Shortcuts

Program Listing for File TensorBody.h

Return to documentation for file (build/aten/src/ATen/core/TensorBody.h)

#pragma once

#include <c10/core/Device.h>
#include <c10/core/Layout.h>
#include <c10/core/MemoryFormat.h>
#include <c10/core/QScheme.h>
#include <c10/core/Stream.h>
#include <c10/core/Scalar.h>
#include <c10/core/ScalarType.h>
#include <c10/core/ScalarTypeToTypeMeta.h>
#include <c10/core/Storage.h>
#include <ATen/core/TensorAccessor.h>
#include <c10/core/TensorImpl.h>
#include <c10/core/UndefinedTensorImpl.h>
#include <c10/util/Exception.h>
#include <c10/util/Deprecated.h>
#include <c10/util/Optional.h>
#include <c10/util/intrusive_ptr.h>
#include <ATen/core/DeprecatedTypePropertiesRegistry.h>
#include <ATen/core/DeprecatedTypeProperties.h>
#include <ATen/core/NamedTensor.h>
#include <ATen/core/QuantizerBase.h>
#include <torch/csrc/WindowsTorchApiMacro.h>

namespace caffe2 {
class Tensor;
}
namespace c10{
struct TensorOptions;
}
namespace at {
struct Generator;
struct Type;
class DeprecatedTypeProperties;
class Tensor;
} // namespace at
namespace at {
namespace indexing {
struct TensorIndex;
} // namespace indexing
} // namespace at

namespace torch { namespace autograd {

struct Node;

}} // namespace torch::autograd

namespace at {

class Tensor;
using TensorList = ArrayRef<Tensor>;

using Stream = c10::Stream;

namespace impl {
inline bool variable_excluded_from_dispatch() {
#ifdef C10_MOBILE
  // Please read the comment in `VariableFallbackKernel.cpp` about the background of this change.
  return true;
#else
  TORCH_INTERNAL_ASSERT_DEBUG_ONLY(!c10::impl::tls_local_dispatch_key_set().excluded_.has(DispatchKey::Autograd));
  return c10::impl::tls_local_dispatch_key_set().excluded_.isSupersetOf(c10::autograd_dispatch_keyset);
#endif
}
}

// Tensor is a "generic" object holding a pointer to the underlying TensorImpl object, which
// has an embedded reference count. In this way, Tensor is similar to boost::intrusive_ptr.
//
// For example:
//
// void func(Tensor a) {
//   Tensor b = a;
//   ...
// }
//
// In this example, when we say Tensor b = a, we are creating a new object that points to the
// same underlying TensorImpl, and bumps its reference count. When b goes out of scope, the
// destructor decrements the reference count by calling release() on the TensorImpl it points to.
// The existing constructors, operator overloads, etc. take care to implement the correct semantics.
//
// Note that Tensor can also be NULL, i.e. it is not associated with any underlying TensorImpl, and
// special care must be taken to handle this.
class CAFFE2_API Tensor {
 public:
  Tensor(){};
  // This constructor should not be used by end users and is an implementation
  // detail invoked by autogenerated code.
  explicit Tensor(
      c10::intrusive_ptr<TensorImpl, UndefinedTensorImpl> tensor_impl)
      : impl_(std::move(tensor_impl)) {
    if (impl_.get() == nullptr) {
      throw std::runtime_error("TensorImpl with nullptr is not supported");
    }
  }
  Tensor(const Tensor&) = default;
  Tensor(Tensor&&) = default;


 public:
  // Creates a new wrapper from TensorImpl. Intentionally a free method because
  // it should be used with care. Checks necessary invariants
  static Tensor wrap_tensor_impl(
      c10::intrusive_ptr<TensorImpl, UndefinedTensorImpl> tensor_impl) {
    Tensor r(std::move(tensor_impl));
    r.enforce_invariants();
    return r;
  }

  int64_t dim() const {
    return impl_->dim();
  }
  int64_t storage_offset() const {
    return impl_->storage_offset();
  }

  TensorImpl * unsafeGetTensorImpl() const {
    return impl_.get();
  }
  TensorImpl * unsafeReleaseTensorImpl() {
    return impl_.release();
  }
  const c10::intrusive_ptr<TensorImpl, UndefinedTensorImpl>& getIntrusivePtr() const {
    return impl_;
  }

  bool defined() const {
    return impl_;
  }

  void reset() {
    impl_.reset();
  }

  // The following overloads are very intruiging.  Consider the following
  // program:
  //
  //    x[1] = 3;
  //
  // We would expect that the first entry of x is written to 3.  But how can we
  // actually achieve this?  x[1] evaluates to a tensor...
  //
  // The answer is, using a ref-qualifier.  x[1] is an rvalue, which cannot be
  // (profitably) assigned to in the traditional sense, so we overload
  // assignment to mean, "Actually, copy 3 into the tensor data."  This is done
  // with an rvalue-reference ref-qualified overload (the methods with && at the
  // end of their type.)
  //
  // There's one more fly in the ointment: We also want
  //
  //    Tensor x = y;
  //
  // to work, and we want it NOT to copy.  So we need a traditional operator=
  // overload.  But we MUST specify a mutable lvalue ref-qualifier, to
  // disambiguate the traditional overload from the rvalue-reference
  // ref-qualified overload.  Otherwise, it will be ambiguous, because
  // a non ref-qualified method is eligible for all situations.

  // Unfortunately, we have to write these constructors out manually
  // to work around an MSVC bug:
  //    error C2580: 'at::Tensor &at::Tensor::operator =(const at::Tensor &) &':
  //    multiple versions of a defaulted special member functions are not allowed
  // Tensor& operator=(const Tensor&) & = default;
  // Tensor& operator=(Tensor&&) & = default;

  // Also MSVC will wrongly issue the following warning with the aforementioned fix
  //    warning C4522: 'at::Tensor': multiple assignment operators specified
  // Let's just skip the warning.

  #ifdef _MSC_VER
  #pragma warning( push )
  #pragma warning( disable : 4522 )
  #endif

  Tensor& operator=(const Tensor& x) & {
    impl_ = x.impl_;
    return *this;
  }
  Tensor& operator=(Tensor&& x) & {
    impl_ = std::move(x.impl_);
    return *this;
  }

  Tensor& operator=(Scalar v) &&;
  Tensor& operator=(const Tensor&) &&;
  Tensor& operator=(Tensor&&) &&;

  #ifdef _MSC_VER
  #pragma warning( pop )
  #endif

  bool is_same(const Tensor& other) const noexcept {
    return impl_ == other.impl_;
  }
  size_t use_count() const noexcept {
    return impl_.use_count();
  }
  size_t weak_use_count() const noexcept {
    return impl_.weak_use_count();
  }

  std::string toString() const;

  IntArrayRef sizes() const {
    return impl_->sizes();
  }
  IntArrayRef strides() const {
    return impl_->strides();
  }
  // See impl::get_opt_names in ATen/NamedTensor.h for docs.
  c10::optional<DimnameList> opt_names() const {
    return impl::get_opt_names(unsafeGetTensorImpl());
  }
  // See impl::get_names in ATen/NamedTensor.h for docs.
  DimnameList names() const {
    return impl::get_names(unsafeGetTensorImpl());
  }
  int64_t ndimension() const {
    return dim();
  }

  bool is_contiguous(at::MemoryFormat memory_format=at::MemoryFormat::Contiguous) const {
    return impl_->is_contiguous(memory_format);
  }

  bool is_non_overlapping_and_dense() const {
    return impl_->is_non_overlapping_and_dense();
  }

  at::MemoryFormat suggest_memory_format(
      bool channels_last_strides_exact_match = false) const {
    // Setting channels_last_strides_exact_match to true forces function to
    // check 0,1 - sized dimension strides.
    if (!is_mkldnn() && !is_sparse()) {
      if (impl_->is_strides_like_channels_last()) {
        if (!channels_last_strides_exact_match ||
            get_channels_last_strides_2d(sizes()) == strides()) {
          return at::MemoryFormat::ChannelsLast;
        }
      }
      else if (impl_->is_strides_like_channels_last_3d()) {
        if (!channels_last_strides_exact_match ||
            get_channels_last_strides_3d(sizes()) == strides()) {
          return at::MemoryFormat::ChannelsLast3d;
        }
      }
    }
    return at::MemoryFormat::Contiguous;
  }

