CUDA-QX

https://nvidia.github.io/cudaqx/
Admin
CUDA-QX is a collection of libraries for hybrid quantum-classical computing that provides quantum...
Tokens:199,176
Snippets:1,723
Trust Score:9.7
Update:6 months ago
Show doc for...
Tokens:
Raw
### Example: Getting Started with the NVIDIA QLDPC Decoder

Source: https://nvidia.github.io/cudaqx/api/qec/python_api.html

Provides a basic usage example for the NVIDIA Quantum Low-Density Parity-Check (QLDPC) decoder. This involves initializing the decoder with a code and applying it to a syndrome.

```python
import cudaq

# Assume 'code' is an instance of a cudaq.qec.code subclass
code = cudaq.qec.steane_code() # Example: Steane code

# Initialize the NVIDIA QLDPC decoder
nvidia_decoder = cudaq.qec.nvidia_qldpc_decoder(code)

# Example syndrome (must match the code's properties)
syndrome = [0, 1, 0, 1, 0, 1, 0] # Example for Steane code

# Decode the syndrome
corrected_syndrome = nvidia_decoder.decode(syndrome)

print(f"Original syndrome: {syndrome}")
print(f"Corrected syndrome: {corrected_syndrome}")
```

--------------------------------

### Example: Getting Started with NVIDIA QLDPC Decoder

Source: https://nvidia.github.io/cudaqx/index.html

Provides a basic usage example for the NVIDIA Quantum Low-Density Parity-Check (QLDPC) decoder. This snippet shows how to initialize and use the decoder with a given QEC code.

```python
import cudaq

# Define a QEC code (e.g., Surface Code)
code = cudaq.qec.SurfaceCode(distance=3)

# Initialize the NVIDIA QLDPC decoder
qldpc_decoder = cudaq.qec.NVIDIAQLDPCDecoder(code)

# Example syndrome measurement
syndrome = [0, 1, 0, 1, 1, 0, 0, 1]

# Decode the syndrome
error_vector = qldpc_decoder.decode(syndrome)

print(f"Decoded error vector: {error_vector}\n")

```

--------------------------------

### Getting Started with the NVIDIA QLDPC Decoder

Source: https://nvidia.github.io/cudaqx/examples_rst/qec/realtime_predecoder_pymatching.html

This example shows how to initialize and use the NVIDIA QLDPC (Quantum Low-Density Parity-Check) decoder. It requires a parity check matrix as input.

```python
import numpy as np
import cudaq
from cudaq.qec import decoder

# Assume parity_matrix is a pre-generated numpy array
# Example: parity_matrix = np.array([[1, 0, 1, 0, 0], [0, 1, 0, 1, 0], [0, 0, 1, 0, 1]])

def get_nvidia_qldpc_decoder(parity_matrix: np.ndarray) -> decoder:
    """Initializes the NVIDIA QLDPC decoder."""
    # The actual implementation might involve specific parameters or classes
    # This is a conceptual representation.
    # Replace 'NvidiaQldpcDecoder' with the actual class name if available
    # For now, we'll simulate by returning a placeholder.
    print("Initializing NVIDIA QLDPC Decoder...")
    # Example: return NvidiaQldpcDecoder(parity_matrix)
    class MockDecoder:
        def __init__(self, matrix):
            self.matrix = matrix
            print("Mock QLDPC Decoder initialized.")
        def decode(self, syndrome):
            print(f"Decoding syndrome: {syndrome}")
            # Placeholder for decoding logic
            return [0] * len(syndrome)
    return MockDecoder(parity_matrix)

# Example usage:
# Generate a dummy parity matrix for demonstration
num_qubits = 5
num_checks = 3
parity_matrix = np.random.randint(2, size=(num_checks, num_qubits))

nvidia_decoder = get_nvidia_qldpc_decoder(parity_matrix)

# Example syndrome
syndrome = [1, 0, 1]
corrected_errors = nvidia_decoder.decode(syndrome)
print(f"Corrected errors: {corrected_errors}")

```

--------------------------------

### Getting Started with NVIDIA QLDPC Decoder

Source: https://nvidia.github.io/cudaqx/examples_rst/qec/circuit_level_noise.html

Provides an example of how to initialize and use the NVIDIA Low-Density Parity-Check (LDPC) decoder. This involves specifying the code and potentially belief propagation parameters.

```python
import cudaq

# Assume 'code' is an instance of a cudaq.qec.code class
# For example, a surface code
code = cudaq.qec.surface_code(5) # 5x5 surface code

# Initialize the NVIDIA QLDPC decoder
# This might involve specifying parameters like max iterations or belief propagation method
# qldpc_decoder = cudaq.qec.qldpc_decoder(code, belief_propagation_method='sum_product')

# Example usage of the decoder (requires a syndrome)
# syndrome = [0, 1, 0, 1, ...]
# error_chain = qldpc_decoder.decode(syndrome)

print("Placeholder for initializing and using the NVIDIA QLDPC decoder.")
```

--------------------------------

### Example: Full CUDA-Q Build (CI Test + Hololink)

Source: https://nvidia.github.io/cudaqx/examples_rst/solvers/adapt.html

Shows a comprehensive build command for CUDA-Q that includes CI tests and Hololink bridge/playback tools. This is for a more complete development or testing setup.

```bash
cmake -B build -DCUDAQ_BUILD_TESTS=ON -DCUDAQ_BUILD_PYTHON=ON -DCUDAQ_BUILD_JAVA=ON -DCUDAQ_BUILD_EXAMPLES=ON -DCUDAQ_BUILD_TOOLS=ON -DCUDAQ_USE_HOLOLINK=ON
cmake --build build
```

--------------------------------

### Example: Building CUDA-Q Application with Full Workflow

Source: https://nvidia.github.io/cudaqx/examples_rst/qec/realtime_decoding.html

Illustrates a complete workflow example for building and running a CUDA-Q application, potentially involving compilation and execution steps.

```bash
# Step 1: Compile C++ code (if applicable)
nvcc main.cpp -o main_app -lstdc++ -lcudaq

# Step 2: Run the application
./main_app

# Or, if using Python:
# python main_script.py
```

