shrew-python 0.1.0

Python bindings for the Shrew deep learning library

shrew-python

Python bindings for Shrew, a modular deep learning framework written in Rust.

Features

• High Performance: Native Rust implementation with minimal Python overhead.
• GPU Support: CUDA acceleration via shrew-cuda (if valid CUDA toolkit is present).
• Declarative Models: Full support for Shrew's .sw intermediate representation.
• Autograd: Reverse-mode automatic differentiation.
• Interoperability: Zero-copy tensor conversion from/to NumPy.

Installation

pip install shrew-python

Usage

import shrew_python as shrew

# Create tensors
x = shrew.tensor([1.0, 2.0, 3.0])
y = shrew.tensor([4.0, 5.0, 6.0])

# Operations
z = x + y
print(z)  # Tensor([5.0, 7.0, 9.0], dtype=F64, dev=Cpu)

# Load a .sw model
executor = shrew.Executor.load("my_model.sw")
result = executor.run("forward", {"input": x})

Building from Source

Requires Rust and Maturin.

git clone https://github.com/ginozza/shrew
cd shrew
maturin develop --manifest-path crates/shrew-python/Cargo.toml --release

License

Apache-2.0