Togglingerror_on_graph_break#
Created On: Sep 03, 2025 | Last Updated On: Sep 03, 2025
Summary:
When
fullgraph=False, we can usetorch._dynamo.error_on_graph_break()for more flexibility indealing with graph breaks.
So far, we have introduced two ways in dealing with graph breaks intorch.compile:
fullgraph=Trueerrors on the first graph break and additionally guarantees that only one graph is traced from the code.fullgraph=Falsecontinues tracing even when encountering graph breaks.
What if we want to disallow graph breaks for most of the code, but there are a few problematic functions where the graph breaks are hard to remove,and we are okay with having those graph breaks? We can usetorch._dynamo.error_on_graph_break() to achieve this.
torch.compile has anerror_on_graph_break setting (initially set toFalse).If a graph break or compiler error occurs in code whileerror_on_graph_break is set toFalse, thentorch.compile will attempt to continue compilation after the graph break/error.Iferror_on_graph_break is set toTrue, thentorch.compile will abort compilation and propagate the error to user code.
A significant difference betweenerror_on_graph_break=True andfullgraph=True is that the formerdoes not guarantee that a single graph will be captured.error_on_graph_breakcan be arbitrarily toggled during compile time by using thetorch._dynamo.error_on_graph_break() context manager/decorator.In comparison, oncefullgraph is set toTrue, it cannot be set back toFalse.Finally,error_on_graph_break has lower precedence thanfullgraph -error_on_graph_break only takes effect whenfullgraph=False.
error_on_graph_break(False) example#
@torch._dynamo.error_on_graph_break(False)defcode_with_a_difficult_graph_break(x):x=x+1torch._dynamo.graph_break()returnx+2definner(x):returncode_with_a_difficult_graph_break(x)# NOTE: fullgraph=False@torch._dynamo.error_on_graph_break(True)@torch.compiledeffn(x):returninner(x)# No error, but there is a graph breakfn(torch.randn(3))
Graph break in user code at /tmp/ipykernel_369/1452578661.py:4Graph 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.htmlUser 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 758, 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/utils.py", line 71, in preserve_context return await f(*args, **kwargs) File "/opt/conda/envs/py_3.10/lib/python3.10/site-packages/ipykernel/kernelbase.py", line 614, in shell_main await self.dispatch_shell(msg, subshell_id=subshell_id) File "/opt/conda/envs/py_3.10/lib/python3.10/site-packages/ipykernel/kernelbase.py", line 471, in dispatch_shell await result File "/opt/conda/envs/py_3.10/lib/python3.10/site-packages/ipykernel/ipkernel.py", line 366, 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 827, in execute_request reply_content = await reply_content File "/opt/conda/envs/py_3.10/lib/python3.10/site-packages/ipykernel/ipkernel.py", line 458, in do_execute res = shell.run_cell( File "/opt/conda/envs/py_3.10/lib/python3.10/site-packages/ipykernel/zmqshell.py", line 663, 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_369/1452578661.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 216, in inner return func(*args, **kwargs) File "/tmp/ipykernel_369/1452578661.py", line 14, in fn return inner(x) File "/tmp/ipykernel_369/1452578661.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 216, in inner return func(*args, **kwargs) File "/tmp/ipykernel_369/1452578661.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_369/1452578661.py:4Graph 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.htmlGraph break (user stack suppressed due to duplicate graph break) in user code at /tmp/ipykernel_369/1452578661.py:4Graph 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.htmlGraph break (user stack suppressed due to duplicate graph break) in user code at /tmp/ipykernel_369/1452578661.py:4Graph 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.htmltensor([3.0113, 3.9169, 0.5099])
Usingerror_on_graph_break(False) undererror_on_graph_break(True) is helpful for when we want to minimize graph breaks (i.e. follow thefullgraph=True programming model),but there are some sections of code with non-performance-critical graph breaks that are difficult to work around.
