Add files using upload-large-folder tool

.gitattributes

@@ -95,3 +95,9 @@ phivenv/Lib/site-packages/sympy/solvers/ode/__pycache__/single.cpython-39.pyc fi
 phivenv/Lib/site-packages/sympy/solvers/tests/__pycache__/test_solvers.cpython-39.pyc filter=lfs diff=lfs merge=lfs -text
 phivenv/Lib/site-packages/sympy/solvers/tests/__pycache__/test_solveset.cpython-39.pyc filter=lfs diff=lfs merge=lfs -text
 phivenv/Lib/site-packages/sympy/solvers/__pycache__/solvers.cpython-39.pyc filter=lfs diff=lfs merge=lfs -text
+phivenv/Lib/site-packages/sympy/solvers/__pycache__/solveset.cpython-39.pyc filter=lfs diff=lfs merge=lfs -text
+phivenv/Lib/site-packages/sympy/stats/__pycache__/crv_types.cpython-39.pyc filter=lfs diff=lfs merge=lfs -text
+phivenv/Lib/site-packages/sympy/tensor/__pycache__/tensor.cpython-39.pyc filter=lfs diff=lfs merge=lfs -text
+phivenv/Lib/site-packages/sympy/utilities/tests/__pycache__/test_wester.cpython-39.pyc filter=lfs diff=lfs merge=lfs -text
+phivenv/Lib/site-packages/tokenizers/tokenizers.pyd filter=lfs diff=lfs merge=lfs -text
+phivenv/Lib/site-packages/torch/bin/asmjit.dll filter=lfs diff=lfs merge=lfs -text

phivenv/Lib/site-packages/sympy/solvers/__pycache__/solveset.cpython-39.pyc

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:9342975557746fd4ee46ca1aa3b9eb809dec3b528282fb743545d529d14dc829
+size 112518

phivenv/Lib/site-packages/sympy/stats/__pycache__/crv_types.cpython-39.pyc

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:263494879393025e37dfdb6734306b383117e0c6da4214c09154aec80e053ea0
+size 132379

phivenv/Lib/site-packages/sympy/tensor/__pycache__/tensor.cpython-39.pyc

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:dbf8730a25907bc5f0428fe546078d92236f12aa51fbfb86bdbd78601614212a
+size 162864

phivenv/Lib/site-packages/sympy/utilities/tests/__pycache__/test_wester.cpython-39.pyc

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:866f6d06f4c81ec07ab4283bdfd4ee8c6be33597944dc13394190625466e603c
+size 112971

phivenv/Lib/site-packages/tokenizers/tokenizers.pyd

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:4e34f54071755a49a90345780b9170ddc3996b29a4c39d3a1fdc4867327f0ef0
+size 7669248

phivenv/Lib/site-packages/torch/bin/asmjit.dll

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:6019433264046fb55b275b0f803d706cf705af3ac06ea97f087b0aa49aacdddd
+size 367104

phivenv/Lib/site-packages/torch/include/ATen/ops/_cummin_helper.h

@@ -0,0 +1,31 @@
+#pragma once
+
+// @generated by torchgen/gen.py from Function.h
+
+#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 <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 <string_view>
+
+
+
+#include <ATen/ops/_cummin_helper_ops.h>
+
+namespace at {
+
+
+// aten::_cummin_helper(Tensor self, Tensor(a!) values, Tensor(b!) indices, int dim) -> ()
+inline void _cummin_helper(const at::Tensor & self, at::Tensor & values, at::Tensor & indices, int64_t dim) {
+    return at::_ops::_cummin_helper::call(self, values, indices, dim);
+}
+
+}

phivenv/Lib/site-packages/torch/include/ATen/ops/_cummin_helper_cpu_dispatch.h

@@ -0,0 +1,23 @@
+#pragma once
+// @generated by torchgen/gen.py from DispatchKeyFunction.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>
+
+// 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>
+
+namespace at {
+
+namespace cpu {
+
+TORCH_API void _cummin_helper(const at::Tensor & self, at::Tensor & values, at::Tensor & indices, int64_t dim);
+
+} // namespace cpu
+} // namespace at

phivenv/Lib/site-packages/torch/include/ATen/ops/_cummin_helper_cuda_dispatch.h

@@ -0,0 +1,23 @@
+#pragma once
+// @generated by torchgen/gen.py from DispatchKeyFunction.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>
+
+// 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>
+
+namespace at {
+
+namespace cuda {
+
+TORCH_API void _cummin_helper(const at::Tensor & self, at::Tensor & values, at::Tensor & indices, int64_t dim);
+
+} // namespace cuda
+} // namespace at

phivenv/Lib/site-packages/torch/include/ATen/ops/_cummin_helper_native.h

@@ -0,0 +1,22 @@
+#pragma once
+
+// @generated by torchgen/gen.py from NativeFunction.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/core/QScheme.h>
+#include <ATen/core/Reduction.h>
+#include <ATen/core/Tensor.h>
+#include <tuple>
+#include <vector>
+
+
+namespace at {
+namespace native {
+TORCH_API void cummin_helper_cpu(const at::Tensor & self, at::Tensor & values, at::Tensor & indices, int64_t dim);
+TORCH_API void cummin_helper_cuda(const at::Tensor & self, at::Tensor & values, at::Tensor & indices, int64_t dim);
+} // namespace native
+} // namespace at

phivenv/Lib/site-packages/torch/include/ATen/ops/_cummin_helper_ops.h

@@ -0,0 +1,29 @@
+#pragma once
+
+// @generated by torchgen/gen.py from Operator.h
+
+#include <string_view>
+#include <tuple>
+#include <vector>
+
+// 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>
+
+namespace at {
+namespace _ops {
+
+
+struct TORCH_API _cummin_helper {
+  using schema = void (const at::Tensor &, at::Tensor &, at::Tensor &, int64_t);
+  using ptr_schema = schema*;
+  // See Note [static constexpr char* members for windows NVCC]
+  static constexpr const char* name = "aten::_cummin_helper";
+  static constexpr const char* overload_name = "";
+  static constexpr const char* schema_str = "_cummin_helper(Tensor self, Tensor(a!) values, Tensor(b!) indices, int dim) -> ()";
+  static void call(const at::Tensor & self, at::Tensor & values, at::Tensor & indices, int64_t dim);
+  static void redispatch(c10::DispatchKeySet dispatchKeySet, const at::Tensor & self, at::Tensor & values, at::Tensor & indices, int64_t dim);
+};
+
+}} // namespace at::_ops

phivenv/Lib/site-packages/torch/include/ATen/ops/_debug_has_internal_overlap.h

@@ -0,0 +1,31 @@
+#pragma once
+
+// @generated by torchgen/gen.py from Function.h
+
+#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 <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 <string_view>
+
+
+
+#include <ATen/ops/_debug_has_internal_overlap_ops.h>
+
+namespace at {
+
+
+// aten::_debug_has_internal_overlap(Tensor self) -> int
+inline int64_t _debug_has_internal_overlap(const at::Tensor & self) {
+    return at::_ops::_debug_has_internal_overlap::call(self);
+}
+
+}

phivenv/Lib/site-packages/torch/include/ATen/ops/_debug_has_internal_overlap_compositeimplicitautograd_dispatch.h

@@ -0,0 +1,23 @@
+#pragma once
+// @generated by torchgen/gen.py from DispatchKeyFunction.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>
+
+// 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>
+
+namespace at {
+
+namespace compositeimplicitautograd {
+
+TORCH_API int64_t _debug_has_internal_overlap(const at::Tensor & self);
+
+} // namespace compositeimplicitautograd
+} // namespace at

phivenv/Lib/site-packages/torch/include/ATen/ops/_debug_has_internal_overlap_native.h

@@ -0,0 +1,21 @@
+#pragma once
+
+// @generated by torchgen/gen.py from NativeFunction.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/core/QScheme.h>
+#include <ATen/core/Reduction.h>
+#include <ATen/core/Tensor.h>
+#include <tuple>
+#include <vector>
+
+
+namespace at {
+namespace native {
+TORCH_API int64_t _debug_has_internal_overlap(const at::Tensor & self);
+} // namespace native
+} // namespace at

phivenv/Lib/site-packages/torch/include/ATen/ops/_debug_has_internal_overlap_ops.h

