+    SCALARTYPE2,                                      \
+    SCALARTYPE3,                                      \
+    SCALARTYPE4,                                      \
+    SCALARTYPE5,                                      \
+    SCALARTYPE6,                                      \
+    SCALARTYPE7,                                      \
+    ...)                                              \
+  AT_DISPATCH_CASE_ALL_TYPES_AND_COMPLEX(__VA_ARGS__) \
+  AT_DISPATCH_CASE(SCALARTYPE1, __VA_ARGS__)          \
+  AT_DISPATCH_CASE(SCALARTYPE2, __VA_ARGS__)          \
+  AT_DISPATCH_CASE(SCALARTYPE3, __VA_ARGS__)          \
+  AT_DISPATCH_CASE(SCALARTYPE4, __VA_ARGS__)          \
+  AT_DISPATCH_CASE(SCALARTYPE5, __VA_ARGS__)          \
+  AT_DISPATCH_CASE(SCALARTYPE6, __VA_ARGS__)          \
+  AT_DISPATCH_CASE(SCALARTYPE7, __VA_ARGS__)
+
+#define AT_DISPATCH_ALL_TYPES_AND_COMPLEX_AND7(    \
+    SCALARTYPE1,                                   \
+    SCALARTYPE2,                                   \
+    SCALARTYPE3,                                   \
+    SCALARTYPE4,                                   \
+    SCALARTYPE5,                                   \
+    SCALARTYPE6,                                   \
+    SCALARTYPE7,                                   \
+    TYPE,                                          \
+    NAME,                                          \
+    ...)                                           \
+  AT_DISPATCH_SWITCH(                              \
+      TYPE,                                        \
+      NAME,                                        \
+      AT_DISPATCH_CASE_ALL_TYPES_AND_COMPLEX_AND7( \
+          SCALARTYPE1,                             \
+          SCALARTYPE2,                             \
+          SCALARTYPE3,                             \
+          SCALARTYPE4,                             \
+          SCALARTYPE5,                             \
+          SCALARTYPE6,                             \
+          SCALARTYPE7,                             \
+          __VA_ARGS__))
+
+#define AT_DISPATCH_CASE_ALL_TYPES_AND_COMPLEX_AND8(  \
+    SCALARTYPE1,                                      \
+    SCALARTYPE2,                                      \
+    SCALARTYPE3,                                      \
+    SCALARTYPE4,                                      \
+    SCALARTYPE5,                                      \
+    SCALARTYPE6,                                      \
+    SCALARTYPE7,                                      \
+    SCALARTYPE8,                                      \
+    ...)                                              \
+  AT_DISPATCH_CASE_ALL_TYPES_AND_COMPLEX(__VA_ARGS__) \
+  AT_DISPATCH_CASE(SCALARTYPE1, __VA_ARGS__)          \
+  AT_DISPATCH_CASE(SCALARTYPE2, __VA_ARGS__)          \
+  AT_DISPATCH_CASE(SCALARTYPE3, __VA_ARGS__)          \
+  AT_DISPATCH_CASE(SCALARTYPE4, __VA_ARGS__)          \
+  AT_DISPATCH_CASE(SCALARTYPE5, __VA_ARGS__)          \
+  AT_DISPATCH_CASE(SCALARTYPE6, __VA_ARGS__)          \
+  AT_DISPATCH_CASE(SCALARTYPE7, __VA_ARGS__)          \
+  AT_DISPATCH_CASE(SCALARTYPE8, __VA_ARGS__)
+
+#define AT_DISPATCH_ALL_TYPES_AND_COMPLEX_AND8(    \
+    SCALARTYPE1,                                   \
+    SCALARTYPE2,                                   \
+    SCALARTYPE3,                                   \
+    SCALARTYPE4,                                   \
+    SCALARTYPE5,                                   \
+    SCALARTYPE6,                                   \
+    SCALARTYPE7,                                   \
+    SCALARTYPE8,                                   \
+    TYPE,                                          \
+    NAME,                                          \
+    ...)                                           \
+  AT_DISPATCH_SWITCH(                              \
+      TYPE,                                        \
+      NAME,                                        \
+      AT_DISPATCH_CASE_ALL_TYPES_AND_COMPLEX_AND8( \
+          SCALARTYPE1,                             \
+          SCALARTYPE2,                             \
+          SCALARTYPE3,                             \
+          SCALARTYPE4,                             \
+          SCALARTYPE5,                             \
+          SCALARTYPE6,                             \
+          SCALARTYPE7,                             \
+          SCALARTYPE8,                             \
+          __VA_ARGS__))
+
+#define AT_DISPATCH_CASE_BIT_TYPES(...)                  \
+  AT_DISPATCH_CASE(at::ScalarType::Bits1x8, __VA_ARGS__) \
+  AT_DISPATCH_CASE(at::ScalarType::Bits2x4, __VA_ARGS__) \
+  AT_DISPATCH_CASE(at::ScalarType::Bits4x2, __VA_ARGS__) \
+  AT_DISPATCH_CASE(at::ScalarType::Bits8, __VA_ARGS__)   \
+  AT_DISPATCH_CASE(at::ScalarType::Bits16, __VA_ARGS__)
+
+#define AT_DISPATCH_BIT_TYPES(TYPE, NAME, ...) \
+  AT_DISPATCH_SWITCH(TYPE, NAME, AT_DISPATCH_CASE_BIT_TYPES(__VA_ARGS__))
+
+#define AT_DISPATCH_INDEX_TYPES(TYPE, NAME, ...)     \
+  AT_DISPATCH_SWITCH(                                \
+      TYPE,                                          \
+      NAME,                                          \
+      AT_PRIVATE_CASE_TYPE_USING_HINT(               \
+          at::ScalarType::Int, index_t, __VA_ARGS__) \
+          AT_PRIVATE_CASE_TYPE_USING_HINT(           \
+              at::ScalarType::Long, index_t, __VA_ARGS__))
+
+// ----------------------------------------------------------------------------
+// DEPRECATED MACROS, DON'T USE THESE
+// ----------------------------------------------------------------------------
+
+#define AT_DISPATCH_ALL_TYPES_AND_HALF(TYPE, NAME, ...) \
+  detail::deprecated_AT_DISPATCH_ALL_TYPES_AND_HALF();  \
+  AT_DISPATCH_SWITCH(                                   \
+      TYPE,                                             \
+      NAME,                                             \
+      AT_DISPATCH_CASE_ALL_TYPES_AND(at::ScalarType::Half, __VA_ARGS__))

.venv/Lib/site-packages/torch/include/ATen/FuncTorchTLS.h

@@ -0,0 +1,46 @@
+#pragma once
+
+#include <c10/macros/Macros.h>
+#include <memory>
+
+namespace at::functorch {
+
+// NOTE [functorch TLS in pytorch/pytorch]
+//
+// functorch lives out-of-tree. However, it has some TLS that needs to be
+// propagated. The solution for that is we store a pointer to the TLS
+// inside pytorch/pytorch and extend FuncTorchTLSBase inside functorch to
+// include whatever functorch needs.
+//
+// We need to store a pointer due to the indirection:
+// inside functorch, we will create a subclass of FunctorchTLSBase called
+// FuncTorchTLSImpl that actually contains metadata, like the DynamicLayerStack.
+// FuncTorchTLSBase doesn't have any metadata because it hasn't been defined
+// yet.
+//
+// Here in pytorch/pytorch, we will pass around FuncTorchTLSBase*, but inside
+// functorch, we will assign a FuncTorchTLSImpl* to the FunctorchTLSBase*.
+// We can't directly pass around FunctorchTLSBase (without a pointer) because
+// FuncTorchTLSImpl does not fit inside a FuncTorchTLSBase by virtue of having
+// more elements.
+struct TORCH_API FuncTorchTLSBase {
+  virtual ~FuncTorchTLSBase() = default;
+  virtual std::unique_ptr<FuncTorchTLSBase> deepcopy() const = 0;
+
+  virtual int64_t checkSupportsSingleLevelAutogradFunction() const = 0;
+  virtual void checkSupportsCppAutogradFunction() const = 0;
+  virtual void checkSupportsInplaceRequiresGrad() const = 0;
+  virtual void checkSupportsRetainGrad() const = 0;
+};
+
+// returns deepcopy of the functorch tls
+TORCH_API std::unique_ptr<FuncTorchTLSBase> getCopyOfFuncTorchTLS();
+
+// sets the functorch tls. always does a deep copy.
+TORCH_API void setFuncTorchTLS(
+    const std::shared_ptr<const FuncTorchTLSBase>& state);
+
+// get a mutable reference to the functorch tls
+TORCH_API std::unique_ptr<FuncTorchTLSBase>& functorchTLSAccessor();
+
+} // namespace at::functorch

.venv/Lib/site-packages/torch/include/ATen/FunctionalTensorWrapper.h

@@ -0,0 +1,454 @@
+
+#pragma once
+
+#include <ATen/ArrayRef.h>
+#include <ATen/FunctionalStorageImpl.h>
+#include <ATen/core/IListRef.h>
+#include <ATen/core/List.h>
+#include <ATen/core/boxing/BoxedKernel.h>
+#include <ATen/core/boxing/impl/boxing.h>
+#include <ATen/core/dispatch/Dispatcher.h>
+
+#include <c10/core/DispatchKey.h>
+
+namespace at {
+
+// Note [Functionalization Pass In Core]
+// The Functionalization pass is used to remove aliasing from a pytorch program.
+//
+// This is useful for backends that don't support aliasing, like XLA and Vulkan.
+// It's also necessary in order to remove mutation from a program, which is
+// needed in Functorch.
+//
+// Consider this program:
+// a = torch.ones(...)
+// b = a.view(...)
+// b.add_(1)
+//
+// In this program, b is meant to alias with a due to the use of view(). At the
+// end of the program, both a and b are full of 2's. However, backends that
+// don't support aliasing aren't able to correctly implement the view()
+// operator. Instead, they can opt into the Functionalization pass, which will
+// sit between the user and the backend, and provide the necessary aliasing
+// logic.
+//
+// The functionalization pass will turn the above program into a slightly
+// different program that has the same semantics, transparently to the user,
+// that backends like XLA/Vulkan are able to implement a = torch.ones(...) b =
+// a.view_copy(...)  # view() replaced with view_copy(). Backends like
+// XLA/Vulkan can implement this! b.add_(1) a.add_(1)  # Our functionalization
+// pass machinery knows that a and b are aliased - it applies b's mutation to a
+// too.
+//
+// So, how does the functionalization pass keep track of which tensors are
+// aliased? The pass works by wrapping EVERY tensor in the program inside of a
+// FunctionalTensorWrapper, which knows about its alias'd tensors.
+//
+// See Note [Functionalization: Alias Removal] for details on the aliasing
+// machinery. See Note [Functionalization: Mutation Removal] for details on
+// mutation removal.
+struct TORCH_API FunctionalTensorWrapper : public c10::TensorImpl {
+  explicit FunctionalTensorWrapper(const Tensor& value);
+  // Additional constructor to create a FunctionalTensorWrapper directly from an
+  // underlying tensor that was created from a view. For example, the code b =
+  // a.view1() will generate a constructor call to FunctionalTensorWrapper(b, a,
+  // view1_meta)
+  explicit FunctionalTensorWrapper(
+      const Tensor& view_value,
+      const FunctionalTensorWrapper* base,
+      const functionalization::ViewMeta& meta);
+
+  // Get the underlying, actual tensor, that doesn't know anything about
+  // functionalization.
+  const Tensor& value() const {
+    return value_;
+  };
+  // The concept of "level" is only ever important to functorch; it's exposed
+  // here as more of a hook for functorch to use.
+  int64_t level() const {
+    return level_;
+  };
+  void set_level(int64_t level) {
+    level_ = level;
+  }
+  bool has_metadata_mutation() const {
+    return has_metadata_mutation_;
+  };
+
+  void mark_mutation() {
+    functional_storage_impl()->mark_mutation();
+  }
+  // Denotes a mutation that's hidden from autograd,
+  // e.g. for the purposes of passing a tensor to a triton kernel
+  void mark_mutation_hidden_from_autograd() {
+    functional_storage_impl()->mark_mutation_hidden_from_autograd();
+  }
+  void mark_mutation_during_no_grad_or_inference_mode() {
+    functional_storage_impl()->mark_mutation_during_no_grad_or_inference_mode();
+  }
+  // Are all the mutations happening to the tensor hidden from autograd
+  bool are_all_mutations_hidden_from_autograd() const {
+    return functional_storage_impl()->are_all_mutations_hidden_from_autograd();
+  }
+  // Did all mutations happen under no_grad or inference_mode
+  // (We also need to ignore mutations fully hidden from autograd here)
+  bool are_all_mutations_under_no_grad_or_inference_mode() const {
+    return functional_storage_impl()
+        ->are_all_mutations_under_no_grad_or_inference_mode();
+  }
+
+  void maybe_mark_symbolic(const functionalization::ViewMeta& meta) {
+    is_symbolic_ = is_symbolic_ | meta.has_symbolic_inputs;
+  }
+
+  bool is_symbolic() const {
+    return is_symbolic_;
+  }
+
+  // Runs the forward_fn of every ViewMeta collected in the current instance
+  // to some other base.
+  Tensor apply_view_metas(const Tensor& base);
+
+  // Sync's the underlying tensor with its alias, if it's out of date. This
+  // involves two steps: 1) Apply any pending updates/mutations to the alias 2)
+  // Replay the views (if any) to regenerate the current tensor off of the
+  // updated alias.
+  void sync_();
+  // Performs step (1) of the sync. This is its own public API because it's
+  // needed by view_inplace ops like transpose_. See Note [Functionalization
+  // Pass - Inplace View Ops]
+  void regenerate_from_base();
+  // Performs step (2) of the sync. This is its own public API because it's
+  // needed by functorch. functorch wants to make sure that all input tensors to
+  // a functionalized program have been properly synced so it can properly
+  // propagate mutations to inputs. It can't just call sync_(), because the
+  // FunctionalTensorWrapper will look like it has no aliases and sync_ will be
+  // a noop. We use the reference count on storage_ to determine if the wrapper
+  // is aliased, and by the time functorch is ready to propagate updates to
+  // inputs, any intermediate views of the input created by the program will
+  // have been deallocated. This function also returns whether or not the base
+  // actually had any updates to apply.
+  bool apply_updates();
+  // Takes the current state of value_ and snapshots it, sending it as a pending
+  // update to the alias.
+  void commit_update();
+  // When any tensor is mutated, the tensor increments its alias's "generation".
+  // Separately, each tensor maintains its own "generation" counter, which is
+  // used to determine if it's up-to-date with its alias. The act of syncing a
+  // tensor will set a tensor's generation equal to its alias's generation.
+  bool is_up_to_date() const;
+  // Freezes the storage of this tensor, preventing subsequent mutations
+  void freeze_storage() const;
+  // Every FunctionalTensorWrapper contains a vector<ViewMeta> objects
+  // describing the series of view ops that ran to generate the current tensor
+  // from the base tensor. This method is used by inplace-view ops like
+  // transpose_. It appends a ViewMeta to the existing stack, and refreshes the
+  // tensor by replaying the views off of the alias.
+  void mutate_view_meta(const at::functionalization::ViewMeta& meta);
+
+  // Custom implementation of self.set_(src)
+  void set__impl(const FunctionalTensorWrapper* other);
+
+  // Custom implementation of resize_storage_bytes_(self, new_size)
+  void storage_resize_(const c10::SymInt& new_size);
+
+  // Returns whether the current tensor's data was ever mutated
+  bool has_data_mutation();
+  //
+  // Returns whether the current FunctionalTensorWrapper
+  // experienced a set_() call.
+  bool was_storage_changed() {
+    return was_storage_changed_;
+  }
+
+  void set_storage_changed() {
+    was_storage_changed_ = true;
+  }
+
+  // A FunctionalTensor is considered a base if its not a view of another
+  // tensor.
+  bool isBaseTensor() const {
+    return view_metas_.empty();
+  }
+
+  c10::SymInt get_storage_size(bool before) {
+    return functional_storage_impl()->get_storage_size(before);
+  }
+
+  // Returns whether the FunctionalTensor experienced an
+  // untyped_storage().resize_() call
+  bool was_inductor_storage_resized() {
+    return functional_storage_impl()->was_inductor_storage_resized();
+  }
+
+  // The functionalization pass can be used to remove mutations.
+  // It does so by replacing any mutation op with it's corresponding
+  // out-of-place op, followed by a call to replace_(). e.g:
+  //
+  // a.add_(1)
+  //
+  // will turn into:
+  //
+  // tmp = a.add(1)
+  // a.replace_(tmp)
+  //
+  // replace_() swaps out the wrapped tensor, value_, with tmp.
+  void replace_(const Tensor& other, bool from_lazy_regenerate = false);
+
+  bool is_multi_output_view() {
+    return is_multi_output_view_;
+  }
+
+  // See Note[resize_() in functionalization pass]
+  void maybe_replace_storage(const Tensor& other);
+
+  // Replaces the storage with a new functional storage,
+  // and clears the view_metas_ stack.
+  // WARNING: Calling this function will sever the aliasing relationship between
+  // the current FunctionalTensorWrapper and any of its outstanding aliases.
+  // Please only call if you know what you're doing.
+  void _unsafe_reset_storage();
+
+  c10::intrusive_ptr<TensorImpl> shallow_copy_and_detach(
+      const c10::VariableVersion& version_counter,
+      bool allow_tensor_metadata_change) const override;
+
+  c10::intrusive_ptr<TensorImpl> shallow_copy_and_detach(
+      c10::VariableVersion&& version_counter,
+      bool allow_tensor_metadata_change) const override;
+
+  ~FunctionalTensorWrapper() override = default;
+
+  // FunctionalTensorWrapper overrides all custom size/stride function,
+  // so that if the inner tensor has a custom implementation
+  // we make sure to call that implementation.
+  at::IntArrayRef sizes_custom() const override;
+  at::IntArrayRef strides_custom() const override;
+  int64_t dim_custom() const override;
+  int64_t numel_custom() const override;
+  bool is_contiguous_custom(at::MemoryFormat memory_format) const override;
+  c10::SymIntArrayRef sym_sizes_custom() const override;
+  c10::SymInt sym_size_custom(int64_t d) const override;
+  c10::SymIntArrayRef sym_strides_custom() const override;
+  c10::SymInt sym_storage_offset_custom() const override;
+  c10::Device device_custom() const override;
+  c10::Layout layout_impl() const override;
+
+ private:
+  const char* tensorimpl_type_name() const override;
+  void set_constructor_metadata();
+  functionalization::FunctionalStorageImpl* functional_storage_impl() const;
+
+  // This is used to re-implement shallow_copy_and_detach for
+  // FunctionalTensorWrapper. The implementation is identical, but we just need
+  // to return a subclass instead of a plain TensorImpl.
+  // TODO: maybe it's possible to arrange for that to happen automatically
+  // without an override here?
+  template <typename VariableVersion>
+  c10::intrusive_ptr<TensorImpl> shallow_copy_and_detach_core(
+      VariableVersion&& version_counter,
+      bool allow_tensor_metadata_change) const;
+
+  void shallow_copy_from(const c10::intrusive_ptr<TensorImpl>& impl) override;
+  void copy_tensor_metadata_and_refresh(
+      const FunctionalTensorWrapper* src_impl,
+      FunctionalTensorWrapper* dest_impl,
+      const c10::VariableVersion& version_counter,
+      bool allow_tensor_metadata_change) const;
+
+  // Note that value is not taken by reference: internally, the wrapper will
+  // change the value tensor that it points to over time.
+  Tensor value_;
+  int64_t level_{};
+  // These two counters are used for identifying
+  // whether all the mutations on a given tensor are hidden from autograd or
+  // not. If we have an input mutation that is hidden from autograd, then once
+  // we convert the input mutation to a copy_() we know it will be safe to hide
+  // the copy_() from autograd as well.
+  bool has_metadata_mutation_ = false;
+  bool is_multi_output_view_ = false;
+  // Did the tensor experience a set_() call.
+  bool was_storage_changed_ = false;
+  // Did the tensor experience any view operation with symbolic int.
+  bool is_symbolic_ = false;
+
+  size_t generation_ = 0;
+  std::vector<at::functionalization::ViewMeta> view_metas_;
+
+ protected:
+  static void copy_tensor_metadata(
+      const FunctionalTensorWrapper* src_impl,
+      FunctionalTensorWrapper* dest_impl,
+      const c10::VariableVersion& version_counter,
+      bool allow_tensor_metadata_change);
+};
+
+// Utility functions for the functionalization pass.
+
+namespace functionalization {
+namespace impl {
+
+TORCH_API inline FunctionalTensorWrapper* unsafeGetFunctionalWrapper(
+    const Tensor& tensor) {
+  auto functional_impl =
+      static_cast<FunctionalTensorWrapper*>(tensor.unsafeGetTensorImpl());
+  TORCH_INTERNAL_ASSERT_DEBUG_ONLY(functional_impl != nullptr);
+  return functional_impl;
+}
+
+TORCH_API bool isBaseTensor(const at::Tensor& tensor);
+
+TORCH_API bool isFunctionalTensor(const at::Tensor& tensor);
+TORCH_API bool isFunctionalTensor(const std::optional<Tensor>& t);
+TORCH_API bool isFunctionalTensor(
+    const c10::List<std::optional<Tensor>>& t_list);
+TORCH_API bool isFunctionalTensor(ITensorListRef list);
+
+TORCH_API Tensor to_functional_tensor(const Tensor& tensor);
+TORCH_API std::optional<Tensor> to_functional_tensor(
+    const std::optional<Tensor>& tensor);
+TORCH_API c10::List<std::optional<Tensor>> to_functional_tensor(
+    const c10::List<std::optional<Tensor>>& t_list);
+TORCH_API std::vector<Tensor> to_functional_tensor(ITensorListRef t_list);
+
+TORCH_API void freeze_functional_tensor(const Tensor& tensor);
+
+TORCH_API Tensor
+from_functional_tensor(const Tensor& tensor, bool assert_functional = true);
+TORCH_API std::optional<Tensor> from_functional_tensor(
+    const std::optional<Tensor>& t,
+    bool assert_functional = true);
+TORCH_API c10::List<std::optional<Tensor>> from_functional_tensor(
+    const c10::List<std::optional<Tensor>>& t_list);
+TORCH_API std::vector<Tensor> from_functional_tensor(ITensorListRef t_list);
+
+TORCH_API void sync(const at::Tensor& t);
+TORCH_API void sync(const std::optional<Tensor>& t);
+TORCH_API void sync(const c10::List<std::optional<Tensor>>& t_list);
+TORCH_API void sync(ITensorListRef t_list);
+
+TORCH_API void replace_(const Tensor& functional_tensor, const Tensor& other);
+TORCH_API void replace_(
+    const ITensorListRef functional_tensor,
+    ITensorListRef other);
+
+TORCH_API void commit_update(const Tensor& functional_tensor);
+TORCH_API void commit_update(ITensorListRef functional_tensor);
+
+TORCH_API void unsafe_reset_storage(const Tensor& functional_tensor);
+
+TORCH_API void mark_mutation_hidden_from_autograd(
+    const Tensor& functional_tensor);
+
+TORCH_API bool are_all_mutations_hidden_from_autograd(
+    const Tensor& functional_tensor);
+
+TORCH_API bool are_all_mutations_under_no_grad_or_inference_mode(
+    const Tensor& functional_tensor);
+
+// These two methods are XLA-specific logic and are no-ops
+// for the normal functionalization flow.
+TORCH_API void propagate_xla_data(
+    const Tensor& functional_tensor,
+    const Tensor& other);
+TORCH_API void propagate_xla_data(
+    const ITensorListRef functional_tensor,
+    ITensorListRef other);
+
+TORCH_API void propagate_xla_data_direct(
+    const Tensor& tensor,
+    const Tensor& other);
+TORCH_API void propagate_xla_data_direct(
+    const ITensorListRef tensor,
+    ITensorListRef other);
+
+Tensor create_functional_tensor_with_view_meta(
+    const Tensor& view_to_wrap,
+    const Tensor& base,
+    functionalization::ViewMeta meta,
+    int64_t out_idx = 0);
+std::vector<Tensor> create_functional_tensor_with_view_meta(
+    ITensorListRef view_to_wrap,
+    const Tensor& base,
+    const functionalization::ViewMeta& meta);
+
+void mutate_view_meta(
+    const Tensor& self,
+    const functionalization::ViewMeta& meta);
+
+void set_sizes_strides_offset(const Tensor& out, const Tensor& meta_out);
+void set_sizes_strides_offset(
+    const std::vector<Tensor>& outs,
+    const std::vector<Tensor>& meta_outs);
+
+//  ~~~~~ TLS used in functionalization ~~~~~
+
+TORCH_API bool getFunctionalizationReapplyViewsTLS();
+TORCH_API void setFunctionalizationReapplyViewsTLS(bool reapply_views);
+
+class TORCH_API FunctionalizationReapplyViewsGuard {
+ public:
+  FunctionalizationReapplyViewsGuard(bool reapply_views)
+      : prev_(getFunctionalizationReapplyViewsTLS()) {
+    setFunctionalizationReapplyViewsTLS(reapply_views);
+  }
+
+  ~FunctionalizationReapplyViewsGuard() {
+    setFunctionalizationReapplyViewsTLS(prev_);
+  }
+
+  FunctionalizationReapplyViewsGuard(
+      const FunctionalizationReapplyViewsGuard&) = delete;
+  FunctionalizationReapplyViewsGuard operator=(
+      const FunctionalizationReapplyViewsGuard&) = delete;
+  FunctionalizationReapplyViewsGuard(FunctionalizationReapplyViewsGuard&&) =
+      delete;
+  FunctionalizationReapplyViewsGuard operator=(
+      FunctionalizationReapplyViewsGuard&&) = delete;
+
+ private:
+  bool prev_;
+};
+
+} // namespace impl
+
+// Helper function to call an out-of-place composite aten kernel that may use
+// mutations / views internally, and functionalize them.
+TORCH_API void functionalize_op_helper(
+    const c10::OperatorHandle& op,
+    torch::jit::Stack* stack);
+
+template <class Op, bool symint, class ReturnType, class... ParameterTypes>
+struct _functionalize_aten_op final {};
+
+template <class Op, bool symint, class ReturnType, class... ParameterTypes>
+struct _functionalize_aten_op<Op, symint, ReturnType(ParameterTypes...)> final {
+  static ReturnType call(
+      typename c10::maybe_keep_symint<symint, ParameterTypes>::type... args) {
+    using FuncType = ReturnType(
+        typename c10::maybe_keep_symint<symint, ParameterTypes>::type...);
+    auto op = c10::Dispatcher::singleton()
+                  .findSchemaOrThrow(
+                      (const char*)Op::name, (const char*)Op::overload_name)
+                  .typed<FuncType>();
+
+    return c10::impl::BoxedKernelWrapper<FuncType>::call(
+        c10::BoxedKernel::makeFromFunction<functionalize_op_helper>(),
+        op,
+        // BoxedKernelWrapper knows to ignore this keyset argument,
+        // because functionalize_op_helper doesn't take in a DispatchKeySet
+        c10::DispatchKeySet(),
+        args...);
+  }
+};
+
+template <class Op>
+using functionalize_aten_op =
+    _functionalize_aten_op<Op, false, typename Op::schema>;
+
+template <class Op>
+using functionalize_aten_op_symint =
+    _functionalize_aten_op<Op, true, typename Op::schema>;
+
+} // namespace functionalization
+} // namespace at

