mplang-nightly 0.1.dev180

Multi-Party Programming Language

MPLang: A Programming Language for Multi-Party Computation

MPLang is a Python-native library for building and executing multi-party and multi-device programs. It simplifies secure computation by allowing developers to write a single program that orchestrates multiple parties in a synchronous, SPMD (Single Program, Multiple Data) fashion.

Features

• Single-Controller SPMD: Write one program that runs across multiple parties in lockstep.
• Explicit Device Placement: Clearly annotate and control where data lives and computation happens (e.g., on party P0, P1, or a secure SPU).
• Function-Level Compilation: Use the @mplang.function decorator to compile Python functions into an auditable, optimizable graph representation.
• Pluggable Architecture: Easily extend MPLang with new frontends (like JAX, Ibis) and backends (like StableHLO, SPU).

Getting Started

Installation

You'll need a modern Python environment (3.10+). We recommend using uv for fast installation.

# Install uv (if not already installed)
curl -LsSf https://astral.sh/uv/install.sh | sh

# Install MPLang from PyPI
uv pip install mplang

Quick Example

Here's a taste of what MPLang looks like. This example shows a "millionaire's problem" where two parties compare their wealth without revealing it.

import mplang
import mplang.device as mpd
from numpy.random import randint

# Use a decorator to compile this function for multi-party execution
@mplang.function
def millionaire():
    # Alice's value, placed on device P0
    x = mpd.device("P0")(randint)(0, 1000000)
    # Bob's value, placed on device P1
    y = mpd.device("P1")(randint)(0, 1000000)
    # The comparison happens on a secure device (SPU)
    z = mpd.device("SPU")(lambda a, b: a < b)(x, y)
    return z

# Set up a local simulator with 2 parties
sim = mplang.Simulator.simple(2)

# Evaluate the compiled function
result = mplang.evaluate(sim, millionaire)

# Securely fetch the result (reveals SPU value)
print("Is Alice poorer than Bob?", mpd.fetch(sim, result))

Learn More

• Tutorials: Check out the tutorials/ directory for in-depth, runnable examples covering conditions, loops, and more.
• Contributing: We welcome contributions! See our Contributing Guide to get started with the development setup.

License

MPLang is licensed under the Apache 2.0 License.