KeyError: 'GPI'

[scarrazza]

After installing qibolab 0.1.10, the the following code:

import qibo
c = qibo.Circuit(17)
c.add(qibo.gates.GPI2(12,0))
c.add(qibo.gates.M(12))
result = c(nshots=100)
print(result.samples())
print(result.frequencies())

executed with module load qibo crashes on qw11q with:

2024-11-03 18:25:04,167 - qm - INFO     - Starting session: 903803b8-8d72-48e9-9dc3-0c438f6b5251
Traceback (most recent call last):
  File "/nfs/users/stefano.carrazza/h.py", line 14, in <module>
    result = c(nshots=100)
  File "/nfs/tools/qibo/lib/python3.10/site-packages/qibo/models/circuit.py", line 1116, in __call__
    return self.execute(initial_state=initial_state, nshots=nshots)
  File "/nfs/tools/qibo/lib/python3.10/site-packages/qibo/models/circuit.py", line 1105, in execute
    transpiler = _Global.transpiler()
  File "/nfs/tools/qibo/lib/python3.10/site-packages/qibo/backends/__init__.py", line 126, in transpiler
    cls._transpiler = cls._default_transpiler()
  File "/nfs/tools/qibo/lib/python3.10/site-packages/qibo/backends/__init__.py", line 163, in _default_transpiler
    Unroller(NativeGates[natives]),
  File "/nfs/tools/qibo/lib/python3.10/site-packages/qibo/transpiler/unroller.py", line 26, in __getitem__
    return reduce(or_, [cls[key] for key in keys])  # pylint: disable=E1136
  File "/nfs/tools/qibo/lib/python3.10/site-packages/qibo/transpiler/unroller.py", line 26, in <listcomp>
    return reduce(or_, [cls[key] for key in keys])  # pylint: disable=E1136
  File "/nfs/tools/qibo/lib/python3.10/site-packages/qibo/transpiler/unroller.py", line 25, in __getitem__
    return super().__getitem__(keys)
  File "/usr/lib/python3.10/enum.py", line 440, in __getitem__
    return cls._member_map_[name]
KeyError: 'GPI'

@csookim could you please check?

[stavros11]

I think the fix is in qiboteam/qibo#1504.

[alecandido]

I think the fix is in qiboteam/qibo#1504.

Indeed, I believe that was the main motivation behind qiboteam/qibo#1504

[scarrazza]

We should do more testing before making releases to avoid these kinds of bugs.

[alecandido]

We should do more testing before making releases to avoid these kinds of bugs.

In this specific case, I believe Qibolab v0.1.10 is fine, we won't need to fix anything.

The problem is that the Qibo side has never been completed (we could have tested more in qiboteam/qibo#1440, but in principle that PR was about the global backend, while we're still waiting for the Qibolab-compatibility support).

[csookim]

We should do more testing before making releases to avoid these kinds of bugs.

The issue is on the Qibo side: Qibo was supposed to raise an error when it received a gate that cannot be used as a native gate.

https://github.com/qiboteam/qibo/blob/22d51946add6739a3e970f32ee8cdd0bd70f5d2d/src/qibo/transpiler/unroller.py#L67-L68

https://github.com/qiboteam/qibo/blob/22d51946add6739a3e970f32ee8cdd0bd70f5d2d/tests/test_transpiler_unroller.py#L19-L21

I have merged qiboteam/qibo#1504, and we should patch it to enable circuit execution. Do we need to wait for qibo 0.2.14?

[alecandido]

I have merged qiboteam/qibo#1504, and we should patch it to enable circuit execution. Do we need to wait for qibo 0.2.14?

We need the .wire_names in any case, so we should complete qiboteam/qibo#1500. No reason to anticipate the next release before that.

Current testing can be done with master. Releasing before that may lead to a similar problem with the next one (missing required features/not enough testing).

[scarrazza]

On top of that, we should try to propose symultaneously a set of runcards (or sliced runcards) that works for at least the lines we are calibrating (D and B) for qw11q. At the moment, setting more than 11 qubits will generate an error message.

[alecandido]

Ok, this is pretty weird, since there are 16 qubits in the platform

https://github.com/qiboteam/qibolab_platforms_qrc/blob/9e80f12ee40436eb9c3582747b09499dbc0f692f/qw11q/parameters.json#L7-L24

(mostly not calibrated, but they should all be exposed)

[scarrazza]

Thanks @alecandido, indeed with 16 qubits it works. I have installed the latest qibo main in the cluster.

[alecandido]

@csookim could you investigate and report about the mentioned error?

[scarrazza]

I believe this is okay @alecandido. My example was using 17 qubits instead of 16. Using 16 works fine.

[csookim]

@csookim could you investigate and report about the mentioned error?

The KeyError: 'GPI' error? Yes, this occurred because there wasn't a specific rule to handle native gates between qibo and qibolab. qiboteam/qibo#1504 resolved this issue. I am testing whether transpiler works well by creating qw5q_dummy and qw11q_dummy.

@scarrazza Could you give me permission to create a branch in qibolab_platforms_qrc? I would like to make a branch that includes the qw5q_dummy and qw11q_dummy models. This branch will not be merged; just for simulating the connection on the CPU.

[scarrazza]

@csookim please go ahead, you should have write access now.

[alecandido]

@scarrazza did you have a chance to test whether this problem is still there? Hopefully, now that qiboteam/qibo#1504 has been merged, it should have been finally solved :)

[scarrazza]

Yes, we can close.