Add sweeper support for `flipping_amplitude` by sorewachigauyo · Pull Request #1508 · qiboteam/qibocal

sorewachigauyo · 2026-05-19T14:53:00Z

As discussed in PR #1504, an implementation of the routine with amplitude sweeper

This probably has to be modified based on how 42aea44 goes and if #1507 gets merged due to

            # First two pulses are the pi/2 pulse and unused alignment and the last pulse is a readout alignment
            # So we add every pulse in-between
            # Technically we can reduce the number of pulses by filtering by unique UUID
            pulses_to_sweep[qubit].extend(
                list(seq.channel(platform.qubits[qubit].drive))[2:-1]
            )

which I'm not super happy about, but it works for now

codecov · 2026-05-19T14:55:08Z

Codecov Report

❌ Patch coverage is 5.26316% with 18 lines in your changes missing coverage. Please review.
✅ Project coverage is 93.71%. Comparing base (97f919c) to head (ffa5908).

Files with missing lines	Patch %	Lines
src/qibocal/protocols/flipping_amplitude.py	5.26%	18 Missing ⚠️

Additional details and impacted files

@@                  Coverage Diff                   @@
##           flipping_amplitude    #1508      +/-   ##
======================================================
- Coverage               93.76%   93.71%   -0.06%     
======================================================
  Files                     137      137              
  Lines                   10799    10803       +4     
======================================================
- Hits                    10126    10124       -2     
- Misses                    673      679       +6

Flag	Coverage Δ
unittests	`93.71% <5.26%> (-0.06%)`	⬇️

Flags with carried forward coverage won't be shown. Click here to find out more.

Files with missing lines	Coverage Δ
src/qibocal/protocols/flipping_amplitude.py	`39.84% <5.26%> (-2.10%)`	⬇️

... and 3 files with indirect coverage changes

🚀 New features to boost your workflow:

❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.
📦 JS Bundle Analysis: Save yourself from yourself by tracking and limiting bundle sizes in JS merges.

lballerio

Thanks @sorewachigauyo for this implementation.
As I said, I think there is still something we can refactor and simplify, but we can postpone this discussion after we merge this PR on PR #1504.
As you mentioned here and in PR #1507, in your implementation we are recycling the same pulse, which I am getting convinced might be the more practical solution since we are sweeping the same way all the pulses in the experiment, so we might actually recycle the pulse so we don't have a huge list of pulses which differ only for the UUID we sweep in the same way...
Btw did you test it?
I am simply waiting for @jevillegasd review.

lballerio · 2026-05-20T07:16:31Z

+    for flips, sequence in zip(flips_range, sequences):
+        for qubit, sweeper in zip(targets, parallel_sweepers):
            acq_channel = platform.qubits[qubit].acquisition
            assert acq_channel is not None
            ro_pulse = list(sequence.channel(acq_channel))[-1]
            assert isinstance(ro_pulse, Readout)
-            prob = results[ro_pulse.id]
-            error = np.sqrt(prob * (1 - prob) / params.nshots)
-            native_amp = data.pulse_amplitudes[qubit]
-            data.register_qubit(
-                FlippingAmplitudeType,
-                qubit,
-                {
-                    "flips": np.array([flips]),
-                    "amplitude": np.array([native_amp + delta_amp]),
-                    "prob": np.array([prob]),
-                    "error": np.array([error]),
-                },
-            )
+            prob_array = results[ro_pulse.id]
+            assert len(prob_array) == len(sweeper.values)
+            for amp, prob in zip(sweeper.values, prob_array):
+                error = np.sqrt(prob * (1 - prob) / params.nshots)
+                data.register_qubit(
+                    FlippingAmplitudeType,
+                    qubit,
+                    {
+                        "flips": np.array([flips]),
+                        "amplitude": np.array([amp]),
+                        "prob": np.array([prob]),
+                        "error": np.array([error]),
+                    },
+                )


maybe can be simplified, but we can do it in PR #1504.

lballerio · 2026-05-20T07:20:25Z

+            # First two pulses are the pi/2 pulse and unused alignment and the last pulse is a readout alignment
+            # So we add every pulse in-between
+            # Technically we can reduce the number of pulses by filtering by unique UUID
+            pulses_to_sweep[qubit].extend(
+                list(seq.channel(platform.qubits[qubit].drive))[2:-1]
+            )


I am not an expert on this experiment, but should we also sweep on the initial RX90 pulse amplitude?

Personally, I don't think we have to.

This particular experiment is looking for the minimum variance between adding multiple $2\pi$ rotations from the initial state, so the initial state doesn't have to be the exact middle, so long as we return to it. (And I guess in that sense, we don't have to start from the middle either)

I think the initial state affects the original experiment more under certain conditions, as that decides under/over-rotation fitting.

I see your point, yeah probably your right, then probably won't make such a difference.

sorewachigauyo · 2026-05-20T07:41:47Z

Thanks @sorewachigauyo for this implementation. As I said, I think there is still something we can refactor and simplify, but we can postpone this discussion after we merge this PR on PR #1504. As you mentioned here and in PR #1507, in your implementation we are recycling the same pulse, which I am getting convinced might be the more practical solution since we are sweeping the same way all the pulses in the experiment, so we might actually recycle the pulse so we don't have a huge list of pulses which differ only for the UUID we sweep in the same way...

