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

Traceback (most recent call last): File "/var/lib/jenkins/workspace/test/test_autograd.py", line 8716, in test_custom_function_boxed_grads out.sum().backward() ~~~~~~~~~~~~~~~~~~^^ File "/opt/conda/envs/py_3.14/lib/python3.14t/site-packages/torch/_tensor.py", line 631, in backward torch.autograd.backward( ~~~~~~~~~~~~~~~~~~~~~~~^ self, gradient, retain_graph, create_graph, inputs=inputs ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ ) ^ File "/opt/conda/envs/py_3.14/lib/python3.14t/site-packages/torch/autograd/init.py", line 379, in backward _engine_run_backward( ~~~~~~~~~~~~~~~~~~~~^ tensors, ^^^^^^^^ ...<5 lines>... accumulate_grad=True, ^^^^^^^^^^^^^^^^^^^^^ ) ^ File "/opt/conda/envs/py_3.14/lib/python3.14t/site-packages/torch/autograd/graph.py", line 882, in _engine_run_backward return Variable._execution_engine.run_backward( # Calls into the C++ engine to run the backward pass ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ t_outputs, *args, **kwargs ^^^^^^^^^^^^^^^^^^^^^^^^^^ ) # Calls into the C++ engine to run the backward pass ^ File "/opt/conda/envs/py_3.14/lib/python3.14t/site-packages/torch/_dynamo/compiled_autograd.py", line 685, in proxy_call_backward raise RuntimeError( ...<2 lines>... ) RuntimeError: boxed_grads_call=True on BoxedFunc is not supported with compiled autograd.

To execute this test, run the following from the base repo dir: python test/test_autograd.py TestAutogradWithCompiledAutograd.test_custom_function_boxed_grads

This message can be suppressed by setting PYTORCH_PRINT_REPRO_ON_FAILURE=0

TL;DR

The test test_custom_function_boxed_grads is failing due to an unsupported feature in compiled autograd, and a workaround or fix is needed to address the RuntimeError caused by boxed_grads_call=True on BoxedFunc.

Guidance

• Review the test file inductor/test_compiled_autograd.py to understand the usage of boxed_grads_call=True and its implications with compiled autograd.
• Investigate the possibility of disabling compiled autograd for this specific test or modifying the test to avoid using boxed_grads_call=True.
• Check the PyTorch documentation and issues for any known workarounds or fixes related to this error message.
• Consider reaching out to the maintainers or experts mentioned in the issue (e.g., @voznesenskym, @penguinwu) for guidance on resolving this issue.

Example

No code snippet is provided as the issue does not contain sufficient information to create a minimal example.

Notes

The issue is specific to the PyTorch framework and its interaction with compiled autograd. The error message suggests that the feature is not supported, and a workaround or fix is needed to address this issue.

Recommendation

Apply a workaround by modifying the test to avoid using boxed_grads_call=True or disabling compiled autograd for this specific test, as the error message indicates that this feature is not supported.