pytorch - 💡(How to fix) Fix DISABLED test_graph_rng_after_failed_capture (__main__.TestCuda) [2 comments, 3 participants]

TL;DR

• Re-enable and re-run the disabled test test_graph_rng_after_failed_capture in test_cuda.py to assess the current failure status.

Guidance

• Review the recent failure examples on torch-ci.com to understand the failure patterns and potential causes.
• Investigate the test test_graph_rng_after_failed_capture in test_cuda.py to identify the root cause of the failure, considering the ROCm platform specifics.
• Consider reaching out to the listed maintainers and experts for additional guidance and support, given their expertise in ROCm and CUDA-related issues.

Notes

• The issue seems to be specific to the ROCm platform, so any solution or workaround might need to consider the unique characteristics and limitations of this platform.
• Without more detailed information about the failure, such as error messages or stack traces, it's challenging to provide a more specific fix or workaround.

Recommendation

• Apply workaround: Given the lack of detailed information, it's safer to apply a workaround, such as modifying the test to accommodate the ROCm platform specifics or temporarily skipping the test until a more permanent fix can be implemented.