torch.utils#

Created On: Jul 21, 2023 | Last Updated On: May 12, 2026

get_cpp_backtrace

Return a string containing the C++ stack trace of the current thread.

set_module

Set the module attribute on a python object for a given object for nicer printing

swap_tensors

This function swaps the content of the two Tensor objects.

torch.utils.backend_registration#

generate_methods_for_privateuse1_backend

Automatically generate attributes and methods for the custom backend after rename privateuse1 backend.

rename_privateuse1_backend

Rename the privateuse1 backend device to make it more convenient to use as a device name within PyTorch APIs.

torch.utils.hooks#

torch.utils.hooks.unserializable_hook(f)[source]#

Mark a function as an unserializable hook with this decorator.

This suppresses warnings that would otherwise arise if you attempt to serialize a tensor that has a hook.

torch.utils.throughput_benchmark#

torch.utils.throughput_benchmark.format_time(time_us=None, time_ms=None, time_s=None)[source]#

Define time formatting.

Return type:: str

torch.utils.collect_env#

`check_release_file`
`is_xnnpack_available`
`main`
`pretty_str`
`run`	Return (return-code, stdout, stderr).
`run_and_parse_first_match`	Run command using run_lambda, returns the first regex match if it exists.
`run_and_read_all`	Run command using run_lambda; reads and returns entire output if rc is 0.
`run_and_return_first_line`	Run command using run_lambda and returns first line if output is not empty.

torch.utils.flop_counter#

`baddbmm_flop`	Count flops for the baddbmm operation.
`bmm_flop`	Count flops for the bmm operation.
`conv_backward_flop`
`conv_flop`	Count flops for convolution.
`conv_flop_count`	Count flops for convolution.
`register_flop_formula`
`sdpa_backward_flop`	Count flops for self-attention backward.
`sdpa_backward_flop_count`
`sdpa_flop`	Count flops for self-attention.
`sdpa_flop_count`	Count flops for self-attention.
`shape_wrapper`

torch.utils.hipify.hipify_python#

The Python Hipify script. ## # Copyright (c) 2015-2016 Advanced Micro Devices, Inc. All rights reserved. # 2017-2018 Advanced Micro Devices, Inc. and # Facebook Inc. All rights reserved. # # Permission is hereby granted, free of charge, to any person obtaining a copy # of this software and associated documentation files (the “Software”), to deal # in the Software without restriction, including without limitation the rights # to use, copy, modify, merge, publish, distribute, sublicense, and/or sell # copies of the Software, and to permit persons to whom the Software is # furnished to do so, subject to the following conditions: # # The above copyright notice and this permission notice shall be included in # all copies or substantial portions of the Software. # # THE SOFTWARE IS PROVIDED “AS IS”, WITHOUT WARRANTY OF ANY KIND, EXPRESS OR # IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, # FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE # AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER # LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, # OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN # THE SOFTWARE.

`add_dim3`	adds dim3() to the second and third arguments in the kernel launch
`compute_stats`
`extract_arguments`	Return the list of arguments in the upcoming function parameter closure.
`file_add_header`
`file_specific_replacement`
`find_bracket_group`	Finds the first balanced parentheses.
`find_closure_group`	Generalization for finding a balancing closure group
`find_parentheses_group`	Finds the first balanced bracket.
`fix_static_global_kernels`	Static global kernels in HIP results in a compilation error.
`get_hip_file_path`	Returns the new name of the hipified file
`hip_header_magic`	If the file makes kernel builtin calls and does not include the cuda_runtime.h header, then automatically add an #include to match the "magic" includes provided by NVCC.
`hipify`
`is_caffe2_gpu_file`
`is_cusparse_file`
`is_out_of_place`
`is_pytorch_file`
`is_special_file`
`match_extensions`	Helper method to see if filename ends with certain extension
`matched_files_iter`
`openf`
`preprocess_file_and_save_result`
`preprocessor`	Executes the CUDA -> HIP conversion on the specified file.
`processKernelLaunches`	Replace the CUDA style Kernel launches with the HIP style kernel launches.
`replace_extern_shared`	Match 'extern __shared__ type foo[];' syntax and use HIP_DYNAMIC_SHARED() MACRO instead.
`replace_math_functions`	FIXME: Temporarily replace std:: invocations of math functions with non-std:: versions to prevent linker errors NOTE: This can lead to correctness issues when running tests, since the correct version of the math function (exp/expf) might not get called.
`str2bool`	ArgumentParser doesn't support type=bool.

TensorWeakRef

Wrapper around a weak ref of a Tensor that handles the _fix_weakref() call required when unwrapping a Tensor weakref.