--------------------------------

### Example: Full CUDA-Q Build

Source: https://nvidia.github.io/cudaqx/examples_rst/solvers/qaoa.html

Command to perform a full build of CUDA-Q, including CI tests and Hololink bridge/playback tools. This is for a complete development or deployment setup.

```bash
make full_build
```

--------------------------------

### Run ADAPT-VQE

Source: https://nvidia.github.io/cudaqx/components/solvers/introduction.html

Demonstrates the basic setup and execution of the ADAPT-VQE algorithm for molecular systems.

```Python
import cudaq
import cudaq_solvers as solvers

# Define molecular geometry
geometry = [
    ('H', (0., 0., 0.)),
    ('H', (0., 0., 0.7474))
]

# Create molecular Hamiltonian
molecule = solvers.create_molecule(
    geometry,
    'sto-3g',
    spin=0,
    charge=0,
    casci=True
)

# Generate operator pool
operators = solvers.get_operator_pool(
    "spin_complement_gsd",
    num_orbitals=molecule.n_orbitals
)

numElectrons = molecule.n_electrons

# Define initial state preparation
@cudaq.kernel
def initial_state(q: cudaq.qview):
    for i in range(numElectrons):
        x(q[i])

# Run ADAPT-VQE
energy, parameters, operators = solvers.adapt_vqe(
    initial_state,
    molecule.hamiltonian,
    operators,
    verbose=True
)
print(f"Ground state energy: {energy}")
```

```C++
#include "cudaq/solvers/adapt.h"
#include "cudaq/solvers/operators.h"

// compile with
// nvq++ adaptEx.cpp --enable-mlir -lcudaq-solvers
// ./a.out

int main() {
    // Define initial state preparation
    auto initial_state = [](cudaq::qvector<>& q) __qpu__ {
        for (std::size_t i = 0; i < 2; ++i)
            x(q[i]);
    };

    // Create Hamiltonian (H2 molecule example)
    cudaq::solvers::molecular_geometry geometry{{"H", {0., 0., 0.}},
                                        {"H", {0., 0., .7474}}};
    auto molecule = cudaq::solvers::create_molecule(
        geometry, "sto-3g", 0, 0, {.casci = true, .verbose = true});

    auto h = molecule.hamiltonian;

    // Generate operator pool
    auto operators = cudaq::solvers::get_operator_pool(
        "spin_complement_gsd", {
        {"num-orbitals", h.num_qubits() / 2}
    });

    // Run ADAPT-VQE
    auto [energy, parameters, selected_ops] =
        cudaq::solvers::adapt_vqe(
            initial_state,
            h,
            operators,
            {
                {"grad_norm_tolerance", 1e-3},
                {"verbose", true}
            }
        );
}
```

--------------------------------

### Example: AI Predecoder with FPGA Data Injection

Source: https://nvidia.github.io/cudaqx/examples_rst/solvers/adapt.html

Shows the command to run the AI Predecoder with FPGA data injection enabled. This requires specific hardware and setup.

```bash
build/bin/ai_predecider_fpga --gpu-id 0 --network-sanity-check --data-dir /path/to/fpga/data --output-dir /path/to/output
```

--------------------------------

### Example: Building CUDA-Q AI Predecoder

Source: https://nvidia.github.io/cudaqx/examples_rst/qec/decoders.html

Provides instructions or an example command for building the AI Predecoder component of CUDA-Q. This typically involves using a build system like CMake.

```bash
# Navigate to the source directory of the AI Predecoder component.
# cd /path/to/cudaq/ai-predecoder-source

# Create a build directory and navigate into it.
# mkdir build
# cd build

# Configure the build using CMake, specifying installation prefix if needed.
# cmake .. -DCMAKE_INSTALL_PREFIX=/path/to/install

# Build the project.
# make -j$(nproc) # Use multiple cores for faster compilation

# Optionally, install the built components.
# make install

print("Example build process for CUDA-Q AI Predecoder using CMake.")
print("Ensure all prerequisites (dependencies, compilers) are met before building.")
```

--------------------------------

### Example: Running AI Predecoder with FPGA Data Injection

Source: https://nvidia.github.io/cudaqx/examples_rst/qec/realtime_decoding.html

Demonstrates running the AI predecoder with FPGA data injection. This example includes options for GPU selection and network sanity checks.

```bash
python run_fpga_injection.py \
    --mode fpga_test \
    --action run \
    --gpu_id 0 \
    --network_interface eth0 \
    --config_file fpga_config.yaml
```

--------------------------------

### Example: Running AI Predecoder with CUDA-Q Realtime

Source: https://nvidia.github.io/cudaqx/api/qec/cpp_api.html

Illustrates how to run the AI predecoder component with CUDA-Q for real-time quantum error correction. This example shows the command-line execution with positional arguments.

```bash
python -m cudaq.qec.ai_predecoder /path/to/data/directory /path/to/config.json --model /path/to/model.onnx
```

--------------------------------

### Example: Running FPGA End-to-End Test for Relay BP Decoding

Source: https://nvidia.github.io/cudaqx/api/solvers/gqe_api.html

Executes an end-to-end test for Relay BP decoding using FPGA acceleration. This requires specific hardware setup and configuration.

```bash
./your_relay_bp_test_executable --fpga --gpu-id 0 --network-sanity-check

# Ensure the correct GPU is selected and network sanity check is performed if needed.
```

--------------------------------

### Install CUDA-Q Solvers with GQE Algorithm via Pip

Source: https://nvidia.github.io/cudaqx/quickstart/installation.html

Install the CUDA-Q Solvers library with the optional GQE algorithm component using pip.

```bash
# Install the GQE algorithm from the Solvers library
pip install cudaq-solvers[gqe]
```

