Pre-Trajectory Sampling with Batch Execution by sacpis · Pull Request #3995 · NVIDIA/cuda-quantum

sacpis · 2026-02-18T05:03:28Z

Adding Pre-Trajectory Sampling with Batch Execution (PTSBE) for noisy circuit sampling. Noise is pre-sampled into trajectories, circuits are run in batches per trajectory and results are aggregated. This enables noisy sampling on simulators.

Refer paper: https://arxiv.org/abs/2504.16297 and https://arxiv.org/pdf/2508.04880

Features

PTSBE pipeline
- Trace capture -> noise-site extraction -> trajectory generation (strategies: probabilistic, exhaustive, ordered, conditional) -> shot allocation (PROPORTIONAL (default), UNIFORM, LOW_WEIGHT_BIAS, HIGH_WEIGHT_BIAS) -> batch execution -> result aggregation.
Dedicated API
- C++
  - cudaq::ptsbe::sample()
  - cudaq::ptsbe::sample_async()
- Python
  - cudaq.ptsbe.sample()
  - cudaq.ptsbe.sample_async()
Noise sources
- Gate-based noise via NoiseModel
  - add_channel, add_all_qubit_channel
- Inline
  - cudaq.apply_noise(...) in the kernel
  - noise_model is optional when using apply_noise
Execution data (optional)
- Circuit structure (Gate/Noise/Measurement), trajectory specs, and per-trajectory counts via
  return_execution_data=True (Python) or options.ptsbe.return_execution_data (C++)
Trace and apply_noise
- Common Trace extended with instruction type enum (Gate / Noise) and appendNoiseInstruction
- NVQIR records apply_noise in tracer mode
- PTSBE skips Noise in gate conversion and resolves channels in the noise extractor

Limitations

Simulator-only (no remote)
No mid-circuit measurements or conditional feedback (deterministic gate sequence only)
Noise channels must be unitary mixtures (for trajectory sampling)

Co-authored-by: @taylorpatti, @1tnguyen, @taalexander

* * Adding the following data structures along with unittests for ptsbe - KrausSelection - KrausTrajectory - PTSBEResult - ShotAllocationStrategy - TrajectoryMetadata Signed-off-by: Sachin Pisal <spisal@nvidia.com> * formatting, spell fix Signed-off-by: Sachin Pisal <spisal@nvidia.com> * removing metadata Signed-off-by: Sachin Pisal <spisal@nvidia.com> * removing description Signed-off-by: Sachin Pisal <spisal@nvidia.com> * renaming KrausOperatorIndex to KrausOperatorType Signed-off-by: Sachin Pisal <spisal@nvidia.com> * pulling out 1e-9 as a constant Signed-off-by: Sachin Pisal <spisal@nvidia.com> * moving countErrors to trajectory Signed-off-by: Sachin Pisal <spisal@nvidia.com> * allocating remaining shots in a batch mode Signed-off-by: Sachin Pisal <spisal@nvidia.com> * allocating shots in one iteration for uniform shots allocation strategy Signed-off-by: Sachin Pisal <spisal@nvidia.com> * adding ordering validation for krausSelection Signed-off-by: Sachin Pisal <spisal@nvidia.com> * * Adding functions for shot allocation strategies * Throwing runtime error on 0 shots Signed-off-by: Sachin Pisal <spisal@nvidia.com> * Updating year in the header Signed-off-by: Sachin Pisal <spisal@nvidia.com> * updating year Signed-off-by: Sachin Pisal <spisal@nvidia.com> * updating year Signed-off-by: Sachin Pisal <spisal@nvidia.com> * adding a builder pattern for building a KrausTrajectory object Signed-off-by: Sachin Pisal <spisal@nvidia.com> * fix spell check Signed-off-by: Sachin Pisal <spisal@nvidia.com> * removing a test Signed-off-by: Sachin Pisal <spisal@nvidia.com> --------- Signed-off-by: Sachin Pisal <spisal@nvidia.com>

* Add interface unittests for TDD. * Complete initial interface definition/testing. * Test tracing intercept. * Add ptsbe injection. * Add core sample integration. * Do a pass over code. * Address review comment. Rename variables. * Fix pre-checks. * Make ptsbe error if no noise model.