.venv/Lib/site-packages/torch/include/ATen/Functions.h

@@ -0,0 +1,1454 @@
+#pragma once
+
+// @generated by torchgen/gen.py from Functions.h
+
+#ifdef TORCH_ASSERT_NO_OPERATORS
+#error This change adds a dependency on native_functions.yaml,            \
+  meaning the file will need to be re-compiled every time an operator     \
+  is changed or added. Consider if your change would be better placed in  \
+  another file, or if a more specific header might achieve the same goal. \
+  See NOTE: [Tensor vs. TensorBase]
+#endif
+
+#if defined(AT_PER_OPERATOR_HEADERS) && defined(TORCH_ASSERT_ONLY_METHOD_OPERATORS)
+#error This change adds a dependency on all pytorch operators, meaning the     \
+  file will need to be re-compiled every time an operator is changed or added. \
+  Consider including a specific operator from <ATen/ops/{my_operator}.h> and   \
+  see NOTE [TORCH_ASSERT_ONLY_METHOD_OPERATORS].
+#endif
+
+// NOTE: [TORCH_ASSERT_ONLY_METHOD_OPERATORS]
+//
+// In ATen, certain generated headers files include the definitions of
+// every single operator in PyTorch. Unfortunately this means every
+// time an operator signature is updated or changed in
+// native_functions.yaml, you (and every other PyTorch developer) need
+// to recompile every source file that includes any of these headers.
+//
+// To break up these header dependencies, and improve incremental
+// build times for all PyTorch developers. These headers are split
+// into per-operator headers in the `ATen/ops` folder. This limits
+// incremental builds to only changes to methods of `Tensor`, or files
+// that use the specific operator being changed. With `at::sum` as an
+// example, you should include
+//
+//   <ATen/ops/sum.h>               // instead of ATen/Functions.h
+//   <ATen/ops/sum_native.h>        // instead of ATen/NativeFunctions.h
+//   <ATen/ops/sum_ops.h>           // instead of ATen/Operators.h
+//   <ATen/ops/sum_cpu_dispatch.h>  // instead of ATen/CPUFunctions.h
+//
+// However, even if you're careful to use this in your own code.
+// `Functions.h` might be included indirectly through another header
+// without you realising. To avoid this, you can add
+//
+//   #define TORCH_ASSERT_ONLY_METHOD_OPERATORS
+//
+// to the top of your source file. This way any time the non-specific
+// headers are included, the compiler will error out.
+//
+// Also, be aware that `ops` are not available in all build
+// configurations (namely fb-internal) so you must guard these
+// includes with `#ifdef AT_PER_OPERATOR_HEADERS`. e.g.
+//
+//   #ifndef AT_PER_OPERATOR_HEADERS
+//   #include <ATen/Functions.h>
+//   #else
+//   #include <ATen/ops/sum.h>
+//   #endif
+
+#include <ATen/Context.h>
+#include <ATen/DeviceGuard.h>
+#include <ATen/TensorUtils.h>
+#include <ATen/TracerMode.h>
+#include <ATen/core/Generator.h>
+#include <ATen/core/Reduction.h>
+#include <c10/core/SymInt.h>
+#include <ATen/core/Tensor.h>
+#include <c10/core/Scalar.h>
+#include <c10/core/Storage.h>
+#include <c10/core/TensorOptions.h>
+#include <c10/util/Deprecated.h>
+#include <optional>
+#include <c10/util/OptionalArrayRef.h>
+
+#include <ATen/ops/from_blob.h>
+#include <ATen/ops/tensor.h>
+
+#include <ATen/ops/_adaptive_avg_pool2d.h>
+#include <ATen/ops/_adaptive_avg_pool2d_backward.h>
+#include <ATen/ops/_adaptive_avg_pool3d.h>
+#include <ATen/ops/_adaptive_avg_pool3d_backward.h>
+#include <ATen/ops/_add_batch_dim.h>
+#include <ATen/ops/_add_relu.h>
+#include <ATen/ops/_addmm_activation.h>
+#include <ATen/ops/_aminmax.h>
+#include <ATen/ops/_amp_foreach_non_finite_check_and_unscale.h>
+#include <ATen/ops/_amp_update_scale.h>
+#include <ATen/ops/_assert_async.h>
+#include <ATen/ops/_assert_scalar.h>
+#include <ATen/ops/_assert_tensor_metadata.h>
+#include <ATen/ops/_autocast_to_full_precision.h>
+#include <ATen/ops/_autocast_to_reduced_precision.h>
+#include <ATen/ops/_backward.h>
+#include <ATen/ops/_batch_norm_impl_index.h>
+#include <ATen/ops/_batch_norm_impl_index_backward.h>
+#include <ATen/ops/_batch_norm_no_update.h>
+#include <ATen/ops/_batch_norm_with_update.h>
+#include <ATen/ops/_cast_Byte.h>
+#include <ATen/ops/_cast_Char.h>
+#include <ATen/ops/_cast_Double.h>
+#include <ATen/ops/_cast_Float.h>
+#include <ATen/ops/_cast_Half.h>
+#include <ATen/ops/_cast_Int.h>
+#include <ATen/ops/_cast_Long.h>
+#include <ATen/ops/_cast_Short.h>
+#include <ATen/ops/_cdist_backward.h>
+#include <ATen/ops/_cdist_forward.h>
+#include <ATen/ops/_cholesky_solve_helper.h>
+#include <ATen/ops/_choose_qparams_per_tensor.h>
+#include <ATen/ops/_chunk_cat.h>
+#include <ATen/ops/_coalesce.h>
+#include <ATen/ops/_coalesced.h>
+#include <ATen/ops/_compute_linear_combination.h>
+#include <ATen/ops/_conj.h>
+#include <ATen/ops/_conj_copy.h>
+#include <ATen/ops/_conj_physical.h>
+#include <ATen/ops/_conv_depthwise2d.h>
+#include <ATen/ops/_convert_indices_from_coo_to_csr.h>
+#include <ATen/ops/_convert_indices_from_csr_to_coo.h>
+#include <ATen/ops/_convert_weight_to_int4pack.h>
+#include <ATen/ops/_convolution.h>
+#include <ATen/ops/_convolution_double_backward.h>
+#include <ATen/ops/_convolution_mode.h>
+#include <ATen/ops/_copy_from.h>
+#include <ATen/ops/_copy_from_and_resize.h>
+#include <ATen/ops/_cslt_compress.h>
+#include <ATen/ops/_cslt_sparse_mm.h>
+#include <ATen/ops/_cslt_sparse_mm_search.h>
+#include <ATen/ops/_ctc_loss.h>
+#include <ATen/ops/_ctc_loss_backward.h>
+#include <ATen/ops/_cudnn_ctc_loss.h>
+#include <ATen/ops/_cudnn_init_dropout_state.h>
+#include <ATen/ops/_cudnn_rnn.h>
+#include <ATen/ops/_cudnn_rnn_backward.h>
+#include <ATen/ops/_cudnn_rnn_flatten_weight.h>
+#include <ATen/ops/_cufft_clear_plan_cache.h>
+#include <ATen/ops/_cufft_get_plan_cache_max_size.h>
+#include <ATen/ops/_cufft_get_plan_cache_size.h>
+#include <ATen/ops/_cufft_set_plan_cache_max_size.h>
+#include <ATen/ops/_cummax_helper.h>
+#include <ATen/ops/_cummin_helper.h>
+#include <ATen/ops/_debug_has_internal_overlap.h>
+#include <ATen/ops/_dimI.h>
+#include <ATen/ops/_dimV.h>
+#include <ATen/ops/_dim_arange.h>
+#include <ATen/ops/_dirichlet_grad.h>
+#include <ATen/ops/_efficient_attention_backward.h>
+#include <ATen/ops/_efficient_attention_forward.h>
+#include <ATen/ops/_efficientzerotensor.h>
+#include <ATen/ops/_embedding_bag.h>
+#include <ATen/ops/_embedding_bag_backward.h>
+#include <ATen/ops/_embedding_bag_dense_backward.h>
+#include <ATen/ops/_embedding_bag_forward_only.h>
+#include <ATen/ops/_embedding_bag_per_sample_weights_backward.h>
+#include <ATen/ops/_embedding_bag_sparse_backward.h>
+#include <ATen/ops/_empty_affine_quantized.h>
+#include <ATen/ops/_empty_per_channel_affine_quantized.h>
+#include <ATen/ops/_euclidean_dist.h>
+#include <ATen/ops/_fake_quantize_learnable_per_channel_affine.h>
+#include <ATen/ops/_fake_quantize_learnable_per_channel_affine_backward.h>
+#include <ATen/ops/_fake_quantize_learnable_per_tensor_affine.h>
+#include <ATen/ops/_fake_quantize_learnable_per_tensor_affine_backward.h>
+#include <ATen/ops/_fake_quantize_per_tensor_affine_cachemask_tensor_qparams.h>
+#include <ATen/ops/_fft_c2c.h>
+#include <ATen/ops/_fft_c2r.h>
+#include <ATen/ops/_fft_r2c.h>
+#include <ATen/ops/_fill_mem_eff_dropout_mask.h>
+#include <ATen/ops/_flash_attention_backward.h>
+#include <ATen/ops/_flash_attention_forward.h>
+#include <ATen/ops/_foobar.h>
+#include <ATen/ops/_foreach_abs.h>
+#include <ATen/ops/_foreach_acos.h>
+#include <ATen/ops/_foreach_add.h>
+#include <ATen/ops/_foreach_addcdiv.h>
+#include <ATen/ops/_foreach_addcmul.h>
+#include <ATen/ops/_foreach_asin.h>
+#include <ATen/ops/_foreach_atan.h>
+#include <ATen/ops/_foreach_ceil.h>
+#include <ATen/ops/_foreach_clamp_max.h>
+#include <ATen/ops/_foreach_clamp_min.h>
+#include <ATen/ops/_foreach_copy.h>
+#include <ATen/ops/_foreach_cos.h>
+#include <ATen/ops/_foreach_cosh.h>
+#include <ATen/ops/_foreach_div.h>
+#include <ATen/ops/_foreach_erf.h>
+#include <ATen/ops/_foreach_erfc.h>
+#include <ATen/ops/_foreach_exp.h>
+#include <ATen/ops/_foreach_expm1.h>
+#include <ATen/ops/_foreach_floor.h>
+#include <ATen/ops/_foreach_frac.h>
+#include <ATen/ops/_foreach_lerp.h>
+#include <ATen/ops/_foreach_lgamma.h>
+#include <ATen/ops/_foreach_log.h>
+#include <ATen/ops/_foreach_log10.h>
+#include <ATen/ops/_foreach_log1p.h>
+#include <ATen/ops/_foreach_log2.h>
+#include <ATen/ops/_foreach_max.h>
+#include <ATen/ops/_foreach_maximum.h>
+#include <ATen/ops/_foreach_minimum.h>
+#include <ATen/ops/_foreach_mul.h>
+#include <ATen/ops/_foreach_neg.h>
+#include <ATen/ops/_foreach_norm.h>
+#include <ATen/ops/_foreach_pow.h>
+#include <ATen/ops/_foreach_reciprocal.h>
+#include <ATen/ops/_foreach_round.h>
+#include <ATen/ops/_foreach_sigmoid.h>
+#include <ATen/ops/_foreach_sign.h>
+#include <ATen/ops/_foreach_sin.h>
+#include <ATen/ops/_foreach_sinh.h>
+#include <ATen/ops/_foreach_sqrt.h>
+#include <ATen/ops/_foreach_sub.h>
+#include <ATen/ops/_foreach_tan.h>
+#include <ATen/ops/_foreach_tanh.h>
+#include <ATen/ops/_foreach_trunc.h>
+#include <ATen/ops/_foreach_zero.h>
+#include <ATen/ops/_functional_assert_async.h>
+#include <ATen/ops/_functional_assert_scalar.h>
+#include <ATen/ops/_functional_sym_constrain_range.h>
+#include <ATen/ops/_functional_sym_constrain_range_for_size.h>
+#include <ATen/ops/_fused_adagrad.h>
+#include <ATen/ops/_fused_adam.h>
+#include <ATen/ops/_fused_adamw.h>
+#include <ATen/ops/_fused_dropout.h>
+#include <ATen/ops/_fused_moving_avg_obs_fq_helper.h>
+#include <ATen/ops/_fused_sdp_choice.h>
+#include <ATen/ops/_fused_sgd.h>
+#include <ATen/ops/_fw_primal.h>
+#include <ATen/ops/_fw_primal_copy.h>
+#include <ATen/ops/_gather_sparse_backward.h>
+#include <ATen/ops/_grid_sampler_2d_cpu_fallback.h>
+#include <ATen/ops/_grid_sampler_2d_cpu_fallback_backward.h>
+#include <ATen/ops/_has_compatible_shallow_copy_type.h>
+#include <ATen/ops/_has_same_storage_numel.h>
+#include <ATen/ops/_histogramdd_bin_edges.h>
+#include <ATen/ops/_histogramdd_from_bin_cts.h>
+#include <ATen/ops/_histogramdd_from_bin_tensors.h>
+#include <ATen/ops/_index_put_impl.h>
+#include <ATen/ops/_indices.h>
+#include <ATen/ops/_indices_copy.h>
+#include <ATen/ops/_int_mm.h>
+#include <ATen/ops/_is_all_true.h>
+#include <ATen/ops/_is_any_true.h>
+#include <ATen/ops/_is_zerotensor.h>
+#include <ATen/ops/_jagged_to_padded_dense_forward.h>
+#include <ATen/ops/_lazy_clone.h>
+#include <ATen/ops/_linalg_check_errors.h>
+#include <ATen/ops/_linalg_det.h>
+#include <ATen/ops/_linalg_eigh.h>
+#include <ATen/ops/_linalg_eigvals.h>
+#include <ATen/ops/_linalg_slogdet.h>
+#include <ATen/ops/_linalg_solve_ex.h>
+#include <ATen/ops/_linalg_svd.h>
+#include <ATen/ops/_local_scalar_dense.h>
+#include <ATen/ops/_log_softmax.h>
+#include <ATen/ops/_log_softmax_backward_data.h>
+#include <ATen/ops/_logcumsumexp.h>
+#include <ATen/ops/_lstm_mps.h>
+#include <ATen/ops/_lu_with_info.h>
+#include <ATen/ops/_make_dep_token.h>
+#include <ATen/ops/_make_dual.h>
+#include <ATen/ops/_make_dual_copy.h>
+#include <ATen/ops/_make_per_channel_quantized_tensor.h>
+#include <ATen/ops/_make_per_tensor_quantized_tensor.h>
+#include <ATen/ops/_masked_scale.h>
+#include <ATen/ops/_masked_softmax.h>
+#include <ATen/ops/_masked_softmax_backward.h>
+#include <ATen/ops/_mixed_dtypes_linear.h>
+#include <ATen/ops/_mkldnn_reshape.h>
+#include <ATen/ops/_mkldnn_transpose.h>
+#include <ATen/ops/_mps_convolution.h>
+#include <ATen/ops/_mps_convolution_transpose.h>
+#include <ATen/ops/_native_batch_norm_legit.h>
+#include <ATen/ops/_native_batch_norm_legit_no_training.h>
+#include <ATen/ops/_native_multi_head_attention.h>
+#include <ATen/ops/_neg_view.h>
+#include <ATen/ops/_neg_view_copy.h>
+#include <ATen/ops/_nested_compute_contiguous_strides_offsets.h>
+#include <ATen/ops/_nested_from_padded.h>
+#include <ATen/ops/_nested_from_padded_and_nested_example.h>
+#include <ATen/ops/_nested_get_jagged_dummy.h>
+#include <ATen/ops/_nested_get_lengths.h>
+#include <ATen/ops/_nested_get_max_seqlen.h>
+#include <ATen/ops/_nested_get_min_seqlen.h>
+#include <ATen/ops/_nested_get_offsets.h>
+#include <ATen/ops/_nested_get_ragged_idx.h>
+#include <ATen/ops/_nested_get_values.h>
+#include <ATen/ops/_nested_get_values_copy.h>
+#include <ATen/ops/_nested_select_backward.h>
+#include <ATen/ops/_nested_sum_backward.h>
+#include <ATen/ops/_nested_tensor_from_mask.h>
+#include <ATen/ops/_nested_tensor_from_mask_left_aligned.h>
+#include <ATen/ops/_nested_tensor_from_tensor_list.h>
+#include <ATen/ops/_nested_tensor_size.h>
+#include <ATen/ops/_nested_tensor_softmax_with_shape.h>
+#include <ATen/ops/_nested_tensor_storage_offsets.h>
+#include <ATen/ops/_nested_tensor_strides.h>
+#include <ATen/ops/_nested_view_from_buffer.h>
+#include <ATen/ops/_nested_view_from_buffer_copy.h>
+#include <ATen/ops/_nested_view_from_jagged.h>
+#include <ATen/ops/_nested_view_from_jagged_copy.h>
+#include <ATen/ops/_new_zeros_with_same_feature_meta.h>
+#include <ATen/ops/_nnpack_available.h>
+#include <ATen/ops/_nnpack_spatial_convolution.h>
+#include <ATen/ops/_nnz.h>
+#include <ATen/ops/_pack_padded_sequence.h>
+#include <ATen/ops/_pack_padded_sequence_backward.h>
+#include <ATen/ops/_pad_circular.h>
+#include <ATen/ops/_pad_enum.h>
+#include <ATen/ops/_pad_packed_sequence.h>
+#include <ATen/ops/_padded_dense_to_jagged_forward.h>
+#include <ATen/ops/_pdist_backward.h>
+#include <ATen/ops/_pdist_forward.h>
+#include <ATen/ops/_pin_memory.h>
+#include <ATen/ops/_prelu_kernel.h>
+#include <ATen/ops/_prelu_kernel_backward.h>
+#include <ATen/ops/_print.h>
+#include <ATen/ops/_propagate_xla_data.h>
+#include <ATen/ops/_remove_batch_dim.h>
+#include <ATen/ops/_reshape_alias.h>
+#include <ATen/ops/_reshape_alias_copy.h>
+#include <ATen/ops/_reshape_copy.h>
+#include <ATen/ops/_reshape_from_tensor.h>
+#include <ATen/ops/_resize_output.h>
+#include <ATen/ops/_rowwise_prune.h>
+#include <ATen/ops/_safe_softmax.h>
+#include <ATen/ops/_sample_dirichlet.h>
+#include <ATen/ops/_saturate_weight_to_fp16.h>
+#include <ATen/ops/_scaled_dot_product_attention_math.h>
+#include <ATen/ops/_scaled_dot_product_attention_math_for_mps.h>
+#include <ATen/ops/_scaled_dot_product_cudnn_attention.h>
+#include <ATen/ops/_scaled_dot_product_cudnn_attention_backward.h>
+#include <ATen/ops/_scaled_dot_product_efficient_attention.h>
+#include <ATen/ops/_scaled_dot_product_efficient_attention_backward.h>
+#include <ATen/ops/_scaled_dot_product_flash_attention.h>
+#include <ATen/ops/_scaled_dot_product_flash_attention_backward.h>
+#include <ATen/ops/_scaled_dot_product_flash_attention_for_cpu.h>
+#include <ATen/ops/_scaled_dot_product_flash_attention_for_cpu_backward.h>
+#include <ATen/ops/_scaled_dot_product_fused_attention_overrideable.h>
+#include <ATen/ops/_scaled_dot_product_fused_attention_overrideable_backward.h>
+#include <ATen/ops/_scaled_mm.h>
+#include <ATen/ops/_segment_reduce_backward.h>
+#include <ATen/ops/_shape_as_tensor.h>
+#include <ATen/ops/_slow_conv2d_backward.h>
+#include <ATen/ops/_slow_conv2d_forward.h>
+#include <ATen/ops/_sobol_engine_draw.h>
+#include <ATen/ops/_sobol_engine_ff.h>
+#include <ATen/ops/_sobol_engine_initialize_state.h>
+#include <ATen/ops/_sobol_engine_scramble.h>
+#include <ATen/ops/_softmax.h>
+#include <ATen/ops/_softmax_backward_data.h>
+#include <ATen/ops/_sparse_addmm.h>
+#include <ATen/ops/_sparse_broadcast_to.h>
+#include <ATen/ops/_sparse_broadcast_to_copy.h>
+#include <ATen/ops/_sparse_bsc_tensor_unsafe.h>
+#include <ATen/ops/_sparse_bsr_tensor_unsafe.h>
+#include <ATen/ops/_sparse_compressed_tensor_unsafe.h>
+#include <ATen/ops/_sparse_compressed_tensor_with_dims.h>
+#include <ATen/ops/_sparse_coo_tensor_unsafe.h>
+#include <ATen/ops/_sparse_coo_tensor_with_dims.h>
+#include <ATen/ops/_sparse_coo_tensor_with_dims_and_tensors.h>
+#include <ATen/ops/_sparse_csc_tensor_unsafe.h>
+#include <ATen/ops/_sparse_csr_prod.h>
+#include <ATen/ops/_sparse_csr_sum.h>
+#include <ATen/ops/_sparse_csr_tensor_unsafe.h>
+#include <ATen/ops/_sparse_log_softmax.h>
+#include <ATen/ops/_sparse_log_softmax_backward_data.h>
+#include <ATen/ops/_sparse_mask_projection.h>
+#include <ATen/ops/_sparse_mm.h>
+#include <ATen/ops/_sparse_mm_reduce_impl.h>
+#include <ATen/ops/_sparse_mm_reduce_impl_backward.h>
+#include <ATen/ops/_sparse_semi_structured_addmm.h>
+#include <ATen/ops/_sparse_semi_structured_apply.h>
+#include <ATen/ops/_sparse_semi_structured_apply_dense.h>
+#include <ATen/ops/_sparse_semi_structured_linear.h>
+#include <ATen/ops/_sparse_semi_structured_mm.h>
+#include <ATen/ops/_sparse_semi_structured_tile.h>
+#include <ATen/ops/_sparse_softmax.h>
+#include <ATen/ops/_sparse_softmax_backward_data.h>
+#include <ATen/ops/_sparse_sparse_matmul.h>
+#include <ATen/ops/_sparse_sum.h>
+#include <ATen/ops/_sparse_sum_backward.h>
+#include <ATen/ops/_spdiags.h>
+#include <ATen/ops/_spsolve.h>
+#include <ATen/ops/_stack.h>
+#include <ATen/ops/_standard_gamma.h>
+#include <ATen/ops/_standard_gamma_grad.h>
+#include <ATen/ops/_test_ambiguous_defaults.h>
+#include <ATen/ops/_test_autograd_multiple_dispatch.h>
+#include <ATen/ops/_test_autograd_multiple_dispatch_view.h>
+#include <ATen/ops/_test_autograd_multiple_dispatch_view_copy.h>
+#include <ATen/ops/_test_check_tensor.h>
+#include <ATen/ops/_test_functorch_fallback.h>
+#include <ATen/ops/_test_optional_filled_intlist.h>
+#include <ATen/ops/_test_optional_floatlist.h>
+#include <ATen/ops/_test_optional_intlist.h>
+#include <ATen/ops/_test_parallel_materialize.h>
+#include <ATen/ops/_test_serialization_subcmul.h>
+#include <ATen/ops/_test_string_default.h>
+#include <ATen/ops/_test_warn_in_autograd.h>
+#include <ATen/ops/_thnn_differentiable_gru_cell_backward.h>
+#include <ATen/ops/_thnn_differentiable_lstm_cell_backward.h>
+#include <ATen/ops/_thnn_fused_gru_cell.h>
+#include <ATen/ops/_thnn_fused_gru_cell_backward.h>
+#include <ATen/ops/_thnn_fused_lstm_cell.h>
+#include <ATen/ops/_thnn_fused_lstm_cell_backward.h>
+#include <ATen/ops/_thnn_fused_lstm_cell_backward_impl.h>
+#include <ATen/ops/_to_copy.h>
+#include <ATen/ops/_to_cpu.h>
+#include <ATen/ops/_to_dense.h>
+#include <ATen/ops/_to_sparse.h>
+#include <ATen/ops/_to_sparse_bsc.h>
+#include <ATen/ops/_to_sparse_bsr.h>
+#include <ATen/ops/_to_sparse_csc.h>
+#include <ATen/ops/_to_sparse_csr.h>
+#include <ATen/ops/_to_sparse_semi_structured.h>
+#include <ATen/ops/_transform_bias_rescale_qkv.h>
+#include <ATen/ops/_transformer_encoder_layer_fwd.h>
+#include <ATen/ops/_trilinear.h>
+#include <ATen/ops/_triton_multi_head_attention.h>
+#include <ATen/ops/_triton_scaled_dot_attention.h>
+#include <ATen/ops/_unique.h>
+#include <ATen/ops/_unique2.h>
+#include <ATen/ops/_unpack_dual.h>
+#include <ATen/ops/_unsafe_index.h>
+#include <ATen/ops/_unsafe_index_put.h>
