Skip to content

Latest commit

 

History

History
140 lines (114 loc) · 3.71 KB

README.md

File metadata and controls

140 lines (114 loc) · 3.71 KB

QML Classifier Mini-App

Overview

The QML Classifier Mini-App is a quantum machine learning benchmarking tool designed to evaluate the performance of quantum classifiers using different optimization techniques and hardware configurations. It supports GPU acceleration, Just-In-Time (JIT) compilation, and vectorized mapping (vmap) for improved performance.

Key Features

  • GPU-accelerated quantum circuit execution
  • JIT compilation support for optimized performance
  • Vectorized batch processing with vmap
  • Configurable circuit parameters:
    • Number of qubits
    • Circuit depth
    • Batch size
    • Number of epochs
  • Comprehensive performance metrics collection
  • Distributed execution capabilities using Ray

Configuration

Basic Configuration

Configuration can be specified either through YAML files or programmatically:

batch_size: 80
depth: 2
device: gpu
jit: true
n_batches: 10
n_epochs: 2
n_qubits: 11
vmap: true

Cluster Configuration

The mini-app supports configuration for HPC environments:

cluster_info = {
    "executor": "pilot",
    "config": {
        "resource": "slurm://localhost",
        "working_directory": "/path/to/work",
        "number_of_nodes": 1,
        "cores_per_node": 256,
        "gpus_per_node": 4,
        "queue": "premium",
        "walltime": 30,
        "type": "ray"
    }
}

Usage

Command Line

python classifier.py

Code Configuration and Execution

from mini_apps.qml_classifier.classifier import QMLClassifierMiniApp

# Initialize mini-app with cluster configuration
app = QMLClassifierMiniApp(cluster_info)

# Define training configurations
configs = [{
    "n_qubits": 13,
    "depth": 1,
    "batch_size": 64,
    "n_batches": 502,
    "n_epochs": 1,
    "jit": True,
    "vmap": True,
    "device": "gpu"
}]

# Run training
app.run(configs)

Performance Optimization

The mini-app supports several optimization techniques:

  1. JIT Compilation:

    • Enables ahead-of-time compilation of quantum circuits
    • Reduces runtime overhead

  2. Vectorized Mapping (vmap):

    • Enables parallel processing of batched inputs
    • Improves GPU utilization

  3. Batch Size Tuning:

    • Configurable batch sizes for optimal performance
    • Automatic batch size adjustment based on hardware capabilities

Implementation Details

The mini-app consists of two main components:

  1. QMLClassifierMiniApp: Main class handling execution and metrics collection
class QMLClassifierMiniApp:
    def __init__(self, pilot_compute_description):
        os.makedirs(pilot_compute_description["config"]["working_directory"], exist_ok=True)
        current_time = datetime.now().strftime("%Y%m%d_%H%M%S")
        file_name = f"result_compression_{current_time}.csv"
        self.result_file = os.path.join(pilot_compute_description["config"]["working_directory"], file_name)
        print(f"Result file: {self.result_file}")
        header = ["column_1", "compute_time_sec"]
        self.metrics_file_writer = MetricsFileWriter(self.result_file, header)
        self.executor = MiniAppExecutor(pilot_compute_description).get_executor()

    def run(self, configs):
        start = perf_counter()
        futures = self.executor.submit_tasks(training, configs)
        self.executor.wait(futures)
        print("Done")
        compute_time_sec = perf_counter() - start
        self.metrics_file_writer.write([
            "worked",
            compute_time_sec])
  1. Configuration Management: Flexible configuration system supporting both YAML and programmatic configuration
batch_size: 80
depth: 2
device: gpu
jit: true
n_batches: 10
n_epochs: 2
n_qubits: 11
vmap: true