* Add interface stubs. * Add gate conversion. Make tests/enqeue tasks public. * Implement task + trajectory merging. * Add generic simulator. * Update testing. * Refactor implementation and add generic invocation. * CI fixes. * CI formatting. * Fix field names in sample_intercept test. Use camelCase field names (kernelTrace, measureQubits) to match PTSBatch struct definition.

* Refactor sampling/ptsbe approaches. * Additional refactor of sampling. * address reviewer comments. * Add async/broadcast support. * Fix clang-format issues. * Update spelling allow list. * Formatting. * Remove changes to spelling allow list.

Signed-off-by: Thomas Alexander <talexander@nvidia.com>

* Add ptsbe generic implementation and early testing with (#3832) * Add interface stubs. * Add gate conversion. Make tests/enqeue tasks public. * Implement task + trajectory merging. * Add generic simulator. * Update testing. * Refactor implementation and add generic invocation. * CI fixes. * CI formatting. * Fix field names in sample_intercept test. Use camelCase field names (kernelTrace, measureQubits) to match PTSBatch struct definition. * Adding sampling strategies Signed-off-by: Sachin Pisal <spisal@nvidia.com> * adding unittests for sampling strategy Signed-off-by: Sachin Pisal <spisal@nvidia.com> * formatting Signed-off-by: Sachin Pisal <spisal@nvidia.com> * fix spell checker Signed-off-by: Sachin Pisal <spisal@nvidia.com> * * Expose isScaledUnitary() and computeUnitaryMixture() as public functions * Added docs to describe full validation requirements * Add a tolerance parameter Signed-off-by: Sachin Pisal <spisal@nvidia.com> * Added overflow protection Signed-off-by: Sachin Pisal <spisal@nvidia.com> * adding PTSSamplingStrategy to cmake Signed-off-by: Sachin Pisal <spisal@nvidia.com> * Modified ConditionalSamplingStartegy logic to be based on probabilistic sampling Signed-off-by: Sachin Pisal <spisal@nvidia.com> * format Signed-off-by: Sachin Pisal <spisal@nvidia.com> * fix spellings Signed-off-by: Sachin Pisal <spisal@nvidia.com> * adding optimized attempts, overflow protection Signed-off-by: Sachin Pisal <spisal@nvidia.com> * format Signed-off-by: Sachin Pisal <spisal@nvidia.com> * adding sampling strategies a sub directory Signed-off-by: Sachin Pisal <spisal@nvidia.com> * * Using cudaq's random seed if set * Using computeTotalTrajectories() in ExhaustiveSamplingStrategy and OrderedSamplingStrategy Signed-off-by: Sachin Pisal <spisal@nvidia.com> * fix spell check Signed-off-by: Sachin Pisal <spisal@nvidia.com> * using constant, removed irrelevant code Signed-off-by: Sachin Pisal <spisal@nvidia.com> --------- Signed-off-by: Sachin Pisal <spisal@nvidia.com> Co-authored-by: Thomas Alexander <thomasalexander2718@gmail.com>

…3870) * Adding NoiseExtractor to extract noise from a circuit trace Signed-off-by: Sachin Pisal <spisal@nvidia.com> * formatting Signed-off-by: Sachin Pisal <spisal@nvidia.com> * * NoisePoint now stores cudaq::kraus_channel instead of separate kraus_operators and probabilities vectors * Remove NoisePoint::isUnitaryMixture() method (validation now handled by kraus_channel itself) * Update computeTotalTrajectories() to use channel.size() Signed-off-by: Sachin Pisal <spisal@nvidia.com> * removing PTSSamplingStrategy.cpp from cmake Signed-off-by: Sachin Pisal <spisal@nvidia.com> * * NoiseExtractor simplified * Strategies use NoisePoint/channel API * NoiseExtractorTester fixes * PTSSamplingStrategyTester fixes Signed-off-by: Sachin Pisal <spisal@nvidia.com> * * Remove redundant NoiseExtractor tests * Add constants to make qubit/gate relationship explicit * Remove unused createParameterizedCircuit() helper Signed-off-by: Sachin Pisal <spisal@nvidia.com> --------- Signed-off-by: Sachin Pisal <spisal@nvidia.com>