error_on_graph_break() can be used as a context manager as well:
# NOTE: fullgraph=False@torch._dynamo.error_on_graph_break(True)@torch.compiledeffn(x):x=x+1withtorch._dynamo.error_on_graph_break(False):torch._dynamo.graph_break()# no errorreturnx+2# No error, but there is a graph breakfn(torch.randn(3))
Graph break in user code at /tmp/ipykernel_369/737485247.py:7Graph 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.htmlUser 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 758, 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/utils.py", line 71, in preserve_context return await f(*args, **kwargs) File "/opt/conda/envs/py_3.10/lib/python3.10/site-packages/ipykernel/kernelbase.py", line 614, in shell_main await self.dispatch_shell(msg, subshell_id=subshell_id) File "/opt/conda/envs/py_3.10/lib/python3.10/site-packages/ipykernel/kernelbase.py", line 471, in dispatch_shell await result File "/opt/conda/envs/py_3.10/lib/python3.10/site-packages/ipykernel/ipkernel.py", line 366, 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 827, in execute_request reply_content = await reply_content File "/opt/conda/envs/py_3.10/lib/python3.10/site-packages/ipykernel/ipkernel.py", line 458, in do_execute res = shell.run_cell( File "/opt/conda/envs/py_3.10/lib/python3.10/site-packages/ipykernel/zmqshell.py", line 663, 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_369/737485247.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 216, in inner return func(*args, **kwargs) File "/tmp/ipykernel_369/737485247.py", line 7, in fn torch._dynamo.graph_break() # no errortensor([2.2028, 3.7649, 4.9742])
You can use monkey patching to toggleerror_on_graph_break for code where you cannot edit the source (e.g. framework code):
classThirdPartyModule(torch.nn.Module):defforward(self,x):x=x+1torch._dynamo.graph_break()returnx+2tp_mod=ThirdPartyModule()tp_mod.forward=torch._dynamo.error_on_graph_break(False)(tp_mod.forward)@torch._dynamo.error_on_graph_break(True)@torch.compiledeffn(x):returntp_mod.forward(x)# No error, but there is a graph breakfn(torch.randn(3))
Graph break in user code at /tmp/ipykernel_369/2112598647.py:4Graph 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.htmlUser 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 758, 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/utils.py", line 71, in preserve_context return await f(*args, **kwargs) File "/opt/conda/envs/py_3.10/lib/python3.10/site-packages/ipykernel/kernelbase.py", line 614, in shell_main await self.dispatch_shell(msg, subshell_id=subshell_id) File "/opt/conda/envs/py_3.10/lib/python3.10/site-packages/ipykernel/kernelbase.py", line 471, in dispatch_shell await result File "/opt/conda/envs/py_3.10/lib/python3.10/site-packages/ipykernel/ipkernel.py", line 366, 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 827, in execute_request reply_content = await reply_content File "/opt/conda/envs/py_3.10/lib/python3.10/site-packages/ipykernel/ipkernel.py", line 458, in do_execute res = shell.run_cell( File "/opt/conda/envs/py_3.10/lib/python3.10/site-packages/ipykernel/zmqshell.py", line 663, 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_369/2112598647.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 216, in inner return func(*args, **kwargs) File "/tmp/ipykernel_369/2112598647.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 216, in inner return func(*args, **kwargs) File "/tmp/ipykernel_369/2112598647.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_369/2112598647.py:4Graph 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.htmlGraph break (user stack suppressed due to duplicate graph break) in user code at /tmp/ipykernel_369/2112598647.py:4Graph 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.htmltensor([3.2821, 4.4465, 2.9630])
error_on_graph_break(True) example#
