- en: torch.profiler
  id: totrans-0
  prefs:
  - PREF_H1
  type: TYPE_NORMAL
  zh: torch.profiler
- en: 原文：[https://pytorch.org/docs/stable/profiler.html](https://pytorch.org/docs/stable/profiler.html)
  id: totrans-1
  prefs:
  - PREF_BQ
  type: TYPE_NORMAL
  zh: 原文：[https://pytorch.org/docs/stable/profiler.html](https://pytorch.org/docs/stable/profiler.html)
- en: '## Overview'
  id: totrans-2
  prefs: []
  type: TYPE_NORMAL
  zh: '## 概述'
- en: PyTorch Profiler is a tool that allows the collection of performance metrics
    during training and inference. Profiler’s context manager API can be used to better
    understand what model operators are the most expensive, examine their input shapes
    and stack traces, study device kernel activity and visualize the execution trace.
  id: totrans-3
  prefs: []
  type: TYPE_NORMAL
  zh: PyTorch Profiler是一个工具，允许在训练和推断过程中收集性能指标。Profiler的上下文管理器API可用于更好地了解哪些模型操作符是最昂贵的，检查它们的输入形状和堆栈跟踪，研究设备内核活动并可视化执行跟踪。
- en: Note
  id: totrans-4
  prefs: []
  type: TYPE_NORMAL
  zh: 注意
- en: An earlier version of the API in [`torch.autograd`](autograd.html#module-torch.autograd
    "torch.autograd") module is considered legacy and will be deprecated.
  id: totrans-5
  prefs: []
  type: TYPE_NORMAL
  zh: '[`torch.autograd`](autograd.html#module-torch.autograd "torch.autograd")模块中的早期版本被视为遗留版本，并将被弃用。'
- en: API Reference
  id: totrans-6
  prefs:
  - PREF_H2
  type: TYPE_NORMAL
  zh: API参考
- en: '[PRE0]'
  id: totrans-7
  prefs: []
  type: TYPE_PRE
  zh: '[PRE0]'
- en: Low-level profiler wrap the autograd profile
  id: totrans-8
  prefs: []
  type: TYPE_NORMAL
  zh: 低级别分析器包装自动梯度分析
- en: Parameters
  id: totrans-9
  prefs: []
  type: TYPE_NORMAL
  zh: 参数
- en: '**activities** (*iterable*) – list of activity groups (CPU, CUDA) to use in
    profiling, supported values: `torch.profiler.ProfilerActivity.CPU`, `torch.profiler.ProfilerActivity.CUDA`.
    Default value: ProfilerActivity.CPU and (when available) ProfilerActivity.CUDA.'
  id: totrans-10
  prefs:
  - PREF_UL
  type: TYPE_NORMAL
  zh: '**activities**（*可迭代对象*）- 要在分析中使用的活动组（CPU、CUDA）列表，支持的值：`torch.profiler.ProfilerActivity.CPU`、`torch.profiler.ProfilerActivity.CUDA`。默认值：ProfilerActivity.CPU和（如果可用）ProfilerActivity.CUDA。'
- en: '**record_shapes** ([*bool*](https://docs.python.org/3/library/functions.html#bool
    "(in Python v3.12)")) – save information about operator’s input shapes.'
  id: totrans-11
  prefs:
  - PREF_UL
  type: TYPE_NORMAL
  zh: '**record_shapes**（[*bool*](https://docs.python.org/3/library/functions.html#bool
    "(in Python v3.12)")）- 保存有关操作符输入形状的信息。'
- en: '**profile_memory** ([*bool*](https://docs.python.org/3/library/functions.html#bool
    "(in Python v3.12)")) – track tensor memory allocation/deallocation (see `export_memory_timeline`
    for more details).'
  id: totrans-12
  prefs:
  - PREF_UL
  type: TYPE_NORMAL
  zh: '**profile_memory**（[*bool*](https://docs.python.org/3/library/functions.html#bool
    "(in Python v3.12)")）- 跟踪张量内存分配/释放（有关更多详细信息，请参阅`export_memory_timeline`）。'
- en: '**with_stack** ([*bool*](https://docs.python.org/3/library/functions.html#bool
    "(in Python v3.12)")) – record source information (file and line number) for the
    ops.'