+#include <ATen/ops/_unsafe_masked_index.h>
+#include <ATen/ops/_unsafe_masked_index_put_accumulate.h>
+#include <ATen/ops/_unsafe_view.h>
+#include <ATen/ops/_upsample_bicubic2d_aa.h>
+#include <ATen/ops/_upsample_bicubic2d_aa_backward.h>
+#include <ATen/ops/_upsample_bilinear2d_aa.h>
+#include <ATen/ops/_upsample_bilinear2d_aa_backward.h>
+#include <ATen/ops/_upsample_nearest_exact1d.h>
+#include <ATen/ops/_upsample_nearest_exact1d_backward.h>
+#include <ATen/ops/_upsample_nearest_exact2d.h>
+#include <ATen/ops/_upsample_nearest_exact2d_backward.h>
+#include <ATen/ops/_upsample_nearest_exact3d.h>
+#include <ATen/ops/_upsample_nearest_exact3d_backward.h>
+#include <ATen/ops/_use_cudnn_ctc_loss.h>
+#include <ATen/ops/_use_cudnn_rnn_flatten_weight.h>
+#include <ATen/ops/_validate_compressed_sparse_indices.h>
+#include <ATen/ops/_validate_sparse_bsc_tensor_args.h>
+#include <ATen/ops/_validate_sparse_bsr_tensor_args.h>
+#include <ATen/ops/_validate_sparse_compressed_tensor_args.h>
+#include <ATen/ops/_validate_sparse_coo_tensor_args.h>
+#include <ATen/ops/_validate_sparse_csc_tensor_args.h>
+#include <ATen/ops/_validate_sparse_csr_tensor_args.h>
+#include <ATen/ops/_values.h>
+#include <ATen/ops/_values_copy.h>
+#include <ATen/ops/_version.h>
+#include <ATen/ops/_weight_int4pack_mm.h>
+#include <ATen/ops/_weight_int8pack_mm.h>
+#include <ATen/ops/_weight_norm.h>
+#include <ATen/ops/_weight_norm_differentiable_backward.h>
+#include <ATen/ops/_weight_norm_interface.h>
+#include <ATen/ops/_weight_norm_interface_backward.h>
+#include <ATen/ops/_wrapped_linear_prepack.h>
+#include <ATen/ops/_wrapped_quantized_linear_prepacked.h>
+#include <ATen/ops/abs.h>
+#include <ATen/ops/absolute.h>
+#include <ATen/ops/acos.h>
+#include <ATen/ops/acosh.h>
+#include <ATen/ops/adaptive_avg_pool1d.h>
+#include <ATen/ops/adaptive_avg_pool2d.h>
+#include <ATen/ops/adaptive_avg_pool3d.h>
+#include <ATen/ops/adaptive_avg_pool3d_backward.h>
+#include <ATen/ops/adaptive_max_pool1d.h>
+#include <ATen/ops/adaptive_max_pool2d.h>
+#include <ATen/ops/adaptive_max_pool2d_backward.h>
+#include <ATen/ops/adaptive_max_pool3d.h>
+#include <ATen/ops/adaptive_max_pool3d_backward.h>
+#include <ATen/ops/add.h>
+#include <ATen/ops/addbmm.h>
+#include <ATen/ops/addcdiv.h>
+#include <ATen/ops/addcmul.h>
+#include <ATen/ops/addmm.h>
+#include <ATen/ops/addmv.h>
+#include <ATen/ops/addr.h>
+#include <ATen/ops/adjoint.h>
+#include <ATen/ops/affine_grid_generator.h>
+#include <ATen/ops/affine_grid_generator_backward.h>
+#include <ATen/ops/alias.h>
+#include <ATen/ops/alias_copy.h>
+#include <ATen/ops/align_as.h>
+#include <ATen/ops/align_tensors.h>
+#include <ATen/ops/align_to.h>
+#include <ATen/ops/all.h>
+#include <ATen/ops/allclose.h>
+#include <ATen/ops/alpha_dropout.h>
+#include <ATen/ops/amax.h>
+#include <ATen/ops/amin.h>
+#include <ATen/ops/aminmax.h>
+#include <ATen/ops/and.h>
+#include <ATen/ops/angle.h>
+#include <ATen/ops/any.h>
+#include <ATen/ops/arange.h>
+#include <ATen/ops/arccos.h>
+#include <ATen/ops/arccosh.h>
+#include <ATen/ops/arcsin.h>
+#include <ATen/ops/arcsinh.h>
+#include <ATen/ops/arctan.h>
+#include <ATen/ops/arctan2.h>
+#include <ATen/ops/arctanh.h>
+#include <ATen/ops/argmax.h>
+#include <ATen/ops/argmin.h>
+#include <ATen/ops/argsort.h>
+#include <ATen/ops/argwhere.h>
+#include <ATen/ops/as_strided.h>
+#include <ATen/ops/as_strided_copy.h>
+#include <ATen/ops/as_strided_scatter.h>
+#include <ATen/ops/asin.h>
+#include <ATen/ops/asinh.h>
+#include <ATen/ops/atan.h>
+#include <ATen/ops/atan2.h>
+#include <ATen/ops/atanh.h>
+#include <ATen/ops/atleast_1d.h>
+#include <ATen/ops/atleast_2d.h>
+#include <ATen/ops/atleast_3d.h>
+#include <ATen/ops/avg_pool1d.h>
+#include <ATen/ops/avg_pool2d.h>
+#include <ATen/ops/avg_pool2d_backward.h>
+#include <ATen/ops/avg_pool3d.h>
+#include <ATen/ops/avg_pool3d_backward.h>
+#include <ATen/ops/baddbmm.h>
+#include <ATen/ops/bartlett_window.h>
+#include <ATen/ops/batch_norm.h>
+#include <ATen/ops/batch_norm_backward.h>
+#include <ATen/ops/batch_norm_backward_elemt.h>
+#include <ATen/ops/batch_norm_backward_reduce.h>
+#include <ATen/ops/batch_norm_elemt.h>
+#include <ATen/ops/batch_norm_gather_stats.h>
+#include <ATen/ops/batch_norm_gather_stats_with_counts.h>
+#include <ATen/ops/batch_norm_stats.h>
+#include <ATen/ops/batch_norm_update_stats.h>
+#include <ATen/ops/bernoulli.h>
+#include <ATen/ops/bilinear.h>
+#include <ATen/ops/binary_cross_entropy.h>
+#include <ATen/ops/binary_cross_entropy_backward.h>
+#include <ATen/ops/binary_cross_entropy_with_logits.h>
+#include <ATen/ops/bincount.h>
+#include <ATen/ops/binomial.h>
+#include <ATen/ops/bitwise_and.h>
+#include <ATen/ops/bitwise_left_shift.h>
+#include <ATen/ops/bitwise_not.h>
+#include <ATen/ops/bitwise_or.h>
+#include <ATen/ops/bitwise_right_shift.h>
+#include <ATen/ops/bitwise_xor.h>
+#include <ATen/ops/blackman_window.h>
+#include <ATen/ops/block_diag.h>
+#include <ATen/ops/bmm.h>
+#include <ATen/ops/broadcast_tensors.h>
+#include <ATen/ops/broadcast_to.h>
+#include <ATen/ops/bucketize.h>
+#include <ATen/ops/can_cast.h>
+#include <ATen/ops/cartesian_prod.h>
+#include <ATen/ops/cat.h>
+#include <ATen/ops/cauchy.h>
+#include <ATen/ops/ccol_indices.h>
+#include <ATen/ops/ccol_indices_copy.h>
+#include <ATen/ops/cdist.h>
+#include <ATen/ops/ceil.h>
+#include <ATen/ops/celu.h>
+#include <ATen/ops/chain_matmul.h>
+#include <ATen/ops/chalf.h>
+#include <ATen/ops/channel_shuffle.h>
+#include <ATen/ops/cholesky.h>
+#include <ATen/ops/cholesky_inverse.h>
+#include <ATen/ops/cholesky_solve.h>
+#include <ATen/ops/choose_qparams_optimized.h>
+#include <ATen/ops/chunk.h>
+#include <ATen/ops/clamp.h>
+#include <ATen/ops/clamp_max.h>
+#include <ATen/ops/clamp_min.h>
+#include <ATen/ops/clip.h>
+#include <ATen/ops/clone.h>
+#include <ATen/ops/coalesce.h>
+#include <ATen/ops/col2im.h>
+#include <ATen/ops/col_indices.h>
+#include <ATen/ops/col_indices_copy.h>
+#include <ATen/ops/column_stack.h>
+#include <ATen/ops/combinations.h>
+#include <ATen/ops/complex.h>
+#include <ATen/ops/concat.h>
+#include <ATen/ops/concatenate.h>
+#include <ATen/ops/conj.h>
+#include <ATen/ops/conj_physical.h>
+#include <ATen/ops/constant_pad_nd.h>
+#include <ATen/ops/contiguous.h>
+#include <ATen/ops/conv1d.h>
+#include <ATen/ops/conv2d.h>
+#include <ATen/ops/conv3d.h>
+#include <ATen/ops/conv_depthwise3d.h>
+#include <ATen/ops/conv_tbc.h>
+#include <ATen/ops/conv_tbc_backward.h>
+#include <ATen/ops/conv_transpose1d.h>
+#include <ATen/ops/conv_transpose2d.h>
+#include <ATen/ops/conv_transpose3d.h>
+#include <ATen/ops/convolution.h>
+#include <ATen/ops/convolution_backward.h>
+#include <ATen/ops/convolution_backward_overrideable.h>
+#include <ATen/ops/convolution_overrideable.h>
+#include <ATen/ops/copy.h>
+#include <ATen/ops/copy_sparse_to_sparse.h>
+#include <ATen/ops/copysign.h>
+#include <ATen/ops/corrcoef.h>
+#include <ATen/ops/cos.h>
+#include <ATen/ops/cosh.h>
+#include <ATen/ops/cosine_embedding_loss.h>
+#include <ATen/ops/cosine_similarity.h>
+#include <ATen/ops/count_nonzero.h>
+#include <ATen/ops/cov.h>
+#include <ATen/ops/cross.h>
+#include <ATen/ops/cross_entropy_loss.h>
+#include <ATen/ops/crow_indices.h>
+#include <ATen/ops/crow_indices_copy.h>
+#include <ATen/ops/ctc_loss.h>
+#include <ATen/ops/cudnn_affine_grid_generator.h>
+#include <ATen/ops/cudnn_affine_grid_generator_backward.h>
+#include <ATen/ops/cudnn_batch_norm.h>
+#include <ATen/ops/cudnn_batch_norm_backward.h>
+#include <ATen/ops/cudnn_convolution.h>
+#include <ATen/ops/cudnn_convolution_add_relu.h>
+#include <ATen/ops/cudnn_convolution_relu.h>
+#include <ATen/ops/cudnn_convolution_transpose.h>
+#include <ATen/ops/cudnn_grid_sampler.h>
+#include <ATen/ops/cudnn_grid_sampler_backward.h>
+#include <ATen/ops/cudnn_is_acceptable.h>
+#include <ATen/ops/cummax.h>
+#include <ATen/ops/cummaxmin_backward.h>
+#include <ATen/ops/cummin.h>
+#include <ATen/ops/cumprod.h>
+#include <ATen/ops/cumprod_backward.h>
+#include <ATen/ops/cumsum.h>
+#include <ATen/ops/cumulative_trapezoid.h>
+#include <ATen/ops/data.h>
+#include <ATen/ops/deg2rad.h>
+#include <ATen/ops/dense_dim.h>
+#include <ATen/ops/dequantize.h>
+#include <ATen/ops/det.h>
+#include <ATen/ops/detach.h>
+#include <ATen/ops/detach_copy.h>
+#include <ATen/ops/diag.h>
+#include <ATen/ops/diag_embed.h>
+#include <ATen/ops/diagflat.h>
+#include <ATen/ops/diagonal.h>
+#include <ATen/ops/diagonal_backward.h>
+#include <ATen/ops/diagonal_copy.h>
+#include <ATen/ops/diagonal_scatter.h>
+#include <ATen/ops/diff.h>
+#include <ATen/ops/digamma.h>
+#include <ATen/ops/dist.h>
+#include <ATen/ops/div.h>
+#include <ATen/ops/divide.h>
+#include <ATen/ops/dot.h>
+#include <ATen/ops/dropout.h>
+#include <ATen/ops/dsplit.h>
+#include <ATen/ops/dstack.h>
+#include <ATen/ops/einsum.h>
+#include <ATen/ops/elu.h>
+#include <ATen/ops/elu_backward.h>
+#include <ATen/ops/embedding.h>
+#include <ATen/ops/embedding_backward.h>
+#include <ATen/ops/embedding_bag.h>
+#include <ATen/ops/embedding_dense_backward.h>
+#include <ATen/ops/embedding_renorm.h>
+#include <ATen/ops/embedding_sparse_backward.h>
+#include <ATen/ops/empty.h>
+#include <ATen/ops/empty_like.h>
+#include <ATen/ops/empty_permuted.h>
+#include <ATen/ops/empty_quantized.h>
+#include <ATen/ops/empty_strided.h>
+#include <ATen/ops/eq.h>
+#include <ATen/ops/equal.h>
+#include <ATen/ops/erf.h>
+#include <ATen/ops/erfc.h>
+#include <ATen/ops/erfinv.h>
+#include <ATen/ops/exp.h>
+#include <ATen/ops/exp2.h>
+#include <ATen/ops/expand.h>
+#include <ATen/ops/expand_as.h>
+#include <ATen/ops/expand_copy.h>
+#include <ATen/ops/expm1.h>
+#include <ATen/ops/exponential.h>
+#include <ATen/ops/eye.h>
+#include <ATen/ops/fake_quantize_per_channel_affine.h>
+#include <ATen/ops/fake_quantize_per_channel_affine_cachemask.h>
+#include <ATen/ops/fake_quantize_per_channel_affine_cachemask_backward.h>
+#include <ATen/ops/fake_quantize_per_tensor_affine.h>
+#include <ATen/ops/fake_quantize_per_tensor_affine_cachemask.h>
+#include <ATen/ops/fake_quantize_per_tensor_affine_cachemask_backward.h>
+#include <ATen/ops/fbgemm_linear_fp16_weight.h>
+#include <ATen/ops/fbgemm_linear_fp16_weight_fp32_activation.h>
+#include <ATen/ops/fbgemm_linear_int8_weight.h>
+#include <ATen/ops/fbgemm_linear_int8_weight_fp32_activation.h>
+#include <ATen/ops/fbgemm_linear_quantize_weight.h>
+#include <ATen/ops/fbgemm_pack_gemm_matrix_fp16.h>
+#include <ATen/ops/fbgemm_pack_quantized_matrix.h>
+#include <ATen/ops/feature_alpha_dropout.h>
+#include <ATen/ops/feature_dropout.h>
+#include <ATen/ops/fft_fft.h>
+#include <ATen/ops/fft_fft2.h>
+#include <ATen/ops/fft_fftfreq.h>
+#include <ATen/ops/fft_fftn.h>
+#include <ATen/ops/fft_fftshift.h>
+#include <ATen/ops/fft_hfft.h>
+#include <ATen/ops/fft_hfft2.h>
+#include <ATen/ops/fft_hfftn.h>
+#include <ATen/ops/fft_ifft.h>
+#include <ATen/ops/fft_ifft2.h>
+#include <ATen/ops/fft_ifftn.h>
+#include <ATen/ops/fft_ifftshift.h>
+#include <ATen/ops/fft_ihfft.h>
+#include <ATen/ops/fft_ihfft2.h>
+#include <ATen/ops/fft_ihfftn.h>
+#include <ATen/ops/fft_irfft.h>
+#include <ATen/ops/fft_irfft2.h>
+#include <ATen/ops/fft_irfftn.h>
+#include <ATen/ops/fft_rfft.h>
+#include <ATen/ops/fft_rfft2.h>
+#include <ATen/ops/fft_rfftfreq.h>
+#include <ATen/ops/fft_rfftn.h>
+#include <ATen/ops/fill.h>
+#include <ATen/ops/fill_diagonal.h>
+#include <ATen/ops/fix.h>
+#include <ATen/ops/flatten.h>
+#include <ATen/ops/flatten_dense_tensors.h>
+#include <ATen/ops/flip.h>
+#include <ATen/ops/fliplr.h>
+#include <ATen/ops/flipud.h>
+#include <ATen/ops/float_power.h>
+#include <ATen/ops/floor.h>
+#include <ATen/ops/floor_divide.h>
+#include <ATen/ops/fmax.h>
+#include <ATen/ops/fmin.h>
+#include <ATen/ops/fmod.h>
+#include <ATen/ops/frac.h>
+#include <ATen/ops/fractional_max_pool2d.h>
+#include <ATen/ops/fractional_max_pool2d_backward.h>
+#include <ATen/ops/fractional_max_pool3d.h>
+#include <ATen/ops/fractional_max_pool3d_backward.h>
+#include <ATen/ops/frexp.h>
+#include <ATen/ops/frobenius_norm.h>
+#include <ATen/ops/from_file.h>
+#include <ATen/ops/full.h>
+#include <ATen/ops/full_like.h>
+#include <ATen/ops/fused_moving_avg_obs_fake_quant.h>
+#include <ATen/ops/gather.h>
+#include <ATen/ops/gather_backward.h>
+#include <ATen/ops/gcd.h>
+#include <ATen/ops/ge.h>
+#include <ATen/ops/gelu.h>
+#include <ATen/ops/gelu_backward.h>
+#include <ATen/ops/geometric.h>
+#include <ATen/ops/geqrf.h>
+#include <ATen/ops/ger.h>
+#include <ATen/ops/glu.h>
+#include <ATen/ops/glu_backward.h>
+#include <ATen/ops/glu_backward_jvp.h>
+#include <ATen/ops/glu_jvp.h>
+#include <ATen/ops/gradient.h>
+#include <ATen/ops/greater.h>
+#include <ATen/ops/greater_equal.h>
+#include <ATen/ops/grid_sampler.h>
+#include <ATen/ops/grid_sampler_2d.h>
+#include <ATen/ops/grid_sampler_2d_backward.h>
+#include <ATen/ops/grid_sampler_3d.h>
+#include <ATen/ops/grid_sampler_3d_backward.h>
+#include <ATen/ops/group_norm.h>
+#include <ATen/ops/gru.h>
+#include <ATen/ops/gru_cell.h>
+#include <ATen/ops/gt.h>
+#include <ATen/ops/hamming_window.h>
+#include <ATen/ops/hann_window.h>
+#include <ATen/ops/hardshrink.h>
+#include <ATen/ops/hardshrink_backward.h>
+#include <ATen/ops/hardsigmoid.h>
+#include <ATen/ops/hardsigmoid_backward.h>
+#include <ATen/ops/hardswish.h>
+#include <ATen/ops/hardswish_backward.h>
+#include <ATen/ops/hardtanh.h>
+#include <ATen/ops/hardtanh_backward.h>
+#include <ATen/ops/heaviside.h>
+#include <ATen/ops/hinge_embedding_loss.h>
+#include <ATen/ops/histc.h>
+#include <ATen/ops/histogram.h>
+#include <ATen/ops/histogramdd.h>
+#include <ATen/ops/hsplit.h>
+#include <ATen/ops/hspmm.h>
+#include <ATen/ops/hstack.h>
+#include <ATen/ops/huber_loss.h>
+#include <ATen/ops/huber_loss_backward.h>
+#include <ATen/ops/hypot.h>
+#include <ATen/ops/i0.h>
+#include <ATen/ops/igamma.h>
+#include <ATen/ops/igammac.h>
+#include <ATen/ops/im2col.h>
+#include <ATen/ops/imag.h>
+#include <ATen/ops/index.h>
+#include <ATen/ops/index_add.h>
+#include <ATen/ops/index_copy.h>
+#include <ATen/ops/index_fill.h>
+#include <ATen/ops/index_put.h>
+#include <ATen/ops/index_reduce.h>
+#include <ATen/ops/index_select.h>
+#include <ATen/ops/index_select_backward.h>
+#include <ATen/ops/indices.h>
+#include <ATen/ops/indices_copy.h>
+#include <ATen/ops/infinitely_differentiable_gelu_backward.h>
+#include <ATen/ops/inner.h>
+#include <ATen/ops/instance_norm.h>
+#include <ATen/ops/int_repr.h>
+#include <ATen/ops/inverse.h>
+#include <ATen/ops/is_coalesced.h>
+#include <ATen/ops/is_complex.h>
+#include <ATen/ops/is_conj.h>
+#include <ATen/ops/is_distributed.h>
+#include <ATen/ops/is_floating_point.h>
+#include <ATen/ops/is_inference.h>
+#include <ATen/ops/is_leaf.h>
+#include <ATen/ops/is_neg.h>
+#include <ATen/ops/is_nonzero.h>
+#include <ATen/ops/is_pinned.h>
+#include <ATen/ops/is_same_size.h>
+#include <ATen/ops/is_set_to.h>
+#include <ATen/ops/is_signed.h>
+#include <ATen/ops/is_vulkan_available.h>
+#include <ATen/ops/isclose.h>
+#include <ATen/ops/isfinite.h>
+#include <ATen/ops/isin.h>
+#include <ATen/ops/isinf.h>
+#include <ATen/ops/isnan.h>
+#include <ATen/ops/isneginf.h>
+#include <ATen/ops/isposinf.h>
+#include <ATen/ops/isreal.h>
+#include <ATen/ops/istft.h>
+#include <ATen/ops/item.h>
+#include <ATen/ops/kaiser_window.h>
+#include <ATen/ops/kl_div.h>
+#include <ATen/ops/kron.h>
+#include <ATen/ops/kthvalue.h>
+#include <ATen/ops/l1_loss.h>
+#include <ATen/ops/layer_norm.h>
+#include <ATen/ops/lcm.h>
+#include <ATen/ops/ldexp.h>
+#include <ATen/ops/le.h>
+#include <ATen/ops/leaky_relu.h>
+#include <ATen/ops/leaky_relu_backward.h>
+#include <ATen/ops/lerp.h>
+#include <ATen/ops/less.h>
+#include <ATen/ops/less_equal.h>
+#include <ATen/ops/lgamma.h>
+#include <ATen/ops/lift.h>
+#include <ATen/ops/lift_fresh.h>
+#include <ATen/ops/lift_fresh_copy.h>
+#include <ATen/ops/linalg_cholesky.h>
+#include <ATen/ops/linalg_cholesky_ex.h>
+#include <ATen/ops/linalg_cond.h>
+#include <ATen/ops/linalg_cross.h>
+#include <ATen/ops/linalg_det.h>
+#include <ATen/ops/linalg_diagonal.h>
+#include <ATen/ops/linalg_eig.h>
+#include <ATen/ops/linalg_eigh.h>
+#include <ATen/ops/linalg_eigvals.h>
+#include <ATen/ops/linalg_eigvalsh.h>
+#include <ATen/ops/linalg_householder_product.h>
+#include <ATen/ops/linalg_inv.h>
+#include <ATen/ops/linalg_inv_ex.h>
+#include <ATen/ops/linalg_ldl_factor.h>
+#include <ATen/ops/linalg_ldl_factor_ex.h>
+#include <ATen/ops/linalg_ldl_solve.h>
+#include <ATen/ops/linalg_lstsq.h>
+#include <ATen/ops/linalg_lu.h>
+#include <ATen/ops/linalg_lu_factor.h>
+#include <ATen/ops/linalg_lu_factor_ex.h>
+#include <ATen/ops/linalg_lu_solve.h>
+#include <ATen/ops/linalg_matmul.h>
+#include <ATen/ops/linalg_matrix_exp.h>
+#include <ATen/ops/linalg_matrix_norm.h>
+#include <ATen/ops/linalg_matrix_power.h>
+#include <ATen/ops/linalg_matrix_rank.h>
+#include <ATen/ops/linalg_multi_dot.h>
+#include <ATen/ops/linalg_norm.h>
+#include <ATen/ops/linalg_pinv.h>
+#include <ATen/ops/linalg_qr.h>
+#include <ATen/ops/linalg_slogdet.h>
+#include <ATen/ops/linalg_solve.h>
+#include <ATen/ops/linalg_solve_ex.h>
+#include <ATen/ops/linalg_solve_triangular.h>
+#include <ATen/ops/linalg_svd.h>
+#include <ATen/ops/linalg_svdvals.h>
+#include <ATen/ops/linalg_tensorinv.h>
+#include <ATen/ops/linalg_tensorsolve.h>
+#include <ATen/ops/linalg_vander.h>
+#include <ATen/ops/linalg_vecdot.h>
+#include <ATen/ops/linalg_vector_norm.h>
+#include <ATen/ops/linear.h>
+#include <ATen/ops/linear_backward.h>
+#include <ATen/ops/linspace.h>
+#include <ATen/ops/log.h>
+#include <ATen/ops/log10.h>
+#include <ATen/ops/log1p.h>
+#include <ATen/ops/log2.h>
+#include <ATen/ops/log_normal.h>
+#include <ATen/ops/log_sigmoid.h>
+#include <ATen/ops/log_sigmoid_backward.h>
+#include <ATen/ops/log_sigmoid_forward.h>
+#include <ATen/ops/log_softmax.h>
+#include <ATen/ops/logaddexp.h>
+#include <ATen/ops/logaddexp2.h>
+#include <ATen/ops/logcumsumexp.h>
+#include <ATen/ops/logdet.h>
+#include <ATen/ops/logical_and.h>
+#include <ATen/ops/logical_not.h>
+#include <ATen/ops/logical_or.h>
+#include <ATen/ops/logical_xor.h>
+#include <ATen/ops/logit.h>
+#include <ATen/ops/logit_backward.h>
+#include <ATen/ops/logspace.h>
+#include <ATen/ops/logsumexp.h>
+#include <ATen/ops/lshift.h>
+#include <ATen/ops/lstm.h>
+#include <ATen/ops/lstm_cell.h>
+#include <ATen/ops/lstm_mps_backward.h>
+#include <ATen/ops/lt.h>
+#include <ATen/ops/lu_solve.h>
+#include <ATen/ops/lu_unpack.h>
+#include <ATen/ops/mH.h>
+#include <ATen/ops/mT.h>
+#include <ATen/ops/margin_ranking_loss.h>
+#include <ATen/ops/masked_fill.h>
+#include <ATen/ops/masked_scatter.h>
+#include <ATen/ops/masked_scatter_backward.h>
+#include <ATen/ops/masked_select.h>
+#include <ATen/ops/masked_select_backward.h>
+#include <ATen/ops/matmul.h>