  // Total bytes consumed by the "view" of elements of the array.  Does not
  // include size of metadata.  The number reported here does not necessarily
  // correspond to the true physical memory consumed by a tensor; instead,
  // it reports the memory the tensor would take *if* it were contiguous.
  // Defined to be numel() * itemsize()
  size_t nbytes() const {
    TORCH_CHECK(layout () != at::kSparse,
                "nbytes is not defined for sparse tensors.  If you want the size of the constituent " \
                "tensors, add the nbytes of the indices and values.  If you want the size of the  " \
                "equivalent dense tensor, multiply numel() by element_size()");
    return impl_->numel() * impl_->itemsize();
  }

  int64_t numel() const {
    return impl_->numel();
  }

  // Length of one array element in bytes.  This is the traditional
  // Numpy naming.
  size_t itemsize() const {
    return impl_->itemsize();
  }

  // Same as itemsize().  This is the PyTorch naming.
  int64_t element_size() const {
    return static_cast<int64_t>(impl_->itemsize());
  }

  C10_DEPRECATED_MESSAGE("Tensor.type() is deprecated. Instead use Tensor.options(), which in many cases (e.g. in a constructor) is a drop-in replacement. If you were using data from type(), that is now available from Tensor itself, so instead of tensor.type().scalar_type(), use tensor.scalar_type() instead and instead of tensor.type().backend() use tensor.device().")
  DeprecatedTypeProperties & type() const {
    return globalDeprecatedTypePropertiesRegistry().getDeprecatedTypeProperties(
        dispatchKeyToBackend(legacyExtractDispatchKey(key_set())),
        scalar_type());
  }
  DispatchKeySet key_set() const {
    return impl_->key_set();
  }
  ScalarType scalar_type() const {
    return typeMetaToScalarType(impl_->dtype());
  }
  bool has_storage() const {
    return defined() && impl_->has_storage();
  }
  const Storage& storage() const {
    return impl_->storage();
  }
  bool is_alias_of(const at::Tensor& other) const{
    return impl_->storage().is_alias_of(other.storage());
  }
  Tensor toType(ScalarType t) const;
  Tensor toBackend(Backend b) const;

  C10_DEPRECATED_MESSAGE("Tensor.is_variable() is deprecated; everything is a variable now. (If you want to assert that variable has been appropriately handled already, use at::impl::variable_excluded_from_dispatch())")
  bool is_variable() const noexcept {
    return !at::impl::variable_excluded_from_dispatch();
  }

  Layout layout() const noexcept;

  caffe2::TypeMeta dtype() const noexcept;

  Device device() const;

  int64_t get_device() const;

  bool is_cuda() const;

  bool is_hip() const;

  bool is_sparse() const;

  bool is_mkldnn() const;

  bool is_vulkan() const;

  bool is_metal() const;

  bool is_quantized() const;

  bool is_meta() const;

  QuantizerPtr quantizer() const;

  bool has_names() const;

  const NamedTensorMeta* get_named_tensor_meta() const;
  NamedTensorMeta* get_named_tensor_meta();

  TensorOptions options() const;

  void* data_ptr() const {
    return this->unsafeGetTensorImpl()->data();
  }

  template <typename T>
  T * data_ptr() const;

  template<typename T>
  C10_DEPRECATED_MESSAGE("Tensor.data<T>() is deprecated. Please use Tensor.data_ptr<T>() instead.")
  T * data() const {
    return data_ptr<T>();
  }

  template <typename T>
  T item() const;

  // Purposely not defined here to avoid inlining
  void print() const;

  // Return a `TensorAccessor` for CPU `Tensor`s. You have to specify scalar type and
  // dimension.
  template<typename T, size_t N>
  TensorAccessor<T,N> accessor() const& {
    static_assert(N > 0, "accessor is used for indexing tensor, for scalars use *data_ptr<T>()");
    TORCH_CHECK(dim() == N, "TensorAccessor expected ", N, " dims but tensor has ", dim());
    return TensorAccessor<T,N>(data_ptr<T>(),sizes().data(),strides().data());
  }
  template<typename T, size_t N>
  TensorAccessor<T,N> accessor() && = delete;

  // Return a `GenericPackedTensorAccessor` for CUDA `Tensor`s. You have to specify scalar type and
  // dimension. You can optionally specify RestrictPtrTraits as a template parameter to
  // cast the data pointer to a __restrict__ pointer.
  // In order to use this, your CUDA kernel has to take a corresponding GenericPackedTensorAccessor
  // as an argument.
  template<typename T, size_t N, template <typename U> class PtrTraits = DefaultPtrTraits, typename index_t = int64_t>
  GenericPackedTensorAccessor<T,N,PtrTraits,index_t> generic_packed_accessor() const& {
    static_assert(N > 0, "accessor is used for indexing tensor, for scalars use *data_ptr<T>()");
    TORCH_CHECK(dim() == N, "TensorAccessor expected ", N, " dims but tensor has ", dim());
    return GenericPackedTensorAccessor<T,N,PtrTraits,index_t>(static_cast<typename PtrTraits<T>::PtrType>(data_ptr<T>()),sizes().data(),strides().data());
  }
  template<typename T, size_t N, template <typename U> class PtrTraits = DefaultPtrTraits, typename index_t = int64_t>
  GenericPackedTensorAccessor<T,N> generic_packed_accessor() && = delete;

  template<typename T, size_t N, template <typename U> class PtrTraits = DefaultPtrTraits>
  PackedTensorAccessor32<T,N,PtrTraits> packed_accessor32() const& {
    return generic_packed_accessor<T,N,PtrTraits,int32_t>();
  }
  template<typename T, size_t N, template <typename U> class PtrTraits = DefaultPtrTraits>
  PackedTensorAccessor32<T,N,PtrTraits> packed_accessor32() && = delete;

  template<typename T, size_t N, template <typename U> class PtrTraits = DefaultPtrTraits>
  PackedTensorAccessor64<T,N,PtrTraits> packed_accessor64() const& {
    return generic_packed_accessor<T,N,PtrTraits,int64_t>();
  }
  template<typename T, size_t N, template <typename U> class PtrTraits = DefaultPtrTraits>
  PackedTensorAccessor64<T,N,PtrTraits> packed_accessor64() && = delete;

  template<typename T, size_t N, template <typename U> class PtrTraits = DefaultPtrTraits, typename index_t = int64_t>
  C10_DEPRECATED_MESSAGE("packed_accessor is deprecated, use packed_accessor32 or packed_accessor64 instead")
  GenericPackedTensorAccessor<T,N,PtrTraits,index_t> packed_accessor() const & {
    return generic_packed_accessor<T,N,PtrTraits,index_t>();
  }
  template<typename T, size_t N, template <typename U> class PtrTraits = DefaultPtrTraits, typename index_t = int64_t>
  C10_DEPRECATED_MESSAGE("packed_accessor is deprecated, use packed_accessor32 or packed_accessor64 instead")
  GenericPackedTensorAccessor<T,N,PtrTraits,index_t> packed_accessor() && = delete;

  Tensor operator~() const;
  Tensor operator-() const;
  Tensor& operator+=(const Tensor & other);
  Tensor& operator+=(Scalar other);
  Tensor& operator-=(const Tensor & other);
  Tensor& operator-=(Scalar other);
  Tensor& operator*=(const Tensor & other);
  Tensor& operator*=(Scalar other);
  Tensor& operator/=(const Tensor & other);
  Tensor& operator/=(Scalar other);
  Tensor& operator&=(const Tensor & other);
  Tensor& operator|=(const Tensor & other);
  Tensor& operator^=(const Tensor & other);
  Tensor operator[](Scalar index) const;
  Tensor operator[](Tensor index) const;
  Tensor operator[](int64_t index) const;

  Tensor index(ArrayRef<at::indexing::TensorIndex> indices) const;
  Tensor index(std::initializer_list<at::indexing::TensorIndex> indices) const;

