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

TL;DR

• The test test_bmm_large_batch_dynamic_cuda in test_aot_inductor.py needs to be fixed or re-enabled after resolving the underlying issue causing its failure on the main branch.

Guidance

• Review the recent failure examples on torch-ci.com to understand the nature of the test failure.
• Investigate the test_bmm_large_batch_dynamic_cuda method in test_aot_inductor.py for potential issues related to CUDA or GPU compatibility.
• Consider reaching out to the listed maintainers or experts (@jeffdaily, @sunway513, etc.) for guidance on resolving the test failure.
• Check if there are any known issues or fixes in the works for similar tests or CUDA-related problems in the project.

Notes

• The issue seems to be specific to the ROCm platform and CUDA compatibility, so any fixes or workarounds may need to be targeted at this environment.
• Without more details on the failure, it's difficult to provide a specific code-level fix.

Recommendation

• Apply workaround: Temporarily disable the failing test or create a workaround to allow the main branch to proceed while the root cause is investigated, given that the test failure is blocking progress.