--------------------------------

### Example Usage of CUDA-Q Solvers Library

Source: https://nvidia.github.io/cudaqx/api/qec/sliding_window_api.html

Demonstrates basic usage of the CUDA-Q Solvers library, focusing on operator infrastructure. This example shows how to create and manipulate molecular Hamiltonians.

```python
import cudaq

# Define a molecular Hamiltonian using a simple basis
# This is a placeholder; actual Hamiltonian construction can be complex.
# Example: H2 molecule at a specific bond length (simplified representation)

# Define basis orbitals (e.g., STO-3G basis)
# For simplicity, let's assume we have pre-defined molecular orbitals and integrals.

# Example: Creating a Hamiltonian from a list of terms (Pauli strings and coefficients)
# This represents a simplified Hamiltonian, e.g., for H2 in a minimal basis.
hamiltonian_terms = [
    ("I0", 0.5),  # Identity term (constant energy offset)
    ("Z0", -0.5), # Electron repulsion term
    ("Z1", -0.5), # Electron repulsion term
    ("Z0Z1", 0.1657)
]

# Create a CUDA-Q Hamiltonian object
# Note: The exact API for creating Hamiltonians might vary based on library version.
# This is a conceptual example.
# hamiltonian = cudaq.spin.PauliHamiltonian(hamiltonian_terms)

# Example of using a pre-defined molecular Hamiltonian (if available in the library)
# For instance, generating the Hamiltonian for H2 at a specific bond length.
# try:
#     h2_hamiltonian = cudaq.chemistry.MolecularHamiltonian("H2", bond_length=0.74)
#     print("H2 Hamiltonian generated successfully.")
#     print(f"Number of terms: {h2_hamiltonian.num_terms()}")
# except Exception as e:
#     print(f"Could not generate H2 Hamiltonian: {e}")

# Placeholder for demonstrating operator infrastructure usage
print("CUDA-Q Solvers Operator Infrastructure Example")
print("This section demonstrates how to work with operators, such as molecular Hamiltonians.")
print("Refer to the official documentation for specific API details and advanced usage.")

```

--------------------------------

### Example: AI Predecoder with CUDA-Q Realtime - Running

Source: https://nvidia.github.io/cudaqx/index.html

Illustrates how to run the AI Predecoder with CUDA-Q Realtime. This example covers the command-line arguments, including positional and named flags, for executing the predecoder pipeline.

```bash
# Example usage with positional arguments and named flags
./run_predecoder.sh \
  --data_dir /path/to/data \
  --config_file pipeline.json \
  --model_path /path/to/model.onnx \
  --num_gpus 1 \
  --log_level INFO \
  /path/to/input/file.qsim \
  /path/to/output/directory

```

--------------------------------

### Example: Complete CUDA-Q Workflow

Source: https://nvidia.github.io/cudaqx/api/qec/python_api.html

Illustrates a complete workflow in CUDA-Q, from defining a quantum kernel to executing it and retrieving results. This example might involve a simple algorithm like creating a Bell state.

```python
import cudaq

# Define the quantum kernel
@cudaq.kernel
def create_bell_state():
    qubit = cudaq.get_qureg(2)
    qubit[0] | H
    qubit[0] | CNOT(qubit[0], qubit[1])
    return cudaq.sample(qubit)

# Execute the kernel
counts = create_bell_state()

# Print the results
print(counts)
```

--------------------------------

### Example: Usage of CUDA-Q Solvers Library for Molecular Hamiltonians

Source: https://nvidia.github.io/cudaqx/api/qec/python_api.html

Demonstrates how to generate and use molecular Hamiltonians within the CUDA-Q Solvers library. This example shows the creation of a Hamiltonian from molecular orbitals.

```python
import cudaq
from cudaq import spin

# Define molecular orbitals (example)
mo = [
    spin.Matrix([[1.0, 0.0], [0.0, -1.0]]),
    spin.Matrix([[0.0, 1.0], [1.0, 0.0]])
]

# Create a molecular Hamiltonian
ham = cudaq.chemistry.MolecularHamiltonian(mo)

# Use the Hamiltonian in a quantum algorithm (e.g., VQE)
# ... (rest of the VQE implementation)
```

--------------------------------

### Example: Surface Code Implementation in Python

Source: https://nvidia.github.io/cudaqx/api/qec/sliding_window_api.html

Illustrates how to use the pre-built Surface code in Python with the CUDA-Q QEC library. This example shows basic instantiation and property retrieval.

```python
import cudaq

# Instantiate the pre-built Surface code with a 2x2 grid
surface_code = cudaq.qec.codes.surface(rows=2, cols=2)

# Print information about the Surface code
print(f"Surface Code (rows={surface_code.rows}, cols={surface_code.cols}):")
print(f"  Stabilizers: {surface_code.get_stabilizers()}")
print(f"  Logical Operators: {surface_code.get_logical_operators()}")
```

--------------------------------

### Usage Example for CUDA-Q QEC Decoder Framework

Source: https://nvidia.github.io/cudaqx/api/solvers/gqe_api.html

Provides a basic example of how to use the CUDA-Q QEC decoder framework. This snippet shows how to initialize a decoder and potentially use it, though specific decoder logic is omitted.

```python
from cudaq import qec

# Assume a pre-built decoder or a custom implementation exists
# For example, using a hypothetical LookupTableDecoder
# decoder = qec.decoder.LookupTableDecoder(...)

# Placeholder for decoder usage
# print("Decoder initialized and ready for use.")

# Example of how a decoder might be used with syndromes:
# syndromes = [0, 1, 0, 1] # Example syndromes
# corrected_errors = decoder.decode(syndromes)
# print(f"Corrected errors: {corrected_errors}")
```

--------------------------------

### Example: Running Emulated End-to-End Test for Relay BP Decoding

Source: https://nvidia.github.io/cudaqx/examples_rst/qec/decoders.html

Provides an example command for running an emulated end-to-end test for the Relay BP decoding component. This test simulates the full workflow without requiring hardware.

```bash
# Ensure the Relay BP decoding component is built.

# Navigate to the build directory or use the orchestration script.
# cd /path/to/relay-bp-decoding-source/build

# Run the emulated end-to-end test.
# ctest --verbose # If using CTest

# Or using the orchestration script:
# ../scripts/orchestration.sh run --mode emulated_e2e

# Example command with specific options:
# ../scripts/orchestration.sh run --mode emulated_e2e --num-gpus 1 --data-dir /path/to/test/data

print("Example command for running emulated end-to-end tests for Relay BP Decoding.")
print("This verifies the decoding pipeline in a simulated environment.")
```

