Old Versions (Retired)

All Versions on This Page Are Retired

The versions documented below (v3.2, v3.3, v3.4, v3.5) are retired and no longer supported. Please upgrade to v4.0 for the latest features, performance improvements, and production-ready optimizations.

Retired Versions Summary

This page contains historical documentation for retired Q-Store versions. These versions are no longer maintained or recommended for production use.

Version	Release	Status	Key Features	Superseded By
v3.2	Q4 2024	❌ Retired	Complete ML training capabilities, quantum neural networks	v3.5
v3.3	Q4 2024	❌ Retired	50-100x algorithmic optimization, SPSA gradient estimator	v3.5
v3.4	Q4 2024	❌ Retired	8-10x speed improvement, true parallelization	v3.5
v3.5	Q1 2025	❌ Retired	2-3x realistic improvements, multi-backend orchestration	v4.0

Migration Recommendation: All users should upgrade directly to v4.0 for the best experience.

---

Example Projects

These example projects work with q-store version 3.5 and below

Example Projects

Q-Store v3.2 (Retired)

Retired Version

This version is retired. Please upgrade to v4.0 for better performance and production-ready optimizations.

Overview

Q-Store v3.2 introduced complete machine learning training capabilities with full hardware abstraction, enabling quantum neural networks that work seamlessly across simulators and quantum hardware.

Key Features

Core ML Components

• Quantum Neural Network Layers: QuantumLayer, QuantumConvolutionalLayer, QuantumPoolingLayer
• Gradient Computation: Parameter Shift Rule, Finite Difference, Natural Gradients
• Data Encoding: Amplitude, Angle, Basis, and ZZ Feature Map encoding
• Training Infrastructure: Complete training orchestration with Adam optimizer, checkpoint management

Hardware Abstraction

Train once, run anywhere across different quantum backends (mock, Cirq, Qiskit, IonQ).

Transfer Learning

Pre-train and fine-tune models with parameter freezing support.

Performance Characteristics

• Qubits: 2-8 qubits
• Parameters: 6-48 trainable parameters
• Training Time: ~45 seconds for 5 epochs (mock backend)

Basic Example

from q_store.core import QuantumTrainer, QuantumModel, TrainingConfig, BackendManager

config = TrainingConfig(
    pinecone_api_key="your-api-key",
    quantum_sdk="mock",
    learning_rate=0.01,
    epochs=10,
    batch_size=5,
    n_qubits=4
)

backend_manager = BackendManager(config)
backend = backend_manager.get_backend("mock_ideal")

model = QuantumModel(
    input_dim=4,
    output_dim=2,
    n_layers=2,
    backend=backend
)

trainer = QuantumTrainer(config, backend_manager)
history = await trainer.train(model, data_loader)

---

Q-Store v3.3 (Retired)

Retired Version

This version is retired. Please upgrade to v4.0 for better performance and production-ready optimizations.

Overview

Q-Store v3.3 delivered 50-100x faster training through algorithmic optimization while maintaining full backward compatibility with v3.2.

Performance Improvements

Metric	v3.2	v3.3	Improvement
Circuits per batch	960	10-20	48-96x
Time per batch	240s	5-10s	24-48x
Time per epoch	40min	50-100s	24-48x
Memory usage	500MB	200MB	2.5x better

Core Components

1. SPSA Gradient Estimator

Simultaneous Perturbation Stochastic Approximation - estimates ALL gradients with just 2 circuit evaluations instead of 2N evaluations for N parameters.

2. Circuit Batch Manager

Batches multiple circuit executions into single API calls, reducing overhead by 5-10x.

3. Intelligent Circuit Cache

Multi-level caching for quantum circuits to avoid redundant computations (2-5x speedup).

4. Hardware-Efficient Ansatz

Optimized quantum layer with reduced gate count (33% fewer parameters).

5. Adaptive Gradient Method Selector

Automatically selects the best gradient method based on training stage.

Migration from v3.2

# v3.2 code
config = TrainingConfig(
    learning_rate=0.01,
    batch_size=32,
    n_qubits=10,
    circuit_depth=4
)

# v3.3 - just add one line for 24x speedup!
config = TrainingConfig(
    learning_rate=0.01,
    batch_size=32,
    n_qubits=10,
    circuit_depth=4,
    gradient_method='spsa'  # Add this!
)

---

Q-Store v3.4 (Retired)

Retired Version

This version is retired. Please upgrade to v4.0 for better performance and production-ready optimizations.

