Skip to content
Quantum.world

The Signal

Pasqal tests HRES scheduling for hybrid QPU jobs

Pasqal describes linking its quantum processors to Slurm’s HRES scheduler through IBM-developed QRMI integration. The company reports tests with an emulator, a simulated on-premises setup and a physical QPU, while noting that Slurm 26.11 or later is required for HRES.

Pasqal tests HRES scheduling for hybrid QPU jobs

Quantum processors as scheduled resources

Slurm 26.11 can treat Pasqal quantum processors as resources assigned to jobs, enabling hybrid workloads to request a QPU through the scheduler, according to a Pasqal technical post published Nov. 16. The integration connects Slurm’s hierarchical resources (HRES) plugin interface to the Quantum Resource Management Interface (QRMI); the HRES–QRMI integration was developed by IBM and Pasqal, the post says.

In the described arrangement, Slurm checks whether a QPU is available and limits how many jobs can use it at once. For an on-premises processor, the scheduler can open a device-access session when a job starts and close it when the job ends or is canceled; if the processor is unavailable, jobs wait rather than start and fail.

How the integration works

HPC users request resources in batch scripts, while administrators define the QPU in Slurm’s configuration. The HRES–QRMI plugin uses QRMI to check device status, acquire and release access, submit quantum tasks and retrieve results. Pasqal says the same integration can serve its QPUs and IBM QPUs.

The change from the earlier QRMI SPANK-plugin method is in the job request: the example switches from a QPU request paired with a site license to a Slurm resource request. Programs written with Pulser or NVIDIA CUDA-Q can keep the same environment, the post says. Slurm 26.11 or later is required for HRES; Pasqal identifies the QRMI SPANK plugin as the production route currently in use and says sites can move to HRES as it matures.

Tests span emulation and hardware

Pasqal says it ran Pulser and CUDA-Q jobs through Slurm with HRES and QRMI against its EMU_FREE cloud emulator and a simulated on-premises deployment. All four combinations passed, and tests of shared access included queued competing jobs, release of access after cancellation and jobs waiting while Warden—the service managing access to an on-premises QPU—was stopped.

The company also says it ran the workflow on FRESNEL_CAN1, a physical Pasqal QPU in Canada accessed through Pasqal Cloud. The sample output the post prints comes from a Pasqal Cloud job assigned the EMU_FREE emulator: two atoms spaced 5 µm apart, 500 shots, P(00)=0.002, P(01)=0.496, P(10)=0.502 and P(11)=0.000.

GPU-to-QPU workflow

Pasqal describes a batch sequence in which a program first runs on a GPU for emulation, then runs unchanged on a QPU only if the first job succeeds; a final job compares the results. The emulation job requests a GPU but no QPU, and the QPU job requests the processor but no GPU. On its test cluster, the GPU and QPU results agreed within the statistical spread at each of six atom distances, according to the company.

Pasqal says HPC sites running Slurm 26.11 can try the HRES–QRMI integration with Pasqal Cloud or an on-premises Pasqal QPU. The company’s companion example includes configuration, programs, tests and the GPU-to-QPU workflow.

Source

  1. Pasqal

Companies in this story

Related stories

Explore

More articles