  id: totrans-13
  prefs:
  - PREF_UL
  type: TYPE_NORMAL
  zh: '**with_stack**（[*bool*](https://docs.python.org/3/library/functions.html#bool
    "(in Python v3.12)")）- 记录操作的源信息（文件和行号）。'
- en: '**with_flops** ([*bool*](https://docs.python.org/3/library/functions.html#bool
    "(in Python v3.12)")) – use formula to estimate the FLOPS of specific operators
    (matrix multiplication and 2D convolution).'
  id: totrans-14
  prefs:
  - PREF_UL
  type: TYPE_NORMAL
  zh: '**with_flops**（[*bool*](https://docs.python.org/3/library/functions.html#bool
    "(in Python v3.12)")）- 使用公式估计特定操作符的FLOPS（矩阵乘法和2D卷积）。'
- en: '**with_modules** ([*bool*](https://docs.python.org/3/library/functions.html#bool
    "(in Python v3.12)")) – record module hierarchy (including function names) corresponding
    to the callstack of the op. e.g. If module A’s forward call’s module B’s forward
    which contains an aten::add op, then aten::add’s module hierarchy is A.B Note
    that this support exist, at the moment, only for TorchScript models and not eager
    mode models.'
  id: totrans-15
  prefs:
  - PREF_UL
  type: TYPE_NORMAL
  zh: '**with_modules**（[*bool*](https://docs.python.org/3/library/functions.html#bool
    "(in Python v3.12)")）- 记录模块层次结构（包括函数名称），对应于操作的调用堆栈。例如，如果模块A的前向调用的模块B的前向包含一个aten::add操作，则aten::add的模块层次结构是A.B
    请注意，此支持目前仅适用于TorchScript模型，而不适用于急切模式模型。'
- en: '**experimental_config** (*_ExperimentalConfig*) – A set of experimental options
    used by profiler libraries like Kineto. Note, backward compatibility is not guaranteed.'
  id: totrans-16
  prefs:
  - PREF_UL
  type: TYPE_NORMAL
  zh: '**experimental_config**（*_ExperimentalConfig*）- 由像Kineto这样的分析器库使用的一组实验选项。请注意，不保证向后兼容性。'
- en: Note
  id: totrans-17
  prefs: []
  type: TYPE_NORMAL
  zh: 注意
- en: This API is experimental and subject to change in the future.
  id: totrans-18
  prefs: []
  type: TYPE_NORMAL
  zh: 此API是实验性的，未来可能会更改。
- en: Enabling shape and stack tracing results in additional overhead. When record_shapes=True
    is specified, profiler will temporarily hold references to the tensors; that may
    further prevent certain optimizations that depend on the reference count and introduce
    extra tensor copies.
  id: totrans-19
  prefs: []
  type: TYPE_NORMAL
  zh: 启用形状和堆栈跟踪会导致额外的开销。当指定record_shapes=True时，分析器将暂时保留对张量的引用；这可能进一步阻止依赖引用计数的某些优化，并引入额外的张量副本。
- en: '[PRE1]'
  id: totrans-20
  prefs: []
  type: TYPE_PRE
  zh: '[PRE1]'
- en: Adds a user defined metadata with a string key and a string value into the trace
    file
  id: totrans-21
  prefs: []
  type: TYPE_NORMAL
  zh: 向跟踪文件中添加具有字符串键和字符串值的用户定义的元数据
- en: '[PRE2]'
  id: totrans-22
  prefs: []
  type: TYPE_PRE
  zh: '[PRE2]'
- en: Adds a user defined metadata with a string key and a valid json value into the
    trace file
  id: totrans-23
  prefs: []
  type: TYPE_NORMAL
  zh: 向跟踪文件中添加具有字符串键和有效json值的用户定义的元数据
- en: '[PRE3]'
  id: totrans-24
  prefs: []
  type: TYPE_PRE
  zh: '[PRE3]'
- en: Returns the list of unaggregated profiler events, to be used in the trace callback
    or after the profiling is finished
  id: totrans-25
  prefs: []
  type: TYPE_NORMAL
  zh: 返回未聚合的分析器事件列表，用于在跟踪回调中使用或在分析完成后使用
- en: '[PRE4]'
  id: totrans-26
  prefs: []
  type: TYPE_PRE
  zh: '[PRE4]'
- en: Exports the collected trace in Chrome JSON format.
  id: totrans-27
  prefs: []
  type: TYPE_NORMAL
  zh: 以Chrome JSON格式导出收集的跟踪信息。
- en: '[PRE5]'
  id: totrans-28
  prefs: []
  type: TYPE_PRE
  zh: '[PRE5]'
- en: Export memory event information from the profiler collected tree for a given
    device, and export a timeline plot. There are 3 exportable files using `export_memory_timeline`,
    each controlled by the `path`’s suffix.