+#include <ATen/ops/matmul_backward.h>
+#include <ATen/ops/matrix_H.h>
+#include <ATen/ops/matrix_exp.h>
+#include <ATen/ops/matrix_exp_backward.h>
+#include <ATen/ops/matrix_power.h>
+#include <ATen/ops/max.h>
+#include <ATen/ops/max_pool1d.h>
+#include <ATen/ops/max_pool1d_with_indices.h>
+#include <ATen/ops/max_pool2d.h>
+#include <ATen/ops/max_pool2d_backward.h>
+#include <ATen/ops/max_pool2d_with_indices.h>
+#include <ATen/ops/max_pool2d_with_indices_backward.h>
+#include <ATen/ops/max_pool3d.h>
+#include <ATen/ops/max_pool3d_with_indices.h>
+#include <ATen/ops/max_pool3d_with_indices_backward.h>
+#include <ATen/ops/max_unpool2d.h>
+#include <ATen/ops/max_unpool3d.h>
+#include <ATen/ops/maximum.h>
+#include <ATen/ops/mean.h>
+#include <ATen/ops/median.h>
+#include <ATen/ops/meshgrid.h>
+#include <ATen/ops/min.h>
+#include <ATen/ops/minimum.h>
+#include <ATen/ops/miopen_batch_norm.h>
+#include <ATen/ops/miopen_batch_norm_backward.h>
+#include <ATen/ops/miopen_convolution.h>
+#include <ATen/ops/miopen_convolution_add_relu.h>
+#include <ATen/ops/miopen_convolution_relu.h>
+#include <ATen/ops/miopen_convolution_transpose.h>
+#include <ATen/ops/miopen_depthwise_convolution.h>
+#include <ATen/ops/miopen_rnn.h>
+#include <ATen/ops/miopen_rnn_backward.h>
+#include <ATen/ops/mish.h>
+#include <ATen/ops/mish_backward.h>
+#include <ATen/ops/mkldnn_adaptive_avg_pool2d.h>
+#include <ATen/ops/mkldnn_adaptive_avg_pool2d_backward.h>
+#include <ATen/ops/mkldnn_convolution.h>
+#include <ATen/ops/mkldnn_linear.h>
+#include <ATen/ops/mkldnn_linear_backward.h>
+#include <ATen/ops/mkldnn_linear_backward_input.h>
+#include <ATen/ops/mkldnn_linear_backward_weights.h>
+#include <ATen/ops/mkldnn_max_pool2d.h>
+#include <ATen/ops/mkldnn_max_pool2d_backward.h>
+#include <ATen/ops/mkldnn_max_pool3d.h>
+#include <ATen/ops/mkldnn_max_pool3d_backward.h>
+#include <ATen/ops/mkldnn_reorder_conv2d_weight.h>
+#include <ATen/ops/mkldnn_reorder_conv3d_weight.h>
+#include <ATen/ops/mkldnn_rnn_layer.h>
+#include <ATen/ops/mkldnn_rnn_layer_backward.h>
+#include <ATen/ops/mm.h>
+#include <ATen/ops/mode.h>
+#include <ATen/ops/moveaxis.h>
+#include <ATen/ops/movedim.h>
+#include <ATen/ops/mps_convolution_backward.h>
+#include <ATen/ops/mps_convolution_transpose_backward.h>
+#include <ATen/ops/mse_loss.h>
+#include <ATen/ops/mse_loss_backward.h>
+#include <ATen/ops/msort.h>
+#include <ATen/ops/mul.h>
+#include <ATen/ops/multi_margin_loss.h>
+#include <ATen/ops/multi_margin_loss_backward.h>
+#include <ATen/ops/multilabel_margin_loss.h>
+#include <ATen/ops/multilabel_margin_loss_backward.h>
+#include <ATen/ops/multilabel_margin_loss_forward.h>
+#include <ATen/ops/multinomial.h>
+#include <ATen/ops/multiply.h>
+#include <ATen/ops/mv.h>
+#include <ATen/ops/mvlgamma.h>
+#include <ATen/ops/nan_to_num.h>
+#include <ATen/ops/nanmean.h>
+#include <ATen/ops/nanmedian.h>
+#include <ATen/ops/nanquantile.h>
+#include <ATen/ops/nansum.h>
+#include <ATen/ops/narrow.h>
+#include <ATen/ops/narrow_copy.h>
+#include <ATen/ops/native_batch_norm.h>
+#include <ATen/ops/native_batch_norm_backward.h>
+#include <ATen/ops/native_channel_shuffle.h>
+#include <ATen/ops/native_dropout.h>
+#include <ATen/ops/native_dropout_backward.h>
+#include <ATen/ops/native_group_norm.h>
+#include <ATen/ops/native_group_norm_backward.h>
+#include <ATen/ops/native_layer_norm.h>
+#include <ATen/ops/native_layer_norm_backward.h>
+#include <ATen/ops/native_norm.h>
+#include <ATen/ops/ne.h>
+#include <ATen/ops/neg.h>
+#include <ATen/ops/negative.h>
+#include <ATen/ops/nested_to_padded_tensor.h>
+#include <ATen/ops/new_empty.h>
+#include <ATen/ops/new_empty_strided.h>
+#include <ATen/ops/new_full.h>
+#include <ATen/ops/new_ones.h>
+#include <ATen/ops/new_zeros.h>
+#include <ATen/ops/nextafter.h>
+#include <ATen/ops/nll_loss.h>
+#include <ATen/ops/nll_loss2d.h>
+#include <ATen/ops/nll_loss2d_backward.h>
+#include <ATen/ops/nll_loss2d_forward.h>
+#include <ATen/ops/nll_loss_backward.h>
+#include <ATen/ops/nll_loss_forward.h>
+#include <ATen/ops/nll_loss_nd.h>
+#include <ATen/ops/nonzero.h>
+#include <ATen/ops/nonzero_numpy.h>
+#include <ATen/ops/nonzero_static.h>
+#include <ATen/ops/norm.h>
+#include <ATen/ops/norm_except_dim.h>
+#include <ATen/ops/normal.h>
+#include <ATen/ops/not_equal.h>
+#include <ATen/ops/nuclear_norm.h>
+#include <ATen/ops/numpy_T.h>
+#include <ATen/ops/one_hot.h>
+#include <ATen/ops/ones.h>
+#include <ATen/ops/ones_like.h>
+#include <ATen/ops/or.h>
+#include <ATen/ops/orgqr.h>
+#include <ATen/ops/ormqr.h>
+#include <ATen/ops/outer.h>
+#include <ATen/ops/output_nr.h>
+#include <ATen/ops/pad.h>
+#include <ATen/ops/pad_sequence.h>
+#include <ATen/ops/pairwise_distance.h>
+#include <ATen/ops/pdist.h>
+#include <ATen/ops/permute.h>
+#include <ATen/ops/permute_copy.h>
+#include <ATen/ops/pin_memory.h>
+#include <ATen/ops/pinverse.h>
+#include <ATen/ops/pixel_shuffle.h>
+#include <ATen/ops/pixel_unshuffle.h>
+#include <ATen/ops/poisson.h>
+#include <ATen/ops/poisson_nll_loss.h>
+#include <ATen/ops/polar.h>
+#include <ATen/ops/polygamma.h>
+#include <ATen/ops/positive.h>
+#include <ATen/ops/pow.h>
+#include <ATen/ops/prelu.h>
+#include <ATen/ops/prod.h>
+#include <ATen/ops/promote_types.h>
+#include <ATen/ops/put.h>
+#include <ATen/ops/q_per_channel_axis.h>
+#include <ATen/ops/q_per_channel_scales.h>
+#include <ATen/ops/q_per_channel_zero_points.h>
+#include <ATen/ops/q_scale.h>
+#include <ATen/ops/q_zero_point.h>
+#include <ATen/ops/qr.h>
+#include <ATen/ops/qscheme.h>
+#include <ATen/ops/quantile.h>
+#include <ATen/ops/quantize_per_channel.h>
+#include <ATen/ops/quantize_per_tensor.h>
+#include <ATen/ops/quantize_per_tensor_dynamic.h>
+#include <ATen/ops/quantized_batch_norm.h>
+#include <ATen/ops/quantized_gru_cell.h>
+#include <ATen/ops/quantized_lstm_cell.h>
+#include <ATen/ops/quantized_max_pool1d.h>
+#include <ATen/ops/quantized_max_pool2d.h>
+#include <ATen/ops/quantized_max_pool3d.h>
+#include <ATen/ops/quantized_rnn_relu_cell.h>
+#include <ATen/ops/quantized_rnn_tanh_cell.h>
+#include <ATen/ops/rad2deg.h>
+#include <ATen/ops/rand.h>
+#include <ATen/ops/rand_like.h>
+#include <ATen/ops/randint.h>
+#include <ATen/ops/randint_like.h>
+#include <ATen/ops/randn.h>
+#include <ATen/ops/randn_like.h>
+#include <ATen/ops/random.h>
+#include <ATen/ops/randperm.h>
+#include <ATen/ops/range.h>
+#include <ATen/ops/ravel.h>
+#include <ATen/ops/real.h>
+#include <ATen/ops/reciprocal.h>
+#include <ATen/ops/record_stream.h>
+#include <ATen/ops/refine_names.h>
+#include <ATen/ops/reflection_pad1d.h>
+#include <ATen/ops/reflection_pad1d_backward.h>
+#include <ATen/ops/reflection_pad2d.h>
+#include <ATen/ops/reflection_pad2d_backward.h>
+#include <ATen/ops/reflection_pad3d.h>
+#include <ATen/ops/reflection_pad3d_backward.h>
+#include <ATen/ops/relu.h>
+#include <ATen/ops/relu6.h>
+#include <ATen/ops/remainder.h>
+#include <ATen/ops/rename.h>
+#include <ATen/ops/renorm.h>
+#include <ATen/ops/repeat.h>
+#include <ATen/ops/repeat_interleave.h>
+#include <ATen/ops/replication_pad1d.h>
+#include <ATen/ops/replication_pad1d_backward.h>
+#include <ATen/ops/replication_pad2d.h>
+#include <ATen/ops/replication_pad2d_backward.h>
+#include <ATen/ops/replication_pad3d.h>
+#include <ATen/ops/replication_pad3d_backward.h>
+#include <ATen/ops/requires_grad.h>
+#include <ATen/ops/reshape.h>
+#include <ATen/ops/reshape_as.h>
+#include <ATen/ops/resize.h>
+#include <ATen/ops/resize_as.h>
+#include <ATen/ops/resize_as_sparse.h>
+#include <ATen/ops/resolve_conj.h>
+#include <ATen/ops/resolve_neg.h>
+#include <ATen/ops/result_type.h>
+#include <ATen/ops/retain_grad.h>
+#include <ATen/ops/retains_grad.h>
+#include <ATen/ops/rms_norm.h>
+#include <ATen/ops/rnn_relu.h>
+#include <ATen/ops/rnn_relu_cell.h>
+#include <ATen/ops/rnn_tanh.h>
+#include <ATen/ops/rnn_tanh_cell.h>
+#include <ATen/ops/roll.h>
+#include <ATen/ops/rot90.h>
+#include <ATen/ops/round.h>
+#include <ATen/ops/row_indices.h>
+#include <ATen/ops/row_indices_copy.h>
+#include <ATen/ops/row_stack.h>
+#include <ATen/ops/rrelu.h>
+#include <ATen/ops/rrelu_with_noise.h>
+#include <ATen/ops/rrelu_with_noise_backward.h>
+#include <ATen/ops/rshift.h>
+#include <ATen/ops/rsqrt.h>
+#include <ATen/ops/rsub.h>
+#include <ATen/ops/scalar_tensor.h>
+#include <ATen/ops/scaled_dot_product_attention.h>
+#include <ATen/ops/scatter.h>
+#include <ATen/ops/scatter_add.h>
+#include <ATen/ops/scatter_reduce.h>
+#include <ATen/ops/searchsorted.h>
+#include <ATen/ops/segment_reduce.h>
+#include <ATen/ops/select.h>
+#include <ATen/ops/select_backward.h>
+#include <ATen/ops/select_copy.h>
+#include <ATen/ops/select_scatter.h>
+#include <ATen/ops/selu.h>
+#include <ATen/ops/set.h>
+#include <ATen/ops/set_data.h>
+#include <ATen/ops/sgn.h>
+#include <ATen/ops/sigmoid.h>
+#include <ATen/ops/sigmoid_backward.h>
+#include <ATen/ops/sign.h>
+#include <ATen/ops/signbit.h>
+#include <ATen/ops/silu.h>
+#include <ATen/ops/silu_backward.h>
+#include <ATen/ops/sin.h>
+#include <ATen/ops/sinc.h>
+#include <ATen/ops/sinh.h>
+#include <ATen/ops/size.h>
+#include <ATen/ops/slice.h>
+#include <ATen/ops/slice_backward.h>
+#include <ATen/ops/slice_copy.h>
+#include <ATen/ops/slice_inverse.h>
+#include <ATen/ops/slice_scatter.h>
+#include <ATen/ops/slogdet.h>
+#include <ATen/ops/slow_conv3d.h>
+#include <ATen/ops/slow_conv3d_forward.h>
+#include <ATen/ops/slow_conv_dilated2d.h>
+#include <ATen/ops/slow_conv_dilated3d.h>
+#include <ATen/ops/slow_conv_transpose2d.h>
+#include <ATen/ops/slow_conv_transpose3d.h>
+#include <ATen/ops/smm.h>
+#include <ATen/ops/smooth_l1_loss.h>
+#include <ATen/ops/smooth_l1_loss_backward.h>
+#include <ATen/ops/soft_margin_loss.h>
+#include <ATen/ops/soft_margin_loss_backward.h>
+#include <ATen/ops/softmax.h>
+#include <ATen/ops/softplus.h>
+#include <ATen/ops/softplus_backward.h>
+#include <ATen/ops/softshrink.h>
+#include <ATen/ops/softshrink_backward.h>
+#include <ATen/ops/sort.h>
+#include <ATen/ops/sparse_bsc_tensor.h>
+#include <ATen/ops/sparse_bsr_tensor.h>
+#include <ATen/ops/sparse_compressed_tensor.h>
+#include <ATen/ops/sparse_coo_tensor.h>
+#include <ATen/ops/sparse_csc_tensor.h>
+#include <ATen/ops/sparse_csr_tensor.h>
+#include <ATen/ops/sparse_dim.h>
+#include <ATen/ops/sparse_mask.h>
+#include <ATen/ops/sparse_resize.h>
+#include <ATen/ops/sparse_resize_and_clear.h>
+#include <ATen/ops/sparse_sampled_addmm.h>
+#include <ATen/ops/special_airy_ai.h>
+#include <ATen/ops/special_bessel_j0.h>
+#include <ATen/ops/special_bessel_j1.h>
+#include <ATen/ops/special_bessel_y0.h>
+#include <ATen/ops/special_bessel_y1.h>
+#include <ATen/ops/special_chebyshev_polynomial_t.h>
+#include <ATen/ops/special_chebyshev_polynomial_u.h>
+#include <ATen/ops/special_chebyshev_polynomial_v.h>
+#include <ATen/ops/special_chebyshev_polynomial_w.h>
+#include <ATen/ops/special_digamma.h>
+#include <ATen/ops/special_entr.h>
+#include <ATen/ops/special_erf.h>
+#include <ATen/ops/special_erfc.h>
+#include <ATen/ops/special_erfcx.h>
+#include <ATen/ops/special_erfinv.h>
+#include <ATen/ops/special_exp2.h>
+#include <ATen/ops/special_expit.h>
+#include <ATen/ops/special_expm1.h>
+#include <ATen/ops/special_gammainc.h>
+#include <ATen/ops/special_gammaincc.h>
+#include <ATen/ops/special_gammaln.h>
+#include <ATen/ops/special_hermite_polynomial_h.h>
+#include <ATen/ops/special_hermite_polynomial_he.h>
+#include <ATen/ops/special_i0.h>
+#include <ATen/ops/special_i0e.h>
+#include <ATen/ops/special_i1.h>
+#include <ATen/ops/special_i1e.h>
+#include <ATen/ops/special_laguerre_polynomial_l.h>
+#include <ATen/ops/special_legendre_polynomial_p.h>
+#include <ATen/ops/special_log1p.h>
+#include <ATen/ops/special_log_ndtr.h>
+#include <ATen/ops/special_log_softmax.h>
+#include <ATen/ops/special_logit.h>
+#include <ATen/ops/special_logsumexp.h>
+#include <ATen/ops/special_modified_bessel_i0.h>
+#include <ATen/ops/special_modified_bessel_i1.h>
+#include <ATen/ops/special_modified_bessel_k0.h>
+#include <ATen/ops/special_modified_bessel_k1.h>
+#include <ATen/ops/special_multigammaln.h>
+#include <ATen/ops/special_ndtr.h>
+#include <ATen/ops/special_ndtri.h>
+#include <ATen/ops/special_polygamma.h>
+#include <ATen/ops/special_psi.h>
+#include <ATen/ops/special_round.h>
+#include <ATen/ops/special_scaled_modified_bessel_k0.h>
+#include <ATen/ops/special_scaled_modified_bessel_k1.h>
+#include <ATen/ops/special_shifted_chebyshev_polynomial_t.h>
+#include <ATen/ops/special_shifted_chebyshev_polynomial_u.h>
+#include <ATen/ops/special_shifted_chebyshev_polynomial_v.h>
+#include <ATen/ops/special_shifted_chebyshev_polynomial_w.h>
+#include <ATen/ops/special_sinc.h>
+#include <ATen/ops/special_softmax.h>
+#include <ATen/ops/special_spherical_bessel_j0.h>
+#include <ATen/ops/special_xlog1py.h>
+#include <ATen/ops/special_xlogy.h>
+#include <ATen/ops/special_zeta.h>
+#include <ATen/ops/split.h>
+#include <ATen/ops/split_copy.h>
+#include <ATen/ops/split_with_sizes.h>
+#include <ATen/ops/split_with_sizes_copy.h>
+#include <ATen/ops/sqrt.h>
+#include <ATen/ops/square.h>
+#include <ATen/ops/squeeze.h>
+#include <ATen/ops/squeeze_copy.h>
+#include <ATen/ops/sspaddmm.h>
+#include <ATen/ops/stack.h>
+#include <ATen/ops/std.h>
+#include <ATen/ops/std_mean.h>
+#include <ATen/ops/stft.h>
+#include <ATen/ops/stride.h>
+#include <ATen/ops/sub.h>
+#include <ATen/ops/subtract.h>
+#include <ATen/ops/sum.h>
+#include <ATen/ops/sum_to_size.h>
+#include <ATen/ops/svd.h>
+#include <ATen/ops/swapaxes.h>
+#include <ATen/ops/swapdims.h>
+#include <ATen/ops/sym_constrain_range.h>
+#include <ATen/ops/sym_constrain_range_for_size.h>
+#include <ATen/ops/sym_numel.h>
+#include <ATen/ops/sym_size.h>
+#include <ATen/ops/sym_storage_offset.h>
+#include <ATen/ops/sym_stride.h>
+#include <ATen/ops/t.h>
+#include <ATen/ops/t_copy.h>
+#include <ATen/ops/take.h>
+#include <ATen/ops/take_along_dim.h>
+#include <ATen/ops/tan.h>
+#include <ATen/ops/tanh.h>
+#include <ATen/ops/tanh_backward.h>
+#include <ATen/ops/tensor_split.h>
+#include <ATen/ops/tensordot.h>
+#include <ATen/ops/thnn_conv2d.h>
+#include <ATen/ops/threshold.h>
+#include <ATen/ops/threshold_backward.h>
+#include <ATen/ops/tile.h>
+#include <ATen/ops/to.h>
+#include <ATen/ops/to_dense.h>
+#include <ATen/ops/to_dense_backward.h>
+#include <ATen/ops/to_mkldnn.h>
+#include <ATen/ops/to_mkldnn_backward.h>
+#include <ATen/ops/to_padded_tensor.h>
+#include <ATen/ops/to_sparse.h>
+#include <ATen/ops/to_sparse_bsc.h>
+#include <ATen/ops/to_sparse_bsr.h>
+#include <ATen/ops/to_sparse_csc.h>
+#include <ATen/ops/to_sparse_csr.h>
+#include <ATen/ops/topk.h>
+#include <ATen/ops/trace.h>
+#include <ATen/ops/trace_backward.h>
+#include <ATen/ops/transpose.h>
+#include <ATen/ops/transpose_copy.h>
+#include <ATen/ops/trapezoid.h>
+#include <ATen/ops/trapz.h>
+#include <ATen/ops/triangular_solve.h>
+#include <ATen/ops/tril.h>
+#include <ATen/ops/tril_indices.h>
+#include <ATen/ops/triplet_margin_loss.h>
+#include <ATen/ops/triu.h>
+#include <ATen/ops/triu_indices.h>
+#include <ATen/ops/true_divide.h>
+#include <ATen/ops/trunc.h>
+#include <ATen/ops/type_as.h>
+#include <ATen/ops/unbind.h>
+#include <ATen/ops/unbind_copy.h>
+#include <ATen/ops/unflatten.h>
+#include <ATen/ops/unflatten_dense_tensors.h>
+#include <ATen/ops/unfold.h>
+#include <ATen/ops/unfold_backward.h>
+#include <ATen/ops/unfold_copy.h>
+#include <ATen/ops/uniform.h>
+#include <ATen/ops/unique_consecutive.h>
+#include <ATen/ops/unique_dim.h>
+#include <ATen/ops/unique_dim_consecutive.h>
+#include <ATen/ops/unsafe_chunk.h>
+#include <ATen/ops/unsafe_split.h>
+#include <ATen/ops/unsafe_split_with_sizes.h>
+#include <ATen/ops/unsqueeze.h>
+#include <ATen/ops/unsqueeze_copy.h>
+#include <ATen/ops/upsample_bicubic2d.h>
+#include <ATen/ops/upsample_bicubic2d_backward.h>
+#include <ATen/ops/upsample_bilinear2d.h>
+#include <ATen/ops/upsample_bilinear2d_backward.h>
+#include <ATen/ops/upsample_linear1d.h>
+#include <ATen/ops/upsample_linear1d_backward.h>
+#include <ATen/ops/upsample_nearest1d.h>
+#include <ATen/ops/upsample_nearest1d_backward.h>
+#include <ATen/ops/upsample_nearest2d.h>
+#include <ATen/ops/upsample_nearest2d_backward.h>
+#include <ATen/ops/upsample_nearest3d.h>
+#include <ATen/ops/upsample_nearest3d_backward.h>
+#include <ATen/ops/upsample_trilinear3d.h>
+#include <ATen/ops/upsample_trilinear3d_backward.h>
+#include <ATen/ops/value_selecting_reduction_backward.h>
+#include <ATen/ops/values.h>
+#include <ATen/ops/values_copy.h>
+#include <ATen/ops/vander.h>
+#include <ATen/ops/var.h>
+#include <ATen/ops/var_mean.h>
+#include <ATen/ops/vdot.h>
+#include <ATen/ops/view.h>
+#include <ATen/ops/view_as.h>
+#include <ATen/ops/view_as_complex.h>
+#include <ATen/ops/view_as_complex_copy.h>
+#include <ATen/ops/view_as_real.h>
+#include <ATen/ops/view_as_real_copy.h>
+#include <ATen/ops/view_copy.h>
+#include <ATen/ops/vsplit.h>
+#include <ATen/ops/vstack.h>
+#include <ATen/ops/where.h>
+#include <ATen/ops/xlogy.h>
+#include <ATen/ops/xor.h>
+#include <ATen/ops/zero.h>
+#include <ATen/ops/zeros.h>
+#include <ATen/ops/zeros_like.h>
+
+namespace at {
+
+
+
+// Special C++ only overloads for std()-like functions (See gh-40287)
+// These are needed because int -> bool conversion takes precedence over int -> IntArrayRef
+// So, for example std(0) would select the std(unbiased=False) overload
+TORCH_API inline Tensor var(const Tensor& self, int dim) {
+  return at::var(self, IntArrayRef{dim});
+}
+TORCH_API inline std::tuple<Tensor, Tensor> var_mean(const Tensor& self, int dim) {
+  return at::var_mean(self, IntArrayRef{dim});
+}
+TORCH_API inline Tensor std(const Tensor& self, int dim) {
+  return at::std(self, IntArrayRef{dim});
+}
+TORCH_API inline std::tuple<Tensor, Tensor> std_mean(const Tensor& self, int dim) {
+  return at::std_mean(self, IntArrayRef{dim});
+}
+
+inline int64_t numel(const Tensor& tensor) {
+  return tensor.numel();
+}
+
+inline int64_t size(const Tensor& tensor, int64_t dim) {
+  return tensor.size(dim);
+}
+
+inline int64_t stride(const Tensor& tensor, int64_t dim) {
+  return tensor.stride(dim);
+}
+
+inline bool is_complex(const Tensor& tensor) {
+  return tensor.is_complex();
+}
+
+inline bool is_floating_point(const Tensor& tensor) {
+  return tensor.is_floating_point();
+}
+
+inline bool is_signed(const Tensor& tensor) {
+  return tensor.is_signed();
+}
+
+inline bool is_inference(const Tensor& tensor) {
+  return tensor.is_inference();
+}
+
+inline bool _is_zerotensor(const Tensor& tensor) {
+  return tensor._is_zerotensor();
+}
+
+inline bool is_conj(const Tensor& tensor) {
+  return tensor.is_conj();
+}
+
+inline Tensor conj(const Tensor& tensor) {
+  return tensor.conj();
+}
+
+inline bool is_neg(const Tensor& tensor) {
+  return tensor.is_neg();
+}
+
+}