#3881) Split PTSBESampler.h into public header (no nvqir dependencies) and internal PTSBESamplerImpl.h. Move template bodies to PTSBESampler.cpp with explicit instantiations for float and double. Remove unnecessary cudaq/simulators.h include from PTSBESampleIntegration.h. Signed-off-by: Thomas Alexander <talexander@nvidia.com>

* Adding trajectory deduplication logic based on kraus selection content Signed-off-by: Sachin Pisal <spisal@nvidia.com> * Adding unittests for trajectory deduplication Signed-off-by: Sachin Pisal <spisal@nvidia.com> * fix spelling Signed-off-by: Sachin Pisal <spisal@nvidia.com> * removing PTSSamplingStrategy Signed-off-by: Sachin Pisal <spisal@nvidia.com> * * Implement hashTrajectoryContent and deduplicateTrajectories keyed by kraus_selections * Representatives keep first prob/id/shots Signed-off-by: Sachin Pisal <spisal@nvidia.com> * fixing spelling Signed-off-by: Sachin Pisal <spisal@nvidia.com> * providing noise model to fix mcm test Signed-off-by: Sachin Pisal <spisal@nvidia.com> * removing duplicate documentation (already defined in PTSBEInterface.h) Signed-off-by: Sachin Pisal <spisal@nvidia.com> * Catching std::exception instead of runtime_error Signed-off-by: Sachin Pisal <spisal@nvidia.com> --------- Signed-off-by: Sachin Pisal <spisal@nvidia.com>

Signed-off-by: Sachin Pisal <spisal@nvidia.com>

* Adding shots allocations in the sample API Signed-off-by: Sachin Pisal <spisal@nvidia.com> * adding end-to-end test Signed-off-by: Sachin Pisal <spisal@nvidia.com> --------- Signed-off-by: Sachin Pisal <spisal@nvidia.com>

* * Adding probability validation to noise channels Signed-off-by: Sachin Pisal <spisal@nvidia.com> * validate in Python to avoid abort in JIT path Signed-off-by: Sachin Pisal <spisal@nvidia.com> * fix spelling and formatting Signed-off-by: Sachin Pisal <spisal@nvidia.com> * skip sample(kernel_invalid) in noise prob test to avoid JIT abort in CI Signed-off-by: Sachin Pisal <spisal@nvidia.com> --------- Signed-off-by: Sachin Pisal <spisal@nvidia.com>

Signed-off-by: Luca Mondada <luca@mondada.net>

* [python] Replace bindings for QIS types with stubs Signed-off-by: Luca Mondada <luca@mondada.net> * fix test Signed-off-by: Luca Mondada <luca@mondada.net> --------- Signed-off-by: Luca Mondada <luca@mondada.net>

* Add PTSBE trace. Signed-off-by: Thomas Alexander <talexander@nvidia.com> * Restructure/extend ptsbe options. Signed-off-by: Thomas Alexander <talexander@nvidia.com> * Add instruction tracing. Signed-off-by: Thomas Alexander <talexander@nvidia.com> * Add trace integration testing. Signed-off-by: Thomas Alexander <talexander@nvidia.com> * Add initial pybind interface for ptsbe. Signed-off-by: Thomas Alexander <talexander@nvidia.com> * Add python tests. Signed-off-by: Thomas Alexander <talexander@nvidia.com> * Fix spelling. Signed-off-by: Thomas Alexander <talexander@nvidia.com> * Spelling. Signed-off-by: Thomas Alexander <talexander@nvidia.com> * Spelling. Signed-off-by: Thomas Alexander <talexander@nvidia.com> * Some PR cleanup/documentation. Signed-off-by: Thomas Alexander <talexander@nvidia.com> * MOre review. Signed-off-by: Thomas Alexander <talexander@nvidia.com> * Spelling. Signed-off-by: Thomas Alexander <talexander@nvidia.com> * Remove tracing for now. Signed-off-by: Thomas Alexander <talexander@nvidia.com> * Wire up PTSBE. --------- Signed-off-by: Thomas Alexander <talexander@nvidia.com>

