OSSForks/pytorch
1
0
Fork 0
mirror of https://github.com/zebrajr/pytorch.git synced 2025-12-06 12:20:52 +01:00
Code Issues Actions 38 Packages Projects Releases Wiki Activity

[BE][16/16] fix typos in torch/ (torch/utils/) (#156606)

Browse Source 
Pull Request resolved: https://github.com/pytorch/pytorch/pull/156606
Approved by: https://github.com/albanD
ghstack dependencies: #156318, #156320, #156602, #156604
This commit is contained in:
Xuehai Pan 2025-07-03 02:12:00 +08:00 committed by PyTorch MergeBot
parent 11c07c848c
commit d40aaa42ee
25 changed files with 54 additions and 52 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

2
.lintrunner.toml
View File

@ -1157,6 +1157,7 @@ exclude_patterns = [
'torch/_vendor/**',
'torch/_inductor/fx_passes/serialized_patterns/**',
'torch/_inductor/autoheuristic/artifacts/**',
'torch/utils/model_dump/preact.mjs',
# These files are all grandfathered in, feel free to remove from this list
# as necessary
# NOTE: remove the patterns in the order they are listed
@ -1174,7 +1175,6 @@ exclude_patterns = [
'torch/**',
'torch/_*/**',
'torch/distributed/tensor/**',
'torch/utils/**',
]
init_command = [
'python3',

8
torch/utils/_config_module.py
View File

@ -52,17 +52,17 @@ class _Config(Generic[T]):
alias: If set, the directly use the value of the alias.
env_name_force: If set, this environment variable has precedence over
everything after this.
If multiple env variables are given, the precendence order is from
If multiple env variables are given, the precedence order is from
left to right.
user_override: If a user sets a value (i.e. foo.bar=True), that
has precedence over everything after this.
env_name_default: If set, this environment variable will override everything
after this.
If multiple env variables are given, the precendence order is from
If multiple env variables are given, the precedence order is from
left to right.
justknob: If this pytorch installation supports justknobs, that will
override defaults, but will not override the user_override precendence.
default: This value is the lowest precendance, and will be used if nothing is
override defaults, but will not override the user_override precedence.
default: This value is the lowest precedence, and will be used if nothing is
set.
Environment Variables:

2
torch/utils/_device.py
View File

@ -82,7 +82,7 @@ class DeviceContext(TorchFunctionMode):
CURRENT_DEVICE = self.old_device
cur_stack = []
# Invariant: there should only be one DeviceContext on the stack at any time
# (At the bottom), pop all mdoes until we hit the bottom, assert it's a DeviceContext
# (At the bottom), pop all modes until we hit the bottom, assert it's a DeviceContext
# or else someone else has popped it!
for _ in range(_len_torch_function_stack() - 1):
mode = _pop_mode()

2
torch/utils/_strobelight/cli_function_profiler.py
View File

@ -58,7 +58,7 @@ class StrobelightCLIFunctionProfiler:
StrobelightCLIFunctionProfiler can be used to profile a python function and
generate a strobelight link with the results. It works on meta servers but
does not requries an fbcode target.
does not requires an fbcode target.
When stop_at_error is false(default), error during profiling does not prevent
the work function from running.

4
torch/utils/_sympy/functions.py
View File

@ -1055,7 +1055,7 @@ class PowByNatural(sympy.Function):
# base is assumed to be nonnegative, thereby prevent complex numbers from
# occuring
# occurring
class FloatPow(sympy.Function):
is_real = True
@ -1315,7 +1315,7 @@ def make_opaque_unary_fn(name):
constant propagation. This helps avoid performing transformations
that are valid for real numbers but are invalid for floating point;
in particular, while we are willing to make optimizations that change
numerics for Tensor compute, we are NOT willing to make optimziations
numerics for Tensor compute, we are NOT willing to make optimizations
that change numerics for size compute.
"""

2
torch/utils/_sympy/value_ranges.py
View File

@ -920,7 +920,7 @@ class SymPyValueRangeAnalysis:
return (a, b)
# piecewise function can be used to convert a SymBool to SymInt:
# int_expr = Piecewise((1, bool_expr), (0, True)), it evalutes to 1 when sym_bool is True and 0 otherwise.
# int_expr = Piecewise((1, bool_expr), (0, True)), it evaluates to 1 when sym_bool is True and 0 otherwise.
#
# ranges is a sequence of (expr_range, condition_range) pairs. The range pair is constructed in expr_cond_pair.
# The ValueRange of Piecewise is just the union of all expr ranges whose condition expr can be True.

