评价此页
★ ★ ★ ★ ★
fullgraph=False">
error_on_graph_break">

切换 error_on_graph_break#

创建日期：2025 年 9 月 3 日 | 最后更新日期：2025 年 9 月 3 日

摘要

  • 当 fullgraph=False 时，我们可以使用 torch._dynamo.error_on_graph_break() 来更灵活地处理图中断。

到目前为止，我们已经介绍了两种处理 torch.compile 中图中断的方法。

  1. fullgraph=True 会在第一次图中断时报错，并额外保证只从代码中跟踪一个图。

  2. fullgraph=False 在遇到图中断时也会继续跟踪。

如果我们希望大部分代码都不允许图中断，但有少数有问题的函数，这些函数的图中断很难消除，而我们对此是可以接受的，该怎么办？我们可以使用 torch._dynamo.error_on_graph_break() 来实现此目的。

torch.compile 有一个 error_on_graph_break 设置（初始值为 False）。如果在 error_on_graph_break 设置为 False 的情况下，在代码中发生图中断或编译器错误，那么 torch.compile 将尝试在图中断/错误后继续编译。如果 error_on_graph_break 设置为 True，那么 torch.compile 将中止编译并将错误传播给用户代码。

error_on_graph_break=True 和 fullgraph=True 之间的显著区别在于，前者**不保证捕获单个图**。error_on_graph_break**可以在编译时任意切换**，通过使用 torch._dynamo.error_on_graph_break() 上下文管理器/装饰器。相比之下，一旦 fullgraph 设置为 True，就无法将其重新设置为 False。最后，error_on_graph_break 的优先级低于 fullgraph — error_on_graph_break 仅在 fullgraph=False 时生效。

error_on_graph_break(False) 示例#

@torch._dynamo.error_on_graph_break(False)
def code_with_a_difficult_graph_break(x):
    x = x + 1
    torch._dynamo.graph_break()
    return x + 2

def inner(x):
    return code_with_a_difficult_graph_break(x)

# NOTE: fullgraph=False
@torch._dynamo.error_on_graph_break(True)
@torch.compile
def fn(x):
    return inner(x)

# No error, but there is a graph break
fn(torch.randn(3))

Graph break in user code at /tmp/ipykernel_880/1452578661.py:4
Graph Break Reason: Call to `torch._dynamo.graph_break()`
  Explanation: User-inserted graph break. Message: None
  Hint: Remove the `torch._dynamo.graph_break()` call.

  Developer debug context: Called `torch._dynamo.graph_break()` with args `[]`, kwargs `{}`

 For more details about this graph break, please visit: https://meta-pytorch.github.io/compile-graph-break-site/gb/gb0025.html
User code traceback:
  File "/opt/conda/envs/py_3.10/lib/python3.10/runpy.py", line 196, in _run_module_as_main
    return _run_code(code, main_globals, None,
  File "/opt/conda/envs/py_3.10/lib/python3.10/runpy.py", line 86, in _run_code
    exec(code, run_globals)
  File "/opt/conda/envs/py_3.10/lib/python3.10/site-packages/ipykernel_launcher.py", line 18, in <module>
    app.launch_new_instance()
  File "/opt/conda/envs/py_3.10/lib/python3.10/site-packages/traitlets/config/application.py", line 1075, in launch_instance
    app.start()
  File "/opt/conda/envs/py_3.10/lib/python3.10/site-packages/ipykernel/kernelapp.py", line 739, in start
    self.io_loop.start()
  File "/opt/conda/envs/py_3.10/lib/python3.10/site-packages/tornado/platform/asyncio.py", line 211, in start
    self.asyncio_loop.run_forever()
  File "/opt/conda/envs/py_3.10/lib/python3.10/asyncio/base_events.py", line 603, in run_forever
    self._run_once()
  File "/opt/conda/envs/py_3.10/lib/python3.10/asyncio/base_events.py", line 1909, in _run_once
    handle._run()
  File "/opt/conda/envs/py_3.10/lib/python3.10/asyncio/events.py", line 80, in _run
    self._context.run(self._callback, *self._args)
  File "/opt/conda/envs/py_3.10/lib/python3.10/site-packages/ipykernel/kernelbase.py", line 519, in dispatch_queue
    await self.process_one()
  File "/opt/conda/envs/py_3.10/lib/python3.10/site-packages/ipykernel/kernelbase.py", line 508, in process_one
    await dispatch(*args)
  File "/opt/conda/envs/py_3.10/lib/python3.10/site-packages/ipykernel/kernelbase.py", line 400, in dispatch_shell
    await result
  File "/opt/conda/envs/py_3.10/lib/python3.10/site-packages/ipykernel/ipkernel.py", line 368, in execute_request
    await super().execute_request(stream, ident, parent)
  File "/opt/conda/envs/py_3.10/lib/python3.10/site-packages/ipykernel/kernelbase.py", line 767, in execute_request
    reply_content = await reply_content
  File "/opt/conda/envs/py_3.10/lib/python3.10/site-packages/ipykernel/ipkernel.py", line 455, in do_execute
    res = shell.run_cell(
  File "/opt/conda/envs/py_3.10/lib/python3.10/site-packages/ipykernel/zmqshell.py", line 577, in run_cell
    return super().run_cell(*args, **kwargs)
  File "/opt/conda/envs/py_3.10/lib/python3.10/site-packages/IPython/core/interactiveshell.py", line 3006, in run_cell
    result = self._run_cell(
  File "/opt/conda/envs/py_3.10/lib/python3.10/site-packages/IPython/core/interactiveshell.py", line 3061, in _run_cell
    result = runner(coro)
  File "/opt/conda/envs/py_3.10/lib/python3.10/site-packages/IPython/core/async_helpers.py", line 129, in _pseudo_sync_runner
    coro.send(None)
  File "/opt/conda/envs/py_3.10/lib/python3.10/site-packages/IPython/core/interactiveshell.py", line 3266, in run_cell_async
    has_raised = await self.run_ast_nodes(code_ast.body, cell_name,
  File "/opt/conda/envs/py_3.10/lib/python3.10/site-packages/IPython/core/interactiveshell.py", line 3445, in run_ast_nodes
    if await self.run_code(code, result, async_=asy):
  File "/opt/conda/envs/py_3.10/lib/python3.10/site-packages/IPython/core/interactiveshell.py", line 3505, in run_code
    exec(code_obj, self.user_global_ns, self.user_ns)
  File "/tmp/ipykernel_880/1452578661.py", line 17, in <module>
    fn(torch.randn(3))
  File "/opt/conda/envs/py_3.10/lib/python3.10/site-packages/torch/_dynamo/external_utils.py", line 209, in inner
    return func(*args, **kwargs)
  File "/tmp/ipykernel_880/1452578661.py", line 14, in fn
    return inner(x)
  File "/tmp/ipykernel_880/1452578661.py", line 8, in inner
    return code_with_a_difficult_graph_break(x)
  File "/opt/conda/envs/py_3.10/lib/python3.10/site-packages/torch/_dynamo/external_utils.py", line 209, in inner
    return func(*args, **kwargs)
  File "/tmp/ipykernel_880/1452578661.py", line 4, in code_with_a_difficult_graph_break
    torch._dynamo.graph_break()

