You’re going to see a significant change in our Pixi / Conda packages in the latest nightlies and next stable release: Mojo is going to be available in its own package! This package includes everything needed for doing Mojo GPU programming. In the latest nightlies, it also includes the Python support necessary for Python → Mojo interoperability.
The Pixi / Conda packaging breakdown is as follows:
-
mojo-compiler: everything you need to build and run Mojo projects -
mojo: depends onmojo-compilerand adds the LSP, debugger, and formatter -
max: depends onmojo-compilerand adds the graph compiler and all of our MAX Python APIs -
modular: an umbrella package that includes themaxpackage, the fullmojopackage (LSP, debugger, formatter), themaxCLI command, and all dependencies necessary to serve MAX AI models
Important: to shrink the size of the max package when used for model development and deployment, we’re unbundling the Mojo LSP, debugger, and formatter into the mojo package. This means that if you used the previous max package for Mojo development in your project, you may start seeing errors about these components being missing, along with a note to install the mojo package. Apologies for that inconvenience, but through this and other package size reductions we’ve been able to almost halve the already-small Docker image size for a full MAX deployment.
For a purely Mojo project, you can migrate by changing your dependency on max to mojo. If you are working on a MAX project with Mojo code, add mojo as a dependency alongside max. Once you do so, the Mojo developer tools should be installed in your project.
For now, this only applies to Pixi / Conda packaging, but we are working to provide these finer-grained packages in PyPI soon. In the latest nightly Conda packages, the module support for Python to Mojo interoperability has moved from max to mojo.
We know that this has been one of our top requests from Mojo developers, and we hope that this makes it even easier to get started with Mojo development!