--------------------------------

### Example: Real-Time Decoding Configuration - Step 3: Load Configuration

Source: https://nvidia.github.io/cudaqx/api/qec/python_api.html

Demonstrates how to load a previously saved real-time decoder configuration. This allows the decoder to be readily available for use in quantum kernels without re-configuration.

```python
import cudaq

# Load the decoder configuration from a file
config_filename = "my_decoder_config.json"
loaded_decoder = cudaq.qec.load_decoder_config(config_filename)

print("Decoder configuration loaded successfully.")
# The 'loaded_decoder' can now be used for decoding.
```

--------------------------------

### Example: Building CUDA-Q with CI Unit Tests

Source: https://nvidia.github.io/cudaqx/examples_rst/solvers/adapt.html

Demonstrates the build process for CUDA-Q specifically for running CI unit tests. This is a minimal build focused on testing infrastructure.

```bash
cmake -B build -DCUDAQ_BUILD_TESTS=ON -DCUDAQ_BUILD_PYTHON=OFF -DCUDAQ_BUILD_JAVA=OFF -DCUDAQ_BUILD_EXAMPLES=OFF -DCUDAQ_BUILD_TOOLS=OFF
cmake --build build --target test
```

--------------------------------

### Usage Example for CUDA-Q QEC Decoder Framework

Source: https://nvidia.github.io/cudaqx/examples_rst/qec/realtime_decoding.html

Illustrates how to use the CUDA-Q QEC decoder framework in C++. This example shows basic setup and usage of a decoder.

```cpp
#include <cudaq/qec/decoder.h>
#include <cudaq/qec/code.h>
#include <iostream>

int main() {
    // Assume a code and stabilizers are defined elsewhere
    // For demonstration, let's create a dummy code and stabilizers
    class DummyCode : public cudaq::qec::code {
    public:
        DummyCode(int n_qubits) : cudaq::qec::code(n_qubits) {}
        std::vector<cudaq::qec::Pauli> get_stabilizers() override {
            return {cudaq::qec::Pauli("XX"), cudaq::qec::Pauli("ZZ")};
        }
        std::vector<cudaq::qec::Pauli> get_logical_operators() override {
            return {cudaq::qec::Pauli("X"), cudaq::qec::Pauli("Z")};
        }
    };

    DummyCode code(2);
    cudaq::qec::decoder decoder(code);

    // Example syndrome measurement (replace with actual measurements)
    std::vector<int> syndrome = {0, 1};

    // Apply the decoder to correct errors
    decoder.apply(syndrome);

    std::cout << "Decoder applied successfully." << std::endl;

    return 0;
}
```

--------------------------------

### Basic VQE Usage Example

Source: https://nvidia.github.io/cudaqx/api/qec/python_realtime_decoding_api.html

A fundamental example demonstrating the Variational Quantum Eigensolver (VQE) using the CUDA-Q Solvers library. This snippet shows the basic setup for finding the ground state energy of a molecule.

```python
from cudaq import solvers
from cudaq.optimization import Adam

# Define Hamiltonian options and create Hamiltonian
ham_opts = solvers.MolecularHamiltonianOptions()
ham_opts.geometry = "H 0.0 0.0 0.0\n H 0.0 0.0 0.74 "
ham_opts.basis = "sto-3g"
hamiltonian = solvers.MolecularHamiltonian(ham_opts)

# Initialize VQE with a basic ansatz and optimizer
vqe = solvers.VQE(ansatz=solvers.ansatz.HardwareEfficient(num_qubits=hamiltonian.qubits), optimizer=Adam(max_iterations=100))

# Run VQE
result = vqe.run(hamiltonian)

print(f"Ground state energy: {result.energy:.5f}")
```

--------------------------------

### Compilation and Execution Commands

Source: https://nvidia.github.io/cudaqx/examples_rst/qec/code_capacity_noise.html

Commands to compile and run the C++ CUDA-QEC example. Ensure you have the necessary compiler and libraries installed.

```bash
nvq++ --enable-mlir --target=stim -lcudaq-qec code_capacity_noise.cpp -o code_capacity_noise
./code_capacity_noise
```

--------------------------------

### Decoder Instantiation (C++)

Source: https://nvidia.github.io/cudaqx/api/qec/cpp_api.html

Example of how to get a decoder instance using the C++ API, supporting default values for parameters.

```APIDOC
## get(name, H, param_map)

### Description
This `get` overload supports default values for parameters.

### Method
`static std::unique_ptr<decoder> get`

### Parameters
- **name** (const std::string &) - The name of the decoder.
- **H** (const cudaqx::tensor<uint8_t> &) - The parity-check matrix.
- **param_map** (const cudaqx::heterogeneous_map &) - Optional map of decoder parameters. Defaults to an empty map.
```

--------------------------------

### Decoder Instantiation (Python)

Source: https://nvidia.github.io/cudaqx/api/qec/cpp_api.html

Example of how to get a decoder instance using the Python API, including specifying the parity-check matrix and options.

```APIDOC
## get_decoder (Python Example)

### Description
Instantiates a decoder using the CUDA-QX extension points API.

### Method
`qec.get_decoder`

### Parameters
- **H** (numpy.ndarray) - The parity-check matrix (PCM) in row-major order (e.g., `order='C'`).
- **opts** (dict) - A dictionary of decoder options.

### Request Example
```python
import cudaq_qec as qec
import numpy as np