Overview

Q-Store v3.4 delivered 8-10x faster training through true parallelization and hardware-native optimizations, achieving sub-60 second training epochs on IonQ hardware.

Performance Improvements

Metric	v3.3	v3.4	Improvement
Batch time (20 circuits)	35s	3-5s	7-12x faster
Circuits per second	0.6	5-8	8-13x faster
Epoch time	392s	30-50s	8-13x faster
Training (5 epochs)	32 min	2.5-4 min	8-13x faster

Core Components

1. IonQBatchClient

True Parallel Batch Submission - submits multiple circuits in a single API call (12x faster submission).

2. IonQNativeGateCompiler

Hardware-Native Gate Compilation - compiles to IonQ native gates (GPi, GPi2, MS) for 30% faster execution.

3. SmartCircuitCache

Template-Based Circuit Caching - caches circuit structure and dynamically binds parameters (10x faster preparation).

4. CircuitBatchManagerV34

Integrated Optimization Pipeline - orchestrates all v3.4 optimizations with comprehensive tracking.

5. AdaptiveQueueManager

Dynamic Batch Optimization - adjusts batch size based on real-time queue conditions.

Configuration

config = TrainingConfig(
    # Enable all v3.4 features
    enable_all_v34_features=True,

    # Or selective enablement
    use_batch_api=True,
    use_native_gates=True,
    enable_smart_caching=True,
    connection_pool_size=5,
)

---

Q-Store v3.5 (Retired)

Retired Version

This version is retired. Please upgrade to v4.0 for better performance and production-ready optimizations.

Overview

Version 3.5 focused on honest performance gains and addressing real bottlenecks with realistic 2-3x performance improvements through proven techniques.

Key Philosophy

• ✅ Realistic 2-3x improvement through proven techniques
• ✅ Multi-backend distribution for true parallel execution
• ✅ Adaptive resource allocation based on training phase
• ✅ Honest documentation with verified benchmarks

What’s New in v3.5

1. Multi-Backend Orchestrator

Distribute quantum circuit execution across multiple backends simultaneously for 2-3x throughput improvement.

2. Adaptive Circuit Optimization

Dynamically adjust circuit complexity during training for 30-40% faster execution.

3. Adaptive Shot Allocation

Use minimum shots needed for gradient estimation, saving 20-30% execution time.

4. Natural Gradient Descent

Replace SPSA with natural gradient for 2-3x fewer iterations to convergence.

Verified Benchmarks

Metric	v3.4 Actual	v3.5 Achieved
Circuits/sec	0.57	1.3
Batch time (20 circuits)	35s	18s
Epoch time	350s	175s
Training (3 epochs)	17.5 min	8.5 min
Overall speedup	1x	2.1x

Complete Configuration Example

from q_store import TrainingConfig, QuantumTrainer

config = TrainingConfig(
    # Multi-backend orchestration
    enable_multi_backend=True,
    backend_configs=[
        {'provider': 'ionq', 'target': 'simulator', 'api_key': key1},
        {'provider': 'ionq', 'target': 'simulator', 'api_key': key2},
        {'provider': 'local', 'simulator': 'qiskit_aer', 'device': 'GPU'},
    ],

    # Adaptive optimizations
    adaptive_circuit_depth=True,
    circuit_depth_schedule='exponential',
    adaptive_shot_allocation=True,

    # Advanced gradient methods
    gradient_method='natural_gradient',

    # Enable all v3.5 features
    enable_all_v35_features=True,
)

trainer = QuantumTrainer(config)
await trainer.train(model, train_loader, val_loader)

Known Limitations

• Accuracy: 70-75% on Fashion MNIST (vs 88-90% classical)
• Inference speed: ~2s per sample (vs <1ms classical)
• Best for: Parameter-limited scenarios, few-shot learning

---

Migration Path

All users of retired versions should upgrade to v4.0:

1. Update package: pip install --upgrade q-store>=4.0.0
2. Review v4.0 documentation: Check the v4.0 release notes
3. Update configuration: Migrate to v4.0 configuration format
4. Test thoroughly: Validate performance and accuracy improvements

For migration assistance, please refer to the v4.0 migration guide or open a GitHub discussion.

---

Last Updated: December 2024 Current Recommended Version: v4.0