Huy Do

For the context, after this change lands in trunk, the new CUDA 12.4 build starts to fails on newly created open PyTorch PR. Here is what happens: * This PR...

@pytorchbot drci

@pytorchbot revert -m 'I think we are using vision in inductor tests with their various models there' -c nosignal

Inductor test failing in trunk https://hud.pytorch.org/pytorch/pytorch/commit/f42ea14c3f795082138421fcef90d24f64c6fd35

@pytorchbot revert -m 'Sorry for reverting your change, I am trying to restore ROCm distributed failures in trunk https://hud.pytorch.org/pytorch/pytorch/commit/9aa7699185e4ec39077e3046dfd63244dffa9ddb' -c weird I'm not entirely sure if the failure is related,...

The increase could very well coming from https://github.com/pytorch/pytorch/pull/118935 given the history of building flash_attention. Here is what I see in the log: In the latest nightly where it timeout, `flash_attention`...

@pytorchbot revert -m 'Sorry for reverting your change, but there are real failures on the PR that sneak in during the log classifier outage' -c weird https://hud.pytorch.org/pytorch/pytorch/commit/2b41e1d6fc05428008875e3cfe8be17184e57491

@pytorchbot revert -m 'Sorry for reverting your change but the newly added test runs out of memory https://hud.pytorch.org/pytorch/pytorch/commit/95b9e981c3ab68fc17f78b8a6bbfd9569745ae4c' -c nosignal (Note for @clee2000, I remember we are still using 2...

@pytorchbot merge -f 'Lint job have passed, the micro benchmark has been working in the previous iteration'

@pytorchbot merge