Skip to content

README.md

Latest commit

 

History

History
110 lines (85 loc) · 2.84 KB

README.md

File metadata and controls

110 lines (85 loc) · 2.84 KB

Here's a README.md for the QML Training Mini-App:

QML Training Mini-App

Overview

The QML Training Mini-App is a quantum machine learning application designed for training quantum circuit-based generative models. It leverages the QuGEN framework and supports distributed execution through Ray/Pilot-Quantum for high-performance computing environments.

Key Features

  • Quantum Circuit Born Machine (QCBM) training
  • Distributed execution capabilities
  • Configurable quantum circuit parameters
  • Performance metrics collection
  • Support for different parallelism frameworks (JAX)

Technical Requirements

Software Dependencies

  • Python 3.10+
  • QuGEN framework
  • JAX
  • Pilot-Quantum
  • Ray (for distributed execution)

Hardware Requirements

  • Multi-core CPU support
  • Optional: GPU acceleration
  • HPC cluster (for distributed training)

Configuration

Cluster Configuration

cluster_info = {
    "executor": "pilot",
    "config": {
        "resource": "slurm://localhost",
        "type": "ray",
        "number_of_nodes": 1,
        "cores_per_node": 2,
        "queue": "debug",
        "walltime": 30
    }
}

Training Parameters

qml_parameters = {
    "build_parameters": {
        "model_type": "discrete",
        "data_set_name": "X_2D",
        "n_qubits": 8,
        "n_registers": 2,
        "circuit_depth": 2,
        "circuit_type": "copula",
        "transformation": "pit",
        "parallelism_framework": "jax"
    },
    "train_parameters": {
        "n_epochs": 3,
        "batch_size": 200,
        "hist_samples": 100000
    }
}

Usage

Basic Execution

from mini_apps.qml_training.qml_training_miniapp import QMLTrainingMiniApp

# Initialize mini-app with configurations
qml_mini_app = QMLTrainingMiniApp(cluster_info, qml_parameters)

# Run training
qml_mini_app.run()

# Clean up
qml_mini_app.close()

Implementation Details

The mini-app consists of two main components:

  1. QMLTrainingMiniApp: Main class handling execution and metrics collection

  2. DiscreteQCBMModelHandler: Handles the quantum circuit model building and training

Performance Metrics

The mini-app collects and records the following metrics:

  • Timestamp
  • Scenario label
  • Number of qubits
  • Computation time
  • Training parameters
  • Cluster configuration
  • Kullback-Leibler divergence

Results are saved in CSV format with timestamps for further analysis.

References