@@ -0,0 +1,29 @@
+#pragma once
+
+// @generated by torchgen/gen.py from Operator.h
+
+#include <string_view>
+#include <tuple>
+#include <vector>
+
+// 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>
+
+namespace at {
+namespace _ops {
+
+
+struct TORCH_API _debug_has_internal_overlap {
+  using schema = int64_t (const at::Tensor &);
+  using ptr_schema = schema*;
+  // See Note [static constexpr char* members for windows NVCC]
+  static constexpr const char* name = "aten::_debug_has_internal_overlap";
+  static constexpr const char* overload_name = "";
+  static constexpr const char* schema_str = "_debug_has_internal_overlap(Tensor self) -> int";
+  static int64_t call(const at::Tensor & self);
+  static int64_t redispatch(c10::DispatchKeySet dispatchKeySet, const at::Tensor & self);
+};
+
+}} // namespace at::_ops

phivenv/Lib/site-packages/torch/include/ATen/ops/_dimI.h

@@ -0,0 +1,27 @@
+#pragma once
+
+// @generated by torchgen/gen.py from Function.h
+
+#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 <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 <string_view>
+
+
+
+#include <ATen/ops/_dimI_ops.h>
+
+namespace at {
+
+
+
+}

phivenv/Lib/site-packages/torch/include/ATen/ops/_dimI_native.h

@@ -0,0 +1,21 @@
+#pragma once
+
+// @generated by torchgen/gen.py from NativeFunction.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/core/QScheme.h>
+#include <ATen/core/Reduction.h>
+#include <ATen/core/Tensor.h>
+#include <tuple>
+#include <vector>
+
+
+namespace at {
+namespace native {
+TORCH_API int64_t sparse_dim_sparse(const at::Tensor & self);
+} // namespace native
+} // namespace at

phivenv/Lib/site-packages/torch/include/ATen/ops/_dimI_ops.h

@@ -0,0 +1,29 @@
+#pragma once
+
+// @generated by torchgen/gen.py from Operator.h
+
+#include <string_view>
+#include <tuple>
+#include <vector>
+
+// 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>
+
+namespace at {
+namespace _ops {
+
+
+struct TORCH_API _dimI {
+  using schema = int64_t (const at::Tensor &);
+  using ptr_schema = schema*;
+  // See Note [static constexpr char* members for windows NVCC]
+  static constexpr const char* name = "aten::_dimI";
+  static constexpr const char* overload_name = "";
+  static constexpr const char* schema_str = "_dimI(Tensor self) -> int";
+  static int64_t call(const at::Tensor & self);
+  static int64_t redispatch(c10::DispatchKeySet dispatchKeySet, const at::Tensor & self);
+};
+
+}} // namespace at::_ops

phivenv/Lib/site-packages/torch/include/ATen/ops/_dimV.h

@@ -0,0 +1,27 @@
+#pragma once
+
+// @generated by torchgen/gen.py from Function.h
+
+#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 <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 <string_view>
+
+
+
+#include <ATen/ops/_dimV_ops.h>
+
+namespace at {
+
+
+
+}

phivenv/Lib/site-packages/torch/include/ATen/ops/_dimV_native.h

@@ -0,0 +1,21 @@
+#pragma once
+
+// @generated by torchgen/gen.py from NativeFunction.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/core/QScheme.h>
+#include <ATen/core/Reduction.h>
+#include <ATen/core/Tensor.h>
+#include <tuple>
+#include <vector>
+
+
+namespace at {
+namespace native {
+TORCH_API int64_t dense_dim_sparse(const at::Tensor & self);
+} // namespace native
+} // namespace at

phivenv/Lib/site-packages/torch/include/ATen/ops/_dimV_ops.h

@@ -0,0 +1,29 @@
+#pragma once
+
+// @generated by torchgen/gen.py from Operator.h
+
+#include <string_view>
+#include <tuple>
+#include <vector>
+
+// 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>
+
+namespace at {
+namespace _ops {
+
+
+struct TORCH_API _dimV {
+  using schema = int64_t (const at::Tensor &);
+  using ptr_schema = schema*;
+  // See Note [static constexpr char* members for windows NVCC]
+  static constexpr const char* name = "aten::_dimV";
+  static constexpr const char* overload_name = "";
+  static constexpr const char* schema_str = "_dimV(Tensor self) -> int";
+  static int64_t call(const at::Tensor & self);
+  static int64_t redispatch(c10::DispatchKeySet dispatchKeySet, const at::Tensor & self);
+};
+
+}} // namespace at::_ops

phivenv/Lib/site-packages/torch/include/ATen/ops/_dim_arange.h

@@ -0,0 +1,31 @@
+#pragma once
+
+// @generated by torchgen/gen.py from Function.h
+
+#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 <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 <string_view>
+
+
+
+#include <ATen/ops/_dim_arange_ops.h>
+
+namespace at {
+
+
+// aten::_dim_arange(Tensor like, int dim) -> Tensor
+inline at::Tensor _dim_arange(const at::Tensor & like, int64_t dim) {
+    return at::_ops::_dim_arange::call(like, dim);
+}
+
+}

phivenv/Lib/site-packages/torch/include/ATen/ops/_dim_arange_compositeimplicitautograd_dispatch.h

@@ -0,0 +1,23 @@
+#pragma once
+// @generated by torchgen/gen.py from DispatchKeyFunction.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>
+
+// 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>
+
+namespace at {
+
+namespace compositeimplicitautograd {
+
+TORCH_API at::Tensor _dim_arange(const at::Tensor & like, int64_t dim);
+
+} // namespace compositeimplicitautograd
+} // namespace at

phivenv/Lib/site-packages/torch/include/ATen/ops/_dim_arange_native.h

@@ -0,0 +1,21 @@
+#pragma once
+
+// @generated by torchgen/gen.py from NativeFunction.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/core/QScheme.h>
+#include <ATen/core/Reduction.h>
+#include <ATen/core/Tensor.h>
+#include <tuple>
+#include <vector>
+
+
+namespace at {
+namespace native {
+TORCH_API at::Tensor _dim_arange(const at::Tensor & like, int64_t dim);
+} // namespace native
+} // namespace at

phivenv/Lib/site-packages/torch/include/ATen/ops/_dim_arange_ops.h

@@ -0,0 +1,29 @@
+#pragma once
+
+// @generated by torchgen/gen.py from Operator.h
+
+#include <string_view>
+#include <tuple>
+#include <vector>
+
+// 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>
+
+namespace at {
+namespace _ops {
+
+
+struct TORCH_API _dim_arange {
+  using schema = at::Tensor (const at::Tensor &, int64_t);
+  using ptr_schema = schema*;
+  // See Note [static constexpr char* members for windows NVCC]
+  static constexpr const char* name = "aten::_dim_arange";
+  static constexpr const char* overload_name = "";
+  static constexpr const char* schema_str = "_dim_arange(Tensor like, int dim) -> Tensor";
+  static at::Tensor call(const at::Tensor & like, int64_t dim);
+  static at::Tensor redispatch(c10::DispatchKeySet dispatchKeySet, const at::Tensor & like, int64_t dim);
+};
+
+}} // namespace at::_ops

phivenv/Lib/site-packages/torch/include/ATen/ops/_dirichlet_grad.h

@@ -0,0 +1,40 @@
+#pragma once
+
+// @generated by torchgen/gen.py from Function.h
+
+#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 <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 <string_view>
+
+
+
+#include <ATen/ops/_dirichlet_grad_ops.h>
+
+namespace at {
+
+
+// aten::_dirichlet_grad(Tensor x, Tensor alpha, Tensor total) -> Tensor
+inline at::Tensor _dirichlet_grad(const at::Tensor & x, const at::Tensor & alpha, const at::Tensor & total) {
+    return at::_ops::_dirichlet_grad::call(x, alpha, total);
+}
+
+// aten::_dirichlet_grad.out(Tensor x, Tensor alpha, Tensor total, *, Tensor(a!) out) -> Tensor(a!)
+inline at::Tensor & _dirichlet_grad_out(at::Tensor & out, const at::Tensor & x, const at::Tensor & alpha, const at::Tensor & total) {
+    return at::_ops::_dirichlet_grad_out::call(x, alpha, total, out);
+}
+// aten::_dirichlet_grad.out(Tensor x, Tensor alpha, Tensor total, *, Tensor(a!) out) -> Tensor(a!)
+inline at::Tensor & _dirichlet_grad_outf(const at::Tensor & x, const at::Tensor & alpha, const at::Tensor & total, at::Tensor & out) {
+    return at::_ops::_dirichlet_grad_out::call(x, alpha, total, out);
+}
+
+}

phivenv/Lib/site-packages/torch/include/ATen/ops/_dirichlet_grad_compositeexplicitautograd_dispatch.h