* More careful checking in SROA Signed-off-by: Adam Geller <adgeller@nvidiae.com> * Add test for issue Signed-off-by: Adam Geller <adgeller@nvidiae.com> * * Missing continue in outer loop * Added Python test Signed-off-by: Pradnya Khalate <pkhalate@nvidia.com> --------- Signed-off-by: Adam Geller <adgeller@nvidiae.com> Signed-off-by: Pradnya Khalate <pkhalate@nvidia.com> Co-authored-by: Adam Geller <adgeller@nvidiae.com> Co-authored-by: Pradnya Khalate <pkhalate@nvidia.com> Co-authored-by: Pradnya Khalate <148914294+khalatepradnya@users.noreply.github.com>

…::ptsbe::sample (#3914)  ### Description Part 1./2. of the successor to #3902. This PR decouples PTSBE from the standard Cuda-Q sampling infrastructure. We treat it as a completely separate execution pathway through Cuda-Q. This mirrors async/broadcasting support for cpp/python. This PR also fixed a bug in the generation of noisy unitary gates. Now we use the noise channel directly indexing into the corresponding operation. This allows us to work with any type of unitary noise (we should probably fix the enum for noise point to an integer in a follow up PR). --------- Signed-off-by: Thomas Alexander <talexander@nvidia.com> Co-authored-by: Sachin Pisal <spisal@nvidia.com>

### Description  There is a sort of circular dependency in `cudaq-python-interop`. It now has explicit dependencies on MLIR (via `ExecutionEngine`), which is available in `CUDAQuantumPythonSources.Extension`. However, `cudaq-python-interop` is a dependency of `CUDAQuantumPythonSources.Extension`. Hence, breaking this loop at `marshal_and_retain_module` in `py_alt_launch_kernel.cpp`, where we have access to the existing `JITExecutionCache` utility. The cache key is returned for clean-up purposes. --------- Signed-off-by: Thien Nguyen <thiennguyen@nvidia.com>

### Description As observed in #3910 `test_run_async` has infrequent issues in CI on macOS Python 3.12. I was unable to replicate locally despite trying a variety of mechanisms. This PR adds a flaky retry decorator to the test for macOS. I will try and run the specific CI test a number of times to see if I can trigger the retry (looking at the output messages). Perhaps a better fix would be to add an exponential backoff for command submission but that seems overkill for this limited CI failure right now. --------- Signed-off-by: Thomas Alexander <talexander@nvidia.com>

### Description  Disable test11 in test_list_of_dataclass_update_failures which is flaky on py3.11 + arm64 + cuda12.6. The struct argument marshaling was fixed in PR #3879, but kernel-internal deep copy and list mutation still produce intermittently wrong results on that configuration. ref: https://nvidia.slack.com/archives/C03KV6A81HT/p1770744080083629 I'll re-enable this test once I find the cause. Signed-off-by: huaweil <huaweil@nvidia.com> Co-authored-by: Mitchell <mitch_dz@hotmail.com>

