Skip to content

MARIMO_ANNOUNCEMENT.md

Latest commit

 

History

History
130 lines (91 loc) · 4.02 KB

MARIMO_ANNOUNCEMENT.md

File metadata and controls

130 lines (91 loc) · 4.02 KB

mojo-marimo: High-Performance Mojo Execution in marimo Notebooks 🔥

Status: Work in progress - Released for community feedback and testing
Version: v0.1.0 (Beta)

Hi marimo community! I built mojo-marimo to bring high-performance Mojo code execution to marimo notebooks.

This is an early release - feedback on the approach and API design is very welcome!

What is Mojo?

Mojo is a systems programming language from Modular that combines Python's ease-of-use with C-level performance, designed for AI/ML infrastructure.

What does mojo-marimo do?

It lets you write and execute Mojo code directly in marimo notebooks with three integration patterns:

1. Decorator Pattern - Clean & Pythonic

Mojo code lives in docstrings with {{param}} template syntax:

from mojo_marimo import mojo

@mojo
def fibonacci(n: int) -> int:
    """[Mojo code here with {{n}} placeholder]"""
    ...

result = fibonacci(10)  # Called like normal Python

2. Executor Pattern - Dynamic Code Generation

Pass Mojo code as strings for runtime flexibility:

from mojo_marimo import run_mojo

mojo_code = f"""[Generated Mojo code]"""
result = run_mojo(mojo_code)

3. Extension Modules - Zero Overhead

Compile to .so files for direct Python imports:

import mojo.importer
import my_mojo_ext

result = my_mojo_ext.compute(100)  # Direct call, no subprocess

Trade-off: More complex Mojo code (requires PythonModuleBuilder)

See working examples in the repository notebooks.

Why marimo?

marimo's reactive execution model pairs perfectly with Mojo:

  • ✅ Reactive sliders/controls trigger Mojo recompilation and execution
  • ✅ Smart caching - Same code reuses cached binaries
  • ✅ Pure Python files - Everything is .py, git-friendly
  • ✅ Visual feedback - Interactive demos with real performance

Use Cases

  • Algorithm prototyping with interactive parameters
  • Side-by-side Python vs Mojo performance comparisons
  • Educational demos (SIMD, compiler behaviour, optimization)
  • Exploring Mandelbrot fractals, Monte Carlo simulations, numerical methods

Features

✅ Real Mojo compilation and execution
✅ Binary caching (SHA256-based)
✅ Three integration patterns
✅ Pre-compilation validation with helpful hints
✅ Example notebooks: Fibonacci, Monte Carlo π, Mandelbrot fractals

Getting Started

Installation:

# Requires Mojo: pip install mojo

git clone https://github.com/DataBooth/mojo-marimo
cd mojo-marimo

# Install with uv (recommended)
uv sync --extra dev

# Or with pixi
pixi install

Try the notebooks:

just notebook-decorator        # Decorator examples
just notebook-mc-extension     # Monte Carlo π estimation  
just notebook-mandelbrot-extension  # Mandelbrot fractals

# Or directly with marimo
marimo edit notebooks/pattern_decorator.py
marimo edit notebooks/monte_carlo_extension.py

How It Works

Mojo code compiles to binaries on first execution and caches them (SHA256 hash). Subsequent calls reuse the cached binary. The @mojo decorator extracts Mojo from docstrings and handles {{param}} template substitution.

Repository

🔗 https://github.com/DataBooth/mojo-marimo

  • 44 passing tests (75% coverage)
  • Interactive example notebooks (Fibonacci, Monte Carlo, Mandelbrot)
  • Complete documentation
  • Supports uv and pixi

🎯 Feedback Welcome!

This is an early release - I'd love your input on:

  • Pattern preferences: Which approach works best for your workflows?
  • Use cases: What problems would you solve with this?
  • marimo integration: How well does this fit the reactive model?
  • API ergonomics: Is the {{param}} syntax intuitive?

All feedback welcome - issues, PRs, or comments! Try it out and let me know what you think! 🚀

Tech Stack: Mojo 0.25.7, marimo 0.19.4+, Python 3.12-3.14
License: Apache-2.0
Tested on: macOS (Apple Silicon), Python 3.12-3.14, Mojo 0.25.7