Graph break (user stack suppressed due to duplicate graph break) in user code at /tmp/ipykernel_880/1452578661.py:4
Graph Break Reason: Call to `torch._dynamo.graph_break()`
  Explanation: User-inserted graph break. Message: None
  Hint: Remove the `torch._dynamo.graph_break()` call.

  Developer debug context: Called `torch._dynamo.graph_break()` with args `[]`, kwargs `{}`

 For more details about this graph break, please visit: https://meta-pytorch.github.io/compile-graph-break-site/gb/gb0025.html
Graph break (user stack suppressed due to duplicate graph break) in user code at /tmp/ipykernel_880/1452578661.py:4
Graph Break Reason: Call to `torch._dynamo.graph_break()`
  Explanation: User-inserted graph break. Message: None
  Hint: Remove the `torch._dynamo.graph_break()` call.

  Developer debug context: Called `torch._dynamo.graph_break()` with args `[]`, kwargs `{}`

 For more details about this graph break, please visit: https://meta-pytorch.github.io/compile-graph-break-site/gb/gb0025.html
Graph break (user stack suppressed due to duplicate graph break) in user code at /tmp/ipykernel_880/1452578661.py:4
Graph Break Reason: Call to `torch._dynamo.graph_break()`
  Explanation: User-inserted graph break. Message: None
  Hint: Remove the `torch._dynamo.graph_break()` call.

  Developer debug context: Called `torch._dynamo.graph_break()` with args `[]`, kwargs `{}`

 For more details about this graph break, please visit: https://meta-pytorch.github.io/compile-graph-break-site/gb/gb0025.html

tensor([1.5304, 1.4924, 0.9838])

在 error_on_graph_break(True) 下使用 error_on_graph_break(False)，有助于在我们希望最小化图中断（即遵循 fullgraph=True 编程模型）但存在某些难以规避的、非性能关键的图中断代码段时。

error_on_graph_break() 也可以用作上下文管理器。

# NOTE: fullgraph=False
@torch._dynamo.error_on_graph_break(True)
@torch.compile
def fn(x):
    x = x + 1
    with torch._dynamo.error_on_graph_break(False):
        torch._dynamo.graph_break()  # no error
    return x + 2

# No error, but there is a graph break
fn(torch.randn(3))

Graph break in user code at /tmp/ipykernel_880/737485247.py:7
Graph Break Reason: Call to `torch._dynamo.graph_break()`
  Explanation: User-inserted graph break. Message: None
  Hint: Remove the `torch._dynamo.graph_break()` call.

  Developer debug context: Called `torch._dynamo.graph_break()` with args `[]`, kwargs `{}`

 For more details about this graph break, please visit: https://meta-pytorch.github.io/compile-graph-break-site/gb/gb0025.html
User code traceback:
  File "/opt/conda/envs/py_3.10/lib/python3.10/runpy.py", line 196, in _run_module_as_main
    return _run_code(code, main_globals, None,
  File "/opt/conda/envs/py_3.10/lib/python3.10/runpy.py", line 86, in _run_code
    exec(code, run_globals)
  File "/opt/conda/envs/py_3.10/lib/python3.10/site-packages/ipykernel_launcher.py", line 18, in <module>
    app.launch_new_instance()
  File "/opt/conda/envs/py_3.10/lib/python3.10/site-packages/traitlets/config/application.py", line 1075, in launch_instance
    app.start()
  File "/opt/conda/envs/py_3.10/lib/python3.10/site-packages/ipykernel/kernelapp.py", line 739, in start
    self.io_loop.start()
  File "/opt/conda/envs/py_3.10/lib/python3.10/site-packages/tornado/platform/asyncio.py", line 211, in start
    self.asyncio_loop.run_forever()
  File "/opt/conda/envs/py_3.10/lib/python3.10/asyncio/base_events.py", line 603, in run_forever
    self._run_once()
  File "/opt/conda/envs/py_3.10/lib/python3.10/asyncio/base_events.py", line 1909, in _run_once
    handle._run()
  File "/opt/conda/envs/py_3.10/lib/python3.10/asyncio/events.py", line 80, in _run
    self._context.run(self._callback, *self._args)
  File "/opt/conda/envs/py_3.10/lib/python3.10/site-packages/ipykernel/kernelbase.py", line 519, in dispatch_queue
    await self.process_one()
  File "/opt/conda/envs/py_3.10/lib/python3.10/site-packages/ipykernel/kernelbase.py", line 508, in process_one
    await dispatch(*args)
  File "/opt/conda/envs/py_3.10/lib/python3.10/site-packages/ipykernel/kernelbase.py", line 400, in dispatch_shell
    await result
  File "/opt/conda/envs/py_3.10/lib/python3.10/site-packages/ipykernel/ipkernel.py", line 368, in execute_request
    await super().execute_request(stream, ident, parent)
  File "/opt/conda/envs/py_3.10/lib/python3.10/site-packages/ipykernel/kernelbase.py", line 767, in execute_request
    reply_content = await reply_content
  File "/opt/conda/envs/py_3.10/lib/python3.10/site-packages/ipykernel/ipkernel.py", line 455, in do_execute
    res = shell.run_cell(
  File "/opt/conda/envs/py_3.10/lib/python3.10/site-packages/ipykernel/zmqshell.py", line 577, in run_cell
    return super().run_cell(*args, **kwargs)
  File "/opt/conda/envs/py_3.10/lib/python3.10/site-packages/IPython/core/interactiveshell.py", line 3006, in run_cell
    result = self._run_cell(
  File "/opt/conda/envs/py_3.10/lib/python3.10/site-packages/IPython/core/interactiveshell.py", line 3061, in _run_cell
    result = runner(coro)
  File "/opt/conda/envs/py_3.10/lib/python3.10/site-packages/IPython/core/async_helpers.py", line 129, in _pseudo_sync_runner
    coro.send(None)
  File "/opt/conda/envs/py_3.10/lib/python3.10/site-packages/IPython/core/interactiveshell.py", line 3266, in run_cell_async
    has_raised = await self.run_ast_nodes(code_ast.body, cell_name,
  File "/opt/conda/envs/py_3.10/lib/python3.10/site-packages/IPython/core/interactiveshell.py", line 3445, in run_ast_nodes
    if await self.run_code(code, result, async_=asy):
  File "/opt/conda/envs/py_3.10/lib/python3.10/site-packages/IPython/core/interactiveshell.py", line 3505, in run_code
    exec(code_obj, self.user_global_ns, self.user_ns)
  File "/tmp/ipykernel_880/737485247.py", line 11, in <module>
    fn(torch.randn(3))
  File "/opt/conda/envs/py_3.10/lib/python3.10/site-packages/torch/_dynamo/external_utils.py", line 209, in inner
    return func(*args, **kwargs)
  File "/tmp/ipykernel_880/737485247.py", line 7, in fn
    torch._dynamo.graph_break()  # no error