2
torch/utils/backend_registration.py
View File

@ -106,7 +106,7 @@ def _normalization_device(
elif isinstance(device, str):
device = torch.device(device)
# variable devcie can only be torch.device type or int type
# variable device can only be torch.device type or int type
if isinstance(device, torch.device):
if device.type != custom_backend_name:
raise RuntimeError(f"Invalid device, must be {custom_backend_name} device")

2
torch/utils/benchmark/utils/compare.py
View File

@ -280,7 +280,7 @@ class Compare:
https://pytorch.org/tutorials/recipes/recipes/benchmark.html
Args:
results: List of Measurment to display.
results: List of Measurement to display.
"""
def __init__(self, results: list[common.Measurement]):
self._results: list[common.Measurement] = []

2
torch/utils/benchmark/utils/compile.py
View File

@ -127,7 +127,7 @@ if HAS_TABULATE:
This is a simple utility that can be used to benchmark torch.compile
In particular it ensures that your GPU is setup to use tensor cores if it supports its
It also tries out all the main backends and prints a table of results so you can easily compare them all
Many of the backendds have their own optional dependencies so please pip install them seperately
Many of the backendds have their own optional dependencies so please pip install them separately
You will get one table for inference and another for training
If you'd like to leverage this utility for training make sure to pass in a torch.optim.Optimizer

2
torch/utils/bundled_inputs.py
View File

@ -116,7 +116,7 @@ def bundle_inputs(
)
# The above cloning function returns a torch._C.scriptmodule and we need a torch.jit.scriptmodule.
# Fortunately theres a function in _recursive that does exactly that conversion.
# Fortunately there is a function in _recursive that does exactly that conversion.
cloned_module = wrap_cpp_module(clone)
if isinstance(inputs, dict):
assert isinstance(info, dict) or info is None

4
torch/utils/checkpoint.py
View File

@ -858,7 +858,7 @@ class _CheckpointFrame:
if not len(self.weak_holders) == self.recomp_counter[gid]:
# 2. During recompute, fewer tensors were saved
#
# We know that everytime we save something do original forward
# We know that every time we save something do original forward
# we append to weak_holder, and every time we save a tensor
# during recompute we increment recompute_counter.
raise CheckpointError(
@ -1270,7 +1270,7 @@ class CheckpointPolicy(enum.Enum):
def _policy_from_bool(b):
# For backward compatability
# For backward compatibility
return CheckpointPolicy.MUST_SAVE if b else CheckpointPolicy.PREFER_RECOMPUTE

2
torch/utils/collect_env.py
View File

@ -493,7 +493,7 @@ def get_env_info():
Caching allocator config, XNNPACK availability and CPU information.
Returns:
SystemEnv (namedtuple): A tuple containining various environment details
SystemEnv (namedtuple): A tuple containing various environment details
and system information.
"""
run_lambda = run

