Resolving inconsistency between attention/attention_bias

[parsifal-47]

This is a fix for:
#17486

Inconsistency between attention/attention_bias operations, after the fix, the script from #17486
is printing proper IR.
Thank you!

[parsifal-47]

@Cookiee235 please take a look when you have a chance, thank you!

[jikechao]

@tvm-bot rerun

[jikechao]

@tqchen Do you know why the CI testing is pending?

[yongwww]

@jikechao we are working on fixing the ci

[yongwww]

@tvm-bot rerun

[yongwww]

@tvm-bot rerun

[Hzfengsy]

@parsifal-47 Please fix the ci checks :)

[parsifal-47]

@parsifal-47 Please fix the ci checks :)

I am going to check that over the weekend, I see that unity/pr-head is failing, but I do not see any mentions of unity here: https://github.com/apache/tvm/blob/main/.github/workflows/main.yml, do you know how to reproduce the failure locally? Thank you!

[Hzfengsy]

Here is the error. You could run the command below :)

pytest tests/python/relax/test_transform_allocate_workspace.py

[parsifal-47]

@tvm-bot rerun

[parsifal-47]

Here is the error. You could run the command below :)

pytest tests/python/relax/test_transform_allocate_workspace.py

thank you for the reference, the test should be fixed now

[parsifal-47]

Here is the error. You could run the command below :)

pytest tests/python/relax/test_transform_allocate_workspace.py

now it fails on tvm.ir.assert_structural_equal in tests/python/relax/test_transform_legalize_ops_nn.py, is there a way to see what is different between two IRs? They look very close