…(cpp+python) (#3915)  ### Description Part 2./2. of #3902. This PR is based on #3914 and should be reviewed/rebased after. This PR adds support for returning execution data (instructions traces, trajectories, individual trajectory shot outcomes). I've put a bit of thought into future extension paths for dynamic circuits/observe which is documented as inline comments. Now we are able to run a demo like below end-to-end with PTSBE execution (ie., this is *mostly* feature complete for the PTSBE demo). ```python import cudaq @cudaq.kernel def bell(): q = cudaq.qvector(2) h(q[0]) cx(q[0], q[1]) mz(q) # Set up noise on the H gate noise = cudaq.NoiseModel() noise.add_all_qubit_channel("h", cudaq.DepolarizationChannel(0.01)) # Sample with execution data result = cudaq.ptsbe.sample( bell, shots_count=1000, noise_model=noise, max_trajectories=50, return_execution_data=True) # Standard sample_result access print(result) # { 00:517 11:483 } # Access execution data data = result.ptsbe_execution_data # Inspect the circuit structure for inst in data.instructions: print(f"{inst.type}: {inst.name} on qubits {inst.targets}") # Gate: h on qubits [0] # Noise: depolarization_channel on qubits [0] # Gate: x on qubits [1] # Measurement: mz on qubits [0] # Measurement: mz on qubits [1] # Count by type print(data.count_instructions(cudaq.ptsbe.TraceInstructionType.Gate)) # 2 print(data.count_instructions(cudaq.ptsbe.TraceInstructionType.Noise)) # 1 # Per-trajectory results for traj in data.trajectories: print(f"Trajectory {traj.trajectory_id}: " f"p={traj.probability:.4f}, " f"shots={traj.num_shots}, " f"counts={dict(traj.measurement_counts)}") ``` --------- Signed-off-by: Adam Geller <adgeller@nvidiae.com> Signed-off-by: Pradnya Khalate <pkhalate@nvidia.com> Signed-off-by: Thomas Alexander <talexander@nvidia.com> Signed-off-by: Mitchell <mitch_dz@hotmail.com> Signed-off-by: Thien Nguyen <thiennguyen@nvidia.com> Signed-off-by: Luca Mondada <luca@mondada.net> Signed-off-by: Luca Mondada <72734770+lmondada@users.noreply.github.com> Co-authored-by: Adam T. Geller <adgeller@nvidia.com> Co-authored-by: Adam Geller <adgeller@nvidiae.com> Co-authored-by: Pradnya Khalate <pkhalate@nvidia.com> Co-authored-by: Pradnya Khalate <148914294+khalatepradnya@users.noreply.github.com> Co-authored-by: Mitchell <mitch_dz@hotmail.com> Co-authored-by: Sachin Pisal <spisal@nvidia.com> Co-authored-by: Thien Nguyen <58006629+1tnguyen@users.noreply.github.com> Co-authored-by: Luca Mondada <72734770+lmondada@users.noreply.github.com>

### Description  As the NVQIR simulator is retrieved as a base class pointer in the runtime, it seems like dynamic casting is the solution for simulator implementations to provide `sampleWithPTSBE`. Signed-off-by: Thien Nguyen <thiennguyen@nvidia.com>

# Bump actions/checkout reusable workflow version Bumps `actions/checkout@v4` -> `actions/checkout@v6` Release notes are located at https://github.com/actions/checkout?tab=readme-ov-file#whats-new The main update to target this is the security posture: * Improved credential security: persist-credentials now stores credentials in a separate file under $RUNNER_TEMP instead of directly in .git/config Signed-off-by: Mitchell <mitch_dz@hotmail.com>

@flaky

### Description  Signed-off-by: Thien Nguyen <thiennguyen@nvidia.com>

github-actions · 2026-02-25T07:00:59Z

CUDA Quantum Docs Bot: A preview of the documentation can be found here.

lmondada

I'll keep reviewing. So far I've mostly been going through the tests. It's looking very clean and nice thank you!

My only major comment so far is: you need to be careful when defining kernels in-line for testing. These kernels do not go through the standard cudaq compilation pipeline, thus you are not actually testing the behaviour of your code on "real" kernels. In theory, it could be possible that the kernels you wrote get compiled to something totally different by nvq++ that your tracing would miss (in practice, behaviour should be broadly similar, given that we support "library mode"; however, the execution path in the runtime is still different in subtle ways).

I think you should add at least a couple of tests that make sure that kernels compiled through nvq++ still get traced properly by capturePTSBatch.