H = np.array([[1, 0, 0, 1, 0, 1, 1],
              [0, 1, 0, 1, 1, 0, 1],
              [0, 0, 1, 0, 1, 1, 1]], dtype=np.uint8) # sample 3x7 PCM

opts = dict() # see below for options

# Note: H must be in row-major order. If you use
# `scipy.sparse.csr_matrix.todense()` to get the parity check
# matrix, you must specify todense(order='C') to get a row-major
```
```

--------------------------------

### Example: Building CUDA-Q Relay BP Decoding (Full Build)

Source: https://nvidia.github.io/cudaqx/index.html

Provides instructions for performing a full build of the CUDA-Q Relay BP Decoding components, including CI tests and Hololink bridge/playback tools. This is necessary for full functionality.

```bash
# Perform a full build including Hololink tools
make -j $(nproc) \
  BUILD_HOLOLINK_TOOLS=ON \
  BUILD_CI_TESTS=ON

```

--------------------------------

### Real-Time Decoding Configuration Example

Source: https://nvidia.github.io/cudaqx/examples_rst/solvers/vqe.html

Shows the steps involved in configuring and saving a decoder for real-time use with CUDA-Q. This includes generating a detector error model and saving the decoder configuration.

```bash
# Step 1: Generate Detector Error Model (Conceptual)
# cudaq-qec generate-dem --code SteaneCode --output steane_dem.json

# Step 2: Configure and Save Decoder (Conceptual)
# cudaq-qec configure-decoder --dem steane_dem.json --decoder NVIDIALDPCDecoder --output steane_decoder.json

# Step 3: Load Configuration (Conceptual)
# decoder_config = cudaq.qec.load_decoder_config("steane_decoder.json")

# Step 4: Use in Quantum Kernels (Conceptual)
# @cudaq.kernel
# def my_kernel():
#     # ... quantum circuit ...
#     syndrome = cudaq.qec.measure_syndrome(...)
#     corrected_error = cudaq.qec.apply_decoder(syndrome, decoder_config)
#     # ... continue ...

print("Conceptual steps for real-time decoding configuration.")
```

--------------------------------

### Example: Real-Time Decoding Configuration - Step 2: Configure and Save Decoder

Source: https://nvidia.github.io/cudaqx/api/qec/python_api.html

This step focuses on configuring a specific decoder (e.g., NVIDIA QLDPC) with the generated detector error model and saving this configuration for later use.

```python
import cudaq

# Assume 'code' and 'dem' are already defined
code = cudaq.qec.surface_code(5)
dem = cudaq.qec.generate_detector_error_model(code)

# Initialize a decoder (e.g., NVIDIA QLDPC)
decoder_config = cudaq.qec.nvidia_qldpc_decoder_config(code, dem)

# Save the decoder configuration to a file
config_filename = "my_decoder_config.json"
decoder_config.save(config_filename)

print(f"Decoder configuration saved to {config_filename}")
```

--------------------------------

### Python Execution Example for CUDA-Q

Source: https://nvidia.github.io/cudaqx/api/solvers/gqe_api.html

Shows how to execute a CUDA-Q Python script. This typically involves running the script using the Python interpreter, assuming CUDA-Q is installed correctly.

```bash
python my_cudaq_script.py

# Ensure that the 'cudaq' Python package is installed in your environment.
```

--------------------------------

### Use Python QEC Implementation

Source: https://nvidia.github.io/cudaqx/components/qec/introduction.html

Demonstrates how to instantiate and use the custom QEC code.

```python
import cudaq_qec as qec
# Either import your code directly (e.g. "import my_steane", assuming
# your my_steane.py file is in the same directory), or you can paste the
# above code here without importing the file.
import my_steane

# Create instance of your code
code = qec.get_code('py-steane-example')

# Use the code for various numerical experiments
```

--------------------------------

### Configuration: Step 3 - Load Configuration

Source: https://nvidia.github.io/cudaqx/api/qec/python_realtime_decoding_api.html

Demonstrates loading a previously saved decoder configuration. This is essential for using a pre-configured decoder in quantum kernels.

```python
from cudaq import qec

# Load the decoder configuration from a file
loaded_config = qec.DecoderConfiguration.load("my_decoder_config.json")

# The loaded_config object can now be used to access the decoder and its settings
print(f"Loaded decoder configuration. Decoder type: {type(loaded_config.decoder).__name__}")
```

--------------------------------

### Complete Workflow Example

Source: https://nvidia.github.io/cudaqx/examples_rst/qec/realtime_decoding.html

Multi-phase workflow for generating DEMs and running real-time decoding in C++ and Python.

```bash
# Phase 1: Generate Detector Error Model (DEM)
# This is done once per code/distance/noise configuration

## C++
./surface_code-1 --distance 3 --num_shots 1000 --p_spam 0.01 \
                 --save_dem config_d3.yaml --num_rounds 12 --decoder_window 6

## Python
python surface_code-1.py --distance 3 --num_shots 1000 --p_spam 0.01 \
                         --save_dem config_d3.yaml --num_rounds 12 --decoder_window 6

# Phase 2: Run with Real-Time Decoding
# Use the saved DEM configuration

## Simulation
./surface_code-1 --distance 3 --num_shots 1000 --load_dem config_d3.yaml \
                 --num_rounds 12 --decoder_window 6

## Quantinuum Emulation
./surface_code-1-quantinuum-emulate --distance 3 --num_shots 1000 --load_dem config_d3.yaml \
                        --num_rounds 12 --decoder_window 6
```

--------------------------------

### Install gfortran on Debian-based Systems

Source: https://nvidia.github.io/cudaqx/quickstart/installation.html

If CUDA-Q Solvers requires `libgfortran` and it's not installed, use this command on Debian-based systems to install it via apt-get.

```bash
apt-get install gfortran
```

--------------------------------

### Configure and Run GQE Algorithm

Source: https://nvidia.github.io/cudaqx/examples_rst/solvers/gqe.html

Sets up the configuration for the GQE algorithm, including logging and verbosity settings, and then executes the algorithm to find the ground state energy.

```python
from lightning.fabric.loggers import CSVLogger

cfg = get_default_config()
cfg.use_fabric_logging = False
logger = CSVLogger("gqe_h2_logs/gqe.csv")
cfg.fabric_logger = logger
cfg.save_trajectory = False
cfg.verbose = True

minE, best_ops = solvers.gqe(cost, op_pool, max_iters=25, ngates=10, config=cfg)
```