  id: totrans-29
  prefs: []
  type: TYPE_NORMAL
  zh: 从收集的树中导出分析器的内存事件信息，用于给定设备，并导出时间线图。使用`export_memory_timeline`有3个可导出的文件，每个文件由`path`的后缀控制。
- en: For an HTML compatible plot, use the suffix `.html`, and a memory timeline plot
    will be embedded as a PNG file in the HTML file.
  id: totrans-30
  prefs:
  - PREF_UL
  type: TYPE_NORMAL
  zh: 要生成HTML兼容的绘图，请使用后缀`.html`，内存时间线图将嵌入到HTML文件中作为PNG文件。
- en: For plot points consisting of `[times, [sizes by category]]`, where `times`
    are timestamps and `sizes` are memory usage for each category. The memory timeline
    plot will be saved a JSON (`.json`) or gzipped JSON (`.json.gz`) depending on
    the suffix.
  id: totrans-31
  prefs:
  - PREF_UL
  type: TYPE_NORMAL
  zh: 对于由`[times, [sizes by category]]`组成的绘图点，其中`times`是时间戳，`sizes`是每个类别的内存使用量。内存时间线图将保存为JSON（`.json`）或经过gzip压缩的JSON（`.json.gz`），具体取决于后缀。
- en: For raw memory points, use the suffix `.raw.json.gz`. Each raw memory event
    will consist of `(timestamp, action, numbytes, category)`, where `action` is one
    of `[PREEXISTING, CREATE, INCREMENT_VERSION, DESTROY]`, and `category` is one
    of the enums from `torch.profiler._memory_profiler.Category`.
  id: totrans-32
  prefs:
  - PREF_UL
  type: TYPE_NORMAL
  zh: 对于原始内存点，请使用后缀`.raw.json.gz`。每个原始内存事件将包括`(时间戳，操作，字节数，类别)`，其中`操作`是`[PREEXISTING,
    CREATE, INCREMENT_VERSION, DESTROY]`之一，`类别`是`torch.profiler._memory_profiler.Category`中的枚举之一。
- en: 'Output: Memory timeline written as gzipped JSON, JSON, or HTML.'
  id: totrans-33
  prefs: []
  type: TYPE_NORMAL
  zh: 输出：内存时间线以gzipped JSON、JSON或HTML形式编写。
- en: '[PRE6]'
  id: totrans-34
  prefs: []
  type: TYPE_PRE
  zh: '[PRE6]'
- en: Save stack traces in a file in a format suitable for visualization.
  id: totrans-35
  prefs: []
  type: TYPE_NORMAL
  zh: 将堆栈跟踪保存在适合可视化的文件中。
- en: Parameters
  id: totrans-36
  prefs: []
  type: TYPE_NORMAL
  zh: 参数
- en: '**path** ([*str*](https://docs.python.org/3/library/stdtypes.html#str "(in
    Python v3.12)")) – save stacks file to this location;'
  id: totrans-37
  prefs:
  - PREF_UL
  type: TYPE_NORMAL
  zh: '**path**（[*str*](https://docs.python.org/3/library/stdtypes.html#str "(在Python
    v3.12中)")）- 将堆栈文件保存到此位置；'
- en: '**metric** ([*str*](https://docs.python.org/3/library/stdtypes.html#str "(in
    Python v3.12)")) – metric to use: “self_cpu_time_total” or “self_cuda_time_total”'
  id: totrans-38
  prefs:
  - PREF_UL
  type: TYPE_NORMAL
  zh: '**metric**（[*str*](https://docs.python.org/3/library/stdtypes.html#str "(在Python
    v3.12中)")）- 要使用的度量标准：“self_cpu_time_total”或“self_cuda_time_total”'
- en: Note
  id: totrans-39
  prefs: []
  type: TYPE_NORMAL
  zh: 注意
- en: 'Example of using FlameGraph tool:'
  id: totrans-40
  prefs: []
  type: TYPE_NORMAL
  zh: 使用FlameGraph工具的示例：
- en: git clone [https://github.com/brendangregg/FlameGraph](https://github.com/brendangregg/FlameGraph)
  id: totrans-41
  prefs:
  - PREF_UL
  type: TYPE_NORMAL
  zh: git clone [https://github.com/brendangregg/FlameGraph](https://github.com/brendangregg/FlameGraph)
- en: cd FlameGraph
  id: totrans-42
  prefs:
  - PREF_UL
  type: TYPE_NORMAL
  zh: cd FlameGraph
- en: ./flamegraph.pl –title “CPU time” –countname “us.” profiler.stacks > perf_viz.svg
  id: totrans-43
  prefs:
  - PREF_UL
  type: TYPE_NORMAL
  zh: ./flamegraph.pl –title “CPU time” –countname “us.” profiler.stacks > perf_viz.svg
- en: '[PRE7]'
  id: totrans-44
  prefs: []
  type: TYPE_PRE
  zh: '[PRE7]'
- en: Averages events, grouping them by operator name and (optionally) input shapes
    and stack.