Whilst I'm on it: consider renaming capturePTSBatch to tracePTSBatch. I am just coming from Python land where we've been doing a lot of work on "capturing" variables, so it is confusing to me.

python/cudaq/runtime/ptsbe.py

python/cudaq/__init__.py

unittests/ptsbe/ExecutePTSBETester.cpp

unittests/ptsbe/KrausSelectionTester.cpp

unittests/ptsbe/PTSBEMultiBackendTester.cpp

unittests/ptsbe/PTSBESampleTester.cpp

unittests/ptsbe/ExecutePTSBETester.cpp

docs/sphinx/examples/python/ptsbe_end_to_end_workflow.ipynb

python/cudaq/runtime/ptsbe.py

python/cudaq/runtime/sample.py

python/runtime/cudaq/algorithms/py_sample_ptsbe.cpp

runtime/common/NoiseModel.cpp

runtime/common/NoiseModel.h

runtime/common/Trace.cpp

runtime/common/Trace.h

runtime/cudaq/ptsbe/strategies/ConditionalSamplingStrategy.cpp

runtime/cudaq/ptsbe/strategies/ConditionalSamplingStrategy.h

runtime/cudaq/ptsbe/strategies/ExhaustiveSamplingStrategy.cpp

### Description Fixes for PTSBE identified in my [second AI-assisted review](#3995 (review)) of the feature branch. Changes: - Update flaky statistical test - PTSBESample.h: runSamplingAsyncPTSBE copies noise model into async lambda, wraps execution in try/catch with set_exception, both sample_async overloads call platform.reset_noise() after enqueue - PTSBESample.h: runSamplingPTSBE builds PTSBE trace once and shares it between execution data and batch construction, removed buildPTSBatchWithTrajectories - PTSBESample.cpp: replaced buildPTSBatchWithTrajectories with buildPTSBatchFromTrace that accepts pre-built trace, removed stub trajectory block in populateExecutionDataTrajectories - py_sample_ptsbe.cpp: pySampleAsyncPTSBE calls platform.reset_noise() after enqueue, added qubits and op_name bindings to KrausSelection - SampleResult.h: made ~sample_result() virtual - PTSSamplingStrategy.h: computeTotalTrajectories skips empty channels instead of dividing by zero - OrderedSamplingStrategy.cpp: sorts operator indices by descending probability before enumeration, delegates to shared enumerateLexicographic - ExhaustiveSamplingStrategy.h: declared enumerateLexicographic free function with optional index mapping - ExhaustiveSamplingStrategy.cpp: implemented shared enumeration loop, ExhaustiveSamplingStrategy delegates to it - ShotAllocationStrategy.cpp: allocateShots zeroes num_shots before dispatching - TrajectoryDeduplication.cpp: accumulates num_shots when merging duplicates - PTSBESampleTester.cpp: added SampleAsyncNoiseModelDestroyed and SampleAsyncPropagatesException tests - PTSSamplingStrategyTester.cpp: added TopKWithNonZeroIdentityIndex and ComputeTotalTrajectoriesSkipsEmptyChannel tests - ShotAllocationTester.cpp: added DoubleAllocationDoesNotAccumulate test - TrajectoryDeduplicationTester.cpp: updated SameContentDifferentIdAndShots to expect accumulated num_shots - test_execution_data.py: augmented test_trajectory_measurement_counts_populated to check qubits and op_name --------- Signed-off-by: Thomas Alexander <talexander@nvidia.com>

taalexander

With the merger of #4043 I have addressed my merge blocking concerns for this PR. All other comments of mine are to be done in follow-up PRs once landing on main.