tensor([4.5572, 3.4907, 1.6146])

您可以使用猴子补丁（monkey patching）来切换 error_on_graph_break，以处理您无法编辑源代码的代码（例如，框架代码）。

class ThirdPartyModule(torch.nn.Module):
    def forward(self, x):
        x = x + 1
        torch._dynamo.graph_break()
        return x + 2

tp_mod = ThirdPartyModule()
tp_mod.forward = torch._dynamo.error_on_graph_break(False)(tp_mod.forward)

@torch._dynamo.error_on_graph_break(True)
@torch.compile
def fn(x):
    return tp_mod.forward(x)

# No error, but there is a graph break
fn(torch.randn(3))

Graph break in user code at /tmp/ipykernel_880/2112598647.py:4
Graph Break Reason: Call to `torch._dynamo.graph_break()`
  Explanation: User-inserted graph break. Message: None
  Hint: Remove the `torch._dynamo.graph_break()` call.

  Developer debug context: Called `torch._dynamo.graph_break()` with args `[]`, kwargs `{}`

 For more details about this graph break, please visit: https://meta-pytorch.github.io/compile-graph-break-site/gb/gb0025.html
User code traceback:
  File "/opt/conda/envs/py_3.10/lib/python3.10/runpy.py", line 196, in _run_module_as_main
    return _run_code(code, main_globals, None,
  File "/opt/conda/envs/py_3.10/lib/python3.10/runpy.py", line 86, in _run_code
    exec(code, run_globals)
  File "/opt/conda/envs/py_3.10/lib/python3.10/site-packages/ipykernel_launcher.py", line 18, in <module>
    app.launch_new_instance()
  File "/opt/conda/envs/py_3.10/lib/python3.10/site-packages/traitlets/config/application.py", line 1075, in launch_instance
    app.start()
  File "/opt/conda/envs/py_3.10/lib/python3.10/site-packages/ipykernel/kernelapp.py", line 739, in start
    self.io_loop.start()
  File "/opt/conda/envs/py_3.10/lib/python3.10/site-packages/tornado/platform/asyncio.py", line 211, in start
    self.asyncio_loop.run_forever()
  File "/opt/conda/envs/py_3.10/lib/python3.10/asyncio/base_events.py", line 603, in run_forever
    self._run_once()
  File "/opt/conda/envs/py_3.10/lib/python3.10/asyncio/base_events.py", line 1909, in _run_once
    handle._run()
  File "/opt/conda/envs/py_3.10/lib/python3.10/asyncio/events.py", line 80, in _run
    self._context.run(self._callback, *self._args)
  File "/opt/conda/envs/py_3.10/lib/python3.10/site-packages/ipykernel/kernelbase.py", line 519, in dispatch_queue
    await self.process_one()
  File "/opt/conda/envs/py_3.10/lib/python3.10/site-packages/ipykernel/kernelbase.py", line 508, in process_one
    await dispatch(*args)
  File "/opt/conda/envs/py_3.10/lib/python3.10/site-packages/ipykernel/kernelbase.py", line 400, in dispatch_shell
    await result
  File "/opt/conda/envs/py_3.10/lib/python3.10/site-packages/ipykernel/ipkernel.py", line 368, in execute_request
    await super().execute_request(stream, ident, parent)
  File "/opt/conda/envs/py_3.10/lib/python3.10/site-packages/ipykernel/kernelbase.py", line 767, in execute_request
    reply_content = await reply_content
  File "/opt/conda/envs/py_3.10/lib/python3.10/site-packages/ipykernel/ipkernel.py", line 455, in do_execute
    res = shell.run_cell(
  File "/opt/conda/envs/py_3.10/lib/python3.10/site-packages/ipykernel/zmqshell.py", line 577, in run_cell
    return super().run_cell(*args, **kwargs)
  File "/opt/conda/envs/py_3.10/lib/python3.10/site-packages/IPython/core/interactiveshell.py", line 3006, in run_cell
    result = self._run_cell(
  File "/opt/conda/envs/py_3.10/lib/python3.10/site-packages/IPython/core/interactiveshell.py", line 3061, in _run_cell
    result = runner(coro)
  File "/opt/conda/envs/py_3.10/lib/python3.10/site-packages/IPython/core/async_helpers.py", line 129, in _pseudo_sync_runner
    coro.send(None)
  File "/opt/conda/envs/py_3.10/lib/python3.10/site-packages/IPython/core/interactiveshell.py", line 3266, in run_cell_async
    has_raised = await self.run_ast_nodes(code_ast.body, cell_name,
  File "/opt/conda/envs/py_3.10/lib/python3.10/site-packages/IPython/core/interactiveshell.py", line 3445, in run_ast_nodes
    if await self.run_code(code, result, async_=asy):
  File "/opt/conda/envs/py_3.10/lib/python3.10/site-packages/IPython/core/interactiveshell.py", line 3505, in run_code
    exec(code_obj, self.user_global_ns, self.user_ns)
  File "/tmp/ipykernel_880/2112598647.py", line 16, in <module>
    fn(torch.randn(3))
  File "/opt/conda/envs/py_3.10/lib/python3.10/site-packages/torch/_dynamo/external_utils.py", line 209, in inner
    return func(*args, **kwargs)
  File "/tmp/ipykernel_880/2112598647.py", line 13, in fn
    return tp_mod.forward(x)
  File "/opt/conda/envs/py_3.10/lib/python3.10/site-packages/torch/_dynamo/external_utils.py", line 209, in inner
    return func(*args, **kwargs)
  File "/tmp/ipykernel_880/2112598647.py", line 4, in forward
    torch._dynamo.graph_break()