  id: totrans-45
  prefs: []
  type: TYPE_NORMAL
  zh: 通过运算符名称和（可选）输入形状和堆栈对事件进行平均分组。
- en: Note
  id: totrans-46
  prefs: []
  type: TYPE_NORMAL
  zh: 注意
- en: To use shape/stack functionality make sure to set record_shapes/with_stack when
    creating profiler context manager.
  id: totrans-47
  prefs: []
  type: TYPE_NORMAL
  zh: 要使用形状/堆栈功能，请确保在创建分析器上下文管理器时设置record_shapes/with_stack。
- en: '[PRE8]'
  id: totrans-48
  prefs: []
  type: TYPE_PRE
  zh: '[PRE8]'
- en: Profiler context manager.
  id: totrans-49
  prefs: []
  type: TYPE_NORMAL
  zh: 分析器上下文管理器。
- en: Parameters
  id: totrans-50
  prefs: []
  type: TYPE_NORMAL
  zh: 参数
- en: '**activities** (*iterable*) – list of activity groups (CPU, CUDA) to use in
    profiling, supported values: `torch.profiler.ProfilerActivity.CPU`, `torch.profiler.ProfilerActivity.CUDA`.
    Default value: ProfilerActivity.CPU and (when available) ProfilerActivity.CUDA.'
  id: totrans-51
  prefs:
  - PREF_UL
  type: TYPE_NORMAL
  zh: '**activities**（*iterable*）- 用于分析的活动组（CPU，CUDA）列表，支持的值：`torch.profiler.ProfilerActivity.CPU`，`torch.profiler.ProfilerActivity.CUDA`。默认值：ProfilerActivity.CPU和（如果可用）ProfilerActivity.CUDA。'
- en: '**schedule** (*Callable*) – callable that takes step (int) as a single parameter
    and returns `ProfilerAction` value that specifies the profiler action to perform
    at each step.'
  id: totrans-52
  prefs:
  - PREF_UL
  type: TYPE_NORMAL
  zh: '**schedule**（*Callable*）- 接受步骤（int）作为单个参数并返回指定在每个步骤执行的分析器操作的`ProfilerAction`值的可调用对象。'
- en: '**on_trace_ready** (*Callable*) – callable that is called at each step when
    `schedule` returns `ProfilerAction.RECORD_AND_SAVE` during the profiling.'
  id: totrans-53
  prefs:
  - PREF_UL
  type: TYPE_NORMAL
  zh: '**on_trace_ready**（*Callable*）- 在分析期间`schedule`返回`ProfilerAction.RECORD_AND_SAVE`时在每个步骤调用的可调用对象。'
- en: '**record_shapes** ([*bool*](https://docs.python.org/3/library/functions.html#bool
    "(in Python v3.12)")) – save information about operator’s input shapes.'
  id: totrans-54
  prefs:
  - PREF_UL
  type: TYPE_NORMAL
  zh: '**record_shapes**（[*bool*](https://docs.python.org/3/library/functions.html#bool
    "(在Python v3.12中)")）- 保存有关运算符输入形状的信息。'
- en: '**profile_memory** ([*bool*](https://docs.python.org/3/library/functions.html#bool
    "(in Python v3.12)")) – track tensor memory allocation/deallocation.'
  id: totrans-55
  prefs:
  - PREF_UL
  type: TYPE_NORMAL
  zh: '**profile_memory**（[*bool*](https://docs.python.org/3/library/functions.html#bool
    "(在Python v3.12中)")）- 跟踪张量内存分配/释放。'
- en: '**with_stack** ([*bool*](https://docs.python.org/3/library/functions.html#bool
    "(in Python v3.12)")) – record source information (file and line number) for the
    ops.'