@@ -0,0 +1,24 @@
+#pragma once
+// @generated by torchgen/gen.py from DispatchKeyFunction.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>
+
+// 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>
+
+namespace at {
+
+namespace compositeexplicitautograd {
+
+TORCH_API at::Tensor & _dirichlet_grad_out(at::Tensor & out, const at::Tensor & x, const at::Tensor & alpha, const at::Tensor & total);
+TORCH_API at::Tensor & _dirichlet_grad_outf(const at::Tensor & x, const at::Tensor & alpha, const at::Tensor & total, at::Tensor & out);
+
+} // namespace compositeexplicitautograd
+} // namespace at

phivenv/Lib/site-packages/torch/include/ATen/ops/_dirichlet_grad_cpu_dispatch.h

@@ -0,0 +1,23 @@
+#pragma once
+// @generated by torchgen/gen.py from DispatchKeyFunction.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>
+
+// 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>
+
+namespace at {
+
+namespace cpu {
+
+TORCH_API at::Tensor _dirichlet_grad(const at::Tensor & x, const at::Tensor & alpha, const at::Tensor & total);
+
+} // namespace cpu
+} // namespace at

phivenv/Lib/site-packages/torch/include/ATen/ops/_dirichlet_grad_cuda_dispatch.h

@@ -0,0 +1,23 @@
+#pragma once
+// @generated by torchgen/gen.py from DispatchKeyFunction.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>
+
+// 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>
+
+namespace at {
+
+namespace cuda {
+
+TORCH_API at::Tensor _dirichlet_grad(const at::Tensor & x, const at::Tensor & alpha, const at::Tensor & total);
+
+} // namespace cuda
+} // namespace at

phivenv/Lib/site-packages/torch/include/ATen/ops/_dirichlet_grad_native.h

@@ -0,0 +1,23 @@
+#pragma once
+
+// @generated by torchgen/gen.py from NativeFunction.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/core/QScheme.h>
+#include <ATen/core/Reduction.h>
+#include <ATen/core/Tensor.h>
+#include <tuple>
+#include <vector>
+
+
+namespace at {
+namespace native {
+TORCH_API at::Tensor & _dirichlet_grad_out(const at::Tensor & x, const at::Tensor & alpha, const at::Tensor & total, at::Tensor & out);
+TORCH_API at::Tensor _dirichlet_grad_cpu(const at::Tensor & x, const at::Tensor & alpha, const at::Tensor & total);
+TORCH_API at::Tensor _dirichlet_grad_cuda(const at::Tensor & x, const at::Tensor & alpha, const at::Tensor & total);
+} // namespace native
+} // namespace at

phivenv/Lib/site-packages/torch/include/ATen/ops/_dirichlet_grad_ops.h

@@ -0,0 +1,40 @@
+#pragma once
+
+// @generated by torchgen/gen.py from Operator.h
+
+#include <string_view>
+#include <tuple>
+#include <vector>
+
+// 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>
+
+namespace at {
+namespace _ops {
+
+
+struct TORCH_API _dirichlet_grad {
+  using schema = at::Tensor (const at::Tensor &, const at::Tensor &, const at::Tensor &);
+  using ptr_schema = schema*;
+  // See Note [static constexpr char* members for windows NVCC]
+  static constexpr const char* name = "aten::_dirichlet_grad";
+  static constexpr const char* overload_name = "";
+  static constexpr const char* schema_str = "_dirichlet_grad(Tensor x, Tensor alpha, Tensor total) -> Tensor";
+  static at::Tensor call(const at::Tensor & x, const at::Tensor & alpha, const at::Tensor & total);
+  static at::Tensor redispatch(c10::DispatchKeySet dispatchKeySet, const at::Tensor & x, const at::Tensor & alpha, const at::Tensor & total);
+};
+
+struct TORCH_API _dirichlet_grad_out {
+  using schema = at::Tensor & (const at::Tensor &, const at::Tensor &, const at::Tensor &, at::Tensor &);
+  using ptr_schema = schema*;
+  // See Note [static constexpr char* members for windows NVCC]
+  static constexpr const char* name = "aten::_dirichlet_grad";
+  static constexpr const char* overload_name = "out";
+  static constexpr const char* schema_str = "_dirichlet_grad.out(Tensor x, Tensor alpha, Tensor total, *, Tensor(a!) out) -> Tensor(a!)";
+  static at::Tensor & call(const at::Tensor & x, const at::Tensor & alpha, const at::Tensor & total, at::Tensor & out);
+  static at::Tensor & redispatch(c10::DispatchKeySet dispatchKeySet, const at::Tensor & x, const at::Tensor & alpha, const at::Tensor & total, at::Tensor & out);
+};
+
+}} // namespace at::_ops

phivenv/Lib/site-packages/torch/include/ATen/ops/_dyn_quant_matmul_4bit.h

@@ -0,0 +1,31 @@
+#pragma once
+
+// @generated by torchgen/gen.py from Function.h
+
+#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 <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 <string_view>
+
+
+
+#include <ATen/ops/_dyn_quant_matmul_4bit_ops.h>
+
+namespace at {
+
+
+// aten::_dyn_quant_matmul_4bit(Tensor inp, Tensor packed_weights, int block_size, int in_features, int out_features) -> Tensor
+inline at::Tensor _dyn_quant_matmul_4bit(const at::Tensor & inp, const at::Tensor & packed_weights, int64_t block_size, int64_t in_features, int64_t out_features) {
+    return at::_ops::_dyn_quant_matmul_4bit::call(inp, packed_weights, block_size, in_features, out_features);
+}
+
+}

phivenv/Lib/site-packages/torch/include/ATen/ops/_dyn_quant_matmul_4bit_cpu_dispatch.h

@@ -0,0 +1,23 @@
+#pragma once
+// @generated by torchgen/gen.py from DispatchKeyFunction.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>
+
+// 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>
+
+namespace at {
+
+namespace cpu {
+
+TORCH_API at::Tensor _dyn_quant_matmul_4bit(const at::Tensor & inp, const at::Tensor & packed_weights, int64_t block_size, int64_t in_features, int64_t out_features);
+
+} // namespace cpu
+} // namespace at

phivenv/Lib/site-packages/torch/include/ATen/ops/_dyn_quant_matmul_4bit_native.h

@@ -0,0 +1,21 @@
+#pragma once
+
+// @generated by torchgen/gen.py from NativeFunction.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/core/QScheme.h>
+#include <ATen/core/Reduction.h>
+#include <ATen/core/Tensor.h>
+#include <tuple>
+#include <vector>
+
+
+namespace at {
+namespace native {
+TORCH_API at::Tensor _dyn_quant_matmul_4bit_cpu(const at::Tensor & inp, const at::Tensor & packed_weights, int64_t block_size, int64_t in_features, int64_t out_features);
+} // namespace native
+} // namespace at

