Handling global backend for protocols involving circuits #1076

andrea-pasquale · 2025-02-05T15:33:21Z

Closes #1065, by directly instatiating a local backend in each acquisition function.

Other changes implemented in this PR:

Drop noise models for RB
Adding to pytest the env variable QIBO_PLATFORM to properly run tests on the cluster
Small fixes related to other tests

codecov · 2025-02-05T15:35:09Z

Codecov Report

All modified and coverable lines are covered by tests ✅

Project coverage is 97.31%. Comparing base (bc7a7ba) to head (d519948).
Report is 15 commits behind head on main.

Additional details and impacted files

@@           Coverage Diff           @@
##             main    #1076   +/-   ##
=======================================
  Coverage   97.30%   97.31%           
=======================================
  Files         124      123    -1     
  Lines        9884     9844   -40     
=======================================
- Hits         9618     9580   -38     
+ Misses        266      264    -2

Flag	Coverage Δ
unittests	`97.31% <100.00%> (+<0.01%)`	⬆️

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

Files with missing lines	Coverage Δ
src/qibocal/auto/execute.py	`97.67% <ø> (ø)`
src/qibocal/auto/runcard.py	`97.50% <100.00%> (-0.07%)`	⬇️
src/qibocal/auto/transpile.py	`90.90% <100.00%> (-0.76%)`	⬇️
...l/protocols/randomized_benchmarking/standard_rb.py	`98.50% <100.00%> (-0.09%)`	⬇️
...qibocal/protocols/randomized_benchmarking/utils.py	`96.01% <100.00%> (+0.21%)`	⬆️
src/qibocal/protocols/readout_mitigation_matrix.py	`100.00% <100.00%> (+2.50%)`	⬆️
src/qibocal/protocols/state_tomography.py	`98.64% <100.00%> (-0.01%)`	⬇️
...l/protocols/two_qubit_interaction/chsh/protocol.py	`97.07% <100.00%> (-0.60%)`	⬇️
...rc/qibocal/protocols/two_qubit_state_tomography.py	`99.28% <100.00%> (-0.01%)`	⬇️

alecandido

Just a couple of comments, but in general I agree with the overall idea (since we're now focusing on the platform, assuming the backend to be Qibolab, there is no need to require everywhere to create the entire backend)

alecandido · 2025-02-05T16:13:59Z

src/qibocal/auto/transpile.py

+    """Qibolab platforms."""
+    try:
+        platforms = list(MetaBackend().list_available())
+    except RuntimeError:


Not a comment on this PR, but just a point raised as a consequence.

Maybe we should actually drop this exception: if there is no path, we should already return an empty list.
It should be sufficient to do it here:
https://github.com/qiboteam/qibolab/blob/6b2a7b790e8d58d52be937c2f41b8606459a3bf1/src/qibolab/_core/platform/load.py#L23-L26

This should result in .list_available() returning an empty list, and .load() raising a different error.

@stavros11

alecandido · 2025-02-05T16:27:38Z

src/qibocal/auto/transpile.py

-            qubit_maps[i] = list(qubit_map)
-    if backend.name == "qibolab":
-        platform_nqubits = backend.platform.nqubits
+    if platform.name in AVAILABLE_PLATFORMS:


Maybe I just forgot, but in which case is this condition supposed to be false?

I think it should be False when we transpile circuits that are executed on simultation so no transpilation at all.

I think it should be False when we transpile circuits that are executed on simultation so no transpilation at all.

Ah, I see. But in that case there should be no Platform at all, since it is the qibolab.Platform.

My advice would be to drop any simulation reference, and assume it is always Qibolab (possibly the emulator, but still Qibolab).
I would reintroduce the option for simulation as part of a larger refactor, taking into account #913. I'd expect there should be other as well not supporting a backend other than Qibolab.

src/qibocal/auto/transpile.py

andrea-pasquale · 2025-02-06T14:07:25Z

I realized that by properly dropping the possibility to run on simultation we don't need many shenaningans involving MetaBackend, for the time being I'm just creating the correct local backend passing the same Platform object.
Let me know if you agree with this solution.

Co-authored-by: Alessandro Candido <candido.ale@gmail.com>

andrea-pasquale · 2025-02-06T14:11:59Z

@stavros11 I believe that with this approach we don't need anymore #1063, feel free to have a look and eventually we can just merge this PR if you agree.

stavros11

@stavros11 I believe that with this approach we don't need anymore #1063, feel free to have a look and eventually we can just merge this PR if you agree.

Thanks @andrea-pasquale. Indeed #1063 was just a temporary solution, not intended to be merged, so feel free to close it after this is merged.

stavros11 · 2025-02-11T10:03:40Z

src/qibocal/auto/transpile.py

@@ -28,20 +28,18 @@ def transpile_circuits(
        are all string or all integers.
    """
    transpiled_circuits = []
-
-    qubits = list(backend.platform.qubits)
+    platform = backend.platform


In principle here you could add something like

from qibolab.platform import Platform if not isinstance(platform, Platform): transpiled_circuits = circuits

to retain the previous behavior when the backend is not qibolab.