  Tensor & index_put_(ArrayRef<at::indexing::TensorIndex> indices, Tensor const & rhs);
  Tensor & index_put_(ArrayRef<at::indexing::TensorIndex> indices, Scalar v);
  Tensor & index_put_(std::initializer_list<at::indexing::TensorIndex> indices, Tensor const & rhs);
  Tensor & index_put_(std::initializer_list<at::indexing::TensorIndex> indices, Scalar v);

  Tensor cpu() const;
  Tensor cuda() const;
  Tensor hip() const;
  Tensor vulkan() const;
  Tensor metal() const;

  // ~~~~~ Autograd API ~~~~~


  void backward(const Tensor & gradient={}, c10::optional<bool> retain_graph=c10::nullopt, bool create_graph=false, c10::optional<TensorList> inputs=c10::nullopt) const {
    // NB: Adding this wrapper to _backward here because we'd like our
    // 'backwards' api to accept the 'inputs' argument optionally. Since code gen
    // currently does not support optional of TensorList our approach is to replace
    // backward in native_functions.yaml with _backward and call it here instead.
    if (inputs.has_value()) {
      TORCH_CHECK(inputs.value().size() > 0, "'inputs' argument to backward cannot be empty")
      this->_backward(inputs.value(), gradient, retain_graph, create_graph);
    } else {
      this->_backward({}, gradient, retain_graph, create_graph);
    }
  }




  Tensor& set_requires_grad(bool requires_grad) {
    impl_->set_requires_grad(requires_grad);
    return *this;
  }
  bool requires_grad() const {
    return impl_->requires_grad();
  }

  Tensor& mutable_grad() {
    return impl_->mutable_grad();
  }

  const Tensor& grad() const {
    return impl_->grad();
  }

  // STOP.  Thinking of adding a method here, which only makes use
  // of other ATen methods?  Define it in native_functions.yaml.