phivenv/Lib/site-packages/torch/include/ATen/ops/_dyn_quant_matmul_4bit_ops.h

@@ -0,0 +1,29 @@
+#pragma once
+
+// @generated by torchgen/gen.py from Operator.h
+
+#include <string_view>
+#include <tuple>
+#include <vector>
+
+// 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>
+
+namespace at {
+namespace _ops {
+
+
+struct TORCH_API _dyn_quant_matmul_4bit {
+  using schema = at::Tensor (const at::Tensor &, const at::Tensor &, int64_t, int64_t, int64_t);
+  using ptr_schema = schema*;
+  // See Note [static constexpr char* members for windows NVCC]
+  static constexpr const char* name = "aten::_dyn_quant_matmul_4bit";
+  static constexpr const char* overload_name = "";
+  static constexpr const char* schema_str = "_dyn_quant_matmul_4bit(Tensor inp, Tensor packed_weights, int block_size, int in_features, int out_features) -> Tensor";
+  static at::Tensor call(const at::Tensor & inp, const at::Tensor & packed_weights, int64_t block_size, int64_t in_features, int64_t out_features);
+  static at::Tensor redispatch(c10::DispatchKeySet dispatchKeySet, const at::Tensor & inp, const at::Tensor & packed_weights, int64_t block_size, int64_t in_features, int64_t out_features);
+};
+
+}} // namespace at::_ops

phivenv/Lib/site-packages/torch/include/ATen/ops/_dyn_quant_pack_4bit_weight.h

@@ -0,0 +1,31 @@
+#pragma once
+
+// @generated by torchgen/gen.py from Function.h
+
+#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 <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 <string_view>
+
+
+
+#include <ATen/ops/_dyn_quant_pack_4bit_weight_ops.h>
+
+namespace at {
+
+
+// aten::_dyn_quant_pack_4bit_weight(Tensor weights, Tensor scales_zeros, Tensor? bias, int block_size, int in_features, int out_features) -> Tensor
+inline at::Tensor _dyn_quant_pack_4bit_weight(const at::Tensor & weights, const at::Tensor & scales_zeros, const ::std::optional<at::Tensor> & bias, int64_t block_size, int64_t in_features, int64_t out_features) {
+    return at::_ops::_dyn_quant_pack_4bit_weight::call(weights, scales_zeros, bias, block_size, in_features, out_features);
+}
+
+}

phivenv/Lib/site-packages/torch/include/ATen/ops/_dyn_quant_pack_4bit_weight_cpu_dispatch.h

@@ -0,0 +1,23 @@
+#pragma once
+// @generated by torchgen/gen.py from DispatchKeyFunction.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>
+
+// 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>
+
+namespace at {
+
+namespace cpu {
+
+TORCH_API at::Tensor _dyn_quant_pack_4bit_weight(const at::Tensor & weights, const at::Tensor & scales_zeros, const ::std::optional<at::Tensor> & bias, int64_t block_size, int64_t in_features, int64_t out_features);
+
+} // namespace cpu
+} // namespace at

phivenv/Lib/site-packages/torch/include/ATen/ops/_dyn_quant_pack_4bit_weight_native.h

@@ -0,0 +1,21 @@
+#pragma once
+
+// @generated by torchgen/gen.py from NativeFunction.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/core/QScheme.h>
+#include <ATen/core/Reduction.h>
+#include <ATen/core/Tensor.h>
+#include <tuple>
+#include <vector>
+
+
+namespace at {
+namespace native {
+TORCH_API at::Tensor _dyn_quant_pack_4bit_weight_cpu(const at::Tensor & weights, const at::Tensor & scales_zeros, const ::std::optional<at::Tensor> & bias, int64_t block_size, int64_t in_features, int64_t out_features);
+} // namespace native
+} // namespace at

phivenv/Lib/site-packages/torch/include/ATen/ops/_dyn_quant_pack_4bit_weight_ops.h

@@ -0,0 +1,29 @@
+#pragma once
+
+// @generated by torchgen/gen.py from Operator.h
+
+#include <string_view>
+#include <tuple>
+#include <vector>
+
+// 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>
+
+namespace at {
+namespace _ops {
+
+
+struct TORCH_API _dyn_quant_pack_4bit_weight {
+  using schema = at::Tensor (const at::Tensor &, const at::Tensor &, const ::std::optional<at::Tensor> &, int64_t, int64_t, int64_t);
+  using ptr_schema = schema*;
+  // See Note [static constexpr char* members for windows NVCC]
+  static constexpr const char* name = "aten::_dyn_quant_pack_4bit_weight";
+  static constexpr const char* overload_name = "";
+  static constexpr const char* schema_str = "_dyn_quant_pack_4bit_weight(Tensor weights, Tensor scales_zeros, Tensor? bias, int block_size, int in_features, int out_features) -> Tensor";
+  static at::Tensor call(const at::Tensor & weights, const at::Tensor & scales_zeros, const ::std::optional<at::Tensor> & bias, int64_t block_size, int64_t in_features, int64_t out_features);
+  static at::Tensor redispatch(c10::DispatchKeySet dispatchKeySet, const at::Tensor & weights, const at::Tensor & scales_zeros, const ::std::optional<at::Tensor> & bias, int64_t block_size, int64_t in_features, int64_t out_features);
+};
+
+}} // namespace at::_ops