24
torch/utils/cpp_extension.py
View File

@ -303,10 +303,10 @@ def _append_sycl_targets_if_missing(cflags):
cflags.append('-fsycl-targets=spir64')
def _get_sycl_device_flags(cflags):
# We need last occurence of -fsycl-targets as it will be the one taking effect.
# We need last occurrence of -fsycl-targets as it will be the one taking effect.
# So searching in reversed list.
flags = [f for f in reversed(cflags) if f.startswith('-fsycl-targets=')]
assert flags, "bug: -fsycl-targets should have been ammended to cflags"
assert flags, "bug: -fsycl-targets should have been amended to cflags"
arch_list = _get_sycl_arch_list()
if arch_list != '':
@ -466,7 +466,7 @@ def get_compiler_abi_compatibility_and_version(compiler) -> tuple[bool, TorchVer
logger.warning('Error checking compiler version for %s: %s', compiler, error)
return (False, TorchVersion('0.0.0'))
# convert alpha-numeric string to numeric string
# convert alphanumeric string to numeric string
# amdclang++ returns str like 0.0.0git, others return 0.0.0
numeric_version = [re.sub(r'\D', '', v) for v in version]
@ -770,7 +770,7 @@ class BuildExtension(build_ext):
r"""Compiles sources by outputting a ninja file and running it."""
# NB: I copied some lines from self.compiler (which is an instance
# of distutils.UnixCCompiler). See the following link.
# https://github.com/python/cpython/blob/f03a8f8d5001963ad5b5b28dbd95497e9cc15596/Lib/distutils/ccompiler.py#L564-L567
# https://github.com/python/cpython/blob/f03a8f8d5001963ad5b5b28dbd95497e9cc15596/Lib/distutils/ccompiler.py#L564-L567 # codespell:ignore
# This can be fragile, but a lot of other repos also do this
# (see https://github.com/search?q=_setup_compile&type=Code)
# so it is probably OK; we'll also get CI signal if/when
@ -899,7 +899,7 @@ class BuildExtension(build_ext):
if m
]
obj_regex = re.compile('/Fo(.*)')
obj_regex = re.compile('/Fo(.*)') # codespell:ignore
obj_list = [
m.group(1) for m in (obj_regex.match(elem) for elem in cmd)
if m
@ -1057,7 +1057,7 @@ class BuildExtension(build_ext):
# Return *all* object filenames, not just the ones we just built.
return objects
# Monkey-patch the _compile or compile method.
# https://github.com/python/cpython/blob/dc0284ee8f7a270b6005467f26d8e5773d76e959/Lib/distutils/ccompiler.py#L511
# https://github.com/python/cpython/blob/dc0284ee8f7a270b6005467f26d8e5773d76e959/Lib/distutils/ccompiler.py#L511 # codespell:ignore
if self.compiler.compiler_type == 'msvc':
if self.use_ninja:
self.compiler.compile = win_wrap_ninja_compile
@ -1744,7 +1744,7 @@ def _check_and_build_extension_h_precompiler_headers(
is_standalone=False):
r'''
Precompiled Headers(PCH) can pre-build the same headers and reduce build time for pytorch load_inline modules.
GCC offical manual: https://gcc.gnu.org/onlinedocs/gcc-4.0.4/gcc/Precompiled-Headers.html
GCC official manual: https://gcc.gnu.org/onlinedocs/gcc-4.0.4/gcc/Precompiled-Headers.html
PCH only works when built pch file(header.h.gch) and build target have the same build parameters. So, We need
add a signature file to record PCH file parameters. If the build parameters(signature) changed, it should rebuild
PCH file.
@ -2447,8 +2447,8 @@ def _get_cuda_arch_flags(cflags: Optional[list[str]] = None) -> list[str]:
# Deal with lists that are ' ' separated (only deal with ';' after)
_arch_list = _arch_list.replace(' ', ';')
# Expand named arches
for named_arch, archval in named_arches.items():
_arch_list = _arch_list.replace(named_arch, archval)
for named_arch, archival in named_arches.items():
_arch_list = _arch_list.replace(named_arch, archival)
arch_list = _arch_list.split(';')
@ -2583,7 +2583,7 @@ def _run_ninja_build(build_directory: str, verbose: bool, error_prefix: str) ->
# subprocess.run assumes that sys.__stdout__ has not been modified and
# attempts to write to it by default. However, when we call _run_ninja_build
# from ahead-of-time cpp extensions, the following happens:
# 1) If the stdout encoding is not utf-8, setuptools detachs __stdout__.
# 1) If the stdout encoding is not utf-8, setuptools detaches __stdout__.
# https://github.com/pypa/setuptools/blob/7e97def47723303fafabe48b22168bbc11bb4821/setuptools/dist.py#L1110
# (it probably shouldn't do this)
# 2) subprocess.run (on POSIX, with no stdout override) relies on
@ -2872,7 +2872,9 @@ e.
if IS_WINDOWS:
compiler_name = "$cxx" if IS_HIP_EXTENSION else "cl"
compile_rule.append(
f' command = {compiler_name} /showIncludes $cflags -c $in /Fo$out $post_cflags')
f' command = {compiler_name} '
'/showIncludes $cflags -c $in /Fo$out $post_cflags' # codespell:ignore
)
if not IS_HIP_EXTENSION:
compile_rule.append(' deps = msvc')
else:

12
torch/utils/data/dataloader.py
View File

@ -480,7 +480,7 @@ class DataLoader(Generic[_T_co]):
def __iter__(self) -> _BaseDataLoaderIter:
# When using a single worker the returned iterator should be
# created everytime to avoid resetting its state
# created every time to avoid resetting its state
# However, in the case of a multiple workers iterator
# the iterator is only created once in the lifetime of the
# DataLoader object so that workers can be reused
@ -556,10 +556,10 @@ class DataLoader(Generic[_T_co]):
# necessary.
#
#
# [Note] Please note that this function repects `cpuset` only when os.sched_getaffinity is
# [Note] Please note that this function respects `cpuset` only when os.sched_getaffinity is
# available (available in most of Linux system, but not OSX and Windows).
# When os.sched_getaffinity is not available, os.cpu_count() is called instead, but
# it doesn't repect cpuset.
# it doesn't respect cpuset.
# We don't take threading into account since each worker process is single threaded
# at this time.
#
@ -886,7 +886,7 @@ class _MultiProcessingDataLoaderIter(_BaseDataLoaderIter):
# 2. A similar issue araises when a `DataLoader` is used in a subprocess.
# When a process ends, it shuts the all its daemonic children
# down with a SIGTERM (instead of joining them without a timeout).
# Simiarly for threads, but by a different mechanism. This fact,
# Similarly for threads, but by a different mechanism. This fact,
# together with a few implementation details of multiprocessing, forces
# us to make workers daemonic. All of our problems arise when a
# DataLoader is used in a subprocess, and are caused by multiprocessing
@ -1017,7 +1017,7 @@ class _MultiProcessingDataLoaderIter(_BaseDataLoaderIter):
# `cancel_join_thread` on that queue if its `IterableDataset` iterator
# happens to exhaust coincidentally, which is out of the control of the
# main process). Thus, since we will exit `pin_memory_thread` before the
# workers (see below), two separete events are used.
# workers (see below), two separate events are used.
#
# NOTE: In short, the protocol is that the main process will set these
# `done_event`s and then the corresponding processes/threads a `None`,
@ -1571,7 +1571,7 @@ class _MultiProcessingDataLoaderIter(_BaseDataLoaderIter):
# (2) since we don't join, the worker may still raise error, and we
# prefer capturing those, rather than ignoring them, even though they
# are raised after the worker has finished its job.
# Joinning is deferred to `_shutdown_workers`, which it is called when
# Joining is deferred to `_shutdown_workers`, which it is called when
# all workers finish their jobs (e.g., `IterableDataset` replicas) or
# when this iterator is garbage collected.

2
torch/utils/data/datapipes/README.md
View File

@ -51,7 +51,7 @@ Note that `__len__` method is optional for `IterDataPipe`.
Like `CSVParserIterDataPipe` in the [Using DataPipe sector](#using-datapipe), `__len__` is not implemented because the size of each file streams is unknown for us before loading it.
Besides, in some special cases, `__len__` method can be provided, but it would either return an integer length or raise Error depending on the arguments of DataPipe.
And, the Error is required to be `TypeError` to support Python's build-in functions like `list(dp)`.
And, the Error is required to be `TypeError` to support Python's built-in functions like `list(dp)`.
Please check NOTE [ Lack of Default `__len__` in Python Abstract Base Classes ] for detailed reason in PyTorch.
### Registering DataPipe with functional API

2
torch/utils/data/datapipes/_typing.py
View File

@ -138,7 +138,7 @@ def _issubtype_with_constraints(variant, constraints, recursive=True):
# - TypeVar[TypeVar[...]]
# So, variant and each constraint may be a TypeVar or a Union.
# In these cases, all of inner types from the variant are required to be
# extraced and verified as a subtype of any constraint. And, all of
# extracted and verified as a subtype of any constraint. And, all of
# inner types from any constraint being a TypeVar or a Union are
# also required to be extracted and verified if the variant belongs to
# any of them.

4
torch/utils/data/datapipes/dataframe/dataframes.py
View File

@ -51,7 +51,7 @@ class DataFrameTracedOps(DFIterDataPipe):
yield self.output_var.apply_ops(item)
# TODO(VitalyFedyunin): Extract this list from the DFIterDataPipe registred functions
# TODO(VitalyFedyunin): Extract this list from the DFIterDataPipe registered functions
DATAPIPES_OPS = [
"_dataframes_as_tuples",
"groupby",
@ -201,7 +201,7 @@ class CaptureLikeMock:
def __init__(self, name):
import unittest.mock as mock
# TODO(VitalyFedyunin): Do not use provate function here, copy own implementation instead.
# TODO(VitalyFedyunin): Do not use private function here, copy own implementation instead.
get_target, attribute = mock._get_target(name) # type: ignore[attr-defined]
self.get_target = get_target
self.attribute = attribute

2
torch/utils/data/datapipes/iter/callable.py
View File

@ -147,7 +147,7 @@ def _collate_helper(conversion, item):
for name in conversion.keys():
if name not in columns_name:
raise RuntimeError("Conversion keys missmatch")
raise RuntimeError("Conversion keys mismatch")
for name in columns_name:
if name in conversion:

4
torch/utils/data/standard_pipes.ipynb
View File

@ -753,7 +753,7 @@
"\n",
"Arguments:\n",
" - `group_key_fn`\n",
" - `group_size` - yeild resulted group as soon as `group_size` elements accumulated\n",
" - `group_size` - yield resulted group as soon as `group_size` elements accumulated\n",
" - `guaranteed_group_size:int = None`\n",
" - `unbatch_level:int = 0` if specified calls `unbatch(unbatch_level=unbatch_level)` on source datapipe before batching (see `unbatch`)\n",
"\n",
@ -962,7 +962,7 @@
"cell_type": "markdown",
"metadata": {},
"source": [
"This behaviour becomes noticable when data is bigger than buffer and some groups getting evicted before gathering all potential items"
"This behaviour becomes noticeable when data is bigger than buffer and some groups getting evicted before gathering all potential items"
]
},
{