Graph break (user stack suppressed due to duplicate graph break) in user code at /tmp/ipykernel_880/2112598647.py:4
Graph Break Reason: Call to `torch._dynamo.graph_break()`
  Explanation: User-inserted graph break. Message: None
  Hint: Remove the `torch._dynamo.graph_break()` call.

  Developer debug context: Called `torch._dynamo.graph_break()` with args `[]`, kwargs `{}`

 For more details about this graph break, please visit: https://meta-pytorch.github.io/compile-graph-break-site/gb/gb0025.html
Graph break (user stack suppressed due to duplicate graph break) in user code at /tmp/ipykernel_880/2112598647.py:4
Graph Break Reason: Call to `torch._dynamo.graph_break()`
  Explanation: User-inserted graph break. Message: None
  Hint: Remove the `torch._dynamo.graph_break()` call.

  Developer debug context: Called `torch._dynamo.graph_break()` with args `[]`, kwargs `{}`

 For more details about this graph break, please visit: https://meta-pytorch.github.io/compile-graph-break-site/gb/gb0025.html

tensor([1.7608, 2.9385, 2.1531])

error_on_graph_break(True) 示例#

@torch._dynamo.error_on_graph_break(True)
def inner2(x):
    x = x + 1
    torch._dynamo.graph_break()  # error
    return x + 2

def inner(x):
    return inner2(x)

# fullgraph=False, error_on_graph_break=False
@torch.compile
def fn(x):
    x = x + 4
    torch._dynamo.graph_break()  # no error
    return inner(x)

try:
    fn(torch.randn(3))
except Exception as e:
    print(e)

Call to `torch._dynamo.graph_break()`
  Explanation: User-inserted graph break. Message: None
  Hint: Remove the `torch._dynamo.graph_break()` call.

  Developer debug context: Called `torch._dynamo.graph_break()` with args `[]`, kwargs `{}`

 For more details about this graph break, please visit: https://meta-pytorch.github.io/compile-graph-break-site/gb/gb0025.html

from user code:
   File "/tmp/ipykernel_880/2379101916.py", line 15, in torch_dynamo_resume_in_fn_at_14
    return inner(x)
  File "/tmp/ipykernel_880/2379101916.py", line 8, in inner
    return inner2(x)
  File "/opt/conda/envs/py_3.10/lib/python3.10/site-packages/torch/_dynamo/external_utils.py", line 209, in inner
    return func(*args, **kwargs)
  File "/tmp/ipykernel_880/2379101916.py", line 4, in inner2
    torch._dynamo.graph_break()  # error

Set TORCHDYNAMO_VERBOSE=1 for the internal stack trace (please do this especially if you're reporting a bug to PyTorch). For even more developer context, set TORCH_LOGS="+dynamo"

Graph break in user code at /tmp/ipykernel_880/2379101916.py:14
Graph Break Reason: Call to `torch._dynamo.graph_break()`
  Explanation: User-inserted graph break. Message: None
  Hint: Remove the `torch._dynamo.graph_break()` call.

  Developer debug context: Called `torch._dynamo.graph_break()` with args `[]`, kwargs `{}`

 For more details about this graph break, please visit: https://meta-pytorch.github.io/compile-graph-break-site/gb/gb0025.html
User code traceback:
  File "/opt/conda/envs/py_3.10/lib/python3.10/runpy.py", line 196, in _run_module_as_main
    return _run_code(code, main_globals, None,
  File "/opt/conda/envs/py_3.10/lib/python3.10/runpy.py", line 86, in _run_code
    exec(code, run_globals)
  File "/opt/conda/envs/py_3.10/lib/python3.10/site-packages/ipykernel_launcher.py", line 18, in <module>
    app.launch_new_instance()
  File "/opt/conda/envs/py_3.10/lib/python3.10/site-packages/traitlets/config/application.py", line 1075, in launch_instance
    app.start()
  File "/opt/conda/envs/py_3.10/lib/python3.10/site-packages/ipykernel/kernelapp.py", line 739, in start
    self.io_loop.start()
  File "/opt/conda/envs/py_3.10/lib/python3.10/site-packages/tornado/platform/asyncio.py", line 211, in start
    self.asyncio_loop.run_forever()
  File "/opt/conda/envs/py_3.10/lib/python3.10/asyncio/base_events.py", line 603, in run_forever
    self._run_once()
  File "/opt/conda/envs/py_3.10/lib/python3.10/asyncio/base_events.py", line 1909, in _run_once
    handle._run()
  File "/opt/conda/envs/py_3.10/lib/python3.10/asyncio/events.py", line 80, in _run
    self._context.run(self._callback, *self._args)
  File "/opt/conda/envs/py_3.10/lib/python3.10/site-packages/ipykernel/kernelbase.py", line 519, in dispatch_queue
    await self.process_one()
  File "/opt/conda/envs/py_3.10/lib/python3.10/site-packages/ipykernel/kernelbase.py", line 508, in process_one
    await dispatch(*args)
  File "/opt/conda/envs/py_3.10/lib/python3.10/site-packages/ipykernel/kernelbase.py", line 400, in dispatch_shell
    await result
  File "/opt/conda/envs/py_3.10/lib/python3.10/site-packages/ipykernel/ipkernel.py", line 368, in execute_request
    await super().execute_request(stream, ident, parent)
  File "/opt/conda/envs/py_3.10/lib/python3.10/site-packages/ipykernel/kernelbase.py", line 767, in execute_request
    reply_content = await reply_content
  File "/opt/conda/envs/py_3.10/lib/python3.10/site-packages/ipykernel/ipkernel.py", line 455, in do_execute
    res = shell.run_cell(
  File "/opt/conda/envs/py_3.10/lib/python3.10/site-packages/ipykernel/zmqshell.py", line 577, in run_cell
    return super().run_cell(*args, **kwargs)
  File "/opt/conda/envs/py_3.10/lib/python3.10/site-packages/IPython/core/interactiveshell.py", line 3006, in run_cell
    result = self._run_cell(
  File "/opt/conda/envs/py_3.10/lib/python3.10/site-packages/IPython/core/interactiveshell.py", line 3061, in _run_cell
    result = runner(coro)
  File "/opt/conda/envs/py_3.10/lib/python3.10/site-packages/IPython/core/async_helpers.py", line 129, in _pseudo_sync_runner
    coro.send(None)
  File "/opt/conda/envs/py_3.10/lib/python3.10/site-packages/IPython/core/interactiveshell.py", line 3266, in run_cell_async
    has_raised = await self.run_ast_nodes(code_ast.body, cell_name,
  File "/opt/conda/envs/py_3.10/lib/python3.10/site-packages/IPython/core/interactiveshell.py", line 3445, in run_ast_nodes
    if await self.run_code(code, result, async_=asy):
  File "/opt/conda/envs/py_3.10/lib/python3.10/site-packages/IPython/core/interactiveshell.py", line 3505, in run_code
    exec(code_obj, self.user_global_ns, self.user_ns)
  File "/tmp/ipykernel_880/2379101916.py", line 18, in <module>
    fn(torch.randn(3))
  File "/tmp/ipykernel_880/2379101916.py", line 14, in fn
    torch._dynamo.graph_break()  # no error

