We’re excited to announce that Magic 🪄 is now the official package and virtual environment manager for MAX and Mojo projects.
Magic builds on the Conda package management system, which we chose for its wide adoption and interoperability. We wanted to make it easier to manage virtual environments to avoid packaging conflicts and improve code reproducibility. magic
gives us a way to deliver MAX and Mojo in a single package, allowing you to instantly launch code examples and easily create new projects.
Starting on September 14, MAX and Mojo packages will be distributed as a single Conda package called max
, available in both stable and nightly builds. While the modular
CLI will still be available, it will not receive any new updates, and we strongly encourage you to migrate to Magic or any other tool that supports Conda package management.
Going forward, we recommend you manage your MAX and Mojo installations using the magic
CLI. It brings a number of improvements specific to Modular pipelines and virtual environments, and we will be adding new features for managing, developing, and deploying pipelines to future magic
releases. The current stable release of magic
is 0.2.3 🚀
You can get started and learn more on our new Magic docs page. A huge thanks to everyone in the community for your feedback and support! Please share questions and feedback in the #magic-cli channel in our Discord server.
Blogs, Tutorials, and Videos
- Bring your own fine-tuned models to MAX pipelines with our latest tutorial.
- Learn about the updated license for Mojo and MAX with our recent video and share questions and feedback in the #licensing channel in our Discord server.
- Chris Lattner spoke with Richard Feldman about Mojo and MAX on the 100th episode of the Software Unscripted podcast.
- Chris Lattner was interviewed by Darrel Etherington on all things Modular for his newly relaunched publication The Angle.
- Maxim Zaks created a video deep dive on CrazyString, a string data structure with small string optimization and code point indexing.
- Anshul Vipat wrote about Mojo for Analytics India Magazine.
Awesome Mojo
Want to learn more about these projects? Chat with their creators in the #community-showcase channel of our Discord server.
- Helehex built a falling sand game in Mojo called Dustbin.
- DobyDabaDu implemented machine learning algorithms from scratch in pure Mojo, in a project called Mojmelo.
- Phomola’s new project, mojolibs, provides a Mojo tokenzier, CSV parser, JSON parser, and more.
- mad alex 1997 released version 0.2 of NuMojo, a library for numerical computing in Mojo (similar to Python’s NumPy). V0.2 includes simplified data type management, Rust-like data type aliases, improved array operations, and updated documentation.
Open-Source Contributions
If you’ve recently had your first PR merged, message Caroline Frasca (@Caroline_Frasca) on Discord to claim your epic Mojo swag!
Check out the recently merged contributions from our valuable community members:
Coming Up
MAX + Mojo Community Meeting
Don’t miss the next MAX + Mojo Community Meeting on September 9th at 10am PST. We’ll hear from Gabriel De Marmiesse on SIMD in complex algorithms and Maxim Zaks on hash functions. Add the schedule of community meetings to your calendar so you can stay up-to-date.
PyTorch Conference
Modular Software Engineer Mikhail Zolutukhin will give a lightning talk at the PyTorch Conference on September 18th at 4:40pm PST, which will focus on how Modular’s inference engine works with PyTorch and defining performant kernels with Mojo.
Modular Engineering Director Nick Kreeger and Software Engineer Jack Clayton will also have a poster presentation on MAX and Mojo. Stop by their poster on September 18th from 5:25pm to 8:25pm PST to say hi! 👋