  //example
  //Tensor * add(Tensor & b);
  void _backward(TensorList inputs, const c10::optional<Tensor>& gradient={}, c10::optional<bool> retain_graph=c10::nullopt, bool create_graph=false) const;
  void set_data(const Tensor & new_data) const;
  Tensor data() const;
  bool is_leaf() const;
  int64_t output_nr() const;
  int64_t _version() const;
  Tensor & requires_grad_(bool requires_grad=true) const;
  void retain_grad() const;
  Tensor & rename_(c10::optional<DimnameList> names) const;
  Tensor rename(c10::optional<DimnameList> names) const;
  Tensor align_to(DimnameList names) const;
  Tensor align_to(DimnameList order, int64_t ellipsis_idx) const;
  Tensor align_as(const Tensor & other) const;
  Tensor refine_names(DimnameList names) const;
  Tensor abs() const;
  Tensor & abs_() const;
  Tensor absolute() const;
  Tensor & absolute_() const;
  Tensor angle() const;
  Tensor sgn() const;
  Tensor & sgn_() const;
  Tensor conj() const;
  Tensor acos() const;
  Tensor & acos_() const;
  Tensor arccos() const;
  Tensor & arccos_() const;
  Tensor add(const Tensor & other, Scalar alpha=1) const;
  Tensor & add_(const Tensor & other, Scalar alpha=1) const;
  Tensor add(Scalar other, Scalar alpha=1) const;
  Tensor & add_(Scalar other, Scalar alpha=1) const;
  Tensor addmv(const Tensor & mat, const Tensor & vec, Scalar beta=1, Scalar alpha=1) const;
  Tensor & addmv_(const Tensor & mat, const Tensor & vec, Scalar beta=1, Scalar alpha=1) const;
  Tensor addr(const Tensor & vec1, const Tensor & vec2, Scalar beta=1, Scalar alpha=1) const;
  Tensor & addr_(const Tensor & vec1, const Tensor & vec2, Scalar beta=1, Scalar alpha=1) const;
  Tensor all(int64_t dim, bool keepdim=false) const;
  Tensor all(Dimname dim, bool keepdim=false) const;
  bool allclose(const Tensor & other, double rtol=1e-05, double atol=1e-08, bool equal_nan=false) const;
  Tensor any(int64_t dim, bool keepdim=false) const;
  Tensor any(Dimname dim, bool keepdim=false) const;
  Tensor argmax(c10::optional<int64_t> dim=c10::nullopt, bool keepdim=false) const;
  Tensor argmin(c10::optional<int64_t> dim=c10::nullopt, bool keepdim=false) const;
  Tensor acosh() const;
  Tensor & acosh_() const;
  Tensor arccosh() const;
  Tensor & arccosh_() const;
  Tensor asinh() const;
  Tensor & asinh_() const;
  Tensor arcsinh() const;
  Tensor & arcsinh_() const;
  Tensor atanh() const;
  Tensor & atanh_() const;
  Tensor arctanh() const;
  Tensor & arctanh_() const;
  Tensor as_strided(IntArrayRef size, IntArrayRef stride, c10::optional<int64_t> storage_offset=c10::nullopt) const;
  Tensor & as_strided_(IntArrayRef size, IntArrayRef stride, c10::optional<int64_t> storage_offset=c10::nullopt) const;
  Tensor asin() const;
  Tensor & asin_() const;
  Tensor arcsin() const;
  Tensor & arcsin_() const;
  Tensor atan() const;
  Tensor & atan_() const;
  Tensor arctan() const;
  Tensor & arctan_() const;
  Tensor baddbmm(const Tensor & batch1, const Tensor & batch2, Scalar beta=1, Scalar alpha=1) const;
  Tensor & baddbmm_(const Tensor & batch1, const Tensor & batch2, Scalar beta=1, Scalar alpha=1) const;
  Tensor bernoulli(c10::optional<Generator> generator=c10::nullopt) const;
  Tensor & bernoulli_(const Tensor & p, c10::optional<Generator> generator=c10::nullopt) const;
  Tensor & bernoulli_(double p=0.5, c10::optional<Generator> generator=c10::nullopt) const;
  Tensor bernoulli(double p, c10::optional<Generator> generator=c10::nullopt) const;
  Tensor bincount(const c10::optional<Tensor>& weights={}, int64_t minlength=0) const;
  Tensor bitwise_not() const;
  Tensor & bitwise_not_() const;
  Tensor copysign(const Tensor & other) const;
  Tensor & copysign_(const Tensor & other) const;
  Tensor copysign(Scalar other) const;
  Tensor & copysign_(Scalar other) const;
  Tensor logical_not() const;
  Tensor & logical_not_() const;
  Tensor logical_xor(const Tensor & other) const;
  Tensor & logical_xor_(const Tensor & other) const;
  Tensor logical_and(const Tensor & other) const;
  Tensor & logical_and_(const Tensor & other) const;
  Tensor logical_or(const Tensor & other) const;
  Tensor & logical_or_(const Tensor & other) const;
  Tensor bmm(const Tensor & mat2) const;
  Tensor ceil() const;
  Tensor & ceil_() const;
  std::vector<Tensor> unsafe_chunk(int64_t chunks, int64_t dim=0) const;
  std::vector<Tensor> chunk(int64_t chunks, int64_t dim=0) const;
  std::vector<Tensor> tensor_split(int64_t sections, int64_t dim=0) const;
  std::vector<Tensor> tensor_split(IntArrayRef indices, int64_t dim=0) const;
  Tensor clamp(c10::optional<Scalar> min=c10::nullopt, c10::optional<Scalar> max=c10::nullopt) const;
  Tensor & clamp_(c10::optional<Scalar> min=c10::nullopt, c10::optional<Scalar> max=c10::nullopt) const;
  Tensor clamp_max(Scalar max) const;
  Tensor & clamp_max_(Scalar max) const;
  Tensor clamp_min(Scalar min) const;
  Tensor & clamp_min_(Scalar min) const;
  Tensor clip(c10::optional<Scalar> min=c10::nullopt, c10::optional<Scalar> max=c10::nullopt) const;
  Tensor & clip_(c10::optional<Scalar> min=c10::nullopt, c10::optional<Scalar> max=c10::nullopt) const;
  Tensor contiguous(MemoryFormat memory_format=MemoryFormat::Contiguous) const;
  Tensor & copy_(const Tensor & src, bool non_blocking=false) const;
  Tensor cos() const;
  Tensor & cos_() const;
  Tensor cosh() const;
  Tensor & cosh_() const;
  Tensor count_nonzero(IntArrayRef dim) const;
  Tensor count_nonzero(c10::optional<int64_t> dim=c10::nullopt) const;
  std::tuple<Tensor,Tensor> cummax(int64_t dim) const;
  std::tuple<Tensor,Tensor> cummax(Dimname dim) const;
  std::tuple<Tensor,Tensor> cummin(int64_t dim) const;
  std::tuple<Tensor,Tensor> cummin(Dimname dim) const;
  Tensor cumprod(int64_t dim, c10::optional<ScalarType> dtype=c10::nullopt) const;
  Tensor & cumprod_(int64_t dim, c10::optional<ScalarType> dtype=c10::nullopt) const;
  Tensor cumprod(Dimname dim, c10::optional<ScalarType> dtype=c10::nullopt) const;
  Tensor & cumprod_(Dimname dim, c10::optional<ScalarType> dtype=c10::nullopt) const;
  Tensor cumsum(int64_t dim, c10::optional<ScalarType> dtype=c10::nullopt) const;
  Tensor & cumsum_(int64_t dim, c10::optional<ScalarType> dtype=c10::nullopt) const;
  Tensor cumsum(Dimname dim, c10::optional<ScalarType> dtype=c10::nullopt) const;
  Tensor & cumsum_(Dimname dim, c10::optional<ScalarType> dtype=c10::nullopt) const;
  Tensor diag_embed(int64_t offset=0, int64_t dim1=-2, int64_t dim2=-1) const;
  Tensor diagflat(int64_t offset=0) const;
  Tensor diagonal(int64_t offset=0, int64_t dim1=0, int64_t dim2=1) const;
  Tensor diagonal(Dimname outdim, Dimname dim1, Dimname dim2, int64_t offset=0) const;
  Tensor & fill_diagonal_(Scalar fill_value, bool wrap=false) const;
  Tensor div(const Tensor & other) const;
  Tensor & div_(const Tensor & other) const;
  Tensor div(Scalar other) const;
  Tensor & div_(Scalar other) const;
  Tensor divide(const Tensor & other) const;
  Tensor & divide_(const Tensor & other) const;
  Tensor divide(Scalar other) const;
  Tensor & divide_(Scalar other) const;
  Tensor true_divide(const Tensor & other) const;
  Tensor & true_divide_(const Tensor & other) const;
  Tensor true_divide(Scalar other) const;
  Tensor & true_divide_(Scalar other) const;
  Tensor dot(const Tensor & tensor) const;
  Tensor vdot(const Tensor & other) const;
  Tensor new_empty(IntArrayRef size, const TensorOptions & options={}) const;
  Tensor new_empty(IntArrayRef size, c10::optional<ScalarType> dtype, c10::optional<Layout> layout, c10::optional<Device> device, c10::optional<bool> pin_memory) const;
  Tensor new_empty_strided(IntArrayRef size, IntArrayRef stride, const TensorOptions & options={}) const;
  Tensor new_empty_strided(IntArrayRef size, IntArrayRef stride, c10::optional<ScalarType> dtype, c10::optional<Layout> layout, c10::optional<Device> device, c10::optional<bool> pin_memory) const;
  Tensor new_full(IntArrayRef size, Scalar fill_value, const TensorOptions & options={}) const;
  Tensor new_full(IntArrayRef size, Scalar fill_value, c10::optional<ScalarType> dtype, c10::optional<Layout> layout, c10::optional<Device> device, c10::optional<bool> pin_memory) const;
  Tensor new_zeros(IntArrayRef size, const TensorOptions & options={}) const;
  Tensor new_zeros(IntArrayRef size, c10::optional<ScalarType> dtype, c10::optional<Layout> layout, c10::optional<Device> device, c10::optional<bool> pin_memory) const;
  Tensor & resize_(IntArrayRef size, c10::optional<MemoryFormat> memory_format=c10::nullopt) const;
  Tensor erf() const;
  Tensor & erf_() const;
  Tensor erfc() const;
  Tensor & erfc_() const;
  Tensor exp() const;
  Tensor & exp_() const;
  Tensor exp2() const;
  Tensor & exp2_() const;
  Tensor expm1() const;
  Tensor & expm1_() const;
  Tensor expand(IntArrayRef size, bool implicit=false) const;
  Tensor expand_as(const Tensor & other) const;
  Tensor flatten(int64_t start_dim=0, int64_t end_dim=-1) const;
  Tensor flatten(int64_t start_dim, int64_t end_dim, Dimname out_dim) const;
  Tensor flatten(Dimname start_dim, Dimname end_dim, Dimname out_dim) const;
  Tensor flatten(DimnameList dims, Dimname out_dim) const;
