barx 0.1.0

Fast, CPU-only AI framework for Python developers

BARX: Fast, CPU-only AI Framework for Python

BARX is a high-performance, CPU-only AI framework that enables Python developers to build, train, and run AI models without GPU requirements.

Key Features

• CPU-only execution: Run models on laptops, edge devices, and servers without GPU dependencies
• High performance: Rust/SIMD kernels for critical operations deliver optimized performance on CPU
• Tensors & Neural Networks: Create and manipulate tensors, build neural networks with standard layers
• Memory efficiency: INT8 quantization for large models keeps memory usage ≤3GB
• Simple API: Clean, intuitive API inspired by popular deep learning frameworks
• Pure Python frontend: Easy to understand and extend with a clean Python interface

Installation

pip install barx

Quick Start

# Create and manipulate tensors
from barx.tensor import T
x = T.randn(32, 128)
y = x.dot(x.T)
print(y.mean())

# Define a neural network
from barx.nn import Linear, ReLU, Softmax, Sequential
model = Sequential(
    Linear(128, 64), ReLU(),
    Linear(64, 10), Softmax()
)

# Train with automatic differentiation
from barx.optim import SGD
optimizer = SGD(model.parameters(), lr=0.01)
for epoch in range(5):
    for x, y in data_loader:
        pred = model(x)
        loss = ((pred - y)**2).mean()
        loss.backward()
        optimizer.step()
        optimizer.zero_grad()

Architecture

BARX is designed for simplicity and efficiency:

• Pure-Python frontend with clear abstractions for tensors and neural networks
• Rust backends for compute-intensive operations via PyO3 bindings
• NumPy fallback ensures all operations work even without Rust kernels
• Automatic differentiation for training neural networks from scratch
• Efficient memory usage with INT8 quantization support
• Multi-threading for operations on large tensors

Examples

The examples/ directory contains sample code demonstrating various features:

• Basic tensor operations
• Neural network training and inference
• INT8 quantization for large models

Comparison with Other Frameworks

Feature	BARX	NumPy	PyTorch	TensorFlow
GPU Support	❌	❌	✅	✅
CPU Performance	✅	⚠️	✅	✅
Easy Installation	✅	✅	⚠️	⚠️
Small Footprint	✅	✅	❌	❌
Autograd	✅	❌	✅	✅
Edge Deployment	✅	⚠️	⚠️	⚠️

Contributing

Contributions are welcome! Feel free to:

1. Report bugs or request features via issues
2. Submit pull requests with improvements
3. Help with documentation or examples
4. Share your experience using BARX

License

MIT License