TL;DR

The test failure in test_scaled_matmul_cuda.py on the main branch may be resolved by investigating and addressing the issues related to the TestFP8MatmulCUDA test case, specifically the test_scaled_mm_deepseek_error_messages_bfloat16_lhs_block_128_rhs_block_1_M_256_N_256_K_512_cuda test.

Guidance

• Review the recent failure examples on torch-ci.com to understand the error messages and potential causes.
• Investigate the TestFP8MatmulCUDA test case and the specific test test_scaled_mm_deepseek_error_messages_bfloat16_lhs_block_128_rhs_block_1_M_256_N_256_K_512_cuda to identify the root cause of the failure.
• Check the test configuration and inputs, such as the block sizes and data types, to ensure they are valid and consistent with the expected behavior.
• Consider re-enabling the test and running it with additional logging or debugging to gather more information about the failure.

Notes

The issue lacks specific technical details, so a more detailed analysis of the test case and failure messages is necessary to determine the root cause and develop a fix.

Recommendation

Apply workaround: Investigate and address the issues related to the TestFP8MatmulCUDA test case, as it is likely that the test failure is due to a specific issue with this test rather than a broader problem.