pytorch - 💡(How to fix) Fix DISABLED test_autograd_function_no_setup_context_transform_grad_cpu (__main__.TestComposabilityCPU) [2 comments, 1 participants]

torch._dynamo.exc.TorchRuntimeError: RuntimeError when making fake tensor call Explanation: Dynamo failed to run FX node with fake tensors: call_method sin(*(GradTrackingTensor(lvl=1, value= FakeTensor(..., size=(3,)) ),), **{}): got RuntimeError('tls_on_entry.has_value() INTERNAL ASSERT FAILED at "/var/lib/jenkins/workspace/aten/src/ATen/core/PythonFallbackKernel.cpp":53, please report a bug to PyTorch. ') Hint: Your code may result in an error when running in eager. Please double check that your code doesn't contain a similar error when actually running eager/uncompiled. You can do this by removing the torch.compile call, or by using torch.compiler.set_stance("force_eager").

Developer debug context:

For more details about this graph break, please visit: https://meta-pytorch.github.io/compile-graph-break-site/gb/gb4315.html

from user code: File "/opt/conda/envs/py_3.14/lib/python3.14t/site-packages/torch/_functorch/apis.py", line 433, in wrapper return eager_transforms.grad_impl(func, argnums, has_aux, args, kwargs) File "/opt/conda/envs/py_3.14/lib/python3.14t/site-packages/torch/_functorch/eager_transforms.py", line 1533, in grad_impl results = grad_and_value_impl(func, argnums, has_aux, args, kwargs) File "/opt/conda/envs/py_3.14/lib/python3.14t/site-packages/torch/_functorch/vmap.py", line 56, in fn return f(*args, **kwargs) File "/opt/conda/envs/py_3.14/lib/python3.14t/site-packages/torch/_functorch/eager_transforms.py", line 1484, in grad_and_value_impl output = func(*args, **kwargs) File "/var/lib/jenkins/workspace/test/functorch/test_eager_transforms.py", line 3784, in forward return x.sin()

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"

During handling of the above exception, another exception occurred:

Traceback (most recent call last): File "/var/lib/jenkins/workspace/test/functorch/test_eager_transforms.py", line 3780, in test_autograd_function_no_setup_context class MySin(torch.autograd.Function): File "/var/lib/jenkins/workspace/test/functorch/test_eager_transforms.py", line 3793, in torch_dynamo_resume_in_test_autograd_function_no_setup_context_at_3780 with self.assertRaisesRegex(RuntimeError, "must override the setup_context"): File "/var/lib/jenkins/workspace/test/functorch/test_eager_transforms.py", line 3793, in torch_dynamo_resume_in_test_autograd_function_no_setup_context_at_3793 with self.assertRaisesRegex(RuntimeError, "must override the setup_context"): ~~~~~~~~~~~~~~~~~~~~~~^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ File "/opt/conda/envs/py_3.14/lib/python3.14t/unittest/case.py", line 285, in exit self._raiseFailure('"{}" does not match "{}"'.format( ~~~~~~~~~~~~~~~~~~^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ expected_regex.pattern, str(exc_value))) ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ File "/opt/conda/envs/py_3.14/lib/python3.14t/unittest/case.py", line 209, in _raiseFailure raise self.test_case.failureException(msg) AssertionError: "must override the setup_context" does not match "RuntimeError when making fake tensor call Explanation: Dynamo failed to run FX node with fake tensors: call_method sin(*(GradTrackingTensor(lvl=1, value= FakeTensor(..., size=(3,)) ),), **{}): got RuntimeError('tls_on_entry.has_value() INTERNAL ASSERT FAILED at "/var/lib/jenkins/workspace/aten/src/ATen/core/PythonFallbackKernel.cpp":53, please report a bug to PyTorch. ') Hint: Your code may result in an error when running in eager. Please double check that your code doesn't contain a similar error when actually running eager/uncompiled. You can do this by removing the torch.compile call, or by using torch.compiler.set_stance("force_eager").

Developer debug context:

For more details about this graph break, please visit: https://meta-pytorch.github.io/compile-graph-break-site/gb/gb4315.html

from user code: File "/opt/conda/envs/py_3.14/lib/python3.14t/site-packages/torch/_functorch/apis.py", line 433, in wrapper return eager_transforms.grad_impl(func, argnums, has_aux, args, kwargs) File "/opt/conda/envs/py_3.14/lib/python3.14t/site-packages/torch/_functorch/eager_transforms.py", line 1533, in grad_impl results = grad_and_value_impl(func, argnums, has_aux, args, kwargs) File "/opt/conda/envs/py_3.14/lib/python3.14t/site-packages/torch/_functorch/vmap.py", line 56, in fn return f(*args, **kwargs) File "/opt/conda/envs/py_3.14/lib/python3.14t/site-packages/torch/_functorch/eager_transforms.py", line 1484, in grad_and_value_impl output = func(*args, **kwargs) File "/var/lib/jenkins/workspace/test/functorch/test_eager_transforms.py", line 3784, in forward return x.sin()

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" "

To execute this test, run the following from the base repo dir: PYTORCH_TEST_WITH_DYNAMO=1 python test/functorch/test_eager_transforms.py TestComposabilityCPU.test_autograd_function_no_setup_context_transform_grad_cpu

This message can be suppressed by setting PYTORCH_PRINT_REPRO_ON_FAILURE=0

TL;DR

The most likely fix for the failing test is to modify the test to handle the RuntimeError exception correctly, possibly by overriding the setup_context method or by using torch.compiler.set_stance("force_eager").

Guidance

• Review the test code in functorch/test_eager_transforms.py to ensure it correctly handles the RuntimeError exception.
• Consider overriding the setup_context method in the MySin class to fix the error.
• Try using torch.compiler.set_stance("force_eager") to run the test in eager mode and see if the error persists.
• Check the PyTorch documentation and the provided links for more information on debugging and fixing the issue.
• Run the test with TORCHDYNAMO_VERBOSE=1 and TORCH_LOGS="+dynamo" to get more detailed error messages and internal stack traces.

Example

# Example of using torch.compiler.set_stance("force_eager")
import torch
torch.compiler.set_stance("force_eager")

Notes

The error message suggests that the test is failing due to a RuntimeError exception when running in compiled mode. The provided links and documentation may help in debugging and fixing the issue. However, without more information about the test code and the expected behavior, it is difficult to provide a more specific solution.

Recommendation

Apply the workaround by using torch.compiler.set_stance("force_eager") to run the test in eager mode, as this may help to identify and fix the underlying issue.