mit-han-lab/torchquantum

Features

• Quantum-Classical Hybrid Models - Builds and trains hybrid quantum-classical neural networks using standard deep learning frameworks and automatic differentiation techniques.
• Quantum Simulators - Performs quantum state vector calculations and unitary operations using multi-dimensional tensor arrays mapped onto classical hardware accelerators.
• Parameter-Shift Rules - Calculates exact derivatives of quantum expectation values by evaluating circuit outputs at shifted positive and negative parameter offsets.
• Gradient Computation - Calculate parameter gradients automatically through dynamic computation graphs to enable training with standard optimization techniques.
• Quantum Parameter Optimizations - Compute gradients automatically through native autograd support to enable backpropagation and optimization of circuit weights.
• Neural Network Layers - Combine classical neural network layers with parameterized quantum circuits inside a unified machine learning pipeline.
• Quantum-Enhanced Training - Optimize parameterized quantum circuit models by computing gradients of expectation values with respect to trainable parameters.
• Quantum Computing - Integrates parameterized quantum circuits and quantum neural network simulation directly into PyTorch training pipelines.
• GPU-Accelerated Quantum Simulators - Executes quantum computational models and statevector simulations on classical hardware using accelerators like GPUs to scale up qubit counts.
• Clifford Circuit Samplers - Executes quantum circuit simulations using stabilizer operations that run exponentially faster than general quantum circuit simulations.
• Quantum Circuit Execution - Simulates quantum computational models and parameterized circuits on classical hardware and accelerators like GPUs for faster execution.
• Autograd Engines - Computes analytical parameter gradients by hooking quantum operations directly into existing deep learning framework automatic differentiation engines.
• Quantum Observable Calculation - Compute analytical parameter gradients for quantum models using shift rules to optimize and train network weights effectively.
• Static Quantization - Applies post-training precision reduction and straight-through estimation estimators to optimise quantum neural networks for constrained deployment environments.
• Quantization-Aware Training - Performs static quantization and quantization-aware finetuning on trained models using straight-through estimation for gradient propagation.
• Quantum State Collapse - Collapses quantum states to extract classical numerical values for classification and regression tasks.
• Quantum Hardware Experimentation - Translates operation histories into standard quantum assembly formats for execution on physical quantum devices and remote processors.
• Classical Data Encoders - Transforms raw classical values into quantum states using dedicated encoding layers for downstream processing.
• QASM Exporters - Generates standard QASM representation strings from recorded quantum operation histories for interoperability with external tools and hardware.
• State Preparation Circuit Trainers - Optimizes parameterized quantum circuit weights using backpropagation to generate specific target quantum states from an initial zero state.
• Batched Quantum Data Processors - Runs inference and training tasks concurrently in batches across processing hardware to accelerate throughput.
• Quantum Circuit Exchange Formats - Translates internal quantum operation histories into standard industry assembly strings for execution on physical quantum processors.

Star history