  Tensor unflatten(int64_t dim, IntArrayRef sizes, c10::optional<DimnameList> names=c10::nullopt) const;
  Tensor unflatten(Dimname dim, IntArrayRef sizes, DimnameList names) const;
  Tensor & fill_(Scalar value) const;
  Tensor & fill_(const Tensor & value) const;
  Tensor floor() const;
  Tensor & floor_() const;
  Tensor floor_divide(const Tensor & other) const;
  Tensor & floor_divide_(const Tensor & other) const;
  Tensor floor_divide(Scalar other) const;
  Tensor & floor_divide_(Scalar other) const;
  Tensor frac() const;
  Tensor & frac_() const;
  Tensor gcd(const Tensor & other) const;
  Tensor & gcd_(const Tensor & other) const;
  Tensor lcm(const Tensor & other) const;
  Tensor & lcm_(const Tensor & other) const;
  Tensor ifft(int64_t signal_ndim, bool normalized=false) const;
  Tensor rfft(int64_t signal_ndim, bool normalized=false, bool onesided=true) const;
  Tensor irfft(int64_t signal_ndim, bool normalized=false, bool onesided=true, IntArrayRef signal_sizes={}) const;
  Tensor index(TensorList indices) const;
  Tensor & index_copy_(int64_t dim, const Tensor & index, const Tensor & source) const;
  Tensor index_copy(int64_t dim, const Tensor & index, const Tensor & source) const;
  Tensor & index_copy_(Dimname dim, const Tensor & index, const Tensor & source) const;
  Tensor index_copy(Dimname dim, const Tensor & index, const Tensor & source) const;
  Tensor & index_put_(TensorList indices, const Tensor & values, bool accumulate=false) const;
  Tensor index_put(TensorList indices, const Tensor & values, bool accumulate=false) const;
  Tensor inverse() const;
  Tensor isclose(const Tensor & other, double rtol=1e-05, double atol=1e-08, bool equal_nan=false) const;
  Tensor isnan() const;
  bool is_distributed() const;
  bool is_floating_point() const;
  bool is_complex() const;
  Tensor isreal() const;
  bool is_nonzero() const;
  bool is_same_size(const Tensor & other) const;
  bool is_signed() const;
  Tensor kron(const Tensor & other) const;
  std::tuple<Tensor,Tensor> kthvalue(int64_t k, int64_t dim=-1, bool keepdim=false) const;
  std::tuple<Tensor,Tensor> kthvalue(int64_t k, Dimname dim, bool keepdim=false) const;
  Tensor nan_to_num(c10::optional<double> nan=c10::nullopt, c10::optional<double> posinf=c10::nullopt, c10::optional<double> neginf=c10::nullopt) const;
  Tensor & nan_to_num_(c10::optional<double> nan=c10::nullopt, c10::optional<double> posinf=c10::nullopt, c10::optional<double> neginf=c10::nullopt) const;
  Tensor ldexp(const Tensor & other) const;
  Tensor & ldexp_(const Tensor & other) const;
  Tensor log() const;
  Tensor & log_() const;
  Tensor log10() const;
  Tensor & log10_() const;
  Tensor log1p() const;
  Tensor & log1p_() const;
  Tensor log2() const;
  Tensor & log2_() const;
  Tensor logaddexp(const Tensor & other) const;
  Tensor logaddexp2(const Tensor & other) const;
  Tensor logdet() const;
  Tensor log_softmax(int64_t dim, c10::optional<ScalarType> dtype=c10::nullopt) const;
  Tensor log_softmax(Dimname dim, c10::optional<ScalarType> dtype=c10::nullopt) const;
  Tensor logcumsumexp(int64_t dim) const;
  Tensor logcumsumexp(Dimname dim) const;
  Tensor logsumexp(IntArrayRef dim, bool keepdim=false) const;
  Tensor logsumexp(DimnameList dim, bool keepdim=false) const;
  Tensor matmul(const Tensor & other) const;
  Tensor matrix_power(int64_t n) const;
  Tensor matrix_exp() const;
  std::tuple<Tensor,Tensor> max(int64_t dim, bool keepdim=false) const;
  std::tuple<Tensor,Tensor> max(Dimname dim, bool keepdim=false) const;
  Tensor amax(IntArrayRef dim={}, bool keepdim=false) const;
  Tensor mean(c10::optional<ScalarType> dtype=c10::nullopt) const;
  Tensor mean(IntArrayRef dim, bool keepdim=false, c10::optional<ScalarType> dtype=c10::nullopt) const;
  Tensor mean(DimnameList dim, bool keepdim=false, c10::optional<ScalarType> dtype=c10::nullopt) const;
  Tensor median() const;
  std::tuple<Tensor,Tensor> median(int64_t dim, bool keepdim=false) const;
  std::tuple<Tensor,Tensor> median(Dimname dim, bool keepdim=false) const;
  Tensor nanmedian() const;
  std::tuple<Tensor,Tensor> nanmedian(int64_t dim, bool keepdim=false) const;
  std::tuple<Tensor,Tensor> nanmedian(Dimname dim, bool keepdim=false) const;
  std::tuple<Tensor,Tensor> min(int64_t dim, bool keepdim=false) const;
  std::tuple<Tensor,Tensor> min(Dimname dim, bool keepdim=false) const;
  Tensor amin(IntArrayRef dim={}, bool keepdim=false) const;
  Tensor mm(const Tensor & mat2) const;
  std::tuple<Tensor,Tensor> mode(int64_t dim=-1, bool keepdim=false) const;
  std::tuple<Tensor,Tensor> mode(Dimname dim, bool keepdim=false) const;
  Tensor mul(const Tensor & other) const;
  Tensor & mul_(const Tensor & other) const;
  Tensor mul(Scalar other) const;
  Tensor & mul_(Scalar other) const;
  Tensor multiply(const Tensor & other) const;
  Tensor & multiply_(const Tensor & other) const;
  Tensor multiply(Scalar other) const;
  Tensor & multiply_(Scalar other) const;
  Tensor mv(const Tensor & vec) const;
  Tensor mvlgamma(int64_t p) const;
  Tensor & mvlgamma_(int64_t p) const;
  Tensor narrow_copy(int64_t dim, int64_t start, int64_t length) const;
  Tensor narrow(int64_t dim, int64_t start, int64_t length) const;
  Tensor narrow(int64_t dim, const Tensor & start, int64_t length) const;
  Tensor permute(IntArrayRef dims) const;
  Tensor movedim(IntArrayRef source, IntArrayRef destination) const;
  Tensor movedim(int64_t source, int64_t destination) const;
  Tensor numpy_T() const;
  bool is_pinned() const;
  Tensor pin_memory() const;
  Tensor pinverse(double rcond=1e-15) const;
  Tensor rad2deg() const;
  Tensor & rad2deg_() const;
  Tensor deg2rad() const;
  Tensor & deg2rad_() const;
  Tensor ravel() const;
  Tensor reciprocal() const;
  Tensor & reciprocal_() const;
  Tensor neg() const;
  Tensor & neg_() const;
  Tensor negative() const;
  Tensor & negative_() const;
  Tensor repeat(IntArrayRef repeats) const;
  Tensor repeat_interleave(const Tensor & repeats, c10::optional<int64_t> dim=c10::nullopt) const;
  Tensor repeat_interleave(int64_t repeats, c10::optional<int64_t> dim=c10::nullopt) const;
  Tensor reshape(IntArrayRef shape) const;
  Tensor reshape_as(const Tensor & other) const;
  Tensor round() const;
  Tensor & round_() const;
  Tensor relu() const;
  Tensor & relu_() const;
  Tensor prelu(const Tensor & weight) const;
  std::tuple<Tensor,Tensor> prelu_backward(const Tensor & grad_output, const Tensor & weight) const;
  Tensor hardshrink(Scalar lambd=0.5) const;
  Tensor hardshrink_backward(const Tensor & grad_out, Scalar lambd) const;
  Tensor rsqrt() const;
  Tensor & rsqrt_() const;
  Tensor select(Dimname dim, int64_t index) const;
  Tensor select(int64_t dim, int64_t index) const;
  Tensor sigmoid() const;
  Tensor & sigmoid_() const;
  Tensor logit(c10::optional<double> eps=c10::nullopt) const;
  Tensor & logit_(c10::optional<double> eps=c10::nullopt) const;
  Tensor sin() const;
  Tensor & sin_() const;
  Tensor sinh() const;
  Tensor & sinh_() const;
  Tensor detach() const;
  Tensor & detach_() const;
  int64_t size(int64_t dim) const;
  int64_t size(Dimname dim) const;
  Tensor slice(int64_t dim=0, int64_t start=0, int64_t end=9223372036854775807, int64_t step=1) const;
  std::tuple<Tensor,Tensor> slogdet() const;
  Tensor smm(const Tensor & mat2) const;
  Tensor softmax(int64_t dim, c10::optional<ScalarType> dtype=c10::nullopt) const;
  Tensor softmax(Dimname dim, c10::optional<ScalarType> dtype=c10::nullopt) const;
  std::vector<Tensor> unsafe_split(int64_t split_size, int64_t dim=0) const;
  std::vector<Tensor> split(int64_t split_size, int64_t dim=0) const;
  std::vector<Tensor> unsafe_split_with_sizes(IntArrayRef split_sizes, int64_t dim=0) const;
  std::vector<Tensor> split_with_sizes(IntArrayRef split_sizes, int64_t dim=0) const;
  Tensor squeeze() const;
  Tensor squeeze(int64_t dim) const;
  Tensor squeeze(Dimname dim) const;
  Tensor & squeeze_() const;
  Tensor & squeeze_(int64_t dim) const;
  Tensor & squeeze_(Dimname dim) const;
  Tensor sspaddmm(const Tensor & mat1, const Tensor & mat2, Scalar beta=1, Scalar alpha=1) const;
  Tensor stft(int64_t n_fft, c10::optional<int64_t> hop_length=c10::nullopt, c10::optional<int64_t> win_length=c10::nullopt, const c10::optional<Tensor>& window={}, bool normalized=false, c10::optional<bool> onesided=c10::nullopt, c10::optional<bool> return_complex=c10::nullopt) const;
  Tensor istft(int64_t n_fft, c10::optional<int64_t> hop_length=c10::nullopt, c10::optional<int64_t> win_length=c10::nullopt, const c10::optional<Tensor>& window={}, bool center=true, bool normalized=false, c10::optional<bool> onesided=c10::nullopt, c10::optional<int64_t> length=c10::nullopt, bool return_complex=false) const;
  int64_t stride(int64_t dim) const;
  int64_t stride(Dimname dim) const;
  Tensor sum(c10::optional<ScalarType> dtype=c10::nullopt) const;
  Tensor sum(IntArrayRef dim, bool keepdim=false, c10::optional<ScalarType> dtype=c10::nullopt) const;
  Tensor sum(DimnameList dim, bool keepdim=false, c10::optional<ScalarType> dtype=c10::nullopt) const;