.venv/Lib/site-packages/torch/include/ATen/Generator.h

@@ -0,0 +1,2 @@
+#pragma once
+#include <ATen/core/Generator.h>

.venv/Lib/site-packages/torch/include/ATen/InferSize.h

@@ -0,0 +1,88 @@
+#pragma once
+
+#include <ATen/DimVector.h>
+#include <c10/core/ScalarType.h>
+#include <c10/core/SymIntArrayRef.h>
+#include <c10/util/DimVector.h>
+#include <optional>
+#include <sstream>
+#include <vector>
+
+namespace at {
+
+// Infers the size of a dim with size -1, if it exists. Also checks that new
+// shape is compatible with the number of elements.
+//
+// templated to handle std::vector<int64_t> and DimVector use cases, see
+// below
+//
+template <typename InputArrayRef, typename NumelType, typename ResultVec>
+inline void infer_size_impl(
+    InputArrayRef shape,
+    NumelType numel,
+    ResultVec& res) {
+  NumelType newsize = 1;
+  // N.B. this is an index, not a sym dim!
+  std::optional<int64_t> infer_dim;
+  for (int64_t dim = 0, ndim = shape.size(); dim != ndim; dim++) {
+    if (shape[dim] == -1) {
+      if (infer_dim) {
+        throw std::runtime_error("only one dimension can be inferred");
+      }
+      infer_dim = dim;
+    } else if (shape[dim] >= 0) {
+      newsize *= shape[dim];
+    } else {
+      AT_ERROR("invalid shape dimension ", shape[dim]);
+    }
+  }
+
+  if (TORCH_GUARD_SIZE_OBLIVIOUS(sym_eq(numel, newsize)) ||
+      (infer_dim && newsize > 0 && numel % newsize == 0)) {
+    if (infer_dim) {
+      // We have a degree of freedom here to select the dimension size; follow
+      // NumPy semantics and just bail.  However, a nice error message is needed
+      // because users often use `view` as a way to flatten & unflatten
+      // dimensions and will otherwise be confused why
+      //   empty_tensor.view( 0, 0)
+      // works yet
+      //   empty_tensor.view(-1, 0)
+      // doesn't.
+      TORCH_CHECK(
+          newsize != 0,
+          "cannot reshape tensor of 0 elements into shape ",
+          shape,
+          " because the unspecified dimension size -1 can be any "
+          "value and is ambiguous");
+      res[*infer_dim] = numel / newsize;
+    }
+    return;
+  }
+
+  std::ostringstream ss;
+  ss << "shape '" << shape << "' is invalid for input of size " << numel;
+  throw std::runtime_error(ss.str());
+}
+
+inline std::vector<int64_t> infer_size(IntArrayRef shape, int64_t numel) {
+  auto res = shape.vec();
+  infer_size_impl(shape, numel, res);
+  return res;
+}
+
+inline at::DimVector infer_size_dv(IntArrayRef shape, int64_t numel) {
+  auto res = at::DimVector(shape);
+  infer_size_impl(shape, numel, res);
+  return res;
+}
+
+inline at::SymDimVector infer_size_dv(
+    c10::SymIntArrayRef shape,
+    c10::SymInt numel) {
+  auto res = at::SymDimVector(shape);
+  infer_size_impl<c10::SymIntArrayRef, c10::SymInt, at::SymDimVector>(
+      shape, std::move(numel), res);
+  return res;
+}
+
+} // namespace at

.venv/Lib/site-packages/torch/include/ATen/InitialTensorOptions.h

@@ -0,0 +1,15 @@
+#pragma once
+
+#include <c10/core/TensorOptions.h>
+
+namespace at {
+
+// Represents the initial TensorOptions, before the "defaults" are ever changed.
+// This is designed to be used in library code, where the explicit devices,
+// dtypes, etc. are known. NOTE: this is not a stable API.
+inline TensorOptions initialTensorOptions() {
+  return TensorOptions(kCPU).dtype(kFloat).layout(kStrided).requires_grad(
+      false);
+}
+
+} // namespace at

.venv/Lib/site-packages/torch/include/ATen/Layout.h

@@ -0,0 +1,2 @@
+#pragma once
+#include <c10/core/Layout.h>

.venv/Lib/site-packages/torch/include/ATen/LegacyBatchedFallback.h

@@ -0,0 +1,25 @@
+#pragma once
+#include <ATen/ATen.h>
+#include <ATen/core/op_registration/op_registration.h>
+#include <torch/library.h>
+
+namespace at {
+
+// If an operator doesn't have a batching rule implemented then we fallback
+// to this implementation. The fallback only works on out-of-place operators
+// that return only tensors with new memory. (e.g., no in-place operators, no
+// view operations).
+//
+// The fallback effectively takes all of the BatchedTensors in `stack`, slices
+// them, and runs `op` on all of the corresponding slices to produce slices
+// of the outputs. The output slices then get `torch.stack`ed to create the
+// final returns.
+//
+// The performance of the fallback is not very good because it introduces an
+// extra copy from stacking the sliced outputs. Because of this, we prefer to
+// write batching rules for operators whenever possible.
+void batchedTensorForLoopFallback(
+    const c10::OperatorHandle& op,
+    torch::jit::Stack* stack);
+
+} // namespace at

.venv/Lib/site-packages/torch/include/ATen/LegacyBatchedTensorImpl.h

@@ -0,0 +1,160 @@
+#pragma once
+
+#include <bitset>
+
+#include <ATen/ArrayRef.h>
+#include <ATen/SmallVector.h>
+#include <ATen/Tensor.h>
+
+namespace at {
+
+// We assume this in a few other places in the codebase,
+// but there isn't a centralized definition.
+constexpr int64_t kVmapMaxTensorDims = 64;
+
+// The valid vmap levels range from [0, 64). This effectively means that we
+// support a maximum of 64 nested vmaps.
+constexpr int64_t kVmapNumLevels = 64;
+
+// Store this number of elements of BatchDims on the stack. Most people will
+// probably use <= 5 nested vmaps, but adjust this number as necessary.
+constexpr int64_t kBatchDimsStackSize = 5;
+
+// a BatchDim represents a "private" dimension on a Tensor created inside of
+// vmap. It is a (level, dim) tuple, with the `dim` indicating which dimension
+// is being vmap'ed over and the `level` being an identifier for which vmap
+// said dimension was created inside. The `dim` corresponds to a "physical
+// dim" - it is a dimension index on the underlying physical tensor that is
+// being vmapped over.
+struct BatchDim {
+  BatchDim(int64_t level, int64_t dim) : dim_(dim), level_(level) {}
+  int64_t dim() const {
+    return dim_;
+  }
+  int64_t level() const {
+    return level_;
+  }
+
+ private:
+  int64_t dim_;
+  int64_t level_;
+};
+
+using BatchDims = SmallVector<BatchDim, kBatchDimsStackSize>;
+using BatchDimsRef = ArrayRef<BatchDim>;
+
+// A BatchedTensorImpl holds an underlying Tensor and a list of BatchDim
+// NB: We use the term "BatchedTensor" to mean a Tensor that is backed with a
+// BatchedTensorImpl.
+//
+// The batch dimensions are treated as being "private"; they are not
+// user-visible. For example, in the following Tensor,
+//    bt = BatchedTensorImpl(ones(2, 3, 5, 7), [(lvl=1, dim=0), (lvl=2, dim=1)])
+// dimensions 0 and 1 are batch dimensions.
+//
+// bt.sizes() returns (5, 7); bt.sum(0) performs a reduction over the (public)
+// dim 0, which is equivalent to dim 3 in the underlying ones(2, 3, 5, 7)
+// tensor.
+struct TORCH_API BatchedTensorImpl : public c10::TensorImpl {
+  explicit BatchedTensorImpl(Tensor value, BatchDims bdims);
+
+  // Returns a reference to BatchDims that represent which dimensions of this
+  // tensor are private.
+  BatchDimsRef bdims() const {
+    return bdims_;
+  }
+
+  // BatchedTensorImpl wraps a Tensor
+  const Tensor& value() const {
+    return value_;
+  };
+
+  // Given a public dimension index, return the dimension index in the
+  // underlying value() tensor. For example, if we have
+  //    bt = BatchedTensorImpl(ones(2, 3, 5, 7), [(lvl=1, dim=0), (lvl=2,
+  //    dim=2)])
+  // bt.actualDim(0) -> 1
+  // bt.actualDim(1) -> 3
+  // bt.actualDim(2) -> Error
+  int64_t actualDim(int64_t dim, bool wrap_dim = true) const;
+
+  // We have to override this because we opted into CustomStrides
+  IntArrayRef strides_custom() const override;
+  // Override a bunch of methods inherited from TensorImpl to return error
+  // messages.
+  bool is_contiguous_custom(at::MemoryFormat memory_format) const override;
+  void set_size(int64_t dim, int64_t new_size) override;
+  void set_stride(int64_t dim, int64_t new_stride) override;
+  void set_storage_offset(int64_t storage_offset) override;
+#ifdef DEBUG
+  bool has_storage() const override;
+#endif
+
+ private:
+  // see NOTE: [BatchedTensorImpl levels invariant]
+  void checkInvariants() const;
+  const char* tensorimpl_type_name() const override;
+
+  Tensor value_;
+
+  // Note: [BatchedTensorImpl levels invariant]
+  // There is an invariant that the BatchDims must be stored in increasing
+  // `level` order. That is, for i < j, bdims_[i].level must be less than
+  // bdims_[j].level.
+  BatchDims bdims_;
+};
+
+// NB: We use the term "BatchedTensor" to mean a Tensor that is backed with a
+// BatchedTensorImpl.
+inline bool isBatchedTensor(const Tensor& tensor) {
+  return tensor.unsafeGetTensorImpl()->key_set().has(DispatchKey::Batched);
+}
+
+// It is unsafe to call this on a Tensor that is not backed by a
+// BatchedTensorImpl. Please use `maybeGetBatchedImpl` whenever possible.
+inline BatchedTensorImpl* unsafeGetBatchedImpl(const Tensor& tensor) {
+  return static_cast<BatchedTensorImpl*>(tensor.unsafeGetTensorImpl());
+}
+
+inline BatchedTensorImpl* maybeGetBatchedImpl(const Tensor& tensor) {
+  if (!isBatchedTensor(tensor)) {
+    return nullptr;
+  }
+  return unsafeGetBatchedImpl(tensor);
+}
+
+// Returns a bitset. If bit i is set, then that means dim i is a batchdim.
+inline std::bitset<kVmapMaxTensorDims> createBatchDimBitset(
+    BatchDimsRef bdims) {
+  std::bitset<kVmapMaxTensorDims> is_bdim;
+  for (const auto& bdim : bdims) {
+    is_bdim.set(bdim.dim());
+  }
+  return is_bdim;
+}
+
+// Creates a bitset for all of the levels present in `bdims`
+inline std::bitset<kVmapNumLevels> createVmapLevelsBitset(BatchDimsRef bdims) {
+  std::bitset<kVmapNumLevels> result;
+  for (const auto& bdim : bdims) {
+    result.set(bdim.level());
+  }
+  return result;
+}
+
+inline std::ostream& operator<<(std::ostream& out, const BatchDim& bdim) {
+  out << "(lvl=" << bdim.level() << ", dim=" << bdim.dim() << ")";
+  return out;
+}
+
+// Use this to construct a BatchedTensor from a regular Tensor
+TORCH_API Tensor makeBatched(const Tensor& tensor, BatchDims bdims);
+
+// Adds a batch dim to `tensor`, returning a BatchedTensor
+TORCH_API Tensor addBatchDim(const Tensor& tensor, int64_t level, int64_t dim);
+
+// Checks if an inplace operation on self and other is "vmap compatible".
+// See NOTE: [vmap-incompatible in-place operations] for the definition of this.
+TORCH_API bool inplaceIsVmapCompatible(const Tensor& self, const Tensor& other);
+
+} // namespace at