在 error_on_graph_break(False) 下使用 error_on_graph_break(True)，有助于在我们希望灵活使用 torch.compile（即遵循 fullgraph=False 编程模型）但某些代码段是性能关键的，并且我们希望确保这些代码段不包含图中断时。

error_on_graph_break 嵌套行为#

torch._dynamo.error_on_graph_break() 也会影响嵌套调用的 error_on_graph_break 设置。

def inner(x):
    x = x + 1
    torch._dynamo.graph_break()
    return x + 2

def inner2(x):
    with torch._dynamo.error_on_graph_break(False):
        return inner(x)

@torch._dynamo.error_on_graph_break(True)
@torch.compile
def fn(x):
    return inner2(x)

# no error
fn(torch.randn(3))

Graph break in user code at /tmp/ipykernel_880/1007149706.py:3
Graph Break Reason: Call to `torch._dynamo.graph_break()`
  Explanation: User-inserted graph break. Message: None
  Hint: Remove the `torch._dynamo.graph_break()` call.

  Developer debug context: Called `torch._dynamo.graph_break()` with args `[]`, kwargs `{}`

 For more details about this graph break, please visit: https://meta-pytorch.github.io/compile-graph-break-site/gb/gb0025.html
User code traceback:
  File "/opt/conda/envs/py_3.10/lib/python3.10/runpy.py", line 196, in _run_module_as_main
    return _run_code(code, main_globals, None,
  File "/opt/conda/envs/py_3.10/lib/python3.10/runpy.py", line 86, in _run_code
    exec(code, run_globals)
  File "/opt/conda/envs/py_3.10/lib/python3.10/site-packages/ipykernel_launcher.py", line 18, in <module>
    app.launch_new_instance()
  File "/opt/conda/envs/py_3.10/lib/python3.10/site-packages/traitlets/config/application.py", line 1075, in launch_instance
    app.start()
  File "/opt/conda/envs/py_3.10/lib/python3.10/site-packages/ipykernel/kernelapp.py", line 739, in start
    self.io_loop.start()
  File "/opt/conda/envs/py_3.10/lib/python3.10/site-packages/tornado/platform/asyncio.py", line 211, in start
    self.asyncio_loop.run_forever()
  File "/opt/conda/envs/py_3.10/lib/python3.10/asyncio/base_events.py", line 603, in run_forever
    self._run_once()
  File "/opt/conda/envs/py_3.10/lib/python3.10/asyncio/base_events.py", line 1909, in _run_once
    handle._run()
  File "/opt/conda/envs/py_3.10/lib/python3.10/asyncio/events.py", line 80, in _run
    self._context.run(self._callback, *self._args)
  File "/opt/conda/envs/py_3.10/lib/python3.10/site-packages/ipykernel/kernelbase.py", line 519, in dispatch_queue
    await self.process_one()
  File "/opt/conda/envs/py_3.10/lib/python3.10/site-packages/ipykernel/kernelbase.py", line 508, in process_one
    await dispatch(*args)
  File "/opt/conda/envs/py_3.10/lib/python3.10/site-packages/ipykernel/kernelbase.py", line 400, in dispatch_shell
    await result
  File "/opt/conda/envs/py_3.10/lib/python3.10/site-packages/ipykernel/ipkernel.py", line 368, in execute_request
    await super().execute_request(stream, ident, parent)
  File "/opt/conda/envs/py_3.10/lib/python3.10/site-packages/ipykernel/kernelbase.py", line 767, in execute_request
    reply_content = await reply_content
  File "/opt/conda/envs/py_3.10/lib/python3.10/site-packages/ipykernel/ipkernel.py", line 455, in do_execute
    res = shell.run_cell(
  File "/opt/conda/envs/py_3.10/lib/python3.10/site-packages/ipykernel/zmqshell.py", line 577, in run_cell
    return super().run_cell(*args, **kwargs)
  File "/opt/conda/envs/py_3.10/lib/python3.10/site-packages/IPython/core/interactiveshell.py", line 3006, in run_cell
    result = self._run_cell(
  File "/opt/conda/envs/py_3.10/lib/python3.10/site-packages/IPython/core/interactiveshell.py", line 3061, in _run_cell
    result = runner(coro)
  File "/opt/conda/envs/py_3.10/lib/python3.10/site-packages/IPython/core/async_helpers.py", line 129, in _pseudo_sync_runner
    coro.send(None)
  File "/opt/conda/envs/py_3.10/lib/python3.10/site-packages/IPython/core/interactiveshell.py", line 3266, in run_cell_async
    has_raised = await self.run_ast_nodes(code_ast.body, cell_name,
  File "/opt/conda/envs/py_3.10/lib/python3.10/site-packages/IPython/core/interactiveshell.py", line 3445, in run_ast_nodes
    if await self.run_code(code, result, async_=asy):
  File "/opt/conda/envs/py_3.10/lib/python3.10/site-packages/IPython/core/interactiveshell.py", line 3505, in run_code
    exec(code_obj, self.user_global_ns, self.user_ns)
  File "/tmp/ipykernel_880/1007149706.py", line 16, in <module>
    fn(torch.randn(3))
  File "/opt/conda/envs/py_3.10/lib/python3.10/site-packages/torch/_dynamo/external_utils.py", line 209, in inner
    return func(*args, **kwargs)
  File "/tmp/ipykernel_880/1007149706.py", line 13, in fn
    return inner2(x)
  File "/tmp/ipykernel_880/1007149706.py", line 8, in inner2
    return inner(x)
  File "/tmp/ipykernel_880/1007149706.py", line 3, in inner
    torch._dynamo.graph_break()