  Tensor nansum(c10::optional<ScalarType> dtype=c10::nullopt) const;
  Tensor nansum(IntArrayRef dim, bool keepdim=false, c10::optional<ScalarType> dtype=c10::nullopt) const;
  Tensor sum_to_size(IntArrayRef size) const;
  Tensor sqrt() const;
  Tensor & sqrt_() const;
  Tensor square() const;
  Tensor & square_() const;
  Tensor std(bool unbiased=true) const;
  Tensor std(IntArrayRef dim, bool unbiased=true, bool keepdim=false) const;
  Tensor std(DimnameList dim, bool unbiased=true, bool keepdim=false) const;
  Tensor prod(c10::optional<ScalarType> dtype=c10::nullopt) const;
  Tensor prod(int64_t dim, bool keepdim=false, c10::optional<ScalarType> dtype=c10::nullopt) const;
  Tensor prod(Dimname dim, bool keepdim=false, c10::optional<ScalarType> dtype=c10::nullopt) const;
  Tensor t() const;
  Tensor & t_() const;
  Tensor tan() const;
  Tensor & tan_() const;
  Tensor tanh() const;
  Tensor & tanh_() const;
  Tensor tile(IntArrayRef dims) const;
  Tensor transpose(int64_t dim0, int64_t dim1) const;
  Tensor transpose(Dimname dim0, Dimname dim1) const;
  Tensor & transpose_(int64_t dim0, int64_t dim1) const;
  Tensor flip(IntArrayRef dims) const;
  Tensor fliplr() const;
  Tensor flipud() const;
  Tensor roll(IntArrayRef shifts, IntArrayRef dims={}) const;
  Tensor rot90(int64_t k=1, IntArrayRef dims={0,1}) const;
  Tensor trunc() const;
  Tensor & trunc_() const;
  Tensor fix() const;
  Tensor & fix_() const;
  Tensor type_as(const Tensor & other) const;
  Tensor unsqueeze(int64_t dim) const;
  Tensor & unsqueeze_(int64_t dim) const;
  Tensor var(bool unbiased=true) const;
  Tensor var(IntArrayRef dim, bool unbiased=true, bool keepdim=false) const;
  Tensor var(DimnameList dim, bool unbiased=true, bool keepdim=false) const;
  Tensor view_as(const Tensor & other) const;
  Tensor where(const Tensor & condition, const Tensor & other) const;
  Tensor norm(c10::optional<Scalar> p, ScalarType dtype) const;
  Tensor norm(Scalar p=2) const;
  Tensor norm(c10::optional<Scalar> p, IntArrayRef dim, bool keepdim, ScalarType dtype) const;
  Tensor norm(c10::optional<Scalar> p, IntArrayRef dim, bool keepdim=false) const;
  Tensor norm(c10::optional<Scalar> p, DimnameList dim, bool keepdim, ScalarType dtype) const;
  Tensor norm(c10::optional<Scalar> p, DimnameList dim, bool keepdim=false) const;
  Tensor clone(c10::optional<MemoryFormat> memory_format=c10::nullopt) const;
  Tensor & resize_as_(const Tensor & the_template, c10::optional<MemoryFormat> memory_format=c10::nullopt) const;
  Tensor & zero_() const;
  Tensor sub(const Tensor & other, Scalar alpha=1) const;
  Tensor & sub_(const Tensor & other, Scalar alpha=1) const;
  Tensor sub(Scalar other, Scalar alpha=1) const;
  Tensor & sub_(Scalar other, Scalar alpha=1) const;
  Tensor subtract(const Tensor & other, Scalar alpha=1) const;
  Tensor & subtract_(const Tensor & other, Scalar alpha=1) const;
  Tensor subtract(Scalar other, Scalar alpha=1) const;
  Tensor & subtract_(Scalar other, Scalar alpha=1) const;
  Tensor heaviside(const Tensor & values) const;
  Tensor & heaviside_(const Tensor & values) const;
  Tensor addmm(const Tensor & mat1, const Tensor & mat2, Scalar beta=1, Scalar alpha=1) const;
  Tensor & addmm_(const Tensor & mat1, const Tensor & mat2, Scalar beta=1, Scalar alpha=1) const;
  Tensor & sparse_resize_(IntArrayRef size, int64_t sparse_dim, int64_t dense_dim) const;
  Tensor & sparse_resize_and_clear_(IntArrayRef size, int64_t sparse_dim, int64_t dense_dim) const;
  Tensor sparse_mask(const Tensor & mask) const;
  Tensor to_dense() const;
  int64_t sparse_dim() const;
  int64_t _dimI() const;
  int64_t dense_dim() const;
  int64_t _dimV() const;
  int64_t _nnz() const;
  Tensor coalesce() const;
  bool is_coalesced() const;
  Tensor _indices() const;
  Tensor _values() const;
  Tensor & _coalesced_(bool coalesced) const;
  Tensor indices() const;
  Tensor values() const;
  std::vector<Tensor> unbind(int64_t dim=0) const;
  std::vector<Tensor> unbind(Dimname dim) const;
  Tensor to_sparse(int64_t sparse_dim) const;
  Tensor to_sparse() const;
  Tensor to_mkldnn() const;
  Tensor dequantize() const;
  double q_scale() const;
  int64_t q_zero_point() const;
  Tensor q_per_channel_scales() const;
  Tensor q_per_channel_zero_points() const;
  int64_t q_per_channel_axis() const;
  Tensor int_repr() const;
  QScheme qscheme() const;
  Tensor to(const TensorOptions & options={}, bool non_blocking=false, bool copy=false, c10::optional<MemoryFormat> memory_format=c10::nullopt) const;
  Tensor to(c10::optional<ScalarType> dtype, c10::optional<Layout> layout, c10::optional<Device> device, c10::optional<bool> pin_memory, bool non_blocking, bool copy, c10::optional<MemoryFormat> memory_format) const;
  Tensor to(Device device, ScalarType dtype, bool non_blocking=false, bool copy=false, c10::optional<MemoryFormat> memory_format=c10::nullopt) const;
  Tensor to(ScalarType dtype, bool non_blocking=false, bool copy=false, c10::optional<MemoryFormat> memory_format=c10::nullopt) const;
  Tensor to(const Tensor & other, bool non_blocking=false, bool copy=false, c10::optional<MemoryFormat> memory_format=c10::nullopt) const;
  Scalar item() const;
  Tensor & set_(Storage source) const;
  Tensor & set_(Storage source, int64_t storage_offset, IntArrayRef size, IntArrayRef stride={}) const;
  Tensor & set_(const Tensor & source) const;
  Tensor & set_() const;
  Tensor & set_quantizer_(ConstQuantizerPtr quantizer) const;
  bool is_set_to(const Tensor & tensor) const;
  Tensor & masked_fill_(const Tensor & mask, Scalar value) const;
  Tensor masked_fill(const Tensor & mask, Scalar value) const;
  Tensor & masked_fill_(const Tensor & mask, const Tensor & value) const;
  Tensor masked_fill(const Tensor & mask, const Tensor & value) const;
  Tensor & masked_scatter_(const Tensor & mask, const Tensor & source) const;
  Tensor masked_scatter(const Tensor & mask, const Tensor & source) const;
  Tensor view(IntArrayRef size) const;
  Tensor & put_(const Tensor & index, const Tensor & source, bool accumulate=false) const;
  Tensor & index_add_(int64_t dim, const Tensor & index, const Tensor & source) const;
  Tensor index_add(int64_t dim, const Tensor & index, const Tensor & source) const;
  Tensor index_add(Dimname dim, const Tensor & index, const Tensor & source) const;
  Tensor & index_fill_(int64_t dim, const Tensor & index, Scalar value) const;
  Tensor index_fill(int64_t dim, const Tensor & index, Scalar value) const;
  Tensor & index_fill_(int64_t dim, const Tensor & index, const Tensor & value) const;
  Tensor index_fill(int64_t dim, const Tensor & index, const Tensor & value) const;
  Tensor & index_fill_(Dimname dim, const Tensor & index, Scalar value) const;
  Tensor & index_fill_(Dimname dim, const Tensor & index, const Tensor & value) const;
  Tensor index_fill(Dimname dim, const Tensor & index, Scalar value) const;
  Tensor index_fill(Dimname dim, const Tensor & index, const Tensor & value) const;
  Tensor & scatter_(int64_t dim, const Tensor & index, const Tensor & src) const;
  Tensor scatter(int64_t dim, const Tensor & index, const Tensor & src) const;
  Tensor & scatter_(int64_t dim, const Tensor & index, Scalar value) const;
  Tensor scatter(int64_t dim, const Tensor & index, Scalar value) const;
  Tensor scatter(Dimname dim, const Tensor & index, const Tensor & src) const;
  Tensor scatter(Dimname dim, const Tensor & index, Scalar value) const;
  Tensor & scatter_(int64_t dim, const Tensor & index, const Tensor & src, std::string reduce) const;
  Tensor & scatter_(int64_t dim, const Tensor & index, Scalar value, std::string reduce) const;
  Tensor & scatter_add_(int64_t dim, const Tensor & index, const Tensor & src) const;
  Tensor scatter_add(int64_t dim, const Tensor & index, const Tensor & src) const;
  Tensor scatter_add(Dimname dim, const Tensor & index, const Tensor & src) const;
  Tensor & eq_(Scalar other) const;
  Tensor & eq_(const Tensor & other) const;
  Tensor bitwise_and(Scalar other) const;
  Tensor bitwise_and(const Tensor & other) const;
  Tensor & bitwise_and_(Scalar other) const;
  Tensor & bitwise_and_(const Tensor & other) const;
  Tensor __and__(Scalar other) const;
  Tensor __and__(const Tensor & other) const;
  Tensor & __iand__(Scalar other) const;
  Tensor & __iand__(const Tensor & other) const;
  Tensor bitwise_or(Scalar other) const;
  Tensor bitwise_or(const Tensor & other) const;
  Tensor & bitwise_or_(Scalar other) const;
  Tensor & bitwise_or_(const Tensor & other) const;
  Tensor __or__(Scalar other) const;
  Tensor __or__(const Tensor & other) const;
  Tensor & __ior__(Scalar other) const;
  Tensor & __ior__(const Tensor & other) const;
  Tensor bitwise_xor(Scalar other) const;
  Tensor bitwise_xor(const Tensor & other) const;
  Tensor & bitwise_xor_(Scalar other) const;
  Tensor & bitwise_xor_(const Tensor & other) const;
  Tensor __xor__(Scalar other) const;
  Tensor __xor__(const Tensor & other) const;
  Tensor & __ixor__(Scalar other) const;
  Tensor & __ixor__(const Tensor & other) const;
  Tensor __lshift__(Scalar other) const;
  Tensor __lshift__(const Tensor & other) const;