.venv/Lib/site-packages/torch/include/ATen/LegacyVmapMode.h

@@ -0,0 +1,26 @@
+#pragma once
+
+#include <c10/core/impl/LocalDispatchKeySet.h>
+
+namespace at::impl {
+
+// VmapMode contains a thread local count of how many nested vmaps
+// we are currently inside. That number is known as the `vmap level`.
+// VmapMode is used in the implementation of the Python `torch.vmap` API.
+//
+// NOTE: this is NOT the c++ api for torch.vmap. That doesn't exist yet.
+
+struct TORCH_API VmapMode {
+  // Returns the vmap level, aka the count of how many nested vmaps we're in.
+  static int64_t current_vmap_level();
+
+  // Increment the count of nested vmaps. If this causes the vmap level to be
+  // greater than 0, then it enables DispatchKey::VmapMode on all tensors.
+  static int64_t increment_nesting();
+
+  // Decrements the count of nested vmaps. If this causes the vmap level to be
+  // equal to 0, then it disables DispatchKey::VmapMode on all tensors.
+  static int64_t decrement_nesting();
+};
+
+} // namespace at::impl

.venv/Lib/site-packages/torch/include/ATen/LegacyVmapTransforms.h

@@ -0,0 +1,183 @@
+#pragma once
+
+#include <ATen/LegacyBatchedTensorImpl.h>
+#include <ATen/core/IListRef.h>
+
+namespace at {
+
+// This file contains abstractions used for transforming *logical* vmap
+// arguments into *physical* arguments. (Keep reading for definitions of these
+// terms).
+
+// NOTE: [Logical vs physical args]
+// Consider the following vmap.
+//   vmap(vmap(func, in_dims=(2,)), in_dims=(0,))(torch.ones(2, 3, 4))
+// This would produce a BatchedTensor wrapping a Tensor of size [2, 3, 4],
+// with batch dims 0 and 2:
+//   BatchedTensor(ones(2, 3, 4), bdims=[(lvl=1,dim=0),(lvl=2,dim=2)])
+//
+// We say the *logical* view of the tensor has size [3] -- tensors inside
+// `func` appear to have size [3].
+// However, the *physical* underlying tensor (the one passed to vmap) has size
+// [2, 3, 4].
+//
+// This notion of logical vs physical also extends to non-tensor arguments.
+// Consider the previous tensor; let's assume the user called
+// `torch.sum(tensor, dim=0)` inside of `func`. Then the logical
+// dimension they are reducing over is dim 0 but the physical dim is dim 1
+// (the first non-batch dimension)
+
+// Forward declared; see NOTE: [What is a VmapPhysicalView?]
+struct VmapPhysicalView;
+
+// Most PyTorch operators take 4 or fewer inputs.
+constexpr int64_t kVmapTransformStaticInputSize = 4;
+using VmapPhysicalViewVec =
+    SmallVector<VmapPhysicalView, kVmapTransformStaticInputSize>;
+
+// Pytorch generally advertises good performance for <= 5 dims.
+// (see ATen/core/DimVector.h). We add a few extra dims (~3) for vmap
+// dimensions to get 8. Adjust this number as necessary
+constexpr int64_t kVmapStaticDimVecSize = 8;
+using VmapDimVector = SmallVector<int64_t, kVmapStaticDimVecSize>;
+using VmapSymDimVector = SmallVector<c10::SymInt, kVmapStaticDimVecSize>;
+
+// NOTE: [What is an VmapTransform?]
+// An *VmapTransform* converts logical views of tensors to physical views.
+//
+// Batching rules use VmapTransforms to convert logical arguments to
+// physical arguments, then call one or more at:: operator that handles the
+// physical arguments, and then converts the physical result back to a logical
+// argument.
+
+// VmapTransform for operators that take tensors with multiple batch dims.
+// Given one or more logical views on Tensors, `logicalToPhysical`
+// permutes all of the batch dims to the front of the tensor, aligns
+// and expands the batch dims to match each other (according to their `level`),
+// and returns a VmapPhysicalView on the tensor(s).
+struct TORCH_API MultiBatchVmapTransform {
+  static VmapPhysicalView logicalToPhysical(const Tensor& logical_tensor);
+  static VmapPhysicalViewVec logicalToPhysical(ITensorListRef logical_tensors);
+};
+
+// VmapTransform for operators that broadcast all inputs.
+// Given some logical views on Tensors, `logicalToPhysical`:
+// - permutes all of the batch dims to the front of the tensors
+// - aligns all the batch dims to the collective levels of all of the tensors.
+//   If a tensor does not have a batch dim for a vmap level, then it receives
+//   a size-one dimension for said level.
+// - aligns the non-batch dims to have the same dimensionality, adding extra
+//   size-1 dimensions in between the batch dimensions and the non-batch
+//   dimensions so that the batch dimensions are lined up from the right.
+//
+// For example: given inputs of size (B, 2) and (B, 3, 2) where B is the batch
+// dimension, BroadcastingVmapTransform returns VmapPhysicalViews that wrap
+// tensors of size (B, 1, 2) and (B, 3, 2).
+//
+// Given inputs of size (B, 2) and (2,), BroadcastingVmapTransform returns
+// VmapPhysicalViews wrapping tensors of size (B, 2) and (1, 2). We don't
+// actually *need* to return a tensor of size (1, 2) for the second tensor
+// because the broadcasting operation takes care of that for us, but we do
+// it anyways to keep things simple.
+struct TORCH_API BroadcastingVmapTransform {
+  static VmapPhysicalViewVec logicalToPhysical(TensorList logical_tensors);
+};
+
+// Forward declared, if you're reading this file head to toe, don't worry about
+// it yet.
+struct VmapPhysicalToLogicalMap;
+
+// NOTE: [What is a VmapPhysicalView?]
+// VmapPhysicalView represents a physical view on a Tensor.
+//
+// One can use it to further convert logical dimension indices, logical shapes,
+// and more to their physical variants, or convert a new (physical) tensor into
+// a logical BatchedTensor. (TODO(rzou): some of these are not yet implemented).
+//
+// VmapPhysicalView stores a physical tensor with all of its batch dimensions at
+// the front and some levels that correspond to said batch dimensions.
+//
+// The levels bitset specifies which vmap levels correspond to the batch
+// dimensions at the front of the tensor. In particular, the number of set bits
+// corresponds to the number of batch dimensions on `tensor` and the rightmost
+// bit of `levels` specifies the maximum number of nested vmaps we are in at
+// this point in time.
+// For example, given:
+//   physical_view = VmapPhysicalView(tensor=ones(2, 3, 4, 5, 6), levels={1, 3})
+//
+// Rightmost bit of `levels` is 3 indicating the number of nested vmaps less
+// than or equal to 3.
+//   bitset: 010100
+//              ^
+//              |
+//   levels: 012345
+struct TORCH_API VmapPhysicalView {
+  VmapPhysicalView(Tensor&& tensor, std::bitset<kVmapNumLevels> levels)
+      : levels_(levels), tensor_(std::move(tensor)) {
+    TORCH_INTERNAL_ASSERT(!isBatchedTensor(tensor_));
+  }
+
+  Tensor& tensor() {
+    return tensor_;
+  }
+  const Tensor& tensor() const {
+    return tensor_;
+  }
+
+  // Maps logical dim indices to physical dim indices. Also does dim wrapping.
+  //
+  // For example, given:
+  //   physical_view = VmapPhysicalView(tensor=ones(2, 3, 4, 5), levels={1, 3})
+  //
+  // Then physical_view.getPhysicalDims({0, 1}) returns {2, 3}.
+  // This is because the size of levels tell us that the first two dimensions
+  // of `tensor_` are batch dimensions, so a logical dim of `n` is actually
+  // a physical dim of `n + 2`.
+  VmapDimVector getPhysicalDims(OptionalIntArrayRef logical_dims) const;
+  int64_t getPhysicalDim(int64_t logical_dim) const;
+
+  // Returns a VmapPhysicalToLogicalMap object. This can be used for
+  // mapping a physical tensor to a new logical tensor (BatchedTensor)
+  VmapPhysicalToLogicalMap getPhysicalToLogicalMap() const;
+
+  // Maps a logical shape to a physical shape by pre-pending the batch
+  // sizes to the logical shape.
+  VmapDimVector getPhysicalShape(IntArrayRef logical_shape) const;
+
+  int64_t numBatchDims() const;
+
+ private:
+  int64_t numLogicalDims() const;
+
+  std::bitset<kVmapNumLevels> levels_;
+  Tensor tensor_;
+};
+
+// Convenience struct used for mapping a physical tensor (a non-BatchedTensor)
+// to a logical one (BatchedTensor). It holds some levels that are used to do
+// the mapping and assumes that the batch dimensions in the physical tensor all
+// occur at the front of the tensor.
+struct TORCH_API VmapPhysicalToLogicalMap {
+  VmapPhysicalToLogicalMap(std::bitset<kVmapNumLevels> levels)
+      : levels_(levels) {}
+
+  // Maps a physical tensor to a new logical tensor (BatchedTensor).
+  // Assumes that all of the "batch dimensions" are at the front
+  // of the physical tensor. For example, given:
+  // - x = rank-4 Tensor with size 2, 3, 5, 7
+  // - levels = (2, 4)
+  // Returns:
+  // - BatchedTensor(x, bdims=[(dim=0,lvl=2), (dim=1, lvl=4)])
+  Tensor apply(const Tensor& physical_tensor) const;
+
+  // Given a vector of physical tensors,
+  // 1. maps each tensor to a new logical tensor. Assumes that all of the
+  //    "batch dimensions" are at the front of the physical tensors.
+  // 2. stores the new logical tensors back into the passed-in vector. This is
+  //    to avoid additional dynamic allocations.
+  void applyInplace(std::vector<Tensor>& physical_tensors) const;
+
+  std::bitset<kVmapNumLevels> levels_;
+};
+
+} // namespace at

.venv/Lib/site-packages/torch/include/ATen/LinalgBackend.h

@@ -0,0 +1,31 @@
+#pragma once
+
+#include <c10/util/Exception.h>
+
+#include <ostream>
+#include <string>
+
+namespace at {
+
+enum class LinalgBackend : int8_t { Default, Cusolver, Magma };
+
+inline std::string LinalgBackendToString(at::LinalgBackend backend) {
+  switch (backend) {
+    case LinalgBackend::Default:
+      return "at::LinalgBackend::Default";
+    case LinalgBackend::Cusolver:
+      return "at::LinalgBackend::Cusolver";
+    case LinalgBackend::Magma:
+      return "at::LinalgBackend::Magma";
+    default:
+      TORCH_CHECK(false, "Unknown linalg backend");
+  }
+}
+
+inline std::ostream& operator<<(
+    std::ostream& stream,
+    at::LinalgBackend backend) {
+  return stream << LinalgBackendToString(backend);
+}
+
+} // namespace at

.venv/Lib/site-packages/torch/include/ATen/MapAllocator.h

@@ -0,0 +1,143 @@
+#pragma once
+
+#include <c10/core/Allocator.h>
+#include <c10/util/string_view.h>
+
+namespace at {
+
+enum MappedAllocatorModes {
+  ALLOCATOR_MAPPED_SHARED = 1,
+  ALLOCATOR_MAPPED_SHAREDMEM = 2,
+  ALLOCATOR_MAPPED_EXCLUSIVE = 4,
+  ALLOCATOR_MAPPED_NOCREATE = 8,
+  ALLOCATOR_MAPPED_KEEPFD = 16,
+  ALLOCATOR_MAPPED_FROMFD = 32,
+  ALLOCATOR_MAPPED_UNLINK = 64
+};
+
+// Sentinel value/type to help distinguish the file descriptor constructor from
+// the non-file descriptor constructor
+enum WithFd { WITH_FD };
+
+TORCH_API std::string NewProcessWideShmHandle();
+
+class TORCH_API MapAllocator {
+ public:
+  MapAllocator(c10::string_view filename, int flags, size_t size);
+  MapAllocator(
+      WithFd,
+      c10::string_view filename,
+      int fd,
+      int flags,
+      size_t size);
+  MapAllocator(const MapAllocator&) = delete;
+  MapAllocator& operator=(const MapAllocator&) = delete;
+  MapAllocator(MapAllocator&&) = delete;
+  MapAllocator& operator=(MapAllocator&&) = delete;
+
+  const char* filename() const {
+    return filename_.c_str();
+  }
+  int fd() const {
+#ifdef _WIN32
+    TORCH_CHECK(false, "MapAllocator::fd() is unsupported on Windows");
+#else
+    return fd_;
+#endif
+  }
+  ptrdiff_t size() const {
+    return size_;
+  }
+  // Return a pointer to the actual data for this allocator
+  // (in the case of the refcounted allocator, this is offset
+  // from the base pointer.)
+  virtual void* data() const {
+    return base_ptr_;
+  }
+
+  int flags() const {
+    return flags_;
+  }
+
+  static MapAllocator* fromDataPtr(const at::DataPtr&);
+  static at::DataPtr makeDataPtr(
+      c10::string_view filename,
+      int flags,
+      size_t size,
+      size_t* actual_size_out);
+  static at::DataPtr makeDataPtr(
+      WithFd,
+      const char* filename,
+      int fd,
+      int flags,
+      size_t size,
+      size_t* actual_size_out);
+
+  // Closes the data.  Helps us avoid destructor shenanigans
+  virtual void close();
+
+  // This is very dangerous.  You have to redefine this destructor for each
+  // subclass
+  virtual ~MapAllocator();
+
+ protected:
+  bool closed_ = false;
+  std::string filename_;
+  int flags_ = 0;
+  ptrdiff_t size_; /* mapped size */
+#ifdef _WIN32
+  void* handle_;
+  void* event_;
+  std::string eventname_;
+#else
+  int fd_ = -1;
+#endif
+  void* base_ptr_ = nullptr;
+};
+
+// Base-from-member idiom
+struct TORCH_API RefcountedMapAllocatorArgCheck {
+  RefcountedMapAllocatorArgCheck(int flags);
+};
+
+class TORCH_API RefcountedMapAllocator : private RefcountedMapAllocatorArgCheck,
+                                         public MapAllocator {
+ public:
+  RefcountedMapAllocator(const char* filename, int flags, size_t size);
+  RefcountedMapAllocator(
+      WithFd,
+      const char* filename,
+      int fd,
+      int flags,
+      size_t size);
+
+  static RefcountedMapAllocator* fromDataPtr(const at::DataPtr&);
+  static at::DataPtr makeDataPtr(
+      const char* filename,
+      int flags,
+      size_t size,
+      size_t* actual_size_out);
+  static at::DataPtr makeDataPtr(
+      WithFd,
+      const char* filename,
+      int fd,
+      int flags,
+      size_t size,
+      size_t* actual_size_out);
+
+  void* data() const override;
+
+  void incref();
+  int decref();
+  void close() override;
+
+  ~RefcountedMapAllocator() override {
+    RefcountedMapAllocator::close();
+  }
+
+ protected:
+  void checkFlags();
+  void initializeAlloc();
+};
+
+} // namespace at

.venv/Lib/site-packages/torch/include/ATen/MatrixRef.h

@@ -0,0 +1,107 @@
+#pragma once
+#include <ATen/Utils.h>
+#include <c10/util/ArrayRef.h>
+
+namespace at {
+/// MatrixRef - Like an ArrayRef, but with an extra recorded strides so that
+/// we can easily view it as a multidimensional array.
+///
+/// Like ArrayRef, this class does not own the underlying data, it is expected
+/// to be used in situations where the data resides in some other buffer.
+///
+/// This is intended to be trivially copyable, so it should be passed by
+/// value.
+///
+/// For now, 2D only (so the copies are actually cheap, without having
+/// to write a SmallVector class) and contiguous only (so we can
+/// return non-strided ArrayRef on index).
+///
+/// P.S. dimension 0 indexes rows, dimension 1 indexes columns
+template <typename T>
+class MatrixRef {
+ public:
+  typedef size_t size_type;
+
+ private:
+  /// Underlying ArrayRef
+  ArrayRef<T> arr;
+
+  /// Stride of dim 0 (outer dimension)
+  size_type stride0;
+
+  // Stride of dim 1 is assumed to be 1
+
+ public:
+  /// Construct an empty Matrixref.
+  /*implicit*/ MatrixRef() : arr(nullptr), stride0(0) {}
+
+  /// Construct an MatrixRef from an ArrayRef and outer stride.
+  /*implicit*/ MatrixRef(ArrayRef<T> arr, size_type stride0)
+      : arr(arr), stride0(stride0) {
+    TORCH_CHECK(
+        arr.size() % stride0 == 0,
+        "MatrixRef: ArrayRef size ",
+        arr.size(),
+        " not divisible by stride ",
+        stride0)
+  }
+
+  /// @}
+  /// @name Simple Operations
+  /// @{
+
+  /// empty - Check if the matrix is empty.
+  bool empty() const {
+    return arr.empty();
+  }
+
+  const T* data() const {
+    return arr.data();
+  }
+
+  /// size - Get size a dimension
+  size_t size(size_t dim) const {
+    if (dim == 0) {
+      return arr.size() / stride0;
+    } else if (dim == 1) {
+      return stride0;
+    } else {
+      TORCH_CHECK(
+          0, "MatrixRef: out of bounds dimension ", dim, "; expected 0 or 1");
+    }
+  }
+
+  size_t numel() const {
+    return arr.size();
+  }
+
+  /// equals - Check for element-wise equality.
+  bool equals(MatrixRef RHS) const {
+    return stride0 == RHS.stride0 && arr.equals(RHS.arr);
+  }
+
+  /// @}
+  /// @name Operator Overloads
+  /// @{
+  ArrayRef<T> operator[](size_t Index) const {
+    return arr.slice(Index * stride0, stride0);
+  }
+
+  /// Disallow accidental assignment from a temporary.
+  ///
+  /// The declaration here is extra complicated so that "arrayRef = {}"
+  /// continues to select the move assignment operator.
+  template <typename U>
+  std::enable_if_t<std::is_same_v<U, T>, MatrixRef<T>>& operator=(
+      U&& Temporary) = delete;
+
+  /// Disallow accidental assignment from a temporary.
+  ///
+  /// The declaration here is extra complicated so that "arrayRef = {}"
+  /// continues to select the move assignment operator.
+  template <typename U>
+  std::enable_if_t<std::is_same_v<U, T>, MatrixRef<T>>& operator=(
+      std::initializer_list<U>) = delete;
+};
+
+} // end namespace at

.venv/Lib/site-packages/torch/include/ATen/MemoryOverlap.h

@@ -0,0 +1,42 @@
+#pragma once
+
+#include <c10/macros/Export.h>
+
+namespace c10 {
+struct TensorImpl;
+}
+
+namespace at {
+class TensorBase;
+
+// MemOverlap: Whether or not there is memory overlap
+//
+// No: Absolutely no memory overlap
+// Yes: Absolutely yes memory overlap
+// TooHard: There might be memory overlap, but it was too expensive to compute.
+//
+// NB: Please update the python test for these if you renumber them.
+enum class MemOverlap { No, Yes, TooHard };
+
+enum class MemOverlapStatus { Full, Partial, No, TooHard };
+
+TORCH_API MemOverlap has_internal_overlap(const TensorBase& t);
+TORCH_API MemOverlap has_internal_overlap(c10::TensorImpl* t);
+
+TORCH_API void assert_no_internal_overlap(const TensorBase& t);
+TORCH_API void assert_no_internal_overlap(c10::TensorImpl* t);
+
+TORCH_API MemOverlapStatus
+get_overlap_status(const TensorBase& a, const TensorBase& b);
+TORCH_API MemOverlapStatus
+get_overlap_status(const c10::TensorImpl* a, const c10::TensorImpl* b);
+
+TORCH_API void assert_no_partial_overlap(
+    const TensorBase& a,
+    const TensorBase& b);
+void assert_no_partial_overlap(c10::TensorImpl* a, c10::TensorImpl* b);
+
+TORCH_API void assert_no_overlap(const TensorBase& a, const TensorBase& b);
+TORCH_API void assert_no_overlap(c10::TensorImpl* a, c10::TensorImpl* b);
+
+} // namespace at

.venv/Lib/site-packages/torch/include/ATen/MetaFunctions.h

@@ -0,0 +1,29 @@
+#include <ATen/core/TensorBody.h>
+
+// TODO Undo all logic introduced for Note [Avoiding Include Cycles In Static Dispatch]
+// Code introduced to avoid cyclic dependency in static dispatch is no longer
+// needed as static dispatch logic is moved from TensorBody.h, which caused cycles in the first place,
+// to Operators.cpp for supporting multiple backends with multiple kernels.
+//
+// Note [Avoiding Include Cycles In Static Dispatch]
+// In order to avoid #include cycles in the static dispatch build, we've carefully split out
+// the static function definition files into {DispatchKey}Functions.h and {DispatchKey}Functions_inl.h.
+//
+// Without this split, the include cycle looks like TensorBody.h -> CPUFunctions.h -> TensorBody.h.
+// - TensorBody.h #includes CPUFunctions.h in the static dispatch build, because the tensor methods
+//   all need to call into the fastpath C++ API defined in CPUFunctions.h. The methods are also all
+//   directly inlined into TensorBody.h.
+// - CPUFunctions.h #includes TensorBody.h because it contains function declarations for the entire C++ API,
+//   which include functions that have defaultable std::optional<Tensor> arguments.
+//   That requires knowing the full Tensor class definition.
+//
+// We break the cycle by doing the following:
+// - Split out CPUFunction.h into two files: CPUFunctions.h and CPUFunctions_inl.h
+// - CPUFunction.h is a dummy file that just includes the Tensor class and includes CPUFunctions_inl.,
+// - CPUFunctions_inl.h includes everything else
+// - (only in the static dispatch build) TensorBody.h makes sure to finish defining the Tensor class,
+//   and then it includes CPUFunctions_inl.h.
+// - All other files that want the cpu fastpath functions can include CPUFunctions.h directly.
+// - This also means that static dispatch build, CPUFunctions.h only needs to
+//   #include TensorBody.h, and it will automatically bring in CPUFunctions_inl.h.
+#include <ATen/MetaFunctions_inl.h>