4
torch/utils/data/typing.ipynb
View File

@ -399,7 +399,7 @@
"\n",
"Note: This decorator is only allowed to be attached to `__iter__` for now. It can be extended into `__getitem__` and further `nonblocking` functions.\n",
"\n",
"`runtime_validation_disabled` is a context manager to turn off the type validaiton during runtime. It's useful for DataLoader to disable the runtime validaiton after the first epoch is finished for better performance. Note: the runtime validation is enabled by default."
"`runtime_validation_disabled` is a context manager to turn off the type validation during runtime. It's useful for DataLoader to disable the runtime validation after the first epoch is finished for better performance. Note: the runtime validation is enabled by default."
]
},
{
@ -684,7 +684,7 @@
"cell_type": "markdown",
"metadata": {},
"source": [
"- Compatible with context mangager to disable validation"
"- Compatible with context manager to disable validation"
]
},
{

2
torch/utils/hipify/cuda_to_hip_mappings.py
View File

@ -8736,7 +8736,7 @@ CAFFE2_SPECIFIC_MAPPINGS = collections.OrderedDict(
]
)
# We must tread very carefully here. Blanket conversions like are done
# We must treat very carefully here. Blanket conversions like are done
# in CAFFE2_SPECIFIC_MAPPINGS are not presently supported on PyTorch,
# because a regex for CUDA will also match a filename like CUDAGuard.h,
# but the HIPIFY script doesn't presently move the file and so the substitution

2
torch/utils/hipify/hipify_python.py
View File

@ -465,7 +465,7 @@ def find_closure_group(input_string, start, group):
def find_bracket_group(input_string, start):
"""Finds the first balanced parantheses."""
"""Finds the first balanced parentheses."""
return find_closure_group(input_string, start, group=["{", "}"])

8
torch/utils/model_dump/__init__.py
View File

@ -232,14 +232,14 @@ def get_model_info(
model_data = get_pickle("data")
constants = get_pickle("constants")
# Intern strings that are likely to be re-used.
# Intern strings that are likely to be reused.
# Pickle automatically detects shared structure,
# so re-used strings are stored efficiently.
# so reused strings are stored efficiently.
# However, JSON has no way of representing this,
# so we have to do it manually.
interned_strings : dict[str, int] = {}
def ist(s):
def intern(s):
if s not in interned_strings:
interned_strings[s] = len(interned_strings)
return interned_strings[s]
@ -293,7 +293,7 @@ def get_model_info(
s_start = 0
s_end = 0
text = raw_code[start:end]
code_parts.append([text.decode("utf-8"), ist(s_file), s_line, ist(s_text), s_start, s_end])
code_parts.append([text.decode("utf-8"), intern(s_file), s_line, intern(s_text), s_start, s_end])
code_files[zi.filename] = code_parts
extra_files_json_pattern = re.compile(re.escape(path_prefix) + "/extra/.*\\.json")

2
torch/utils/tensorboard/_utils.py
View File

@ -45,7 +45,7 @@ def _prepare_video(V):
Convesrion is done from [batchsize, time(frame), channel(color), height, width] (5D tensor)
to [time(frame), new_width, new_height, channel] (4D tensor).
A batch of images are spreaded to a grid, which forms a frame.
A batch of images are spread to a grid, which forms a frame.
e.g. Video with batchsize 16 will have a 4x4 grid.
"""
b, t, c, h, w = V.shape

4
torch/utils/viz/_cycles.py
View File

@ -70,7 +70,7 @@ def observe_garbage(observer):
gc.callbacks.remove(gc_callback)
return remove
# Function to visualize cycles adapated from refcycle:
# Function to visualize cycles adapted from refcycle:
# Copyright 2013 Mark Dickinson
#
# Licensed under the Apache License, Version 2.0 (the "License");
@ -482,7 +482,7 @@ def warn_tensor_cycles():
Install a warning that reports whenever a cycle that is holding CUDA memory is observed.
The warning produces an .html file that visualizes the cycle,
and links it to the stack frame that allocted the CUDA tensor.
and links it to the stack frame that allocated the CUDA tensor.
Reference cycles are freed by the cycle collector rather than being cleaned up
when the objects in the cycle first become unreachable. If a cycle points to a tensor,