  Tensor & __ilshift__(Scalar other) const;
  Tensor & __ilshift__(const Tensor & other) const;
  Tensor __rshift__(Scalar other) const;
  Tensor __rshift__(const Tensor & other) const;
  Tensor & __irshift__(Scalar other) const;
  Tensor & __irshift__(const Tensor & other) const;
  Tensor & lgamma_() const;
  Tensor & atan2_(const Tensor & other) const;
  Tensor & tril_(int64_t diagonal=0) const;
  Tensor & triu_(int64_t diagonal=0) const;
  Tensor & digamma_() const;
  Tensor & polygamma_(int64_t n) const;
  Tensor & renorm_(Scalar p, int64_t dim, Scalar maxnorm) const;
  Tensor & pow_(Scalar exponent) const;
  Tensor & pow_(const Tensor & exponent) const;
  Tensor & lerp_(const Tensor & end, Scalar weight) const;
  Tensor & lerp_(const Tensor & end, const Tensor & weight) const;
  Tensor & fmod_(Scalar other) const;
  Tensor & fmod_(const Tensor & other) const;
  Tensor & remainder_(Scalar other) const;
  Tensor & remainder_(const Tensor & other) const;
  Tensor & addbmm_(const Tensor & batch1, const Tensor & batch2, Scalar beta=1, Scalar alpha=1) const;
  Tensor addbmm(const Tensor & batch1, const Tensor & batch2, Scalar beta=1, Scalar alpha=1) const;
  Tensor & addcdiv_(const Tensor & tensor1, const Tensor & tensor2, Scalar value=1) const;
  Tensor & random_(int64_t from, c10::optional<int64_t> to, c10::optional<Generator> generator=c10::nullopt) const;
  Tensor & random_(int64_t to, c10::optional<Generator> generator=c10::nullopt) const;
  Tensor & random_(c10::optional<Generator> generator=c10::nullopt) const;
  Tensor & uniform_(double from=0, double to=1, c10::optional<Generator> generator=c10::nullopt) const;
  Tensor & cauchy_(double median=0, double sigma=1, c10::optional<Generator> generator=c10::nullopt) const;
  Tensor & log_normal_(double mean=1, double std=2, c10::optional<Generator> generator=c10::nullopt) const;
  Tensor & exponential_(double lambd=1, c10::optional<Generator> generator=c10::nullopt) const;
  Tensor & geometric_(double p, c10::optional<Generator> generator=c10::nullopt) const;
  Tensor diag(int64_t diagonal=0) const;
  Tensor cross(const Tensor & other, c10::optional<int64_t> dim=c10::nullopt) const;
  Tensor triu(int64_t diagonal=0) const;
  Tensor tril(int64_t diagonal=0) const;
  Tensor trace() const;
  Tensor ne(Scalar other) const;
  Tensor ne(const Tensor & other) const;
  Tensor & ne_(Scalar other) const;
  Tensor & ne_(const Tensor & other) const;
  Tensor not_equal(Scalar other) const;
  Tensor not_equal(const Tensor & other) const;
  Tensor & not_equal_(Scalar other) const;
  Tensor & not_equal_(const Tensor & other) const;
  Tensor eq(Scalar other) const;
  Tensor eq(const Tensor & other) const;
  Tensor ge(Scalar other) const;
  Tensor ge(const Tensor & other) const;
  Tensor & ge_(Scalar other) const;
  Tensor & ge_(const Tensor & other) const;
  Tensor greater_equal(Scalar other) const;
  Tensor greater_equal(const Tensor & other) const;
  Tensor & greater_equal_(Scalar other) const;
  Tensor & greater_equal_(const Tensor & other) const;
  Tensor le(Scalar other) const;
  Tensor le(const Tensor & other) const;
  Tensor & le_(Scalar other) const;
  Tensor & le_(const Tensor & other) const;
  Tensor less_equal(Scalar other) const;
  Tensor less_equal(const Tensor & other) const;
  Tensor & less_equal_(Scalar other) const;
  Tensor & less_equal_(const Tensor & other) const;
  Tensor gt(Scalar other) const;
  Tensor gt(const Tensor & other) const;
  Tensor & gt_(Scalar other) const;
  Tensor & gt_(const Tensor & other) const;
  Tensor greater(Scalar other) const;
  Tensor greater(const Tensor & other) const;
  Tensor & greater_(Scalar other) const;
  Tensor & greater_(const Tensor & other) const;
  Tensor lt(Scalar other) const;
  Tensor lt(const Tensor & other) const;
  Tensor & lt_(Scalar other) const;
  Tensor & lt_(const Tensor & other) const;
  Tensor less(Scalar other) const;
  Tensor less(const Tensor & other) const;
  Tensor & less_(Scalar other) const;
  Tensor & less_(const Tensor & other) const;
  Tensor take(const Tensor & index) const;
  Tensor index_select(int64_t dim, const Tensor & index) const;
  Tensor index_select(Dimname dim, const Tensor & index) const;
  Tensor masked_select(const Tensor & mask) const;
  Tensor nonzero() const;
  std::vector<Tensor> nonzero_numpy() const;
  Tensor gather(int64_t dim, const Tensor & index, bool sparse_grad=false) const;
  Tensor gather(Dimname dim, const Tensor & index, bool sparse_grad=false) const;
  Tensor addcmul(const Tensor & tensor1, const Tensor & tensor2, Scalar value=1) const;
  Tensor & addcmul_(const Tensor & tensor1, const Tensor & tensor2, Scalar value=1) const;
  Tensor addcdiv(const Tensor & tensor1, const Tensor & tensor2, Scalar value=1) const;
  std::tuple<Tensor,Tensor> lstsq(const Tensor & A) const;
  std::tuple<Tensor,Tensor> triangular_solve(const Tensor & A, bool upper=true, bool transpose=false, bool unitriangular=false) const;
  std::tuple<Tensor,Tensor> symeig(bool eigenvectors=false, bool upper=true) const;
  std::tuple<Tensor,Tensor> eig(bool eigenvectors=false) const;
  std::tuple<Tensor,Tensor,Tensor> svd(bool some=true, bool compute_uv=true) const;
  Tensor swapaxes(int64_t axis0, int64_t axis1) const;
  Tensor & swapaxes_(int64_t axis0, int64_t axis1) const;
  Tensor swapdims(int64_t dim0, int64_t dim1) const;
  Tensor & swapdims_(int64_t dim0, int64_t dim1) const;
  Tensor cholesky(bool upper=false) const;
  Tensor cholesky_solve(const Tensor & input2, bool upper=false) const;
  std::tuple<Tensor,Tensor> solve(const Tensor & A) const;
  Tensor cholesky_inverse(bool upper=false) const;
  std::tuple<Tensor,Tensor> qr(bool some=true) const;
  std::tuple<Tensor,Tensor> geqrf() const;
  Tensor orgqr(const Tensor & input2) const;
  Tensor ormqr(const Tensor & input2, const Tensor & input3, bool left=true, bool transpose=false) const;
  Tensor lu_solve(const Tensor & LU_data, const Tensor & LU_pivots) const;
  Tensor multinomial(int64_t num_samples, bool replacement=false, c10::optional<Generator> generator=c10::nullopt) const;
  Tensor lgamma() const;
  Tensor digamma() const;
  Tensor polygamma(int64_t n) const;
  Tensor erfinv() const;
  Tensor & erfinv_() const;
  Tensor i0() const;
  Tensor & i0_() const;
  Tensor sign() const;
  Tensor & sign_() const;
  Tensor signbit() const;
  Tensor dist(const Tensor & other, Scalar p=2) const;
  Tensor atan2(const Tensor & other) const;
  Tensor lerp(const Tensor & end, Scalar weight) const;
  Tensor lerp(const Tensor & end, const Tensor & weight) const;
  Tensor histc(int64_t bins=100, Scalar min=0, Scalar max=0) const;
  Tensor fmod(Scalar other) const;
  Tensor fmod(const Tensor & other) const;
  Tensor hypot(const Tensor & other) const;
  Tensor & hypot_(const Tensor & other) const;
  Tensor igamma(const Tensor & other) const;
  Tensor & igamma_(const Tensor & other) const;
  Tensor igammac(const Tensor & other) const;
  Tensor & igammac_(const Tensor & other) const;
  Tensor nextafter(const Tensor & other) const;
  Tensor & nextafter_(const Tensor & other) const;
  Tensor remainder(Scalar other) const;
  Tensor remainder(const Tensor & other) const;
  Tensor min() const;
  Tensor max() const;
  Tensor maximum(const Tensor & other) const;
  Tensor max(const Tensor & other) const;
  Tensor minimum(const Tensor & other) const;
  Tensor min(const Tensor & other) const;
  Tensor quantile(double q, c10::optional<int64_t> dim=c10::nullopt, bool keepdim=false) const;
  Tensor quantile(const Tensor & q, c10::optional<int64_t> dim=c10::nullopt, bool keepdim=false) const;
  Tensor nanquantile(double q, c10::optional<int64_t> dim=c10::nullopt, bool keepdim=false) const;
  Tensor nanquantile(const Tensor & q, c10::optional<int64_t> dim=c10::nullopt, bool keepdim=false) const;
  std::tuple<Tensor,Tensor> sort(int64_t dim=-1, bool descending=false) const;
  std::tuple<Tensor,Tensor> sort(Dimname dim, bool descending=false) const;
  Tensor argsort(int64_t dim=-1, bool descending=false) const;
  Tensor argsort(Dimname dim, bool descending=false) const;
  std::tuple<Tensor,Tensor> topk(int64_t k, int64_t dim=-1, bool largest=true, bool sorted=true) const;
  Tensor all() const;
  Tensor any() const;
  Tensor renorm(Scalar p, int64_t dim, Scalar maxnorm) const;
  Tensor unfold(int64_t dimension, int64_t size, int64_t step) const;
  bool equal(const Tensor & other) const;
  Tensor pow(const Tensor & exponent) const;
  Tensor pow(Scalar exponent) const;
  Tensor float_power(const Tensor & exponent) const;
  Tensor float_power(Scalar exponent) const;
  Tensor & float_power_(Scalar exponent) const;
  Tensor & float_power_(const Tensor & exponent) const;
  Tensor & normal_(double mean=0, double std=1, c10::optional<Generator> generator=c10::nullopt) const;
  Tensor alias() const;
  Tensor isfinite() const;
  Tensor isinf() const;
  void record_stream(Stream s) const;
  Tensor isposinf() const;
  Tensor isneginf() const;
  Tensor fft(int64_t signal_ndim, bool normalized=false) const;
  Tensor det() const;
  Tensor outer(const Tensor & vec2) const;
  Tensor ger(const Tensor & vec2) const;