Graph break (user stack suppressed due to duplicate graph break) in user code at /tmp/ipykernel_880/1007149706.py:3
Graph Break Reason: Call to `torch._dynamo.graph_break()`
  Explanation: User-inserted graph break. Message: None
  Hint: Remove the `torch._dynamo.graph_break()` call.

  Developer debug context: Called `torch._dynamo.graph_break()` with args `[]`, kwargs `{}`

 For more details about this graph break, please visit: https://meta-pytorch.github.io/compile-graph-break-site/gb/gb0025.html
Graph break (user stack suppressed due to duplicate graph break) in user code at /tmp/ipykernel_880/1007149706.py:3
Graph Break Reason: Call to `torch._dynamo.graph_break()`
  Explanation: User-inserted graph break. Message: None
  Hint: Remove the `torch._dynamo.graph_break()` call.

  Developer debug context: Called `torch._dynamo.graph_break()` with args `[]`, kwargs `{}`

 For more details about this graph break, please visit: https://meta-pytorch.github.io/compile-graph-break-site/gb/gb0025.html

tensor([3.5942, 2.0674, 2.8445])

torch._dynamo.error_on_graph_break() 可以在另一个 torch._dynamo.error_on_graph_break() 区域下使用。

def inner(x):
    x = x + 1
    with torch._dynamo.error_on_graph_break(False):
        torch._dynamo.graph_break()
    return x + 2

def inner2(x):
    with torch._dynamo.error_on_graph_break(True):
        return inner(x)

@torch.compile
def fn(x):
    return inner2(x)

# no error
fn(torch.randn(3))

Graph break in user code at /tmp/ipykernel_880/1343774799.py:4
Graph Break Reason: Call to `torch._dynamo.graph_break()`
  Explanation: User-inserted graph break. Message: None
  Hint: Remove the `torch._dynamo.graph_break()` call.

  Developer debug context: Called `torch._dynamo.graph_break()` with args `[]`, kwargs `{}`

 For more details about this graph break, please visit: https://meta-pytorch.github.io/compile-graph-break-site/gb/gb0025.html
User code traceback:
  File "/opt/conda/envs/py_3.10/lib/python3.10/runpy.py", line 196, in _run_module_as_main
    return _run_code(code, main_globals, None,
  File "/opt/conda/envs/py_3.10/lib/python3.10/runpy.py", line 86, in _run_code
    exec(code, run_globals)
  File "/opt/conda/envs/py_3.10/lib/python3.10/site-packages/ipykernel_launcher.py", line 18, in <module>
    app.launch_new_instance()
  File "/opt/conda/envs/py_3.10/lib/python3.10/site-packages/traitlets/config/application.py", line 1075, in launch_instance
    app.start()
  File "/opt/conda/envs/py_3.10/lib/python3.10/site-packages/ipykernel/kernelapp.py", line 739, in start
    self.io_loop.start()
  File "/opt/conda/envs/py_3.10/lib/python3.10/site-packages/tornado/platform/asyncio.py", line 211, in start
    self.asyncio_loop.run_forever()
  File "/opt/conda/envs/py_3.10/lib/python3.10/asyncio/base_events.py", line 603, in run_forever
    self._run_once()
  File "/opt/conda/envs/py_3.10/lib/python3.10/asyncio/base_events.py", line 1909, in _run_once
    handle._run()
  File "/opt/conda/envs/py_3.10/lib/python3.10/asyncio/events.py", line 80, in _run
    self._context.run(self._callback, *self._args)
  File "/opt/conda/envs/py_3.10/lib/python3.10/site-packages/ipykernel/kernelbase.py", line 519, in dispatch_queue
    await self.process_one()
  File "/opt/conda/envs/py_3.10/lib/python3.10/site-packages/ipykernel/kernelbase.py", line 508, in process_one
    await dispatch(*args)
  File "/opt/conda/envs/py_3.10/lib/python3.10/site-packages/ipykernel/kernelbase.py", line 400, in dispatch_shell
    await result
  File "/opt/conda/envs/py_3.10/lib/python3.10/site-packages/ipykernel/ipkernel.py", line 368, in execute_request
    await super().execute_request(stream, ident, parent)
  File "/opt/conda/envs/py_3.10/lib/python3.10/site-packages/ipykernel/kernelbase.py", line 767, in execute_request
    reply_content = await reply_content
  File "/opt/conda/envs/py_3.10/lib/python3.10/site-packages/ipykernel/ipkernel.py", line 455, in do_execute
    res = shell.run_cell(
  File "/opt/conda/envs/py_3.10/lib/python3.10/site-packages/ipykernel/zmqshell.py", line 577, in run_cell
    return super().run_cell(*args, **kwargs)
  File "/opt/conda/envs/py_3.10/lib/python3.10/site-packages/IPython/core/interactiveshell.py", line 3006, in run_cell
    result = self._run_cell(
  File "/opt/conda/envs/py_3.10/lib/python3.10/site-packages/IPython/core/interactiveshell.py", line 3061, in _run_cell
    result = runner(coro)
  File "/opt/conda/envs/py_3.10/lib/python3.10/site-packages/IPython/core/async_helpers.py", line 129, in _pseudo_sync_runner
    coro.send(None)
  File "/opt/conda/envs/py_3.10/lib/python3.10/site-packages/IPython/core/interactiveshell.py", line 3266, in run_cell_async
    has_raised = await self.run_ast_nodes(code_ast.body, cell_name,
  File "/opt/conda/envs/py_3.10/lib/python3.10/site-packages/IPython/core/interactiveshell.py", line 3445, in run_ast_nodes
    if await self.run_code(code, result, async_=asy):
  File "/opt/conda/envs/py_3.10/lib/python3.10/site-packages/IPython/core/interactiveshell.py", line 3505, in run_code
    exec(code_obj, self.user_global_ns, self.user_ns)
  File "/tmp/ipykernel_880/1343774799.py", line 16, in <module>
    fn(torch.randn(3))
  File "/tmp/ipykernel_880/1343774799.py", line 13, in fn
    return inner2(x)
  File "/tmp/ipykernel_880/1343774799.py", line 9, in inner2
    return inner(x)
  File "/tmp/ipykernel_880/1343774799.py", line 4, in inner
    torch._dynamo.graph_break()

