static-embed 0.1.1

Static sentence embedding via Rust + Candle wrapped for Python

static_embed

static_embed is an educational (but fully operatioal) library and and repository that shows how to use Static Embedding with Rust. Its a library written in Rust (Candle + tokenizers) and exported to Python via PyO3 and maturin.

Features

• Pure-Rust embedding implementation (no Python runtime dependencies at inference time)
• High performance embeddings
• CPU-only, self-contained model weights (downloaded on first use)
• Python bindings that expose an easy-to-use Embedder class

---

Installation (for users)

Prerequisites

• Python ≥ 3.8 (CPython)
• A Rust toolchain (stable) – install with rustup

Quick install into the current virtual env

pip install maturin  # once
# From the repository root
maturin develop --release

This builds the Rust crate as a Python extension and installs it into the environment.

---

Usage

from static_embed import Embedder

# 1. Use the default public model (no args)
embedder = Embedder()

# 2. OR specify your own base-URL that hosts the weights/tokeniser
#    (must contain the same two files: ``model.safetensors`` & ``tokenizer.json``)
# custom_url = "https://my-cdn.example.com/static-retrieval-mrl-en-v1"
# embedder = Embedder(custom_url)

texts = ["Hello world!", "Rust + Python via PyO3"]
embeddings = embedder.embed(texts)

print(len(embeddings), "embeddings", "dimension", len(embeddings[0]))

---

Development workflow

1. Set up a virtual environment

python -m venv .venv && source .venv/bin/activate
pip install maturin pytest

2. Build the extension in editable mode

maturin develop                      # or `--release` for optimised builds

3. Run the Python example

python python_example.py

4. Run the test-suite

pytest -q

---

Project layout

.
├── Cargo.toml         # Rust crate manifest
├── src/               # Rust source (embedder logic + PyO3 bindings)
├── models/            # (Auto-downloaded) model weights live here
├── python_example.py  # Minimal demo script
├── test_embed.py      # Pytest verifying the binding
└── pyproject.toml     # Build configuration for maturin

---

Publishing to PyPI

(Requires an API token in $POETRY_PYPI_TOKEN_PYPI or ~/.pypirc)

maturin build --release --skip-auditwheel  # wheels in target/wheels/
maturin publish --skip-existing           # upload

---

License

MIT © Your Name