phivenv/Lib/site-packages/torch/include/ATen/ops/_efficient_attention_backward.h

@@ -0,0 +1,48 @@
+#pragma once
+
+// @generated by torchgen/gen.py from Function.h
+
+#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 <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 <string_view>
+
+
+
+#include <ATen/ops/_efficient_attention_backward_ops.h>
+
+namespace at {
+
+
+// aten::_efficient_attention_backward(Tensor grad_out_, Tensor query, Tensor key, Tensor value, Tensor? bias, Tensor out, Tensor? cu_seqlens_q, Tensor? cu_seqlens_k, SymInt max_seqlen_q, SymInt max_seqlen_k, Tensor logsumexp, float dropout_p, Tensor philox_seed, Tensor philox_offset, int custom_mask_type, bool bias_requires_grad, *, float? scale=None, int? num_splits_key=None, int? window_size=None, bool shared_storage_dqdkdv=False) -> (Tensor, Tensor, Tensor, Tensor)
+inline ::std::tuple<at::Tensor,at::Tensor,at::Tensor,at::Tensor> _efficient_attention_backward(const at::Tensor & grad_out_, const at::Tensor & query, const at::Tensor & key, const at::Tensor & value, const ::std::optional<at::Tensor> & bias, const at::Tensor & out, const ::std::optional<at::Tensor> & cu_seqlens_q, const ::std::optional<at::Tensor> & cu_seqlens_k, int64_t max_seqlen_q, int64_t max_seqlen_k, const at::Tensor & logsumexp, double dropout_p, const at::Tensor & philox_seed, const at::Tensor & philox_offset, int64_t custom_mask_type, bool bias_requires_grad, ::std::optional<double> scale=::std::nullopt, ::std::optional<int64_t> num_splits_key=::std::nullopt, ::std::optional<int64_t> window_size=::std::nullopt, bool shared_storage_dqdkdv=false) {
+    return at::_ops::_efficient_attention_backward::call(grad_out_, query, key, value, bias, out, cu_seqlens_q, cu_seqlens_k, max_seqlen_q, max_seqlen_k, logsumexp, dropout_p, philox_seed, philox_offset, custom_mask_type, bias_requires_grad, scale, num_splits_key, window_size, shared_storage_dqdkdv);
+}
+namespace symint {
+  template <typename T, typename = std::enable_if_t<std::is_same_v<T, int64_t>>>
+  ::std::tuple<at::Tensor,at::Tensor,at::Tensor,at::Tensor> _efficient_attention_backward(const at::Tensor & grad_out_, const at::Tensor & query, const at::Tensor & key, const at::Tensor & value, const ::std::optional<at::Tensor> & bias, const at::Tensor & out, const ::std::optional<at::Tensor> & cu_seqlens_q, const ::std::optional<at::Tensor> & cu_seqlens_k, int64_t max_seqlen_q, int64_t max_seqlen_k, const at::Tensor & logsumexp, double dropout_p, const at::Tensor & philox_seed, const at::Tensor & philox_offset, int64_t custom_mask_type, bool bias_requires_grad, ::std::optional<double> scale=::std::nullopt, ::std::optional<int64_t> num_splits_key=::std::nullopt, ::std::optional<int64_t> window_size=::std::nullopt, bool shared_storage_dqdkdv=false) {
+    return at::_ops::_efficient_attention_backward::call(grad_out_, query, key, value, bias, out, cu_seqlens_q, cu_seqlens_k, max_seqlen_q, max_seqlen_k, logsumexp, dropout_p, philox_seed, philox_offset, custom_mask_type, bias_requires_grad, scale, num_splits_key, window_size, shared_storage_dqdkdv);
+  }
+}
+
+// aten::_efficient_attention_backward(Tensor grad_out_, Tensor query, Tensor key, Tensor value, Tensor? bias, Tensor out, Tensor? cu_seqlens_q, Tensor? cu_seqlens_k, SymInt max_seqlen_q, SymInt max_seqlen_k, Tensor logsumexp, float dropout_p, Tensor philox_seed, Tensor philox_offset, int custom_mask_type, bool bias_requires_grad, *, float? scale=None, int? num_splits_key=None, int? window_size=None, bool shared_storage_dqdkdv=False) -> (Tensor, Tensor, Tensor, Tensor)
+inline ::std::tuple<at::Tensor,at::Tensor,at::Tensor,at::Tensor> _efficient_attention_backward_symint(const at::Tensor & grad_out_, const at::Tensor & query, const at::Tensor & key, const at::Tensor & value, const ::std::optional<at::Tensor> & bias, const at::Tensor & out, const ::std::optional<at::Tensor> & cu_seqlens_q, const ::std::optional<at::Tensor> & cu_seqlens_k, c10::SymInt max_seqlen_q, c10::SymInt max_seqlen_k, const at::Tensor & logsumexp, double dropout_p, const at::Tensor & philox_seed, const at::Tensor & philox_offset, int64_t custom_mask_type, bool bias_requires_grad, ::std::optional<double> scale=::std::nullopt, ::std::optional<int64_t> num_splits_key=::std::nullopt, ::std::optional<int64_t> window_size=::std::nullopt, bool shared_storage_dqdkdv=false) {
+    return at::_ops::_efficient_attention_backward::call(grad_out_, query, key, value, bias, out, cu_seqlens_q, cu_seqlens_k, max_seqlen_q, max_seqlen_k, logsumexp, dropout_p, philox_seed, philox_offset, custom_mask_type, bias_requires_grad, scale, num_splits_key, window_size, shared_storage_dqdkdv);
+}
+namespace symint {
+  template <typename T, typename = std::enable_if_t<std::is_same_v<T, c10::SymInt>>>
+  ::std::tuple<at::Tensor,at::Tensor,at::Tensor,at::Tensor> _efficient_attention_backward(const at::Tensor & grad_out_, const at::Tensor & query, const at::Tensor & key, const at::Tensor & value, const ::std::optional<at::Tensor> & bias, const at::Tensor & out, const ::std::optional<at::Tensor> & cu_seqlens_q, const ::std::optional<at::Tensor> & cu_seqlens_k, c10::SymInt max_seqlen_q, c10::SymInt max_seqlen_k, const at::Tensor & logsumexp, double dropout_p, const at::Tensor & philox_seed, const at::Tensor & philox_offset, int64_t custom_mask_type, bool bias_requires_grad, ::std::optional<double> scale=::std::nullopt, ::std::optional<int64_t> num_splits_key=::std::nullopt, ::std::optional<int64_t> window_size=::std::nullopt, bool shared_storage_dqdkdv=false) {
+    return at::_ops::_efficient_attention_backward::call(grad_out_, query, key, value, bias, out, cu_seqlens_q, cu_seqlens_k, max_seqlen_q, max_seqlen_k, logsumexp, dropout_p, philox_seed, philox_offset, custom_mask_type, bias_requires_grad, scale, num_splits_key, window_size, shared_storage_dqdkdv);
+  }
+}
+
+}

phivenv/Lib/site-packages/torch/include/ATen/ops/_efficient_attention_backward_cuda_dispatch.h

@@ -0,0 +1,24 @@
+#pragma once
+// @generated by torchgen/gen.py from DispatchKeyFunction.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>
+
+// 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>
+
+namespace at {
+
+namespace cuda {
+
+TORCH_API ::std::tuple<at::Tensor,at::Tensor,at::Tensor,at::Tensor> _efficient_attention_backward(const at::Tensor & grad_out_, const at::Tensor & query, const at::Tensor & key, const at::Tensor & value, const ::std::optional<at::Tensor> & bias, const at::Tensor & out, const ::std::optional<at::Tensor> & cu_seqlens_q, const ::std::optional<at::Tensor> & cu_seqlens_k, int64_t max_seqlen_q, int64_t max_seqlen_k, const at::Tensor & logsumexp, double dropout_p, const at::Tensor & philox_seed, const at::Tensor & philox_offset, int64_t custom_mask_type, bool bias_requires_grad, ::std::optional<double> scale=::std::nullopt, ::std::optional<int64_t> num_splits_key=::std::nullopt, ::std::optional<int64_t> window_size=::std::nullopt, bool shared_storage_dqdkdv=false);
+TORCH_API ::std::tuple<at::Tensor,at::Tensor,at::Tensor,at::Tensor> _efficient_attention_backward_symint(const at::Tensor & grad_out_, const at::Tensor & query, const at::Tensor & key, const at::Tensor & value, const ::std::optional<at::Tensor> & bias, const at::Tensor & out, const ::std::optional<at::Tensor> & cu_seqlens_q, const ::std::optional<at::Tensor> & cu_seqlens_k, c10::SymInt max_seqlen_q, c10::SymInt max_seqlen_k, const at::Tensor & logsumexp, double dropout_p, const at::Tensor & philox_seed, const at::Tensor & philox_offset, int64_t custom_mask_type, bool bias_requires_grad, ::std::optional<double> scale=::std::nullopt, ::std::optional<int64_t> num_splits_key=::std::nullopt, ::std::optional<int64_t> window_size=::std::nullopt, bool shared_storage_dqdkdv=false);
+
+} // namespace cuda
+} // namespace at

phivenv/Lib/site-packages/torch/include/ATen/ops/_efficient_attention_backward_native.h

@@ -0,0 +1,21 @@
+#pragma once
+
+// @generated by torchgen/gen.py from NativeFunction.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/core/QScheme.h>
+#include <ATen/core/Reduction.h>
+#include <ATen/core/Tensor.h>
+#include <tuple>
+#include <vector>
+
+
+namespace at {
+namespace native {
+TORCH_API ::std::tuple<at::Tensor,at::Tensor,at::Tensor,at::Tensor> _efficient_attention_backward(const at::Tensor & grad_out_, const at::Tensor & query, const at::Tensor & key, const at::Tensor & value, const ::std::optional<at::Tensor> & bias, const at::Tensor & out, const ::std::optional<at::Tensor> & cu_seqlens_q, const ::std::optional<at::Tensor> & cu_seqlens_k, int64_t max_seqlen_q, int64_t max_seqlen_k, const at::Tensor & logsumexp, double dropout_p, const at::Tensor & philox_seed, const at::Tensor & philox_offset, int64_t custom_mask_type, bool bias_requires_grad, ::std::optional<double> scale=::std::nullopt, ::std::optional<int64_t> num_splits_key=::std::nullopt, ::std::optional<int64_t> window_size=::std::nullopt, bool shared_storage_dqdkdv=false);
+} // namespace native
+} // namespace at

phivenv/Lib/site-packages/torch/include/ATen/ops/_efficient_attention_backward_ops.h