Graph break (user stack suppressed due to duplicate graph break) in user code at /tmp/ipykernel_880/1343774799.py:4
Graph Break Reason: Call to `torch._dynamo.graph_break()`
  Explanation: User-inserted graph break. Message: None
  Hint: Remove the `torch._dynamo.graph_break()` call.

  Developer debug context: Called `torch._dynamo.graph_break()` with args `[]`, kwargs `{}`

 For more details about this graph break, please visit: https://meta-pytorch.github.io/compile-graph-break-site/gb/gb0025.html
Graph break (user stack suppressed due to duplicate graph break) in user code at /tmp/ipykernel_880/1343774799.py:4
Graph Break Reason: Call to `torch._dynamo.graph_break()`
  Explanation: User-inserted graph break. Message: None
  Hint: Remove the `torch._dynamo.graph_break()` call.

  Developer debug context: Called `torch._dynamo.graph_break()` with args `[]`, kwargs `{}`

 For more details about this graph break, please visit: https://meta-pytorch.github.io/compile-graph-break-site/gb/gb0025.html

tensor([1.3759, 2.1866, 3.8692])

与 fullgraph 的交互#

fullgraph=True 的优先级高于 error_on_graph_break。

@torch._dynamo.error_on_graph_break(False)
def inner(x):
    x = x + 1
    torch._dynamo.graph_break()
    return x + 2

@torch.compile(fullgraph=True)
def fn(x):
    return inner(x)

try:
    fn(torch.randn(3))
except Exception as e:
    print(e)

Call to `torch._dynamo.graph_break()`
  Explanation: User-inserted graph break. Message: None
  Hint: Remove the `torch._dynamo.graph_break()` call.

  Developer debug context: Called `torch._dynamo.graph_break()` with args `[]`, kwargs `{}`

 For more details about this graph break, please visit: https://meta-pytorch.github.io/compile-graph-break-site/gb/gb0025.html

from user code:
   File "/tmp/ipykernel_880/2331424258.py", line 9, in fn
    return inner(x)
  File "/opt/conda/envs/py_3.10/lib/python3.10/site-packages/torch/_dynamo/external_utils.py", line 209, in inner
    return func(*args, **kwargs)
  File "/tmp/ipykernel_880/2331424258.py", line 3, in inner
    x = x + 1

Set TORCHDYNAMO_VERBOSE=1 for the internal stack trace (please do this especially if you're reporting a bug to PyTorch). For even more developer context, set TORCH_LOGS="+dynamo"

fullgraph=True 无法切换回 fullgraph=False。

@torch.compile(fullgraph=False)
def inner(x):
    x = x + 1
    torch._dynamo.graph_break()
    return x + 2

@torch.compile(fullgraph=True)
def fn(x):
    return inner(x)

try:
    fn(torch.randn(3))
except Exception as e:
    print(e)

Call to `torch._dynamo.graph_break()`
  Explanation: User-inserted graph break. Message: None
  Hint: Remove the `torch._dynamo.graph_break()` call.

  Developer debug context: Called `torch._dynamo.graph_break()` with args `[]`, kwargs `{}`

 For more details about this graph break, please visit: https://meta-pytorch.github.io/compile-graph-break-site/gb/gb0025.html

from user code:
   File "/tmp/ipykernel_880/262151723.py", line 9, in fn
    return inner(x)
  File "/opt/conda/envs/py_3.10/lib/python3.10/site-packages/torch/_dynamo/polyfills/__init__.py", line 264, in getattr_and_trace
    return fn(*args[2:], **kwargs)
  File "/tmp/ipykernel_880/262151723.py", line 3, in inner
    x = x + 1

Set TORCHDYNAMO_VERBOSE=1 for the internal stack trace (please do this especially if you're reporting a bug to PyTorch). For even more developer context, set TORCH_LOGS="+dynamo"

@torch.compile(fullgraph=True)
def inner(x):
    x = x + 1
    torch._dynamo.graph_break()
    return x + 2

@torch.compile(fullgraph=False)
def fn(x):
    return inner(x)

try:
    fn(torch.randn(3))
except Exception as e:
    print(e)

Call to `torch._dynamo.graph_break()`
  Explanation: User-inserted graph break. Message: None
  Hint: Remove the `torch._dynamo.graph_break()` call.

  Developer debug context: Called `torch._dynamo.graph_break()` with args `[]`, kwargs `{}`

 For more details about this graph break, please visit: https://meta-pytorch.github.io/compile-graph-break-site/gb/gb0025.html

from user code:
   File "/tmp/ipykernel_880/2119173801.py", line 3, in inner
    x = x + 1

Set TORCHDYNAMO_VERBOSE=1 for the internal stack trace (please do this especially if you're reporting a bug to PyTorch). For even more developer context, set TORCH_LOGS="+dynamo"

Graph break in user code at /tmp/ipykernel_880/2119173801.py:3
Graph Break Reason: Call to `torch._dynamo.graph_break()`
  Explanation: User-inserted graph break. Message: None
  Hint: Remove the `torch._dynamo.graph_break()` call.

