Pasqal Cloud QRMI - CUDA-Q Examples#

GitHub

Prerequisites#

  • Rust 1.85.1 or above

  • Python 3.11 or 3.12

  • Install the QRMI Python package

  • CUDA-Q installed with the Pasqal backend built

Install dependencies#

source ~/py311_qrmi_venv/bin/activate
pip install -r ../requirements.txt
pip install cudaq

Set environment variables#

QRMI supports Pasqal Cloud configuration via environment variables. For Pasqal Cloud auth, QRMI also supports reading ~/.pasqal/config (token or username/password). PASQAL_CONFIG_ROOT may point elsewhere and takes priority over <backend_name>_PASQAL_CONFIG_ROOT; QRMI expands ~, $VAR, and ${VAR} before appending .pasqal/config.

The required environment variables are listed below. They are populated automatically by the spank plugin.

Environment variables

Descriptions

<backend_name>_QRMI_PASQAL_CLOUD_PROJECT_ID

Pasqal Cloud Project ID to access the QPU

<backend_name>_QRMI_PASQAL_CLOUD_AUTH_TOKEN

Pasqal Cloud Auth Token

<backend_name>_QRMI_PASQAL_CLOUD_CLIENT_ID

Pasqal Cloud service account client ID (optional)

<backend_name>_QRMI_PASQAL_CLOUD_CLIENT_SECRET

Pasqal Cloud service account client secret (optional)

<backend_name>_QRMI_PASQAL_CLOUD_AUTH_ENDPOINT

(Optional) Auth endpoint URL/path for token retrieval. Default: authenticate.pasqal.cloud/oauth/token

~/.pasqal/config (optional)#

Create ~/.pasqal/config:

config#
username=<your username>
password=<your password>
# or:
# token=<your token>
# or:
# client_id=<your client id>
# client_secret=<your client secret>  # pragma: allowlist secret

# optional override:
# project_id=<your project id>
# auth_endpoint=<auth endpoint URL/path>

Using this backend from CUDA-Q (pasqal)#

When CUDA-Q is configured with target pasqal and machine in cudaq.set_target(..., machine=...), it should match qrmi. This way it picks up the machine target from QRMI, as populated by (for example) the SPANK plugin’s --qpu argument, or manually set by QRMI_JOB_QPU_RESOURCES.

In pasqal.py:

1import cudaq
2cudaq.set_target("pasqal", machine="qrmi")

See the CUDA-Q docs to see how to send a C++ job. To use QRMI, simply set the target and machine as above.

How to run this example#

All information is baked into the Python script. Run pasqal.py:

python pasqal.py

Build from source#

For up-to-date information on how to build the latest version, we suggest you follow CUDA-Q’s official build docs and scripts.

We assume Slurm containers have been set up as per the spank-plugin development documentation and the CUDA-Q repository has been cloned into /shared.

# 1) Rebuild QRMI
cd /shared/qrmi
cargo build --release --lib
# For pasqal-local support
cargo build --release --lib --features munge

# 2) Build CUDA-Q with local QRMI
cd /shared/cuda-quantum
QRMI_INSTALL_PREFIX=/shared/qrmi bash scripts/build_cudaq.sh -p -i -j nproc

# 3) Install CUDA-Q Python package (non-editable!)
source /shared/pyenv/bin/activate
pip uninstall -y cuda-quantum-cu13 || true
pip install --no-build-isolation /shared/cuda-quantum

Do not use an editable install for CUDA-Q in this workspace (pip install -e .) as it further requires manually specifying paths to get a working environment.

The CUDA-Q build configuration used during development was as follows:

dnf install epel-release
dnf makecache
dnf install ccache
source /shared/pyenv/bin/activate && cd /shared/cuda-quantum
PATH=/opt/llvm/bin:$PATH Python3_EXECUTABLE=/shared/pyenv/bin/python ./scripts/install_prerequisites.sh -e "aws;qrmi"
PATH=/opt/llvm/bin:$PATH Python3_EXECUTABLE=/shared/pyenv/bin/python QRMI_INSTALL_PREFIX=/shared/qrmi CUDAQ_BUILD_TESTS=FALSE CUDAQ_WERROR=OFF ./scripts/build_cudaq.sh -j nproc -- -DCUDAQ_ENABLE_PASQAL_QRMI_CONNECTOR=ON -DCUDAQ_ENABLE_BRAKET_BACKEND=OFF -DCUDAQ_ENABLE_QCI_BACKEND=OFF -DCUDAQ_ENABLE_QUANTUM_MACHINES_BACKEND=OFF

Troubleshooting#

To be sure that CUDA-Q is detected and is using the QRMI library that you just built, checkout the QRMI_LIBRARY var in cuda-quantum/build/CMakeCache.txt. By default, that QRMI library build is located in qrmi/target/release/libqrmi.so, so you can copy it to where QRMI_LIBRARY is pointing if there is a mismatch.