Alance SDK

NumPy + Autodiff + Quantum Simulator — in one clean, lightweight Python SDK.

Alance is a minimal, hackable, open-source SDK that brings together:

Array computing (NumPy-like)

Automatic differentiation (PyTorch/JAX-style)

Quantum circuit simulation (Qiskit-style basics)

Optional GPU acceleration

Built for learning, experimentation, research, and building small ML/physics tools without heavy dependencies.

Features

• Alance Array

• NumPy-like API

• Supports Python lists, scalars

• Elementwise ops: + - * /

• Broadcasting

• Optional GPU support (via CuPy, if available)

Reverse-Mode Autodiff

• Define variables, build expressions

• Compute gradients automatically

• Similar conceptual model to PyTorch but far simpler

• Perfect for building your own neural nets or optimizers

Quantum Simulator

• Up to small 1–5 qubit states

• Common gates: H, X, Y, Z, CNOT, RX, RY, RZ

• Apply gates, simulate circuits, measure qubits

• Educational and extendable

Installation

• PyPI (recommended)

• pip install alance

• Local Development

• git clone https://github.com//Alance.git cd Alance pip install -e .

Quick Start Demo (All Features in One)

from alance import Array, get_device, set_device from alance.autodiff import Variable from alance import Simulator, H, CNOT, RX

print("Device:", get_device())

Arrays

a = Array([1, 2]) b = Array([3, 4]) print("a+b =", a + b) print("a*b =", a * b)

Autodiff

x, y = Variable(2.0), Variable(5.0) z = x * y + x z.backward() print("z =", z.value) print("dz/dx =", x.grad, "dz/dy =", y.grad)

Quantum

sim = Simulator(2) sim.apply_single_qubit_gate(H, 0) sim.apply_cnot(0, 1) sim.apply_param_gate(RX, 3.14159/2, 0) print("State:", sim.state) print("Measure qubit 0:", sim.measure(0))

Project Structure

alance/ │ ├── array.py # NumPy-like Array class ├── autodiff/ # Reverse-mode autodiff engine │ ├── variable.py │ ├── functions.py │ ├── quantum/ # Quantum simulator + gates │ ├── simulator.py │ ├── gates.py │ ├── utils/ # Helper utilities │ ├── init.py └── pyproject.toml # Build configuration

A clean and simple architecture so you can read, understand, and extend everything.

Architecture Diagram

┌─────────────────────┐ │ Alance SDK │ └──────────┬──────────┘ │ ┌──────────────────┼──────────────────┐ │ │ │ ┌────────────┐ ┌─────────────┐ ┌──────────────┐ │ Array │ │ Autodiff │ │ Quantum │ │ (CPU/GPU) │ │ Engine │ │ Simulator │ └────────────┘ └─────────────┘ └──────────────┘ │ │ │ ┌──────────┐ ┌────────────┐ ┌──────────────┐ │ Ops: +-* │ │ Graph-based│ │ H, X, CNOT │ │ shapes │ │ derivatives│ │ RX, RY, RZ │ └──────────┘ └────────────┘ └──────────────┘

Contributing

Contributions are welcome!

1. Fork the repo

2. Create a feature branch

3. Add tests if possible

4. Open a Pull Request

To run tests:

pytest .

Roadmap

v0.2

• Vectorized math ops

• More quantum gates

• More autodiff functions (sin, cos, exp)

• Better GPU switching

v0.3

• Simple neural network module

• Quantum circuit builder API

• Sparse simulator option

v1.0

• Stable public API

• Documentation website

• Interactive notebooks

📄 License

This project is licensed under the MIT License (see LICENSE file).