@@ -0,0 +1,29 @@
+#pragma once
+
+// @generated by torchgen/gen.py from Operator.h
+
+#include <string_view>
+#include <tuple>
+#include <vector>
+
+// 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>
+
+namespace at {
+namespace _ops {
+
+
+struct TORCH_API _efficient_attention_backward {
+  using schema = ::std::tuple<at::Tensor,at::Tensor,at::Tensor,at::Tensor> (const at::Tensor &, const at::Tensor &, const at::Tensor &, const at::Tensor &, const ::std::optional<at::Tensor> &, const at::Tensor &, const ::std::optional<at::Tensor> &, const ::std::optional<at::Tensor> &, c10::SymInt, c10::SymInt, const at::Tensor &, double, const at::Tensor &, const at::Tensor &, int64_t, bool, ::std::optional<double>, ::std::optional<int64_t>, ::std::optional<int64_t>, bool);
+  using ptr_schema = schema*;
+  // See Note [static constexpr char* members for windows NVCC]
+  static constexpr const char* name = "aten::_efficient_attention_backward";
+  static constexpr const char* overload_name = "";
+  static constexpr const char* schema_str = "_efficient_attention_backward(Tensor grad_out_, Tensor query, Tensor key, Tensor value, Tensor? bias, Tensor out, Tensor? cu_seqlens_q, Tensor? cu_seqlens_k, SymInt max_seqlen_q, SymInt max_seqlen_k, Tensor logsumexp, float dropout_p, Tensor philox_seed, Tensor philox_offset, int custom_mask_type, bool bias_requires_grad, *, float? scale=None, int? num_splits_key=None, int? window_size=None, bool shared_storage_dqdkdv=False) -> (Tensor, Tensor, Tensor, Tensor)";
+  static ::std::tuple<at::Tensor,at::Tensor,at::Tensor,at::Tensor> call(const at::Tensor & grad_out_, const at::Tensor & query, const at::Tensor & key, const at::Tensor & value, const ::std::optional<at::Tensor> & bias, const at::Tensor & out, const ::std::optional<at::Tensor> & cu_seqlens_q, const ::std::optional<at::Tensor> & cu_seqlens_k, c10::SymInt max_seqlen_q, c10::SymInt max_seqlen_k, const at::Tensor & logsumexp, double dropout_p, const at::Tensor & philox_seed, const at::Tensor & philox_offset, int64_t custom_mask_type, bool bias_requires_grad, ::std::optional<double> scale, ::std::optional<int64_t> num_splits_key, ::std::optional<int64_t> window_size, bool shared_storage_dqdkdv);
+  static ::std::tuple<at::Tensor,at::Tensor,at::Tensor,at::Tensor> redispatch(c10::DispatchKeySet dispatchKeySet, const at::Tensor & grad_out_, const at::Tensor & query, const at::Tensor & key, const at::Tensor & value, const ::std::optional<at::Tensor> & bias, const at::Tensor & out, const ::std::optional<at::Tensor> & cu_seqlens_q, const ::std::optional<at::Tensor> & cu_seqlens_k, c10::SymInt max_seqlen_q, c10::SymInt max_seqlen_k, const at::Tensor & logsumexp, double dropout_p, const at::Tensor & philox_seed, const at::Tensor & philox_offset, int64_t custom_mask_type, bool bias_requires_grad, ::std::optional<double> scale, ::std::optional<int64_t> num_splits_key, ::std::optional<int64_t> window_size, bool shared_storage_dqdkdv);
+};
+
+}} // namespace at::_ops

phivenv/Lib/site-packages/torch/include/ATen/ops/_efficient_attention_forward.h

@@ -0,0 +1,48 @@
+#pragma once
+
+// @generated by torchgen/gen.py from Function.h
+
+#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 <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 <string_view>
+
+
+
+#include <ATen/ops/_efficient_attention_forward_ops.h>
+
+namespace at {
+
+
+// aten::_efficient_attention_forward(Tensor query, Tensor key, Tensor value, Tensor? bias, Tensor? cu_seqlens_q, Tensor? cu_seqlens_k, SymInt? max_seqlen_q, SymInt? max_seqlen_k, float dropout_p, int custom_mask_type, bool compute_log_sumexp=False, *, float? scale=None, Tensor? seqlen_k=None, int? window_size=None) -> (Tensor output, Tensor logsumexp, Tensor philox_seed, Tensor philox_offset, SymInt max_seqlen_batch_q, SymInt max_seqlen_batch_k)
+inline ::std::tuple<at::Tensor,at::Tensor,at::Tensor,at::Tensor,c10::SymInt,c10::SymInt> _efficient_attention_forward(const at::Tensor & query, const at::Tensor & key, const at::Tensor & value, const ::std::optional<at::Tensor> & bias, const ::std::optional<at::Tensor> & cu_seqlens_q, const ::std::optional<at::Tensor> & cu_seqlens_k, ::std::optional<int64_t> max_seqlen_q, ::std::optional<int64_t> max_seqlen_k, double dropout_p, int64_t custom_mask_type, bool compute_log_sumexp=false, ::std::optional<double> scale=::std::nullopt, const ::std::optional<at::Tensor> & seqlen_k={}, ::std::optional<int64_t> window_size=::std::nullopt) {
+    return at::_ops::_efficient_attention_forward::call(query, key, value, bias, cu_seqlens_q, cu_seqlens_k, max_seqlen_q.has_value() ? ::std::make_optional(c10::SymInt(*max_seqlen_q)) : ::std::nullopt, max_seqlen_k.has_value() ? ::std::make_optional(c10::SymInt(*max_seqlen_k)) : ::std::nullopt, dropout_p, custom_mask_type, compute_log_sumexp, scale, seqlen_k, window_size);
+}
+namespace symint {
+  template <typename T, typename = std::enable_if_t<std::is_same_v<T, int64_t>>>
+  ::std::tuple<at::Tensor,at::Tensor,at::Tensor,at::Tensor,c10::SymInt,c10::SymInt> _efficient_attention_forward(const at::Tensor & query, const at::Tensor & key, const at::Tensor & value, const ::std::optional<at::Tensor> & bias, const ::std::optional<at::Tensor> & cu_seqlens_q, const ::std::optional<at::Tensor> & cu_seqlens_k, ::std::optional<int64_t> max_seqlen_q, ::std::optional<int64_t> max_seqlen_k, double dropout_p, int64_t custom_mask_type, bool compute_log_sumexp=false, ::std::optional<double> scale=::std::nullopt, const ::std::optional<at::Tensor> & seqlen_k={}, ::std::optional<int64_t> window_size=::std::nullopt) {
+    return at::_ops::_efficient_attention_forward::call(query, key, value, bias, cu_seqlens_q, cu_seqlens_k, max_seqlen_q.has_value() ? ::std::make_optional(c10::SymInt(*max_seqlen_q)) : ::std::nullopt, max_seqlen_k.has_value() ? ::std::make_optional(c10::SymInt(*max_seqlen_k)) : ::std::nullopt, dropout_p, custom_mask_type, compute_log_sumexp, scale, seqlen_k, window_size);
+  }
+}
+
+// aten::_efficient_attention_forward(Tensor query, Tensor key, Tensor value, Tensor? bias, Tensor? cu_seqlens_q, Tensor? cu_seqlens_k, SymInt? max_seqlen_q, SymInt? max_seqlen_k, float dropout_p, int custom_mask_type, bool compute_log_sumexp=False, *, float? scale=None, Tensor? seqlen_k=None, int? window_size=None) -> (Tensor output, Tensor logsumexp, Tensor philox_seed, Tensor philox_offset, SymInt max_seqlen_batch_q, SymInt max_seqlen_batch_k)
+inline ::std::tuple<at::Tensor,at::Tensor,at::Tensor,at::Tensor,c10::SymInt,c10::SymInt> _efficient_attention_forward_symint(const at::Tensor & query, const at::Tensor & key, const at::Tensor & value, const ::std::optional<at::Tensor> & bias, const ::std::optional<at::Tensor> & cu_seqlens_q, const ::std::optional<at::Tensor> & cu_seqlens_k, ::std::optional<c10::SymInt> max_seqlen_q, ::std::optional<c10::SymInt> max_seqlen_k, double dropout_p, int64_t custom_mask_type, bool compute_log_sumexp=false, ::std::optional<double> scale=::std::nullopt, const ::std::optional<at::Tensor> & seqlen_k={}, ::std::optional<int64_t> window_size=::std::nullopt) {
+    return at::_ops::_efficient_attention_forward::call(query, key, value, bias, cu_seqlens_q, cu_seqlens_k, max_seqlen_q, max_seqlen_k, dropout_p, custom_mask_type, compute_log_sumexp, scale, seqlen_k, window_size);
+}
+namespace symint {
+  template <typename T, typename = std::enable_if_t<std::is_same_v<T, c10::SymInt>>>
+  ::std::tuple<at::Tensor,at::Tensor,at::Tensor,at::Tensor,c10::SymInt,c10::SymInt> _efficient_attention_forward(const at::Tensor & query, const at::Tensor & key, const at::Tensor & value, const ::std::optional<at::Tensor> & bias, const ::std::optional<at::Tensor> & cu_seqlens_q, const ::std::optional<at::Tensor> & cu_seqlens_k, ::std::optional<c10::SymInt> max_seqlen_q, ::std::optional<c10::SymInt> max_seqlen_k, double dropout_p, int64_t custom_mask_type, bool compute_log_sumexp=false, ::std::optional<double> scale=::std::nullopt, const ::std::optional<at::Tensor> & seqlen_k={}, ::std::optional<int64_t> window_size=::std::nullopt) {
+    return at::_ops::_efficient_attention_forward::call(query, key, value, bias, cu_seqlens_q, cu_seqlens_k, max_seqlen_q, max_seqlen_k, dropout_p, custom_mask_type, compute_log_sumexp, scale, seqlen_k, window_size);
+  }
+}
+
+}