1tnguyen

Approved changes to .github config

This PR introduces "deferred compilation" in Python. By default, kernels are no longer compiled to MLIR (aka pre-compiled) at kernel definition time, but only when the kernel is invoked for the first time. As before, the MLIR module is then cached for further invocation. For example, ```python import cudaq @cudaq.kernel def foo(): pass ``` would not trigger any compilation, until `foo()` is called for the first time. This enables: 1. Faster package load times: `import cudaq` takes ~4s now, versus `~8s` before, as the kernels included with `cudaq` are no longer compiled when the package is loaded. This will also apply to any other library that provides kernel definitions. 2. Kernels can be defined out-of-order: previously, calling kernel B from kernel A would error if B was defined after A. This is now supported (see tests). 3. Captured variables that aren't defined at kernel definition time but are defined at invocation time are now supported. Compilation can be forced (thus recovering the old behaviour) using the `defer_compilation=False` flag. The following: ```python @cudaq.kernel(defer_compilation=False) def foo(): pass ``` triggers compilation immediately. ## Limitations To limit the scope of this PR, I introduced two limitations (see tests for examples): 1. Recursive kernel calls are still not supported. The way kernels are captured and lifted into arguments would have to handle recursive calls specially. 2. A kernel builder A cannot call another kernel B using `apply_call(B)` if B wasn't compiled beforehand. In that case, the user will receive a clear error suggesting to either use `defer_compilation=False` or calling `B.compile()` directly, e.g. ```python @cudaq.kernel(defer_compilation=True) def notPrecompiledKernel(): pass kernel = cudaq.make_kernel() kernel.apply_call(notPrecompiledKernel) # fails. User must set `defer_compilation=False` # or call `notPrecompiledKernel.compile()` ``` --------- Signed-off-by: Luca Mondada <luca@mondada.net>

…n` (#4017) This PR adds a migration guide for transitioning CUDA-Q quantum kernels from the `sample` API to the `run` API, following a breaking change in PR #3731. The update clarifies when and how to use each API, provides migration steps and examples in both Python and C++, and adds supporting code samples and references in the documentation. --------- Signed-off-by: Pradnya Khalate <pkhalate@nvidia.com>

github-actions · 2026-02-26T01:29:08Z

CUDA Quantum Docs Bot: A preview of the documentation can be found here.

github-actions · 2026-02-26T10:57:39Z

CUDA Quantum Docs Bot: A preview of the documentation can be found here.

lmondada

Thanks for the great work!!

github-actions · 2026-02-26T18:54:05Z

CUDA Quantum Docs Bot: A preview of the documentation can be found here.

sacpis and others added 30 commits February 7, 2026 16:36

Fix rebasing issues surrounding context handling updates for PTSBE.

e11d51a

Signed-off-by: Thomas Alexander <talexander@nvidia.com>

Fix mcm_rejection: catch (...) for dynamic circuit rejection

5ac3373

Signed-off-by: Sachin Pisal <spisal@nvidia.com>

* Accept any exception type in throw to fix CI jobs

63d7fb4

Signed-off-by: Sachin Pisal <spisal@nvidia.com>

Merge remote-tracking branch 'upstream/main' into features/ptsbe

ad170bb

[PTSBE] Allocate shots to trajectories (#3899)

3be6898

* Adding shots allocations in the sample API Signed-off-by: Sachin Pisal <spisal@nvidia.com> * adding end-to-end test Signed-off-by: Sachin Pisal <spisal@nvidia.com> --------- Signed-off-by: Sachin Pisal <spisal@nvidia.com>

[python] Remove GlobalAstRegistry (#3896)

a07423c

Signed-off-by: Luca Mondada <luca@mondada.net>

[python] Replace bindings for QIS types with stubs (#3882)

846fec6

* [python] Replace bindings for QIS types with stubs Signed-off-by: Luca Mondada <luca@mondada.net> * fix test Signed-off-by: Luca Mondada <luca@mondada.net> --------- Signed-off-by: Luca Mondada <luca@mondada.net>

Merge remote-tracking branch 'upstream/main' into features/ptsbe

4e244b6

Merge remote-tracking branch 'upstream/main' into features/ptsbe

443c7d4

Merge remote-tracking branch 'upstream/main' into features/ptsbe

7ee202f

sacpis added 2 commits February 24, 2026 21:04

Merge remote-tracking branch 'upstream/main' into features/ptsbe

ff279ab

Merge remote-tracking branch 'upstream/main' into features/ptsbe

183a81d

lmondada reviewed Feb 25, 2026

View reviewed changes

taalexander mentioned this pull request Feb 25, 2026

[PTSBE] Core code fixes to merge #4043

Merged