I would keep assuming it is Qibolab, to make everything simpler. Even though in some places may be simpler to deal with other backends, it may fail in other places. We will overhaul this consistently later on.

Indeed, the expected behavior would be to have a qibolab platform

qibocal/src/qibocal/cli/run.py

Lines 22 to 23 in 59f95e3

if platform is None:

raise ValueError("Qibocal requires a Qibolab platform to run.")

stavros11 · 2025-02-11T10:04:18Z

tests/conftest.py

+def pytest_configure():
+    os.environ["QIBO_PLATFORM"] = "dummy"


If we are avoiding the GlobalBackend, why is this needed?

There are some tests which are specifically looking at this env variable:

qibocal/tests/test_executor.py

Line 179 in bc7a7ba

def test_default_executor(tmp_path: Path, fake_platform: str, monkeypatch):

I noticed it because when I wan running tests on the cluster they were trying to connect to QM.

Sorry, my bad.

I introduced that test when I implemented the scripting mechanism, because that was the context at the time. However, there is no special reason to keep it, so you could just drop the test.

To avoid trigerring the error we can also decide to not rely on QIBO_PLATFORM anymore and just use dummy as default platform given that the behavior of QIBO_PLATFORM is mostly undocumented.

andrea-pasquale added 3 commits January 14, 2025 17:12

fix: get backend from platform

cf08f45

fix: Catch error when getting platforms

8c52ce5

feat: Use MetaBackend

53ba1be

andrea-pasquale requested review from Edoardo-Pedicillo and alecandido February 5, 2025 15:33

alecandido reviewed Feb 5, 2025

View reviewed changes

andrea-pasquale added 3 commits February 6, 2025 16:30

feat\!: Drop simulation

713956b

test: Add QIBO_PLATFORM

a86f91d

fix: More test fixes spotted on cluster

8ea8342

andrea-pasquale force-pushed the fix_backend branch from 0674c61 to 8ea8342 Compare February 6, 2025 13:19

andrea-pasquale added 3 commits February 6, 2025 17:23

Merge branch 'main' into fix_backend

b7195db

feat: Simplify backend handling by setting local backend

9edcba8

chore: Remove noise models related files

7c5aed5

alecandido approved these changes Feb 6, 2025

View reviewed changes

src/qibocal/auto/transpile.py Outdated Show resolved Hide resolved

Update src/qibocal/auto/transpile.py

61039a9

Co-authored-by: Alessandro Candido <candido.ale@gmail.com>

andrea-pasquale requested a review from stavros11 February 6, 2025 14:12

andrea-pasquale force-pushed the fix_backend branch 2 times, most recently from c2d5df9 to 29bc178 Compare February 6, 2025 16:14

test: More fixes for tests

59f95e3

andrea-pasquale force-pushed the fix_backend branch from 29bc178 to 59f95e3 Compare February 6, 2025 16:21

stavros11 approved these changes Feb 11, 2025

View reviewed changes

refactor: Drop QIBO_PLATFORM reference

d519948

andrea-pasquale added this pull request to the merge queue Feb 12, 2025

Merged via the queue into main with commit d8e8dda Feb 12, 2025
21 checks passed

andrea-pasquale deleted the fix_backend branch February 12, 2025 09:45

andrea-pasquale mentioned this pull request Feb 12, 2025

Temporary fix of error mitigation routine if QIBO_PLATFORM is not set #1063

Closed

This was referenced Feb 19, 2025

Compatibility with the new Qibo global backend handling #1022

Closed

Refactor after update to qibolab 0.2 #1095

Open

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Handling global backend for protocols involving circuits #1076

Handling global backend for protocols involving circuits #1076

andrea-pasquale commented Feb 5, 2025 •

edited

Loading

codecov bot commented Feb 5, 2025 •

edited

Loading

alecandido left a comment

alecandido Feb 5, 2025

alecandido Feb 5, 2025

andrea-pasquale Feb 5, 2025

alecandido Feb 5, 2025

andrea-pasquale commented Feb 6, 2025

andrea-pasquale commented Feb 6, 2025

stavros11 left a comment

stavros11 Feb 11, 2025

alecandido Feb 11, 2025

andrea-pasquale Feb 11, 2025

stavros11 Feb 11, 2025

andrea-pasquale Feb 11, 2025

alecandido Feb 11, 2025

andrea-pasquale Feb 11, 2025

	if platform is None:
	raise ValueError("Qibocal requires a Qibolab platform to run.")

		def pytest_configure():
		os.environ["QIBO_PLATFORM"] = "dummy"

Handling global backend for protocols involving circuits #1076

Handling global backend for protocols involving circuits #1076

Conversation

andrea-pasquale commented Feb 5, 2025 • edited Loading

codecov bot commented Feb 5, 2025 • edited Loading

Codecov Report

alecandido left a comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

andrea-pasquale commented Feb 6, 2025

andrea-pasquale commented Feb 6, 2025

stavros11 left a comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

andrea-pasquale commented Feb 5, 2025 •

edited

Loading

codecov bot commented Feb 5, 2025 •

edited

Loading