--------------------------------

### Install CUDA-Q Solvers Library via Pip

Source: https://nvidia.github.io/cudaqx/quickstart/installation.html

Use this command to install the CUDA-Q Solvers library using pip. This is the simplest installation method for the Solvers component.

```bash
# Install Solvers library
pip install cudaq-solvers
```

--------------------------------

### Initialize nv-qldpc-decoder with various BP methods

Source: https://nvidia.github.io/cudaqx/examples_rst/qec/decoders.html

Demonstrates the configuration of different BP methods using the qec.get_decoder interface. Requires a parity check matrix H and appropriate parameters for the chosen method.

```python
# Simple 3x7 parity check matrix for demonstration
H_list = [[1, 0, 0, 1, 0, 1, 1], [0, 1, 0, 1, 1, 0, 1],
          [0, 0, 1, 0, 1, 1, 1]]
H = np.array(H_list, dtype=np.uint8)

print("=" * 60)
print("Demonstrating BP Methods in nv-qldpc-decoder")
print("=" * 60)

# Method 0: Sum-Product BP (default)
print("\n1. Sum-Product BP (bp_method=0, default):")
try:
    decoder_sp = qec.get_decoder("nv-qldpc-decoder",
                                 H,
                                 bp_method=0,
                                 max_iterations=30)
except Exception as e:
    print(
        'The nv-qldpc-decoder is not available with your current CUDA-Q ' +
        'QEC installation.')
    exit(0)
print("   Created decoder with sum-product BP")

# Method 1: Min-Sum BP
print("\n2. Min-Sum BP (bp_method=1):")
decoder_ms = qec.get_decoder("nv-qldpc-decoder",
                             H,
                             bp_method=1,
                             max_iterations=30,
                             scale_factor=1.0)
print("   Created decoder with min-sum BP")

# Method 2: Min-Sum with uniform Memory (Mem-BP)
print("\n3. Mem-BP (bp_method=2, uniform memory strength):")
decoder_mem = qec.get_decoder("nv-qldpc-decoder",
                              H,
                              bp_method=2,
                              max_iterations=30,
                              use_sparsity=True,
                              gamma0=0.5)
print("   Created decoder with Mem-BP (gamma0=0.5)")

# Method 3: Min-Sum with Disordered Memory (DMem-BP)
print("\n4. DMem-BP (bp_method=3, disordered memory strength):")
# Option A: Using gamma_dist (random gammas in range)
decoder_dmem = qec.get_decoder("nv-qldpc-decoder",
                               H,
                               bp_method=3,
                               max_iterations=30,
                               use_sparsity=True,
                               gamma_dist=[0.1, 0.5],
                               bp_seed=42)
print("   Created decoder with DMem-BP (gamma_dist=[0.1, 0.5])")

# Option B: Using explicit_gammas (specify exact gamma for each variable)
block_size = H.shape[1]
explicit_gammas = [[0.1 + 0.05 * i for i in range(block_size)]]
decoder_dmem_explicit = qec.get_decoder("nv-qldpc-decoder",
                                        H,
                                        bp_method=3,
                                        max_iterations=30,
                                        use_sparsity=True,
                                        explicit_gammas=explicit_gammas)
print("   Created decoder with DMem-BP (explicit gammas)")

# Method 4: Sequential Relay BP (composition=1)
print("\n5. Sequential Relay BP (composition=1):")
print("   Requires bp_method=3 and srelay_config")

# Configure relay parameters
srelay_config = {
    'pre_iter': 5,  # Run 5 iterations with gamma0 before relay legs
    'num_sets': 3,  # Use 3 relay legs
    'stopping_criterion': 'FirstConv'  # Stop after first convergence
}

# Option A: Using gamma_dist for relay legs
decoder_relay = qec.get_decoder("nv-qldpc-decoder",
                                H,
                                bp_method=3,
                                composition=1,
                                max_iterations=50,
                                use_sparsity=True,
                                gamma0=0.3,
                                gamma_dist=[0.1, 0.5],
                                srelay_config=srelay_config,
                                bp_seed=42)
print("   Created decoder with Relay-BP (gamma_dist, FirstConv stopping)")

# Option B: Using explicit gammas for each relay leg
num_relay_legs = 3
explicit_relay_gammas = [
    [0.1 + 0.02 * i for i in range(block_size)],  # First relay leg
    [0.2 + 0.03 * i for i in range(block_size)],  # Second relay leg
    [0.3 + 0.04 * i for i in range(block_size)]  # Third relay leg
]

srelay_config_all = {
    'pre_iter': 10,
    'num_sets': 3,
    'stopping_criterion': 'All'  # Run all relay legs
}

decoder_relay_explicit = qec.get_decoder(
    "nv-qldpc-decoder",
    H,
    bp_method=3,
    composition=1,
    max_iterations=50,
    use_sparsity=True,
    gamma0=0.3,
    explicit_gammas=explicit_relay_gammas,
    srelay_config=srelay_config_all)
print("   Created decoder with Relay-BP (explicit gammas, All legs)")
```

--------------------------------

### Install CUDA-Q QEC Library via Pip

Source: https://nvidia.github.io/cudaqx/quickstart/installation.html

Use this command to install the CUDA-Q QEC library using pip. This is the simplest installation method for the QEC component.

```bash
# Install QEC library
pip install cudaq-qec
```

--------------------------------

### Step 3: Load Configuration for Real-Time Decoding

Source: https://nvidia.github.io/cudaqx/api/qec/sliding_window_api.html

Demonstrates how to load a previously saved decoder configuration for use in real-time quantum error correction with CUDA-Q. This allows quick deployment of decoders.

```python
import cudaq

# Define the path to the saved decoder configuration file
config_path = "./my_decoder_config.json"

# Load the decoder configuration
# This function typically returns a decoder object ready for use.
# loaded_decoder = cudaq.qec.load_decoder_configuration(config_path)

print("Step 3: Load Configuration for Real-Time Decoding")
print(f"This step loads the decoder configuration from {config_path}.")
print("The loaded decoder is now ready to process syndromes in real-time.")
print("Refer to CUDA-Q documentation for the exact loading function.")