  id: totrans-56
  prefs:
  - PREF_UL
  type: TYPE_NORMAL
  zh: '**with_stack**（[*bool*](https://docs.python.org/3/library/functions.html#bool
    "(在Python v3.12中)")）- 记录操作的源信息（文件和行号）。'
- en: '**with_flops** ([*bool*](https://docs.python.org/3/library/functions.html#bool
    "(in Python v3.12)")) – use formula to estimate the FLOPs (floating point operations)
    of specific operators (matrix multiplication and 2D convolution).'
  id: totrans-57
  prefs:
  - PREF_UL
  type: TYPE_NORMAL
  zh: '**with_flops**（[*bool*](https://docs.python.org/3/library/functions.html#bool
    "(在Python v3.12中)")）- 使用公式估算特定运算符（矩阵乘法和2D卷积）的FLOPs（浮点运算）。'
- en: '**with_modules** ([*bool*](https://docs.python.org/3/library/functions.html#bool
    "(in Python v3.12)")) – record module hierarchy (including function names) corresponding
    to the callstack of the op. e.g. If module A’s forward call’s module B’s forward
    which contains an aten::add op, then aten::add’s module hierarchy is A.B Note
    that this support exist, at the moment, only for TorchScript models and not eager
    mode models.'
  id: totrans-58
  prefs:
  - PREF_UL
  type: TYPE_NORMAL
  zh: '**with_modules**（[*bool*](https://docs.python.org/3/library/functions.html#bool
    "(在Python v3.12中)")）- 记录与操作的调用堆栈对应的模块层次结构（包括函数名称）。例如，如果模块A的前向调用模块B的前向，其中包含一个aten::add操作，则aten::add的模块层次结构是A.B。请注意，此支持目前仅适用于TorchScript模型，而不适用于急切模式模型。'
- en: '**experimental_config** (*_ExperimentalConfig*) – A set of experimental options
    used for Kineto library features. Note, backward compatibility is not guaranteed.'
  id: totrans-59
  prefs:
  - PREF_UL
  type: TYPE_NORMAL
  zh: '**experimental_config**（*_ExperimentalConfig*）- 用于Kineto库功能的一组实验选项。请注意，不保证向后兼容性。'
- en: '**use_cuda** ([*bool*](https://docs.python.org/3/library/functions.html#bool
    "(in Python v3.12)")) –'
  id: totrans-60
  prefs:
  - PREF_UL
  type: TYPE_NORMAL
  zh: '**use_cuda**（[*bool*](https://docs.python.org/3/library/functions.html#bool
    "(在Python v3.12中)")）-'
- en: 'Deprecated since version 1.8.1: use `activities` instead.'
  id: totrans-61
  prefs:
  - PREF_IND
  type: TYPE_NORMAL
  zh: 自1.8.1版本起已弃用：请改用`activities`。
- en: Note
  id: totrans-62
  prefs: []
  type: TYPE_NORMAL
  zh: 注意
- en: Use [`schedule()`](#torch.profiler.schedule "torch.profiler.schedule") to generate
    the callable schedule. Non-default schedules are useful when profiling long training
    jobs and allow the user to obtain multiple traces at the different iterations
    of the training process. The default schedule simply records all the events continuously
    for the duration of the context manager.