@torch._dynamo.error_on_graph_break(True)definner2(x):x=x+1torch._dynamo.graph_break()# errorreturnx+2definner(x):returninner2(x)# fullgraph=False, error_on_graph_break=False@torch.compiledeffn(x):x=x+4torch._dynamo.graph_break()# no errorreturninner(x)try:fn(torch.randn(3))exceptExceptionase: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.htmlfrom user code: File "/tmp/ipykernel_369/2379101916.py", line 15, in torch_dynamo_resume_in_fn_at_14 return inner(x) File "/tmp/ipykernel_369/2379101916.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 216, in inner return func(*args, **kwargs) File "/tmp/ipykernel_369/2379101916.py", line 4, in inner2 torch._dynamo.graph_break() # errorSet 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_369/2379101916.py:14Graph 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.htmlUser 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 758, 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/utils.py", line 71, in preserve_context return await f(*args, **kwargs) File "/opt/conda/envs/py_3.10/lib/python3.10/site-packages/ipykernel/kernelbase.py", line 614, in shell_main await self.dispatch_shell(msg, subshell_id=subshell_id) File "/opt/conda/envs/py_3.10/lib/python3.10/site-packages/ipykernel/kernelbase.py", line 471, in dispatch_shell await result File "/opt/conda/envs/py_3.10/lib/python3.10/site-packages/ipykernel/ipkernel.py", line 366, 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 827, in execute_request reply_content = await reply_content File "/opt/conda/envs/py_3.10/lib/python3.10/site-packages/ipykernel/ipkernel.py", line 458, in do_execute res = shell.run_cell( File "/opt/conda/envs/py_3.10/lib/python3.10/site-packages/ipykernel/zmqshell.py", line 663, 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_369/2379101916.py", line 18, in <module> fn(torch.randn(3)) File "/tmp/ipykernel_369/2379101916.py", line 14, in fn torch._dynamo.graph_break() # no errorUsingerror_on_graph_break(True) undererror_on_graph_break(False) is helpful for when we want to usetorch.compile flexibly (i.e. follow thefullgraph=False programming model),but there are some sections of the code that are performance-critical and we want to ensure that those sections do not contain graph breaks.
error_on_graph_break nesting behavior#
torch._dynamo.error_on_graph_break() affects theerror_on_graph_break setting of nested calls as well:
definner(x):x=x+1torch._dynamo.graph_break()returnx+2definner2(x):withtorch._dynamo.error_on_graph_break(False):returninner(x)@torch._dynamo.error_on_graph_break(True)@torch.compiledeffn(x):returninner2(x)# no errorfn(torch.randn(3))
Graph break in user code at /tmp/ipykernel_369/1007149706.py:3Graph 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.htmlUser 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 758, 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/utils.py", line 71, in preserve_context return await f(*args, **kwargs) File "/opt/conda/envs/py_3.10/lib/python3.10/site-packages/ipykernel/kernelbase.py", line 614, in shell_main await self.dispatch_shell(msg, subshell_id=subshell_id) File "/opt/conda/envs/py_3.10/lib/python3.10/site-packages/ipykernel/kernelbase.py", line 471, in dispatch_shell await result File "/opt/conda/envs/py_3.10/lib/python3.10/site-packages/ipykernel/ipkernel.py", line 366, 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 827, in execute_request reply_content = await reply_content File "/opt/conda/envs/py_3.10/lib/python3.10/site-packages/ipykernel/ipkernel.py", line 458, in do_execute res = shell.run_cell( File "/opt/conda/envs/py_3.10/lib/python3.10/site-packages/ipykernel/zmqshell.py", line 663, 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_369/1007149706.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 216, in inner return func(*args, **kwargs) File "/tmp/ipykernel_369/1007149706.py", line 13, in fn return inner2(x) File "/tmp/ipykernel_369/1007149706.py", line 8, in inner2 return inner(x) File "/tmp/ipykernel_369/1007149706.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_369/1007149706.py:3Graph 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.htmlGraph break (user stack suppressed due to duplicate graph break) in user code at /tmp/ipykernel_369/1007149706.py:3Graph 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.htmltensor([2.7700, 2.0265, 1.2761])
torch._dynamo.error_on_graph_break() can be used under anothertorch._dynamo.error_on_graph_break() region:
definner(x):x=x+1withtorch._dynamo.error_on_graph_break(False):torch._dynamo.graph_break()returnx+2definner2(x):withtorch._dynamo.error_on_graph_break(True):returninner(x)@torch.compiledeffn(x):returninner2(x)# no errorfn(torch.randn(3))