phivenv/Lib/site-packages/torch/include/ATen/ops/_efficient_attention_forward_cuda_dispatch.h

@@ -0,0 +1,24 @@
+#pragma once
+// @generated by torchgen/gen.py from DispatchKeyFunction.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>
+
+// 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>
+
+namespace at {
+
+namespace cuda {
+
+TORCH_API ::std::tuple<at::Tensor,at::Tensor,at::Tensor,at::Tensor,c10::SymInt,c10::SymInt> _efficient_attention_forward(const at::Tensor & query, const at::Tensor & key, const at::Tensor & value, const ::std::optional<at::Tensor> & bias, const ::std::optional<at::Tensor> & cu_seqlens_q, const ::std::optional<at::Tensor> & cu_seqlens_k, ::std::optional<int64_t> max_seqlen_q, ::std::optional<int64_t> max_seqlen_k, double dropout_p, int64_t custom_mask_type, bool compute_log_sumexp=false, ::std::optional<double> scale=::std::nullopt, const ::std::optional<at::Tensor> & seqlen_k={}, ::std::optional<int64_t> window_size=::std::nullopt);
+TORCH_API ::std::tuple<at::Tensor,at::Tensor,at::Tensor,at::Tensor,c10::SymInt,c10::SymInt> _efficient_attention_forward_symint(const at::Tensor & query, const at::Tensor & key, const at::Tensor & value, const ::std::optional<at::Tensor> & bias, const ::std::optional<at::Tensor> & cu_seqlens_q, const ::std::optional<at::Tensor> & cu_seqlens_k, ::std::optional<c10::SymInt> max_seqlen_q, ::std::optional<c10::SymInt> max_seqlen_k, double dropout_p, int64_t custom_mask_type, bool compute_log_sumexp=false, ::std::optional<double> scale=::std::nullopt, const ::std::optional<at::Tensor> & seqlen_k={}, ::std::optional<int64_t> window_size=::std::nullopt);
+
+} // namespace cuda
+} // namespace at

phivenv/Lib/site-packages/torch/include/ATen/ops/_efficient_attention_forward_native.h

@@ -0,0 +1,21 @@
+#pragma once
+
+// @generated by torchgen/gen.py from NativeFunction.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/core/QScheme.h>
+#include <ATen/core/Reduction.h>
+#include <ATen/core/Tensor.h>
+#include <tuple>
+#include <vector>
+
+
+namespace at {
+namespace native {
+TORCH_API ::std::tuple<at::Tensor,at::Tensor,at::Tensor,at::Tensor,c10::SymInt,c10::SymInt> _efficient_attention_forward(const at::Tensor & query, const at::Tensor & key, const at::Tensor & value, const ::std::optional<at::Tensor> & bias, const ::std::optional<at::Tensor> & cu_seqlens_q, const ::std::optional<at::Tensor> & cu_seqlens_k, ::std::optional<int64_t> max_seqlen_q, ::std::optional<int64_t> max_seqlen_k, double dropout_p, int64_t custom_mask_type, bool compute_log_sumexp=false, ::std::optional<double> scale=::std::nullopt, const ::std::optional<at::Tensor> & seqlen_k={}, ::std::optional<int64_t> window_size=::std::nullopt);
+} // namespace native
+} // namespace at

phivenv/Lib/site-packages/torch/include/ATen/ops/_efficient_attention_forward_ops.h

@@ -0,0 +1,29 @@
+#pragma once
+
+// @generated by torchgen/gen.py from Operator.h
+
+#include <string_view>
+#include <tuple>
+#include <vector>
+
+// 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>
+
+namespace at {
+namespace _ops {
+
+
+struct TORCH_API _efficient_attention_forward {
+  using schema = ::std::tuple<at::Tensor,at::Tensor,at::Tensor,at::Tensor,c10::SymInt,c10::SymInt> (const at::Tensor &, const at::Tensor &, const at::Tensor &, const ::std::optional<at::Tensor> &, const ::std::optional<at::Tensor> &, const ::std::optional<at::Tensor> &, ::std::optional<c10::SymInt>, ::std::optional<c10::SymInt>, double, int64_t, bool, ::std::optional<double>, const ::std::optional<at::Tensor> &, ::std::optional<int64_t>);
+  using ptr_schema = schema*;
+  // See Note [static constexpr char* members for windows NVCC]
+  static constexpr const char* name = "aten::_efficient_attention_forward";
+  static constexpr const char* overload_name = "";
+  static constexpr const char* schema_str = "_efficient_attention_forward(Tensor query, Tensor key, Tensor value, Tensor? bias, Tensor? cu_seqlens_q, Tensor? cu_seqlens_k, SymInt? max_seqlen_q, SymInt? max_seqlen_k, float dropout_p, int custom_mask_type, bool compute_log_sumexp=False, *, float? scale=None, Tensor? seqlen_k=None, int? window_size=None) -> (Tensor output, Tensor logsumexp, Tensor philox_seed, Tensor philox_offset, SymInt max_seqlen_batch_q, SymInt max_seqlen_batch_k)";
+  static ::std::tuple<at::Tensor,at::Tensor,at::Tensor,at::Tensor,c10::SymInt,c10::SymInt> call(const at::Tensor & query, const at::Tensor & key, const at::Tensor & value, const ::std::optional<at::Tensor> & bias, const ::std::optional<at::Tensor> & cu_seqlens_q, const ::std::optional<at::Tensor> & cu_seqlens_k, ::std::optional<c10::SymInt> max_seqlen_q, ::std::optional<c10::SymInt> max_seqlen_k, double dropout_p, int64_t custom_mask_type, bool compute_log_sumexp, ::std::optional<double> scale, const ::std::optional<at::Tensor> & seqlen_k, ::std::optional<int64_t> window_size);
+  static ::std::tuple<at::Tensor,at::Tensor,at::Tensor,at::Tensor,c10::SymInt,c10::SymInt> redispatch(c10::DispatchKeySet dispatchKeySet, const at::Tensor & query, const at::Tensor & key, const at::Tensor & value, const ::std::optional<at::Tensor> & bias, const ::std::optional<at::Tensor> & cu_seqlens_q, const ::std::optional<at::Tensor> & cu_seqlens_k, ::std::optional<c10::SymInt> max_seqlen_q, ::std::optional<c10::SymInt> max_seqlen_k, double dropout_p, int64_t custom_mask_type, bool compute_log_sumexp, ::std::optional<double> scale, const ::std::optional<at::Tensor> & seqlen_k, ::std::optional<int64_t> window_size);
+};
+
+}} // namespace at::_ops