  Developer debug context: Called `torch._dynamo.graph_break()` with args `[]`, kwargs `{}`

 For more details about this graph break, please visit: https://meta-pytorch.github.io/compile-graph-break-site/gb/gb0025.html
User code traceback:
  File "/opt/conda/envs/py_3.10/lib/python3.10/runpy.py", line 196, in _run_module_as_main
    return _run_code(code, main_globals, None,
  File "/opt/conda/envs/py_3.10/lib/python3.10/runpy.py", line 86, in _run_code
    exec(code, run_globals)
  File "/opt/conda/envs/py_3.10/lib/python3.10/site-packages/ipykernel_launcher.py", line 18, in <module>
    app.launch_new_instance()
  File "/opt/conda/envs/py_3.10/lib/python3.10/site-packages/traitlets/config/application.py", line 1075, in launch_instance
    app.start()
  File "/opt/conda/envs/py_3.10/lib/python3.10/site-packages/ipykernel/kernelapp.py", line 739, in start
    self.io_loop.start()
  File "/opt/conda/envs/py_3.10/lib/python3.10/site-packages/tornado/platform/asyncio.py", line 211, in start
    self.asyncio_loop.run_forever()
  File "/opt/conda/envs/py_3.10/lib/python3.10/asyncio/base_events.py", line 603, in run_forever
    self._run_once()
  File "/opt/conda/envs/py_3.10/lib/python3.10/asyncio/base_events.py", line 1909, in _run_once
    handle._run()
  File "/opt/conda/envs/py_3.10/lib/python3.10/asyncio/events.py", line 80, in _run
    self._context.run(self._callback, *self._args)
  File "/opt/conda/envs/py_3.10/lib/python3.10/site-packages/ipykernel/kernelbase.py", line 519, in dispatch_queue
    await self.process_one()
  File "/opt/conda/envs/py_3.10/lib/python3.10/site-packages/ipykernel/kernelbase.py", line 508, in process_one
    await dispatch(*args)
  File "/opt/conda/envs/py_3.10/lib/python3.10/site-packages/ipykernel/kernelbase.py", line 400, in dispatch_shell
    await result
  File "/opt/conda/envs/py_3.10/lib/python3.10/site-packages/ipykernel/ipkernel.py", line 368, in execute_request
    await super().execute_request(stream, ident, parent)
  File "/opt/conda/envs/py_3.10/lib/python3.10/site-packages/ipykernel/kernelbase.py", line 767, in execute_request
    reply_content = await reply_content
  File "/opt/conda/envs/py_3.10/lib/python3.10/site-packages/ipykernel/ipkernel.py", line 455, in do_execute
    res = shell.run_cell(
  File "/opt/conda/envs/py_3.10/lib/python3.10/site-packages/ipykernel/zmqshell.py", line 577, in run_cell
    return super().run_cell(*args, **kwargs)
  File "/opt/conda/envs/py_3.10/lib/python3.10/site-packages/IPython/core/interactiveshell.py", line 3006, in run_cell
    result = self._run_cell(
  File "/opt/conda/envs/py_3.10/lib/python3.10/site-packages/IPython/core/interactiveshell.py", line 3061, in _run_cell
    result = runner(coro)
  File "/opt/conda/envs/py_3.10/lib/python3.10/site-packages/IPython/core/async_helpers.py", line 129, in _pseudo_sync_runner
    coro.send(None)
  File "/opt/conda/envs/py_3.10/lib/python3.10/site-packages/IPython/core/interactiveshell.py", line 3266, in run_cell_async
    has_raised = await self.run_ast_nodes(code_ast.body, cell_name,
  File "/opt/conda/envs/py_3.10/lib/python3.10/site-packages/IPython/core/interactiveshell.py", line 3445, in run_ast_nodes
    if await self.run_code(code, result, async_=asy):
  File "/opt/conda/envs/py_3.10/lib/python3.10/site-packages/IPython/core/interactiveshell.py", line 3505, in run_code
    exec(code_obj, self.user_global_ns, self.user_ns)
  File "/tmp/ipykernel_880/2119173801.py", line 12, in <module>
    fn(torch.randn(3))
  File "/tmp/ipykernel_880/2119173801.py", line 9, in fn
    return inner(x)
  File "/opt/conda/envs/py_3.10/lib/python3.10/site-packages/torch/_dynamo/polyfills/__init__.py", line 264, in getattr_and_trace
    return fn(*args[2:], **kwargs)
  File "/tmp/ipykernel_880/2119173801.py", line 3, in inner
    x = x + 1

fullgraph=True/False 与 error_on_graph_break 总结#

下表总结了 fullgraph=True/False 和 error_on_graph_break 之间的区别。

error_on_graph_break=True

error_on_graph_break=False (默认)

fullgraph=True

图中断会导致错误。只报告第一次图中断。保证一个图。

无法将 fullgraph 切换到 False。error_on_graph_break 无效。

用户代码必须完全兼容 torch.compile。保证不会因图中断产生性能损失（因为没有图中断）。

最适合对图中断敏感的代码：框架/库代码或需要最大性能的场景。防止下游用户代码无意中允许图中断。

与 fullgraph=True 和 error_on_graph_break=True 相同，因为当 fullgraph=True 时，error_on_graph_break 无效。

fullgraph=False (默认)

图中断会导致错误。只报告第一次图中断。不保证一个图。

可以将 error_on_graph_break 切换到 False。

用户代码必须完全兼容 torch.compile。保证不会因图中断产生性能损失（因为没有图中断）。

最适合对图中断敏感的用户代码。error_on_graph_break 可以切换到 False 来处理那些难以规避的、有图中断的代码段。

遇到图中断后会继续编译。所有图中断都会被报告。

可以将 error_on_graph_break 切换到 True。

不需要大量用户代码更改即可工作。性能可能会因图中断而受到负面影响。

最适合开箱即用的场景，“非特殊”代码，或不需要榨取最大性能的场景。
在此页面上
在 GitHub 上编辑
显示源代码
PyTorch 库