  id: totrans-63
  prefs: []
  type: TYPE_NORMAL
  zh: 使用[`schedule()`](#torch.profiler.schedule "torch.profiler.schedule")生成可调度的调度。非默认调度在分析长时间训练作业时很有用，并允许用户在训练过程的不同迭代中获取多个跟踪。默认调度仅在上下文管理器的持续时间内连续记录所有事件。
- en: Note
  id: totrans-64
  prefs: []
  type: TYPE_NORMAL
  zh: 注意
- en: 'Use [`tensorboard_trace_handler()`](#torch.profiler.tensorboard_trace_handler
    "torch.profiler.tensorboard_trace_handler") to generate result files for TensorBoard:'
  id: totrans-65
  prefs: []
  type: TYPE_NORMAL
  zh: 使用[`tensorboard_trace_handler()`](#torch.profiler.tensorboard_trace_handler
    "torch.profiler.tensorboard_trace_handler")生成TensorBoard的结果文件：
- en: '`on_trace_ready=torch.profiler.tensorboard_trace_handler(dir_name)`'
  id: totrans-66
  prefs: []
  type: TYPE_NORMAL
  zh: '`on_trace_ready=torch.profiler.tensorboard_trace_handler(dir_name)`'
- en: 'After profiling, result files can be found in the specified directory. Use
    the command:'
  id: totrans-67
  prefs: []
  type: TYPE_NORMAL
  zh: 分析后，结果文件可以在指定目录中找到。使用命令：
- en: '`tensorboard --logdir dir_name`'
  id: totrans-68
  prefs: []
  type: TYPE_NORMAL
  zh: '`tensorboard --logdir dir_name`'
- en: to see the results in TensorBoard. For more information, see [PyTorch Profiler
    TensorBoard Plugin](https://github.com/pytorch/kineto/tree/master/tb_plugin)
  id: totrans-69
  prefs: []
  type: TYPE_NORMAL
  zh: 在TensorBoard中查看结果。有关更多信息，请参阅[PyTorch Profiler TensorBoard插件](https://github.com/pytorch/kineto/tree/master/tb_plugin)
- en: Note
  id: totrans-70
  prefs: []
  type: TYPE_NORMAL
  zh: 注意
- en: Enabling shape and stack tracing results in additional overhead. When record_shapes=True
    is specified, profiler will temporarily hold references to the tensors; that may
    further prevent certain optimizations that depend on the reference count and introduce
    extra tensor copies.
  id: totrans-71
  prefs: []
  type: TYPE_NORMAL
  zh: 启用形状和堆栈跟踪会导致额外的开销。当指定record_shapes=True时，分析器将暂时保留对张量的引用；这可能进一步阻止依赖引用计数的某些优化，并引入额外的张量副本。
- en: 'Examples:'
  id: totrans-72
  prefs: []
  type: TYPE_NORMAL
  zh: 示例：
- en: '[PRE9]'
  id: totrans-73
  prefs: []
  type: TYPE_PRE
  zh: '[PRE9]'
- en: 'Using the profiler’s `schedule`, `on_trace_ready` and `step` functions:'
  id: totrans-74
  prefs: []
  type: TYPE_NORMAL
  zh: 使用分析器的`schedule`、`on_trace_ready`和`step`函数：
- en: '[PRE10]'
  id: totrans-75
  prefs: []
  type: TYPE_PRE
  zh: '[PRE10]'
- en: '[PRE11]'
  id: totrans-76
  prefs: []
  type: TYPE_PRE
  zh: '[PRE11]'
- en: Signals the profiler that the next profiling step has started.
  id: totrans-77
  prefs: []
  type: TYPE_NORMAL
  zh: 信号分析器下一个分析步骤已经开始。
- en: '[PRE12]'
  id: totrans-78
  prefs: []
  type: TYPE_PRE
  zh: '[PRE12]'
- en: Profiler actions that can be taken at the specified intervals
  id: totrans-79
  prefs: []
  type: TYPE_NORMAL
  zh: 在指定间隔可以执行的分析器操作
- en: '[PRE13]'
  id: totrans-80
  prefs: []
  type: TYPE_PRE
  zh: '[PRE13]'
- en: 'Members:'
  id: totrans-81
  prefs: []
  type: TYPE_NORMAL
  zh: 成员：
- en: CPU
  id: totrans-82
  prefs: []
  type: TYPE_NORMAL
  zh: CPU
- en: XPU
  id: totrans-83
  prefs: []
  type: TYPE_NORMAL
  zh: XPU
- en: MTIA
  id: totrans-84
  prefs: []
  type: TYPE_NORMAL
  zh: MTIA
- en: CUDA
  id: totrans-85
  prefs: []
  type: TYPE_NORMAL
  zh: CUDA
- en: '[PRE14]'
  id: totrans-86
  prefs: []
  type: TYPE_PRE
  zh: '[PRE14]'
- en: '[PRE15]'
  id: totrans-87
  prefs: []
  type: TYPE_PRE
  zh: '[PRE15]'
- en: Returns a callable that can be used as profiler `schedule` argument. The profiler
    will skip the first `skip_first` steps, then wait for `wait` steps, then do the
    warmup for the next `warmup` steps, then do the active recording for the next
    `active` steps and then repeat the cycle starting with `wait` steps. The optional
    number of cycles is specified with the `repeat` parameter, the zero value means
    that the cycles will continue until the profiling is finished.