.venv/Lib/site-packages/torch/include/ATen/MetaFunctions_inl.h

@@ -0,0 +1,325 @@
+#pragma once
+// @generated by torchgen/gen.py from DispatchKeyFunctions_inl.h
+
+// NB: The implementing C++ file is RegisterDispatchKey.cpp
+
+// The only #includes we need are for custom classes that have defaults in the C++ API
+#include <c10/core/MemoryFormat.h>
+#include <c10/core/Scalar.h>
+#include <ATen/core/Reduction.h>
+
+#if defined(AT_PER_OPERATOR_HEADERS) && defined(TORCH_ASSERT_ONLY_METHOD_OPERATORS)
+#error This change adds a dependency on all pytorch operators, meaning the     \
+  file will need to be re-compiled every time an operator is changed or added. \
+  Consider including a specific operator from                                  \
+  <ATen/ops/{my_operator}_meta_dispatch.h>.                   \
+  See NOTE [TORCH_ASSERT_ONLY_METHOD_OPERATORS].
+#endif
+
+#include <ATen/ops/_add_relu_meta_dispatch.h>
+#include <ATen/ops/_addmm_activation_meta_dispatch.h>
+#include <ATen/ops/_amp_update_scale_meta_dispatch.h>
+#include <ATen/ops/_coalesced_meta_dispatch.h>
+#include <ATen/ops/_convert_indices_from_coo_to_csr_meta_dispatch.h>
+#include <ATen/ops/_convert_indices_from_csr_to_coo_meta_dispatch.h>
+#include <ATen/ops/_ctc_loss_meta_dispatch.h>
+#include <ATen/ops/_efficientzerotensor_meta_dispatch.h>
+#include <ATen/ops/_fill_mem_eff_dropout_mask_meta_dispatch.h>
+#include <ATen/ops/_fused_sdp_choice_meta_dispatch.h>
+#include <ATen/ops/_index_put_impl_meta_dispatch.h>
+#include <ATen/ops/_linalg_det_meta_dispatch.h>
+#include <ATen/ops/_linalg_eigh_meta_dispatch.h>
+#include <ATen/ops/_linalg_slogdet_meta_dispatch.h>
+#include <ATen/ops/_linalg_solve_ex_meta_dispatch.h>
+#include <ATen/ops/_linalg_svd_meta_dispatch.h>
+#include <ATen/ops/_log_softmax_meta_dispatch.h>
+#include <ATen/ops/_log_softmax_backward_data_meta_dispatch.h>
+#include <ATen/ops/_mkldnn_transpose_meta_dispatch.h>
+#include <ATen/ops/_reshape_alias_meta_dispatch.h>
+#include <ATen/ops/_resize_output_meta_dispatch.h>
+#include <ATen/ops/_softmax_meta_dispatch.h>
+#include <ATen/ops/_softmax_backward_data_meta_dispatch.h>
+#include <ATen/ops/_sparse_coo_tensor_with_dims_meta_dispatch.h>
+#include <ATen/ops/_sparse_coo_tensor_with_dims_and_tensors_meta_dispatch.h>
+#include <ATen/ops/_upsample_bicubic2d_aa_meta_dispatch.h>
+#include <ATen/ops/_upsample_bicubic2d_aa_backward_meta_dispatch.h>
+#include <ATen/ops/_upsample_bilinear2d_aa_meta_dispatch.h>
+#include <ATen/ops/_upsample_bilinear2d_aa_backward_meta_dispatch.h>
+#include <ATen/ops/_upsample_nearest_exact1d_meta_dispatch.h>
+#include <ATen/ops/_upsample_nearest_exact1d_backward_meta_dispatch.h>
+#include <ATen/ops/_upsample_nearest_exact2d_meta_dispatch.h>
+#include <ATen/ops/_upsample_nearest_exact2d_backward_meta_dispatch.h>
+#include <ATen/ops/_upsample_nearest_exact3d_meta_dispatch.h>
+#include <ATen/ops/_upsample_nearest_exact3d_backward_meta_dispatch.h>
+#include <ATen/ops/acos_meta_dispatch.h>
+#include <ATen/ops/acosh_meta_dispatch.h>
+#include <ATen/ops/adaptive_max_pool2d_meta_dispatch.h>
+#include <ATen/ops/adaptive_max_pool2d_backward_meta_dispatch.h>
+#include <ATen/ops/adaptive_max_pool3d_meta_dispatch.h>
+#include <ATen/ops/adaptive_max_pool3d_backward_meta_dispatch.h>
+#include <ATen/ops/add_meta_dispatch.h>
+#include <ATen/ops/addbmm_meta_dispatch.h>
+#include <ATen/ops/addcdiv_meta_dispatch.h>
+#include <ATen/ops/addcmul_meta_dispatch.h>
+#include <ATen/ops/addmm_meta_dispatch.h>
+#include <ATen/ops/addmv_meta_dispatch.h>
+#include <ATen/ops/all_meta_dispatch.h>
+#include <ATen/ops/amax_meta_dispatch.h>
+#include <ATen/ops/amin_meta_dispatch.h>
+#include <ATen/ops/aminmax_meta_dispatch.h>
+#include <ATen/ops/any_meta_dispatch.h>
+#include <ATen/ops/arange_meta_dispatch.h>
+#include <ATen/ops/argmax_meta_dispatch.h>
+#include <ATen/ops/argmin_meta_dispatch.h>
+#include <ATen/ops/as_strided_meta_dispatch.h>
+#include <ATen/ops/asin_meta_dispatch.h>
+#include <ATen/ops/asinh_meta_dispatch.h>
+#include <ATen/ops/atan_meta_dispatch.h>
+#include <ATen/ops/atan2_meta_dispatch.h>
+#include <ATen/ops/atanh_meta_dispatch.h>
+#include <ATen/ops/avg_pool2d_meta_dispatch.h>
+#include <ATen/ops/avg_pool2d_backward_meta_dispatch.h>
+#include <ATen/ops/avg_pool3d_meta_dispatch.h>
+#include <ATen/ops/avg_pool3d_backward_meta_dispatch.h>
+#include <ATen/ops/baddbmm_meta_dispatch.h>
+#include <ATen/ops/bernoulli_meta_dispatch.h>
+#include <ATen/ops/bitwise_and_meta_dispatch.h>
+#include <ATen/ops/bitwise_left_shift_meta_dispatch.h>
+#include <ATen/ops/bitwise_not_meta_dispatch.h>
+#include <ATen/ops/bitwise_or_meta_dispatch.h>
+#include <ATen/ops/bitwise_right_shift_meta_dispatch.h>
+#include <ATen/ops/bitwise_xor_meta_dispatch.h>
+#include <ATen/ops/bmm_meta_dispatch.h>
+#include <ATen/ops/cat_meta_dispatch.h>
+#include <ATen/ops/cauchy_meta_dispatch.h>
+#include <ATen/ops/ceil_meta_dispatch.h>
+#include <ATen/ops/clamp_meta_dispatch.h>
+#include <ATen/ops/clamp_max_meta_dispatch.h>
+#include <ATen/ops/clamp_min_meta_dispatch.h>
+#include <ATen/ops/copy_meta_dispatch.h>
+#include <ATen/ops/copy_sparse_to_sparse_meta_dispatch.h>
+#include <ATen/ops/copysign_meta_dispatch.h>
+#include <ATen/ops/cos_meta_dispatch.h>
+#include <ATen/ops/cosh_meta_dispatch.h>
+#include <ATen/ops/cumprod_meta_dispatch.h>
+#include <ATen/ops/cumsum_meta_dispatch.h>
+#include <ATen/ops/digamma_meta_dispatch.h>
+#include <ATen/ops/div_meta_dispatch.h>
+#include <ATen/ops/elu_meta_dispatch.h>
+#include <ATen/ops/elu_backward_meta_dispatch.h>
+#include <ATen/ops/embedding_renorm_meta_dispatch.h>
+#include <ATen/ops/empty_meta_dispatch.h>
+#include <ATen/ops/empty_strided_meta_dispatch.h>
+#include <ATen/ops/eq_meta_dispatch.h>
+#include <ATen/ops/erf_meta_dispatch.h>
+#include <ATen/ops/erfc_meta_dispatch.h>
+#include <ATen/ops/erfinv_meta_dispatch.h>
+#include <ATen/ops/exp_meta_dispatch.h>
+#include <ATen/ops/exp2_meta_dispatch.h>
+#include <ATen/ops/expm1_meta_dispatch.h>
+#include <ATen/ops/exponential_meta_dispatch.h>
+#include <ATen/ops/eye_meta_dispatch.h>
+#include <ATen/ops/fill_meta_dispatch.h>
+#include <ATen/ops/floor_meta_dispatch.h>
+#include <ATen/ops/floor_divide_meta_dispatch.h>
+#include <ATen/ops/fmax_meta_dispatch.h>
+#include <ATen/ops/fmin_meta_dispatch.h>
+#include <ATen/ops/fmod_meta_dispatch.h>
+#include <ATen/ops/frac_meta_dispatch.h>
+#include <ATen/ops/fractional_max_pool2d_meta_dispatch.h>
+#include <ATen/ops/fractional_max_pool2d_backward_meta_dispatch.h>
+#include <ATen/ops/fractional_max_pool3d_meta_dispatch.h>
+#include <ATen/ops/gather_meta_dispatch.h>
+#include <ATen/ops/gcd_meta_dispatch.h>
+#include <ATen/ops/ge_meta_dispatch.h>
+#include <ATen/ops/gelu_meta_dispatch.h>
+#include <ATen/ops/gelu_backward_meta_dispatch.h>
+#include <ATen/ops/geometric_meta_dispatch.h>
+#include <ATen/ops/glu_meta_dispatch.h>
+#include <ATen/ops/gt_meta_dispatch.h>
+#include <ATen/ops/hardshrink_meta_dispatch.h>
+#include <ATen/ops/hardshrink_backward_meta_dispatch.h>
+#include <ATen/ops/hardsigmoid_meta_dispatch.h>
+#include <ATen/ops/hardsigmoid_backward_meta_dispatch.h>
+#include <ATen/ops/hardswish_meta_dispatch.h>
+#include <ATen/ops/hardtanh_meta_dispatch.h>
+#include <ATen/ops/heaviside_meta_dispatch.h>
+#include <ATen/ops/hypot_meta_dispatch.h>
+#include <ATen/ops/i0_meta_dispatch.h>
+#include <ATen/ops/igamma_meta_dispatch.h>
+#include <ATen/ops/igammac_meta_dispatch.h>
+#include <ATen/ops/index_meta_dispatch.h>
+#include <ATen/ops/index_add_meta_dispatch.h>
+#include <ATen/ops/index_copy_meta_dispatch.h>
+#include <ATen/ops/index_fill_meta_dispatch.h>
+#include <ATen/ops/index_reduce_meta_dispatch.h>
+#include <ATen/ops/isin_meta_dispatch.h>
+#include <ATen/ops/isneginf_meta_dispatch.h>
+#include <ATen/ops/isposinf_meta_dispatch.h>
+#include <ATen/ops/lcm_meta_dispatch.h>
+#include <ATen/ops/le_meta_dispatch.h>
+#include <ATen/ops/leaky_relu_meta_dispatch.h>
+#include <ATen/ops/leaky_relu_backward_meta_dispatch.h>
+#include <ATen/ops/lerp_meta_dispatch.h>
+#include <ATen/ops/lgamma_meta_dispatch.h>
+#include <ATen/ops/linalg_cholesky_ex_meta_dispatch.h>
+#include <ATen/ops/linalg_cross_meta_dispatch.h>
+#include <ATen/ops/linalg_inv_ex_meta_dispatch.h>
+#include <ATen/ops/linalg_ldl_factor_ex_meta_dispatch.h>
+#include <ATen/ops/linalg_ldl_solve_meta_dispatch.h>
+#include <ATen/ops/linalg_lu_meta_dispatch.h>
+#include <ATen/ops/linalg_lu_factor_ex_meta_dispatch.h>
+#include <ATen/ops/linalg_lu_solve_meta_dispatch.h>
+#include <ATen/ops/linalg_qr_meta_dispatch.h>
+#include <ATen/ops/linalg_vector_norm_meta_dispatch.h>
+#include <ATen/ops/linspace_meta_dispatch.h>
+#include <ATen/ops/log_meta_dispatch.h>
+#include <ATen/ops/log10_meta_dispatch.h>
+#include <ATen/ops/log1p_meta_dispatch.h>
+#include <ATen/ops/log2_meta_dispatch.h>
+#include <ATen/ops/log_normal_meta_dispatch.h>
+#include <ATen/ops/logaddexp_meta_dispatch.h>
+#include <ATen/ops/logaddexp2_meta_dispatch.h>
+#include <ATen/ops/logit_meta_dispatch.h>
+#include <ATen/ops/logit_backward_meta_dispatch.h>
+#include <ATen/ops/logspace_meta_dispatch.h>
+#include <ATen/ops/lshift_meta_dispatch.h>
+#include <ATen/ops/lt_meta_dispatch.h>
+#include <ATen/ops/lu_unpack_meta_dispatch.h>
+#include <ATen/ops/masked_fill_meta_dispatch.h>
+#include <ATen/ops/masked_scatter_meta_dispatch.h>
+#include <ATen/ops/max_meta_dispatch.h>
+#include <ATen/ops/max_pool2d_with_indices_meta_dispatch.h>
+#include <ATen/ops/max_pool2d_with_indices_backward_meta_dispatch.h>
+#include <ATen/ops/maximum_meta_dispatch.h>
+#include <ATen/ops/mean_meta_dispatch.h>
+#include <ATen/ops/min_meta_dispatch.h>
+#include <ATen/ops/minimum_meta_dispatch.h>
+#include <ATen/ops/mish_meta_dispatch.h>
+#include <ATen/ops/mm_meta_dispatch.h>
+#include <ATen/ops/mse_loss_meta_dispatch.h>
+#include <ATen/ops/mul_meta_dispatch.h>
+#include <ATen/ops/ne_meta_dispatch.h>
+#include <ATen/ops/neg_meta_dispatch.h>
+#include <ATen/ops/nextafter_meta_dispatch.h>
+#include <ATen/ops/nll_loss_backward_meta_dispatch.h>
+#include <ATen/ops/nll_loss_forward_meta_dispatch.h>
+#include <ATen/ops/norm_meta_dispatch.h>
+#include <ATen/ops/normal_meta_dispatch.h>
+#include <ATen/ops/polygamma_meta_dispatch.h>
+#include <ATen/ops/pow_meta_dispatch.h>
+#include <ATen/ops/prod_meta_dispatch.h>
+#include <ATen/ops/put_meta_dispatch.h>
+#include <ATen/ops/random_meta_dispatch.h>
+#include <ATen/ops/range_meta_dispatch.h>
+#include <ATen/ops/reciprocal_meta_dispatch.h>
+#include <ATen/ops/reflection_pad1d_meta_dispatch.h>
+#include <ATen/ops/reflection_pad1d_backward_meta_dispatch.h>
+#include <ATen/ops/reflection_pad3d_meta_dispatch.h>
+#include <ATen/ops/reflection_pad3d_backward_meta_dispatch.h>
+#include <ATen/ops/relu_meta_dispatch.h>
+#include <ATen/ops/remainder_meta_dispatch.h>
+#include <ATen/ops/renorm_meta_dispatch.h>
+#include <ATen/ops/replication_pad1d_meta_dispatch.h>
+#include <ATen/ops/replication_pad1d_backward_meta_dispatch.h>
+#include <ATen/ops/replication_pad2d_meta_dispatch.h>
+#include <ATen/ops/replication_pad3d_meta_dispatch.h>
+#include <ATen/ops/resize_meta_dispatch.h>
+#include <ATen/ops/resize_as_sparse_meta_dispatch.h>
+#include <ATen/ops/round_meta_dispatch.h>
+#include <ATen/ops/rrelu_with_noise_meta_dispatch.h>
+#include <ATen/ops/rshift_meta_dispatch.h>
+#include <ATen/ops/rsqrt_meta_dispatch.h>
+#include <ATen/ops/scatter_meta_dispatch.h>
+#include <ATen/ops/scatter_add_meta_dispatch.h>
+#include <ATen/ops/scatter_reduce_meta_dispatch.h>
+#include <ATen/ops/set_meta_dispatch.h>
+#include <ATen/ops/sgn_meta_dispatch.h>
+#include <ATen/ops/sigmoid_meta_dispatch.h>
+#include <ATen/ops/sigmoid_backward_meta_dispatch.h>
+#include <ATen/ops/sign_meta_dispatch.h>
+#include <ATen/ops/signbit_meta_dispatch.h>
+#include <ATen/ops/silu_meta_dispatch.h>
+#include <ATen/ops/silu_backward_meta_dispatch.h>
+#include <ATen/ops/sin_meta_dispatch.h>
+#include <ATen/ops/sinc_meta_dispatch.h>
+#include <ATen/ops/sinh_meta_dispatch.h>
+#include <ATen/ops/slow_conv_transpose2d_meta_dispatch.h>
+#include <ATen/ops/smooth_l1_loss_meta_dispatch.h>
+#include <ATen/ops/softplus_meta_dispatch.h>
+#include <ATen/ops/softplus_backward_meta_dispatch.h>
+#include <ATen/ops/softshrink_meta_dispatch.h>
+#include <ATen/ops/softshrink_backward_meta_dispatch.h>
+#include <ATen/ops/sort_meta_dispatch.h>
+#include <ATen/ops/sparse_resize_meta_dispatch.h>
+#include <ATen/ops/sparse_resize_and_clear_meta_dispatch.h>
+#include <ATen/ops/special_airy_ai_meta_dispatch.h>
+#include <ATen/ops/special_bessel_j0_meta_dispatch.h>
+#include <ATen/ops/special_bessel_j1_meta_dispatch.h>
+#include <ATen/ops/special_bessel_y0_meta_dispatch.h>
+#include <ATen/ops/special_bessel_y1_meta_dispatch.h>
+#include <ATen/ops/special_chebyshev_polynomial_t_meta_dispatch.h>
+#include <ATen/ops/special_chebyshev_polynomial_u_meta_dispatch.h>
+#include <ATen/ops/special_chebyshev_polynomial_v_meta_dispatch.h>
+#include <ATen/ops/special_chebyshev_polynomial_w_meta_dispatch.h>
+#include <ATen/ops/special_entr_meta_dispatch.h>
+#include <ATen/ops/special_erfcx_meta_dispatch.h>
+#include <ATen/ops/special_hermite_polynomial_h_meta_dispatch.h>
+#include <ATen/ops/special_hermite_polynomial_he_meta_dispatch.h>
+#include <ATen/ops/special_i0e_meta_dispatch.h>
+#include <ATen/ops/special_i1_meta_dispatch.h>
+#include <ATen/ops/special_i1e_meta_dispatch.h>
+#include <ATen/ops/special_laguerre_polynomial_l_meta_dispatch.h>
+#include <ATen/ops/special_legendre_polynomial_p_meta_dispatch.h>
+#include <ATen/ops/special_log_ndtr_meta_dispatch.h>
+#include <ATen/ops/special_modified_bessel_i0_meta_dispatch.h>
+#include <ATen/ops/special_modified_bessel_i1_meta_dispatch.h>
+#include <ATen/ops/special_modified_bessel_k0_meta_dispatch.h>
+#include <ATen/ops/special_modified_bessel_k1_meta_dispatch.h>
+#include <ATen/ops/special_ndtri_meta_dispatch.h>
+#include <ATen/ops/special_scaled_modified_bessel_k0_meta_dispatch.h>
+#include <ATen/ops/special_scaled_modified_bessel_k1_meta_dispatch.h>
+#include <ATen/ops/special_shifted_chebyshev_polynomial_t_meta_dispatch.h>
+#include <ATen/ops/special_shifted_chebyshev_polynomial_u_meta_dispatch.h>
+#include <ATen/ops/special_shifted_chebyshev_polynomial_v_meta_dispatch.h>
+#include <ATen/ops/special_shifted_chebyshev_polynomial_w_meta_dispatch.h>
+#include <ATen/ops/special_spherical_bessel_j0_meta_dispatch.h>
+#include <ATen/ops/special_xlog1py_meta_dispatch.h>
+#include <ATen/ops/special_zeta_meta_dispatch.h>
+#include <ATen/ops/sqrt_meta_dispatch.h>
+#include <ATen/ops/sub_meta_dispatch.h>
+#include <ATen/ops/sum_meta_dispatch.h>
+#include <ATen/ops/tan_meta_dispatch.h>
+#include <ATen/ops/tanh_meta_dispatch.h>
+#include <ATen/ops/tanh_backward_meta_dispatch.h>
+#include <ATen/ops/threshold_meta_dispatch.h>
+#include <ATen/ops/threshold_backward_meta_dispatch.h>
+#include <ATen/ops/topk_meta_dispatch.h>
+#include <ATen/ops/triangular_solve_meta_dispatch.h>
+#include <ATen/ops/tril_meta_dispatch.h>
+#include <ATen/ops/triu_meta_dispatch.h>
+#include <ATen/ops/trunc_meta_dispatch.h>
+#include <ATen/ops/unfold_meta_dispatch.h>
+#include <ATen/ops/uniform_meta_dispatch.h>
+#include <ATen/ops/upsample_bicubic2d_meta_dispatch.h>
+#include <ATen/ops/upsample_bicubic2d_backward_meta_dispatch.h>
+#include <ATen/ops/upsample_bilinear2d_meta_dispatch.h>
+#include <ATen/ops/upsample_bilinear2d_backward_meta_dispatch.h>
+#include <ATen/ops/upsample_linear1d_meta_dispatch.h>
+#include <ATen/ops/upsample_linear1d_backward_meta_dispatch.h>
+#include <ATen/ops/upsample_nearest1d_meta_dispatch.h>
+#include <ATen/ops/upsample_nearest1d_backward_meta_dispatch.h>
+#include <ATen/ops/upsample_nearest2d_meta_dispatch.h>
+#include <ATen/ops/upsample_nearest2d_backward_meta_dispatch.h>
+#include <ATen/ops/upsample_nearest3d_meta_dispatch.h>
+#include <ATen/ops/upsample_nearest3d_backward_meta_dispatch.h>
+#include <ATen/ops/upsample_trilinear3d_meta_dispatch.h>
+#include <ATen/ops/upsample_trilinear3d_backward_meta_dispatch.h>
+#include <ATen/ops/view_meta_dispatch.h>
+#include <ATen/ops/view_as_complex_meta_dispatch.h>
+#include <ATen/ops/view_as_real_meta_dispatch.h>
+#include <ATen/ops/xlogy_meta_dispatch.h>
+#include <ATen/ops/zero_meta_dispatch.h>
+
+
+