```

--------------------------------

### Install PyTorch directly

Source: https://nvidia.github.io/cudaqx/quickstart/installation.html

Installs the standard PyTorch package independently of CUDA-QX components.

```bash
pip install torch
```

--------------------------------

### Example: Real-Time Decoding Configuration - Step 4: Use in Quantum Kernels

Source: https://nvidia.github.io/cudaqx/api/qec/python_api.html

Shows how to integrate a loaded real-time decoder into a CUDA-Q quantum kernel for live error correction during quantum computation.

```python
import cudaq

# Assume 'loaded_decoder' is already loaded from a configuration file
# loaded_decoder = cudaq.qec.load_decoder_config("my_decoder_config.json")

@cudaq.kernel
def my_quantum_kernel(decoder: cudaq.qec.decoder):
    q = cudaq.get_qureg(5)
    # ... perform quantum operations ...
    syndrome = cudaq.qec.measure_syndrome(q, code) # Assuming 'code' is accessible
    corrected_syndrome = decoder.decode(syndrome)
    # ... continue computation with corrected syndrome ...

# Example of how to pass the loaded decoder to the kernel
# Note: The actual 'code' object needs to be accessible or passed appropriately
# For simplicity, assume 'code' is globally available or part of the kernel context
# my_quantum_kernel(loaded_decoder)
```

--------------------------------

### Install PyTorch via CUDA-QX optional components

Source: https://nvidia.github.io/cudaqx/quickstart/installation.html

Installs PyTorch as a dependency when setting up the QEC or Solvers libraries.

```bash
# Installs PyTorch as a dependency
pip install cudaq-qec[tensor-network-decoder]
pip install cudaq-solvers[gqe]
```

--------------------------------

### Install CUDA-Q with pip

Source: https://nvidia.github.io/cudaqx/api/qec/nv_qldpc_decoder_api.html

Install the CUDA-Q Python package using pip. Ensure you have a compatible Python environment.

```bash
pip install cudaq
```

--------------------------------

### Example: Running AI Predecoder with CUDA-Q Realtime

Source: https://nvidia.github.io/cudaqx/api/solvers/gqe_api.html

Demonstrates how to run the AI predecoder, specifying positional arguments and named flags. This is crucial for setting up the predecoder pipeline for real-time quantum error correction.

```bash
./your_predecoder_executable \
  --data-dir /path/to/data \
  --model /path/to/model.onnx \
  --num-gpus 1 \
  --pipeline-config config.json

# Replace placeholders with actual paths and configuration details.
```

--------------------------------

### hololink_qldpc_graph_decoder_test.sh - FPGA Mode Example

Source: https://nvidia.github.io/cudaqx/examples_rst/qec/realtime_relay_bp.html

Example of running the decoder test script in FPGA mode with specific network and device configurations.

```APIDOC
## Running in FPGA Mode

### Description
This section demonstrates how to run the `hololink_qldpc_graph_decoder_test.sh` script in FPGA mode, specifying various parameters for network interfaces, devices, and GPU selection.

### Method
Shell Script Execution

### Endpoint
`./libs/qec/unittests/utils/hololink_qldpc_graph_decoder_test.sh`

### Parameters
#### Path Parameters
None

#### Query Parameters
None

#### Request Body
None

### Command Example
```bash
./libs/qec/unittests/utils/hololink_qldpc_graph_decoder_test.sh \
  --build \
  --setup-network \
  --device mlx5_5 \
  --bridge-ip 192.168.0.1 \
  --fpga-ip 192.168.0.2 \
  --gpu 2 \
  --page-size 512 \
  --hsb-dir /path/to/holoscan-sensor-bridge
```

### Key Parameters for FPGA Mode

| Parameter         | Description                                       |
|-------------------|---------------------------------------------------|
| `--device`        | ConnectX IB device name (e.g., `mlx5_5`)          |
| `--bridge-ip`     | IP address assigned to the ConnectX interface     |
| `--fpga-ip`       | FPGA’s IP address                                 |
| `--gpu`           | GPU device ID (choose NUMA-local GPU for lowest latency) |
| `--page-size`     | Ring buffer slot size in bytes (use `512` on GB200 for alignment) |
| `--spacing`       | Inter-shot spacing in microseconds                |

### Note on `--spacing`
The `--spacing` value should be set to at least the per-shot decode time to avoid overrunning the input ring buffer. If syndromes arrive faster than the decoder can process them, the buffer fills up and messages are lost. Use a `--spacing` value at or above the observed decode time for sustained operation.

### GPU Selection
For lowest latency, choose a GPU that is NUMA-local to the ConnectX NIC. For example, on a GB200 system where `mlx5_5` is on NUMA node 1, use `--gpu 2` or `--gpu 3`. Check NUMA locality with:

```bash
cat /sys/class/infiniband/<device>/device/numa_node
```

### Network Sanity Check
Before running, verify that the bridge IP is assigned to exactly one interface:

```bash
ip addr show | grep 192.168.0.1
```

If multiple interfaces show the same IP, remove the duplicate to avoid routing ambiguity that silently drops RDMA packets.
```

--------------------------------

### Build FPGA Demo and Realtime Tools

Source: https://nvidia.github.io/cudaqx/examples_rst/qec/realtime_predecoder_fpga.html