  id: totrans-88
  prefs: []
  type: TYPE_NORMAL
  zh: 返回一个可用作分析器`schedule`参数的可调用对象。分析器将跳过前`skip_first`步，然后等待`wait`步，然后为接下来的`warmup`步进行预热，然后为接下来的`active`步进行活动记录，然后重复以`wait`步开始的循环。循环的可选次数由`repeat`参数指定，零值表示循环将持续直到分析完成。
- en: Return type
  id: totrans-89
  prefs: []
  type: TYPE_NORMAL
  zh: 返回类型
- en: '[*Callable*](https://docs.python.org/3/library/typing.html#typing.Callable
    "(in Python v3.12)")'
  id: totrans-90
  prefs: []
  type: TYPE_NORMAL
  zh: '[*Callable*](https://docs.python.org/3/library/typing.html#typing.Callable
    "(在 Python v3.12 中)")'
- en: '[PRE16]'
  id: totrans-91
  prefs: []
  type: TYPE_PRE
  zh: '[PRE16]'
- en: Outputs tracing files to directory of `dir_name`, then that directory can be
    directly delivered to tensorboard as logdir. `worker_name` should be unique for
    each worker in distributed scenario, it will be set to ‘[hostname]_[pid]’ by default.
  id: totrans-92
  prefs: []
  type: TYPE_NORMAL
  zh: 将跟踪文件输出到`dir_name`目录，然后该目录可以直接作为logdir传递给tensorboard。在分布式场景中，`worker_name`应该对每个worker是唯一的，默认情况下将设置为‘[hostname]_[pid]’。
- en: Intel Instrumentation and Tracing Technology APIs[](#intel-instrumentation-and-tracing-technology-apis
    "Permalink to this heading")
  id: totrans-93
  prefs:
  - PREF_H2
  type: TYPE_NORMAL
  zh: 英特尔仪器和跟踪技术APIs[](#intel-instrumentation-and-tracing-technology-apis "跳转到此标题")
- en: '[PRE17]'
  id: totrans-94
  prefs: []
  type: TYPE_PRE
  zh: '[PRE17]'
- en: Check if ITT feature is available or not
  id: totrans-95
  prefs: []
  type: TYPE_NORMAL
  zh: 检查ITT功能是否可用
- en: '[PRE18]'
  id: totrans-96
  prefs: []
  type: TYPE_PRE
  zh: '[PRE18]'
- en: Describe an instantaneous event that occurred at some point.
  id: totrans-97
  prefs: []
  type: TYPE_NORMAL
  zh: 描述在某个时间点发生的瞬时事件。
- en: Parameters
  id: totrans-98
  prefs: []
  type: TYPE_NORMAL
  zh: 参数
- en: '**msg** ([*str*](https://docs.python.org/3/library/stdtypes.html#str "(in Python
    v3.12)")) – ASCII message to associate with the event.'
  id: totrans-99
  prefs: []
  type: TYPE_NORMAL
  zh: '**msg** ([*str*](https://docs.python.org/3/library/stdtypes.html#str "(在 Python
    v3.12 中)")) – 与事件关联的ASCII消息。'
- en: '[PRE19]'
  id: totrans-100
  prefs: []
  type: TYPE_PRE
  zh: '[PRE19]'
- en: Pushes a range onto a stack of nested range span. Returns zero-based depth of
    the range that is started.
  id: totrans-101
  prefs: []
  type: TYPE_NORMAL
  zh: 将范围推送到嵌套范围堆栈上。返回开始的范围的从零开始的深度。
- en: Parameters
  id: totrans-102
  prefs: []
  type: TYPE_NORMAL
  zh: 参数
- en: '**msg** ([*str*](https://docs.python.org/3/library/stdtypes.html#str "(in Python
    v3.12)")) – ASCII message to associate with range'
  id: totrans-103
  prefs: []
  type: TYPE_NORMAL
  zh: '**msg** ([*str*](https://docs.python.org/3/library/stdtypes.html#str "(在 Python
    v3.12 中)")) – 与范围关联的ASCII消息'
- en: '[PRE20]'
  id: totrans-104
  prefs: []
  type: TYPE_PRE
  zh: '[PRE20]'
- en: Pops a range off of a stack of nested range spans. Returns the zero-based depth
    of the range that is ended.
  id: totrans-105
  prefs: []
  type: TYPE_NORMAL
  zh: 从嵌套范围跨度堆栈中弹出一个范围。返回结束的范围的从零开始的深度。