.venv/Lib/site-packages/torch/include/ATen/MethodOperators.h

@@ -0,0 +1,443 @@
+#pragma once
+
+// @generated by torchgen/gen.py from MethodOperators.h
+
+#ifdef TORCH_ASSERT_NO_OPERATORS
+#error This change adds a dependency on native_functions.yaml,             \
+  meaning the file will need to be re-compiled every time an operator      \
+  is changed or added. Consider if your change would be better placed in   \
+  another file, or if a more specific header might achieve the same goal.  \
+  See NOTE: [Tensor vs. TensorBase]
+#endif
+
+// Forward declarations of any types needed in the operator signatures.
+// We can't directly include these classes because it will cause circular include dependencies.
+// This file is included by TensorBody.h, which defines the Tensor class.
+#include <ATen/core/ATen_fwd.h>
+
+#include <ATen/ops/_addmm_activation_ops.h>
+#include <ATen/ops/_autocast_to_full_precision_ops.h>
+#include <ATen/ops/_autocast_to_reduced_precision_ops.h>
+#include <ATen/ops/_backward_ops.h>
+#include <ATen/ops/_coalesced_ops.h>
+#include <ATen/ops/_conj_ops.h>
+#include <ATen/ops/_conj_physical_ops.h>
+#include <ATen/ops/_dimI_ops.h>
+#include <ATen/ops/_dimV_ops.h>
+#include <ATen/ops/_fw_primal_ops.h>
+#include <ATen/ops/_indices_ops.h>
+#include <ATen/ops/_is_all_true_ops.h>
+#include <ATen/ops/_is_any_true_ops.h>
+#include <ATen/ops/_is_zerotensor_ops.h>
+#include <ATen/ops/_lazy_clone_ops.h>
+#include <ATen/ops/_neg_view_ops.h>
+#include <ATen/ops/_nested_tensor_size_ops.h>
+#include <ATen/ops/_nested_tensor_storage_offsets_ops.h>
+#include <ATen/ops/_nested_tensor_strides_ops.h>
+#include <ATen/ops/_nnz_ops.h>
+#include <ATen/ops/_reshape_alias_ops.h>
+#include <ATen/ops/_sparse_mask_projection_ops.h>
+#include <ATen/ops/_to_dense_ops.h>
+#include <ATen/ops/_to_sparse_bsc_ops.h>
+#include <ATen/ops/_to_sparse_bsr_ops.h>
+#include <ATen/ops/_to_sparse_csc_ops.h>
+#include <ATen/ops/_to_sparse_csr_ops.h>
+#include <ATen/ops/_to_sparse_ops.h>
+#include <ATen/ops/_values_ops.h>
+#include <ATen/ops/_version_ops.h>
+#include <ATen/ops/abs_ops.h>
+#include <ATen/ops/absolute_ops.h>
+#include <ATen/ops/acos_ops.h>
+#include <ATen/ops/acosh_ops.h>
+#include <ATen/ops/add_ops.h>
+#include <ATen/ops/addbmm_ops.h>
+#include <ATen/ops/addcdiv_ops.h>
+#include <ATen/ops/addcmul_ops.h>
+#include <ATen/ops/addmm_ops.h>
+#include <ATen/ops/addmv_ops.h>
+#include <ATen/ops/addr_ops.h>
+#include <ATen/ops/adjoint_ops.h>
+#include <ATen/ops/alias_ops.h>
+#include <ATen/ops/align_as_ops.h>
+#include <ATen/ops/align_to_ops.h>
+#include <ATen/ops/all_ops.h>
+#include <ATen/ops/allclose_ops.h>
+#include <ATen/ops/amax_ops.h>
+#include <ATen/ops/amin_ops.h>
+#include <ATen/ops/aminmax_ops.h>
+#include <ATen/ops/and_ops.h>
+#include <ATen/ops/angle_ops.h>
+#include <ATen/ops/any_ops.h>
+#include <ATen/ops/arccos_ops.h>
+#include <ATen/ops/arccosh_ops.h>
+#include <ATen/ops/arcsin_ops.h>
+#include <ATen/ops/arcsinh_ops.h>
+#include <ATen/ops/arctan2_ops.h>
+#include <ATen/ops/arctan_ops.h>
+#include <ATen/ops/arctanh_ops.h>
+#include <ATen/ops/argmax_ops.h>
+#include <ATen/ops/argmin_ops.h>
+#include <ATen/ops/argsort_ops.h>
+#include <ATen/ops/argwhere_ops.h>
+#include <ATen/ops/as_strided_ops.h>
+#include <ATen/ops/as_strided_scatter_ops.h>
+#include <ATen/ops/asin_ops.h>
+#include <ATen/ops/asinh_ops.h>
+#include <ATen/ops/atan2_ops.h>
+#include <ATen/ops/atan_ops.h>
+#include <ATen/ops/atanh_ops.h>
+#include <ATen/ops/baddbmm_ops.h>
+#include <ATen/ops/bernoulli_ops.h>
+#include <ATen/ops/bincount_ops.h>
+#include <ATen/ops/bitwise_and_ops.h>
+#include <ATen/ops/bitwise_left_shift_ops.h>
+#include <ATen/ops/bitwise_not_ops.h>
+#include <ATen/ops/bitwise_or_ops.h>
+#include <ATen/ops/bitwise_right_shift_ops.h>
+#include <ATen/ops/bitwise_xor_ops.h>
+#include <ATen/ops/bmm_ops.h>
+#include <ATen/ops/broadcast_to_ops.h>
+#include <ATen/ops/cauchy_ops.h>
+#include <ATen/ops/ccol_indices_ops.h>
+#include <ATen/ops/ceil_ops.h>
+#include <ATen/ops/chalf_ops.h>
+#include <ATen/ops/cholesky_inverse_ops.h>
+#include <ATen/ops/cholesky_ops.h>
+#include <ATen/ops/cholesky_solve_ops.h>
+#include <ATen/ops/chunk_ops.h>
+#include <ATen/ops/clamp_max_ops.h>
+#include <ATen/ops/clamp_min_ops.h>
+#include <ATen/ops/clamp_ops.h>
+#include <ATen/ops/clip_ops.h>
+#include <ATen/ops/clone_ops.h>
+#include <ATen/ops/coalesce_ops.h>
+#include <ATen/ops/col_indices_ops.h>
+#include <ATen/ops/conj_ops.h>
+#include <ATen/ops/conj_physical_ops.h>
+#include <ATen/ops/contiguous_ops.h>
+#include <ATen/ops/copy_ops.h>
+#include <ATen/ops/copysign_ops.h>
+#include <ATen/ops/corrcoef_ops.h>
+#include <ATen/ops/cos_ops.h>
+#include <ATen/ops/cosh_ops.h>
+#include <ATen/ops/count_nonzero_ops.h>
+#include <ATen/ops/cov_ops.h>
+#include <ATen/ops/cross_ops.h>
+#include <ATen/ops/crow_indices_ops.h>
+#include <ATen/ops/cummax_ops.h>
+#include <ATen/ops/cummin_ops.h>
+#include <ATen/ops/cumprod_ops.h>
+#include <ATen/ops/cumsum_ops.h>
+#include <ATen/ops/data_ops.h>
+#include <ATen/ops/deg2rad_ops.h>
+#include <ATen/ops/dense_dim_ops.h>
+#include <ATen/ops/dequantize_ops.h>
+#include <ATen/ops/det_ops.h>
+#include <ATen/ops/detach_ops.h>
+#include <ATen/ops/diag_embed_ops.h>
+#include <ATen/ops/diag_ops.h>
+#include <ATen/ops/diagflat_ops.h>
+#include <ATen/ops/diagonal_ops.h>
+#include <ATen/ops/diagonal_scatter_ops.h>
+#include <ATen/ops/diff_ops.h>
+#include <ATen/ops/digamma_ops.h>
+#include <ATen/ops/dist_ops.h>
+#include <ATen/ops/div_ops.h>
+#include <ATen/ops/divide_ops.h>
+#include <ATen/ops/dot_ops.h>
+#include <ATen/ops/dsplit_ops.h>
+#include <ATen/ops/eq_ops.h>
+#include <ATen/ops/equal_ops.h>
+#include <ATen/ops/erf_ops.h>
+#include <ATen/ops/erfc_ops.h>
+#include <ATen/ops/erfinv_ops.h>
+#include <ATen/ops/exp2_ops.h>
+#include <ATen/ops/exp_ops.h>
+#include <ATen/ops/expand_as_ops.h>
+#include <ATen/ops/expand_ops.h>
+#include <ATen/ops/expm1_ops.h>
+#include <ATen/ops/exponential_ops.h>
+#include <ATen/ops/fill_diagonal_ops.h>
+#include <ATen/ops/fill_ops.h>
+#include <ATen/ops/fix_ops.h>
+#include <ATen/ops/flatten_ops.h>
+#include <ATen/ops/flip_ops.h>
+#include <ATen/ops/fliplr_ops.h>
+#include <ATen/ops/flipud_ops.h>
+#include <ATen/ops/float_power_ops.h>
+#include <ATen/ops/floor_divide_ops.h>
+#include <ATen/ops/floor_ops.h>
+#include <ATen/ops/fmax_ops.h>
+#include <ATen/ops/fmin_ops.h>
+#include <ATen/ops/fmod_ops.h>
+#include <ATen/ops/frac_ops.h>
+#include <ATen/ops/frexp_ops.h>
+#include <ATen/ops/gather_ops.h>
+#include <ATen/ops/gcd_ops.h>
+#include <ATen/ops/ge_ops.h>
+#include <ATen/ops/geometric_ops.h>
+#include <ATen/ops/geqrf_ops.h>
+#include <ATen/ops/ger_ops.h>
+#include <ATen/ops/greater_equal_ops.h>
+#include <ATen/ops/greater_ops.h>
+#include <ATen/ops/gt_ops.h>
+#include <ATen/ops/hardshrink_backward_ops.h>
+#include <ATen/ops/hardshrink_ops.h>
+#include <ATen/ops/heaviside_ops.h>
+#include <ATen/ops/histc_ops.h>
+#include <ATen/ops/histogram_ops.h>
+#include <ATen/ops/hsplit_ops.h>
+#include <ATen/ops/hypot_ops.h>
+#include <ATen/ops/i0_ops.h>
+#include <ATen/ops/igamma_ops.h>
+#include <ATen/ops/igammac_ops.h>
+#include <ATen/ops/index_add_ops.h>
+#include <ATen/ops/index_copy_ops.h>
+#include <ATen/ops/index_fill_ops.h>
+#include <ATen/ops/index_ops.h>
+#include <ATen/ops/index_put_ops.h>
+#include <ATen/ops/index_reduce_ops.h>
+#include <ATen/ops/index_select_ops.h>
+#include <ATen/ops/indices_ops.h>
+#include <ATen/ops/inner_ops.h>
+#include <ATen/ops/int_repr_ops.h>
+#include <ATen/ops/inverse_ops.h>
+#include <ATen/ops/is_coalesced_ops.h>
+#include <ATen/ops/is_complex_ops.h>
+#include <ATen/ops/is_conj_ops.h>
+#include <ATen/ops/is_distributed_ops.h>
+#include <ATen/ops/is_floating_point_ops.h>
+#include <ATen/ops/is_inference_ops.h>
+#include <ATen/ops/is_leaf_ops.h>
+#include <ATen/ops/is_neg_ops.h>
+#include <ATen/ops/is_nonzero_ops.h>
+#include <ATen/ops/is_pinned_ops.h>
+#include <ATen/ops/is_same_size_ops.h>
+#include <ATen/ops/is_set_to_ops.h>
+#include <ATen/ops/is_signed_ops.h>
+#include <ATen/ops/isclose_ops.h>
+#include <ATen/ops/isfinite_ops.h>
+#include <ATen/ops/isinf_ops.h>
+#include <ATen/ops/isnan_ops.h>
+#include <ATen/ops/isneginf_ops.h>
+#include <ATen/ops/isposinf_ops.h>
+#include <ATen/ops/isreal_ops.h>
+#include <ATen/ops/istft_ops.h>
+#include <ATen/ops/item_ops.h>
+#include <ATen/ops/kron_ops.h>
+#include <ATen/ops/kthvalue_ops.h>
+#include <ATen/ops/lcm_ops.h>
+#include <ATen/ops/ldexp_ops.h>
+#include <ATen/ops/le_ops.h>
+#include <ATen/ops/lerp_ops.h>
+#include <ATen/ops/less_equal_ops.h>
+#include <ATen/ops/less_ops.h>
+#include <ATen/ops/lgamma_ops.h>
+#include <ATen/ops/log10_ops.h>
+#include <ATen/ops/log1p_ops.h>
+#include <ATen/ops/log2_ops.h>
+#include <ATen/ops/log_normal_ops.h>
+#include <ATen/ops/log_ops.h>
+#include <ATen/ops/log_softmax_ops.h>
+#include <ATen/ops/logaddexp2_ops.h>
+#include <ATen/ops/logaddexp_ops.h>
+#include <ATen/ops/logcumsumexp_ops.h>
+#include <ATen/ops/logdet_ops.h>
+#include <ATen/ops/logical_and_ops.h>
+#include <ATen/ops/logical_not_ops.h>
+#include <ATen/ops/logical_or_ops.h>
+#include <ATen/ops/logical_xor_ops.h>
+#include <ATen/ops/logit_ops.h>
+#include <ATen/ops/logsumexp_ops.h>
+#include <ATen/ops/lshift_ops.h>
+#include <ATen/ops/lt_ops.h>
+#include <ATen/ops/lu_solve_ops.h>
+#include <ATen/ops/mH_ops.h>
+#include <ATen/ops/mT_ops.h>
+#include <ATen/ops/masked_fill_ops.h>
+#include <ATen/ops/masked_scatter_ops.h>
+#include <ATen/ops/masked_select_ops.h>
+#include <ATen/ops/matmul_ops.h>
+#include <ATen/ops/matrix_H_ops.h>
+#include <ATen/ops/matrix_exp_ops.h>
+#include <ATen/ops/matrix_power_ops.h>
+#include <ATen/ops/max_ops.h>
+#include <ATen/ops/maximum_ops.h>
+#include <ATen/ops/mean_ops.h>
+#include <ATen/ops/median_ops.h>
+#include <ATen/ops/min_ops.h>
+#include <ATen/ops/minimum_ops.h>
+#include <ATen/ops/mm_ops.h>
+#include <ATen/ops/mode_ops.h>
+#include <ATen/ops/moveaxis_ops.h>
+#include <ATen/ops/movedim_ops.h>
+#include <ATen/ops/msort_ops.h>
+#include <ATen/ops/mul_ops.h>
+#include <ATen/ops/multinomial_ops.h>
+#include <ATen/ops/multiply_ops.h>
+#include <ATen/ops/mv_ops.h>
+#include <ATen/ops/mvlgamma_ops.h>
+#include <ATen/ops/nan_to_num_ops.h>
+#include <ATen/ops/nanmean_ops.h>
+#include <ATen/ops/nanmedian_ops.h>
+#include <ATen/ops/nanquantile_ops.h>
+#include <ATen/ops/nansum_ops.h>
+#include <ATen/ops/narrow_copy_ops.h>
+#include <ATen/ops/narrow_ops.h>
+#include <ATen/ops/ne_ops.h>
+#include <ATen/ops/neg_ops.h>
+#include <ATen/ops/negative_ops.h>
+#include <ATen/ops/new_empty_ops.h>
+#include <ATen/ops/new_empty_strided_ops.h>
+#include <ATen/ops/new_full_ops.h>
+#include <ATen/ops/new_ones_ops.h>
+#include <ATen/ops/new_zeros_ops.h>
+#include <ATen/ops/nextafter_ops.h>
+#include <ATen/ops/nonzero_numpy_ops.h>
+#include <ATen/ops/nonzero_ops.h>
+#include <ATen/ops/nonzero_static_ops.h>
+#include <ATen/ops/norm_ops.h>
+#include <ATen/ops/normal_ops.h>
+#include <ATen/ops/not_equal_ops.h>
+#include <ATen/ops/numpy_T_ops.h>
+#include <ATen/ops/or_ops.h>
+#include <ATen/ops/orgqr_ops.h>
+#include <ATen/ops/ormqr_ops.h>
+#include <ATen/ops/outer_ops.h>
+#include <ATen/ops/output_nr_ops.h>
+#include <ATen/ops/permute_ops.h>
+#include <ATen/ops/pin_memory_ops.h>
+#include <ATen/ops/pinverse_ops.h>
+#include <ATen/ops/polygamma_ops.h>
+#include <ATen/ops/positive_ops.h>
+#include <ATen/ops/pow_ops.h>
+#include <ATen/ops/prelu_ops.h>
+#include <ATen/ops/prod_ops.h>
+#include <ATen/ops/put_ops.h>
+#include <ATen/ops/q_per_channel_axis_ops.h>
+#include <ATen/ops/q_per_channel_scales_ops.h>
+#include <ATen/ops/q_per_channel_zero_points_ops.h>
+#include <ATen/ops/q_scale_ops.h>
+#include <ATen/ops/q_zero_point_ops.h>
+#include <ATen/ops/qr_ops.h>
+#include <ATen/ops/qscheme_ops.h>
+#include <ATen/ops/quantile_ops.h>
+#include <ATen/ops/rad2deg_ops.h>
+#include <ATen/ops/random_ops.h>
+#include <ATen/ops/ravel_ops.h>
+#include <ATen/ops/reciprocal_ops.h>
+#include <ATen/ops/record_stream_ops.h>
+#include <ATen/ops/refine_names_ops.h>
+#include <ATen/ops/relu_ops.h>
+#include <ATen/ops/remainder_ops.h>
+#include <ATen/ops/rename_ops.h>
+#include <ATen/ops/renorm_ops.h>
+#include <ATen/ops/repeat_interleave_ops.h>
+#include <ATen/ops/repeat_ops.h>
+#include <ATen/ops/requires_grad_ops.h>
+#include <ATen/ops/reshape_as_ops.h>
+#include <ATen/ops/reshape_ops.h>
+#include <ATen/ops/resize_as_ops.h>
+#include <ATen/ops/resize_as_sparse_ops.h>
+#include <ATen/ops/resize_ops.h>
+#include <ATen/ops/resolve_conj_ops.h>
+#include <ATen/ops/resolve_neg_ops.h>
+#include <ATen/ops/retain_grad_ops.h>
+#include <ATen/ops/retains_grad_ops.h>
+#include <ATen/ops/roll_ops.h>
+#include <ATen/ops/rot90_ops.h>
+#include <ATen/ops/round_ops.h>
+#include <ATen/ops/row_indices_ops.h>
+#include <ATen/ops/rshift_ops.h>
+#include <ATen/ops/rsqrt_ops.h>
+#include <ATen/ops/scatter_add_ops.h>
+#include <ATen/ops/scatter_ops.h>
+#include <ATen/ops/scatter_reduce_ops.h>
+#include <ATen/ops/select_ops.h>
+#include <ATen/ops/select_scatter_ops.h>
+#include <ATen/ops/set_data_ops.h>
+#include <ATen/ops/set_ops.h>
+#include <ATen/ops/sgn_ops.h>
+#include <ATen/ops/sigmoid_ops.h>
+#include <ATen/ops/sign_ops.h>
+#include <ATen/ops/signbit_ops.h>
+#include <ATen/ops/sin_ops.h>
+#include <ATen/ops/sinc_ops.h>
+#include <ATen/ops/sinh_ops.h>
+#include <ATen/ops/size_ops.h>
+#include <ATen/ops/slice_inverse_ops.h>
+#include <ATen/ops/slice_ops.h>
+#include <ATen/ops/slice_scatter_ops.h>
+#include <ATen/ops/slogdet_ops.h>
+#include <ATen/ops/smm_ops.h>
+#include <ATen/ops/softmax_ops.h>
+#include <ATen/ops/sort_ops.h>
+#include <ATen/ops/sparse_dim_ops.h>
+#include <ATen/ops/sparse_mask_ops.h>
+#include <ATen/ops/sparse_resize_and_clear_ops.h>
+#include <ATen/ops/sparse_resize_ops.h>
+#include <ATen/ops/split_ops.h>
+#include <ATen/ops/split_with_sizes_ops.h>
+#include <ATen/ops/sqrt_ops.h>
+#include <ATen/ops/square_ops.h>
+#include <ATen/ops/squeeze_ops.h>
+#include <ATen/ops/sspaddmm_ops.h>
+#include <ATen/ops/std_ops.h>
+#include <ATen/ops/stft_ops.h>
+#include <ATen/ops/stride_ops.h>
+#include <ATen/ops/sub_ops.h>
+#include <ATen/ops/subtract_ops.h>
+#include <ATen/ops/sum_ops.h>
+#include <ATen/ops/sum_to_size_ops.h>
+#include <ATen/ops/svd_ops.h>
+#include <ATen/ops/swapaxes_ops.h>
+#include <ATen/ops/swapdims_ops.h>
+#include <ATen/ops/t_ops.h>
+#include <ATen/ops/take_along_dim_ops.h>
+#include <ATen/ops/take_ops.h>
+#include <ATen/ops/tan_ops.h>
+#include <ATen/ops/tanh_ops.h>
+#include <ATen/ops/tensor_split_ops.h>
+#include <ATen/ops/tile_ops.h>
+#include <ATen/ops/to_dense_ops.h>
+#include <ATen/ops/to_mkldnn_ops.h>
+#include <ATen/ops/to_ops.h>
+#include <ATen/ops/to_padded_tensor_ops.h>
+#include <ATen/ops/to_sparse_bsc_ops.h>
+#include <ATen/ops/to_sparse_bsr_ops.h>
+#include <ATen/ops/to_sparse_csc_ops.h>
+#include <ATen/ops/to_sparse_csr_ops.h>
+#include <ATen/ops/to_sparse_ops.h>
+#include <ATen/ops/topk_ops.h>
+#include <ATen/ops/trace_ops.h>
+#include <ATen/ops/transpose_ops.h>
+#include <ATen/ops/triangular_solve_ops.h>
+#include <ATen/ops/tril_ops.h>
+#include <ATen/ops/triu_ops.h>
+#include <ATen/ops/true_divide_ops.h>
+#include <ATen/ops/trunc_ops.h>
+#include <ATen/ops/type_as_ops.h>
+#include <ATen/ops/unbind_ops.h>
+#include <ATen/ops/unflatten_ops.h>
+#include <ATen/ops/unfold_ops.h>
+#include <ATen/ops/uniform_ops.h>
+#include <ATen/ops/unsafe_chunk_ops.h>
+#include <ATen/ops/unsafe_split_ops.h>
+#include <ATen/ops/unsafe_split_with_sizes_ops.h>
+#include <ATen/ops/unsqueeze_ops.h>
+#include <ATen/ops/values_ops.h>
+#include <ATen/ops/var_ops.h>
+#include <ATen/ops/vdot_ops.h>
+#include <ATen/ops/view_as_ops.h>
+#include <ATen/ops/view_ops.h>
+#include <ATen/ops/vsplit_ops.h>
+#include <ATen/ops/where_ops.h>
+#include <ATen/ops/xlogy_ops.h>
+#include <ATen/ops/xor_ops.h>
+#include <ATen/ops/zero_ops.h>
+
+namespace at {
+namespace _ops {
+
+} // namespace _ops
+} // namespace at