Provides the sequence of commands to clone, configure, and build the cuda-quantum realtime components and the holoscan-sensor-bridge dependency.

```bash
# 1. Clone cuda-quantum (realtime)
git clone --filter=blob:none --no-checkout \
  https://github.com/NVIDIA/cuda-quantum.git cudaq-realtime-src
cd cudaq-realtime-src
git sparse-checkout init --cone
git sparse-checkout set realtime
git checkout releases/v0.14.1
cd ..

# 2. Build holoscan-sensor-bridge (tag 2.6.0-EA2)
#    Requires cmake >= 3.30.4
git clone --branch 2.6.0-EA2 \
  https://github.com/nvidia-holoscan/holoscan-sensor-bridge.git
cd holoscan-sensor-bridge

# Strip operators we don't need to avoid configure failures
sed -i '/add_subdirectory(audio_packetizer)/d; /add_subdirectory(compute_crc)/d;
        /add_subdirectory(csi_to_bayer)/d; /add_subdirectory(image_processor)/d;
        /add_subdirectory(iq_dec)/d; /add_subdirectory(iq_enc)/d;
        /add_subdirectory(linux_coe_receiver)/d; /add_subdirectory(linux_receiver)/d;
        /add_subdirectory(packed_format_converter)/d; /add_subdirectory(sub_frame_combiner)/d;
        /add_subdirectory(udp_transmitter)/d; /add_subdirectory(emulator)/d;
        /add_subdirectory(sig_gen)/d; /add_subdirectory(sig_viewer)/d' \
  src/hololink/operators/CMakeLists.txt

mkdir -p build && cd build
cmake -G Ninja -DCMAKE_BUILD_TYPE=Release \
  -DHOLOLINK_BUILD_ONLY_NATIVE=OFF \
  -DHOLOLINK_BUILD_PYTHON=OFF \
  -DHOLOLINK_BUILD_TESTS=OFF \
  -DHOLOLINK_BUILD_TOOLS=OFF \
  -DHOLOLINK_BUILD_EXAMPLES=OFF \
  -DHOLOLINK_BUILD_EMULATOR=OFF ..
cmake --build . --target gpu_roce_transceiver hololink_core
cd ../..

# 3. Build libcudaq-realtime with Hololink tools enabled
cd cudaq-realtime-src/realtime && mkdir -p build && cd build
cmake -G Ninja -DCMAKE_INSTALL_PREFIX=/tmp/cudaq-realtime \
  -DCUDAQ_REALTIME_ENABLE_HOLOLINK_TOOLS=ON \
  -DHOLOSCAN_SENSOR_BRIDGE_SOURCE_DIR=../../holoscan-sensor-bridge \
  -DHOLOSCAN_SENSOR_BRIDGE_BUILD_DIR=../../holoscan-sensor-bridge/build \
  ..
ninja && ninja install
cd ../../..
```

--------------------------------

### Install Both CUDA-Q QEC and Solvers Libraries via Pip

Source: https://nvidia.github.io/cudaqx/quickstart/installation.html

Install both the CUDA-Q QEC and Solvers libraries simultaneously using a single pip command.

```bash
# Install both libraries
pip install cudaq-qec cudaq-solvers
```

--------------------------------

### C++ Compilation Example

Source: https://nvidia.github.io/cudaqx/api/qec/python_realtime_decoding_api.html

Provides a basic example of how to compile a CUDA-Q C++ application. This typically involves using nvcc and linking against CUDA-Q libraries.

```bash
nvcc my_cudaq_app.cpp -o my_cudaq_app -l:libcudaq.so -L/path/to/cudaq/lib -I/path/to/cudaq/include
```

--------------------------------

### Step 3: Load Configuration

Source: https://nvidia.github.io/cudaqx/api/qec/tensor_network_decoder_api.html

Explains how to load a previously saved decoder configuration. This is necessary to utilize a pre-trained or pre-configured decoder within a quantum kernel or simulation.

```python
import cudaq

# Conceptual function to load a decoder configuration
def load_decoder_configuration(config_path):
    print(f"Loading decoder configuration from: {config_path}")
    # This function would read the configuration file and instantiate
    # the appropriate decoder object with its trained parameters.
    # loaded_decoder = cudaq.qec.load_decoder(config_path)
    # return loaded_decoder
    return "loaded_decoder_object" # Placeholder

# Example usage:
# decoder_config_path = "./my_qldpc_decoder.config"
# my_decoder = load_decoder_configuration(decoder_config_path)
# print(f"Decoder loaded: {my_decoder}")

```

--------------------------------

### AI Predecoder Example

Source: https://nvidia.github.io/cudaqx/examples_rst/qec/realtime_predecoder_pymatching.html

This example demonstrates the usage of an AI-based predecoder with CUDA-Q Realtime. It outlines the necessary prerequisites, data directory layout, and how to run the predecoder.

```bash
# Example command to run the AI predecoder
# python run_predecoder.py \
#   --data_dir /path/to/your/data \
#   --model_path /path/to/your/model.onnx \
#   --num_qubits 4 \
#   --num_rounds 10 \
#   --threshold 0.05

# Positional Arguments:
# data_dir: Path to the directory containing simulation data.
# model_path: Path to the ONNX model file for the predecoder.

# Named Flags:
# --num_qubits: Number of qubits in the quantum system.
# --num_rounds: Number of QEC rounds to simulate.
# --threshold: Error rate threshold for decoding.

```

--------------------------------

### Basic VQE Example in CUDA-Q

Source: https://nvidia.github.io/cudaqx/examples_rst/qec/realtime_predecoder_fpga.html

A fundamental example showcasing the Variational Quantum Eigensolver (VQE) implementation in CUDA-Q, suitable for basic quantum chemistry calculations.

```python
import cudaq
from cudaq import spin

# Define the Hamiltonian (e.g., from a molecular model)
hamiltonian = spin.FermionicModel.from_qiskit_hamiltonian(...)

# Define the ansatz circuit
ansatz = cudaq.qiskit.circuit.QuantumCircuit(hamiltonian.num_qubits)
# ... add gates to ansatz ...

# Run VQE
result = cudaq.minimize(ansatz, hamiltonian=hamiltonian, optimizer=cudaq.optimizers.Adam(50), sampler=cudaq.SampleResult())

print(f"Minimum energy found: {result.optimal_value}")
```
Tokens:
Raw