phivenv/Lib/site-packages/torch/include/ATen/ops/_efficientzerotensor.h

@@ -0,0 +1,114 @@
+#pragma once
+
+// @generated by torchgen/gen.py from Function.h
+
+#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 <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 <string_view>
+
+
+
+#include <ATen/ops/_efficientzerotensor_ops.h>
+
+namespace at {
+
+
+// aten::_efficientzerotensor(SymInt[] size, *, ScalarType? dtype=None, Layout? layout=None, Device? device=None, bool? pin_memory=None) -> Tensor
+inline at::Tensor _efficientzerotensor(at::IntArrayRef size, at::TensorOptions options={}) {
+    return at::_ops::_efficientzerotensor::call(c10::fromIntArrayRefSlow(size), c10::optTypeMetaToScalarType(options.dtype_opt()), options.layout_opt(), options.device_opt(), options.pinned_memory_opt());
+}
+namespace symint {
+  template <typename T, typename = std::enable_if_t<std::is_same_v<T, int64_t>>>
+  at::Tensor _efficientzerotensor(at::IntArrayRef size, at::TensorOptions options={}) {
+    return at::_ops::_efficientzerotensor::call(c10::fromIntArrayRefSlow(size), c10::optTypeMetaToScalarType(options.dtype_opt()), options.layout_opt(), options.device_opt(), options.pinned_memory_opt());
+  }
+}
+
+// aten::_efficientzerotensor(SymInt[] size, *, ScalarType? dtype=None, Layout? layout=None, Device? device=None, bool? pin_memory=None) -> Tensor
+inline at::Tensor _efficientzerotensor(at::IntArrayRef size, ::std::optional<at::ScalarType> dtype, ::std::optional<at::Layout> layout, ::std::optional<at::Device> device, ::std::optional<bool> pin_memory) {
+    return at::_ops::_efficientzerotensor::call(c10::fromIntArrayRefSlow(size), dtype, layout, device, pin_memory);
+}
+namespace symint {
+  template <typename T, typename = std::enable_if_t<std::is_same_v<T, int64_t>>>
+  at::Tensor _efficientzerotensor(at::IntArrayRef size, ::std::optional<at::ScalarType> dtype, ::std::optional<at::Layout> layout, ::std::optional<at::Device> device, ::std::optional<bool> pin_memory) {
+    return at::_ops::_efficientzerotensor::call(c10::fromIntArrayRefSlow(size), dtype, layout, device, pin_memory);
+  }
+}
+
+// aten::_efficientzerotensor(SymInt[] size, *, ScalarType? dtype=None, Layout? layout=None, Device? device=None, bool? pin_memory=None) -> Tensor
+inline at::Tensor _efficientzerotensor_symint(c10::SymIntArrayRef size, at::TensorOptions options={}) {
+    return at::_ops::_efficientzerotensor::call(size, c10::optTypeMetaToScalarType(options.dtype_opt()), options.layout_opt(), options.device_opt(), options.pinned_memory_opt());
+}
+namespace symint {
+  template <typename T, typename = std::enable_if_t<std::is_same_v<T, c10::SymInt>>>
+  at::Tensor _efficientzerotensor(c10::SymIntArrayRef size, at::TensorOptions options={}) {
+    return at::_ops::_efficientzerotensor::call(size, c10::optTypeMetaToScalarType(options.dtype_opt()), options.layout_opt(), options.device_opt(), options.pinned_memory_opt());
+  }
+}
+
+// aten::_efficientzerotensor(SymInt[] size, *, ScalarType? dtype=None, Layout? layout=None, Device? device=None, bool? pin_memory=None) -> Tensor
+inline at::Tensor _efficientzerotensor_symint(c10::SymIntArrayRef size, ::std::optional<at::ScalarType> dtype, ::std::optional<at::Layout> layout, ::std::optional<at::Device> device, ::std::optional<bool> pin_memory) {
+    return at::_ops::_efficientzerotensor::call(size, dtype, layout, device, pin_memory);
+}
+namespace symint {
+  template <typename T, typename = std::enable_if_t<std::is_same_v<T, c10::SymInt>>>
+  at::Tensor _efficientzerotensor(c10::SymIntArrayRef size, ::std::optional<at::ScalarType> dtype, ::std::optional<at::Layout> layout, ::std::optional<at::Device> device, ::std::optional<bool> pin_memory) {
+    return at::_ops::_efficientzerotensor::call(size, dtype, layout, device, pin_memory);
+  }
+}
+
+// aten::_efficientzerotensor.out(SymInt[] size, *, Tensor(a!) out) -> Tensor(a!)
+inline at::Tensor & _efficientzerotensor_out(at::Tensor & out, at::IntArrayRef size) {
+    return at::_ops::_efficientzerotensor_out::call(c10::fromIntArrayRefSlow(size), out);
+}
+namespace symint {
+  template <typename T, typename = std::enable_if_t<std::is_same_v<T, int64_t>>>
+  at::Tensor & _efficientzerotensor_out(at::Tensor & out, at::IntArrayRef size) {
+    return at::_ops::_efficientzerotensor_out::call(c10::fromIntArrayRefSlow(size), out);
+  }
+}
+
+// aten::_efficientzerotensor.out(SymInt[] size, *, Tensor(a!) out) -> Tensor(a!)
+inline at::Tensor & _efficientzerotensor_outf(at::IntArrayRef size, at::Tensor & out) {
+    return at::_ops::_efficientzerotensor_out::call(c10::fromIntArrayRefSlow(size), out);
+}
+namespace symint {
+  template <typename T, typename = std::enable_if_t<std::is_same_v<T, int64_t>>>
+  at::Tensor & _efficientzerotensor_outf(at::IntArrayRef size, at::Tensor & out) {
+    return at::_ops::_efficientzerotensor_out::call(c10::fromIntArrayRefSlow(size), out);
+  }
+}
+
+// aten::_efficientzerotensor.out(SymInt[] size, *, Tensor(a!) out) -> Tensor(a!)
+inline at::Tensor & _efficientzerotensor_symint_out(at::Tensor & out, c10::SymIntArrayRef size) {
+    return at::_ops::_efficientzerotensor_out::call(size, out);
+}
+namespace symint {
+  template <typename T, typename = std::enable_if_t<std::is_same_v<T, c10::SymInt>>>
+  at::Tensor & _efficientzerotensor_out(at::Tensor & out, c10::SymIntArrayRef size) {
+    return at::_ops::_efficientzerotensor_out::call(size, out);
+  }
+}
+
+// aten::_efficientzerotensor.out(SymInt[] size, *, Tensor(a!) out) -> Tensor(a!)
+inline at::Tensor & _efficientzerotensor_symint_outf(c10::SymIntArrayRef size, at::Tensor & out) {
+    return at::_ops::_efficientzerotensor_out::call(size, out);
+}
+namespace symint {
+  template <typename T, typename = std::enable_if_t<std::is_same_v<T, c10::SymInt>>>
+  at::Tensor & _efficientzerotensor_outf(c10::SymIntArrayRef size, at::Tensor & out) {
+    return at::_ops::_efficientzerotensor_out::call(size, out);
+  }
+}
+
+}

phivenv/Lib/site-packages/torch/include/ATen/ops/_efficientzerotensor_compositeexplicitautograd_dispatch.h

@@ -0,0 +1,26 @@
+#pragma once
+// @generated by torchgen/gen.py from DispatchKeyFunction.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>
+
+// 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>
+
+namespace at {
+
+namespace compositeexplicitautograd {
+
+TORCH_API at::Tensor & _efficientzerotensor_out(at::Tensor & out, at::IntArrayRef size);
+TORCH_API at::Tensor & _efficientzerotensor_outf(at::IntArrayRef size, at::Tensor & out);
+TORCH_API at::Tensor & _efficientzerotensor_symint_out(at::Tensor & out, c10::SymIntArrayRef size);
+TORCH_API at::Tensor & _efficientzerotensor_symint_outf(c10::SymIntArrayRef size, at::Tensor & out);
+
+} // namespace compositeexplicitautograd
+} // namespace at