On my driver implementation, sweeping/playing multiple pulses with the same UUID is not an issue. My concern from the other PR is if that was an issue with QM/Qblox, which I don't have in my lab to test. If this isn't a problem, then yeah, we should go ahead with the recycling, because in the current implementation (prior to #1507), each call of flipping_sequence generates flips number of UUIDs due to replace making a new pulse and replace being used in a loop.

Btw did you test it?

Yeah, it was actually in the other PR, but here is a new test
flipping_2d.tar.gz

lballerio · 2026-05-20T11:34:39Z

Hi @sorewachigauyo, I just tried to run this implementation on qblox and actually the sweeper is too big and does not fit, it raises an error.
I am trying to modify a bit the code.

…ment

sorewachigauyo · 2026-05-20T13:53:30Z

Hi @lballerio
Could you try this branch https://github.com/qiboteam/qibocal/tree/flipping_amplitude_sweeper_minpulses?

Its a bit more extreme rewrite that doesn't reuse the existing sequence generation and should make qibolab happier
QPU data flipping_2d.tar.gz

If this doesn't work, then I'll probably change this into a separate PR for sweepers

lballerio · 2026-05-20T14:00:05Z

This is the memory issue on the sweepers, as you can see from the error message:

I also tried to run by repeating always the same pulse, trying to see if inserting only once the pulse to sweep it effectively sweeps over all the pulses in the sequence, but at the time being it seems it is not working as it should, but I need to see in more detail.

Ok let me see your branch, thanks!

lballerio · 2026-05-20T14:45:03Z

Hi @sorewachigauyo I tried your new branch and the results I think a re very satifying!
In this way you are simply using the same UUID throughout the entire experiment, right?
Maybe this is better.

I also worked on this sweeper implementation and ended up with https://github.com/qiboteam/qibocal/tree/flipping_amplitude_sweeper_lballerio.
In my case I was calling flipping_sequence for each nflips values, hence I am sweeping over N pulses, with N total number of nflips, since within the same sequence the UUID is the same. However these modifications didn't measure good signal, as you can see from the picture below:

if you want to have a look, but I think you're implementation is more than fine...maybe we should move the PR on that branch.
I think in my implementation there is probably some error in the sweeper usage, better to use your which is just the same pulse

sorewachigauyo · 2026-05-20T14:55:35Z

Yeah, its the same UUID/pulse object for the entire experiment

Okay, good to see that it works with Qblox
I'll just push the commit here

lballerio · 2026-05-20T15:36:20Z

perfect, yeah I like your implementation!

lballerio

tested with Qblox on TII qw5q_platinum QPU, I think is good

…cation. This is still however unable to run for long pulse durations in the spin lock pulse.

feat(flipping): add sweeper support for amp protocol

a879949

sorewachigauyo changed the title ~~feat(flipping): add sweeper support for amp protocol~~ Add sweeper support for flipping_amplitude May 19, 2026

sorewachigauyo requested a review from jevillegasd May 19, 2026 14:59

lballerio reviewed May 20, 2026

View reviewed changes

style(flipping): update type hint

001c80c

perf(flipping): optimize sequence generation and pulse-sweeper assign…

776a94d

…ment

lballerio approved these changes May 20, 2026

View reviewed changes

scarrazza requested a review from a team May 21, 2026 08:15

scarrazza approved these changes May 21, 2026

View reviewed changes

Merge branch 'flipping_amplitude' into flipping_amplitude_sweeper

ffa5908

lballerio mentioned this pull request Jun 1, 2026

Improve flipping sequence generation #1507

Merged

lballerio merged commit 89cae79 into flipping_amplitude Jun 1, 2026
38 of 39 checks passed

lballerio deleted the flipping_amplitude_sweeper branch June 1, 2026 14:03

jevillegasd added a commit that referenced this pull request Jun 15, 2026

Use a pulse library like in #1508 to try a more efficient memory allo…

2ab0cd7

…cation. This is still however unable to run for long pulse durations in the spin lock pulse.

Uh oh!

Conversation

sorewachigauyo commented May 19, 2026

Uh oh!

codecov Bot commented May 19, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Codecov Report

Uh oh!

lballerio left a comment

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Uh oh!

lballerio May 20, 2026

Choose a reason for hiding this comment

Uh oh!

lballerio May 20, 2026

Choose a reason for hiding this comment

Uh oh!

sorewachigauyo May 20, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Choose a reason for hiding this comment

Uh oh!

lballerio May 20, 2026

Choose a reason for hiding this comment

Uh oh!

sorewachigauyo commented May 20, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

lballerio commented May 20, 2026

Uh oh!

sorewachigauyo commented May 20, 2026

Uh oh!

lballerio commented May 20, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

lballerio commented May 20, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

sorewachigauyo commented May 20, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

lballerio commented May 20, 2026

Uh oh!

lballerio left a comment

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

3 participants

codecov Bot commented May 19, 2026 •

edited

Loading

sorewachigauyo May 20, 2026 •

edited

Loading

sorewachigauyo commented May 20, 2026 •

edited

Loading

lballerio commented May 20, 2026 •

edited

Loading

lballerio commented May 20, 2026 •

edited

Loading

sorewachigauyo commented May 20, 2026 •

edited

Loading