Fix mixer forwarding | Fix MPI implementation | Add test cases

[kaiqiy-nv]

Fix mixer (referenceHamiltonian) forwarding in qaoa Python binder.
Fix MPI implementation in adapt simulator.
Add test cases to validate the above modifications.

[copy-pr-bot]

This pull request requires additional validation before any workflows can run on NVIDIA's runners.

Pull request vetters can view their responsibilities here.

Contributors can view more details about this message here.

[bmhowe23]

/ok to test ebb568f

[bmhowe23]

This CI failure looks unrelated to this PR. It may have been caused by a previous PR. Will need to investigate.

[bmhowe23]

/ok to test 000960a

[bmhowe23]

This CI failure looks unrelated to this PR. It may have been caused by a previous PR. Will need to investigate.

This should be fixed by #453

[kaiqiy-nv]

Just a kind reminder that this PR is for NVBug 5933231 and NVBug 5933312.

[bmhowe23]

/ok to test d4f1930

[bmhowe23]

Just a kind reminder that this PR is for NVBug 5933231 and NVBug 5933312.

Hi @kaiqiy-nv - could you please investigate these CI failures from the "Build wheels" job for this PR? https://github.com/NVIDIA/cudaqx/actions/runs/22824100177

[kaiqiy-nv]

==================================== ERRORS ====================================
_________ ERROR collecting libs/solvers/python/tests/test_adapt_mpi.py _________
libs/solvers/python/tests/test_adapt_mpi.py:20: in <module>
    cudaq.mpi.initialize()
E   RuntimeError: No MPI support can be found when attempted to use cudaq::mpi APIs. Please refer to the documentation for instructions to activate MPI support.

Hello @bmhowe23 , those errors are caused by the new MPI tests. I am not sure if the CI supports MPI already, but I have updated the test cases (both gtest and pytest) to let them more compatible with the environment. Please have another /ok to test. Thanks.

[bmhowe23]

/ok to test c283051

[bmhowe23]

It looks like there are some per-PR CI failures now.

• E.g. https://github.com/NVIDIA/cudaqx/actions/runs/22869263614/job/66344419087?pr=452

There are also some different "Build wheels" failures here:

• https://github.com/NVIDIA/cudaqx/actions/runs/22869280837/job/66348387094#step:11:6237

[kaiqiy-nv]

Again both of them are related to MPI tests.
For the PR workflow job, it is because the UCX PML is not supported on the CPU-only CI runner.
For the Build Wheels job, it seems that the CUDA-Q installed by the automation script was not built with MPI supports.
Therefore,
(1) I have reformed the gtests and pytests involving MPI again to make sure it can skip testing if the platform does not support MPI or cudaq::mpi fully.
(2) I have made a small change on the PR workflow config to let the ctest use ob1 instead of the default ucx.

Please have another try. Thanks.

[bmhowe23]

/ok to test 3029dcc

[bmhowe23]

Thanks, @kaiqiy-nv. Now all tests are passing, which is great. It looks good to me, but @wsttiger or @kvmto, could you please give it a quick review to make sure you are ok with it, too?

[kvmto]

Nice! Looks good to me!
Just two minor comments about the tests :) if possible to address.
I would have added a test to exercise qaoa with an asymmetric mixer like 3.0X0 + 0.1X1 and comparing at fixed parameters.

[kaiqiy-nv]

Hello @kvmto , I have updated the python tests accordingly. Please /ok to test and review the latest. Thanks!

[kvmto]

/ok to test 52fd669

[kvmto]

looks good to me, we can merge!