Graph break in user code at /tmp/ipykernel_369/1343774799.py:4Graph 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.htmlUser 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 758, 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/utils.py", line 71, in preserve_context return await f(*args, **kwargs) File "/opt/conda/envs/py_3.10/lib/python3.10/site-packages/ipykernel/kernelbase.py", line 614, in shell_main await self.dispatch_shell(msg, subshell_id=subshell_id) File "/opt/conda/envs/py_3.10/lib/python3.10/site-packages/ipykernel/kernelbase.py", line 471, in dispatch_shell await result File "/opt/conda/envs/py_3.10/lib/python3.10/site-packages/ipykernel/ipkernel.py", line 366, 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 827, in execute_request reply_content = await reply_content File "/opt/conda/envs/py_3.10/lib/python3.10/site-packages/ipykernel/ipkernel.py", line 458, in do_execute res = shell.run_cell( File "/opt/conda/envs/py_3.10/lib/python3.10/site-packages/ipykernel/zmqshell.py", line 663, 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_369/1343774799.py", line 16, in <module> fn(torch.randn(3)) File "/tmp/ipykernel_369/1343774799.py", line 13, in fn return inner2(x) File "/tmp/ipykernel_369/1343774799.py", line 9, in inner2 return inner(x) File "/tmp/ipykernel_369/1343774799.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_369/1343774799.py:4Graph 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.htmlGraph break (user stack suppressed due to duplicate graph break) in user code at /tmp/ipykernel_369/1343774799.py:4Graph 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.htmltensor([2.9136, 2.4328, 3.4279])
Interaction withfullgraph#
fullgraph=True takes higher precedence thanerror_on_graph_break:
@torch._dynamo.error_on_graph_break(False)definner(x):x=x+1torch._dynamo.graph_break()returnx+2@torch.compile(fullgraph=True)deffn(x):returninner(x)try:fn(torch.randn(3))exceptExceptionase: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.htmlfrom user code: File "/tmp/ipykernel_369/2331424258.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 216, in inner return func(*args, **kwargs) File "/tmp/ipykernel_369/2331424258.py", line 3, in inner x = x + 1Set 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 cannot be toggled back tofullgraph=False:
@torch.compile(fullgraph=False)definner(x):x=x+1torch._dynamo.graph_break()returnx+2@torch.compile(fullgraph=True)deffn(x):returninner(x)try:fn(torch.randn(3))exceptExceptionase: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.htmlfrom user code: File "/tmp/ipykernel_369/262151723.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 276, in getattr_and_trace return fn(*args[2:], **kwargs) File "/tmp/ipykernel_369/262151723.py", line 3, in inner x = x + 1Set 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)definner(x):x=x+1torch._dynamo.graph_break()returnx+2@torch.compile(fullgraph=False)deffn(x):returninner(x)try:fn(torch.randn(3))exceptExceptionase: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.htmlfrom user code: File "/tmp/ipykernel_369/2119173801.py", line 3, in inner x = x + 1Set 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_369/2119173801.py:3Graph 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.htmlUser 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 758, 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/utils.py", line 71, in preserve_context return await f(*args, **kwargs) File "/opt/conda/envs/py_3.10/lib/python3.10/site-packages/ipykernel/kernelbase.py", line 614, in shell_main await self.dispatch_shell(msg, subshell_id=subshell_id) File "/opt/conda/envs/py_3.10/lib/python3.10/site-packages/ipykernel/kernelbase.py", line 471, in dispatch_shell await result File "/opt/conda/envs/py_3.10/lib/python3.10/site-packages/ipykernel/ipkernel.py", line 366, 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 827, in execute_request reply_content = await reply_content File "/opt/conda/envs/py_3.10/lib/python3.10/site-packages/ipykernel/ipkernel.py", line 458, in do_execute res = shell.run_cell( File "/opt/conda/envs/py_3.10/lib/python3.10/site-packages/ipykernel/zmqshell.py", line 663, 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_369/2119173801.py", line 12, in <module> fn(torch.randn(3)) File "/tmp/ipykernel_369/2119173801.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 276, in getattr_and_trace return fn(*args[2:], **kwargs) File "/tmp/ipykernel_369/2119173801.py", line 3, in inner x = x + 1Summary offullgraph=True/False vserror_on_graph_break#
Here is a table summarizing the differences betweenfullgraph=True/False anderror_on_graph_break:
|
| |
|---|---|---|
| Graph breaks result in errors. Only the first graph break will be reported.One graph guarantee. | Same as |
| Graph breaks result in errors. Only the first graph break will be reported.No one graph guarantee. | Will continue to compile after encountering graph breaks. All graph breaks will be reported. |