  // 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

  Tensor var(int dim) const {
    return var(IntArrayRef{dim});
  }

  Tensor std(int dim) const {
    return std(IntArrayRef{dim});
  }

  // We changed .dtype() to return a TypeMeta in #12766. Ideally, we want the
  // at::kDouble and its friends to be TypeMeta's, but that hasn't happened yet.
  // Before that change, we make this method to maintain BC for C++ usage like
  // `x.to(y.dtype)`.
  // TODO: remove following two after at::kDouble and its friends are TypeMeta's.
  inline Tensor to(caffe2::TypeMeta type_meta, bool non_blocking=false, bool copy=false) const {
    return this->to(/*scalar_type=*/typeMetaToScalarType(type_meta), non_blocking, copy);
  }
  inline Tensor to(Device device, caffe2::TypeMeta type_meta, bool non_blocking=false, bool copy=false) const {
    return this->to(device, /*scalar_type=*/typeMetaToScalarType(type_meta), non_blocking, copy);
  }

  template <typename F, typename... Args>
  decltype(auto) m(F func, Args&&... params) const {
    return func(*this, std::forward<Args>(params)...);
  }

  at::Tensor tensor_data() const;

  at::Tensor variable_data() const;

  // Gradient Node and Edges
  //~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~

  const std::shared_ptr<torch::autograd::Node>& grad_fn() const;

  // Hooks
  //~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~

  template <typename T>
  using hook_return_void_t = std::enable_if_t<std::is_void<typename std::result_of<T&(Tensor)>::type>::value, unsigned>;
  template <typename T>
  using hook_return_var_t = std::enable_if_t<std::is_same<typename std::result_of<T&(Tensor)>::type, Tensor>::value, unsigned>;

  template <typename T>
  hook_return_void_t<T> register_hook(T&& hook) const;
  template <typename T>
  hook_return_var_t<T> register_hook(T&& hook) const;

private:
  unsigned _register_hook(std::function<Tensor(const Tensor&)> hook) const;

public:

  void remove_hook(unsigned pos) const;

  // View Variables
  //~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~

  bool is_view() const;

  const Tensor& _base() const;

  // Miscellaneous
  //~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~

  const std::string& name() const;

protected:
  friend class ::caffe2::Tensor;

  void enforce_invariants();
  c10::intrusive_ptr<TensorImpl, UndefinedTensorImpl> impl_;
};

int64_t get_device(Tensor self);

template <typename T>
auto Tensor::register_hook(T&& hook) const -> Tensor::hook_return_void_t<T> {
  // Return the grad argument in case of a hook with void return type to have an
  // std::function with Tensor return type
  std::function<void(Tensor)> fn(hook);
  return _register_hook([fn](const Tensor& grad) {
    fn(grad);
    return Tensor();
  });
}

template <typename T>
auto Tensor::register_hook(T&& hook) const -> Tensor::hook_return_var_t<T> {
  return _register_hook(hook);
}

namespace detail {
// Helper creator for Tensor class which doesn't requires the users to pass
// in an intrusive_ptr instead it just converts the argument passed to
// requested intrusive_ptr type.
template <typename T, typename... Args>
Tensor make_tensor(Args&&... args) {
  return Tensor(c10::make_intrusive<T>(std::forward<Args>(args)...));
}

} // namespace detail

static inline DispatchKey legacyExtractDispatchKey(const Tensor& t) {
  return legacyExtractDispatchKey(t.key_set());
}

} // namespace at

Docs

Access comprehensive developer documentation for PyTorch

View Docs

Tutorials

Get in-depth tutorials for beginners and advanced developers

View Tutorials

Resources

Find development resources and get your questions answered

View Resources