Where did parallelize go :)?

Mojo Mojo 1.0 Beta
1

In Mojo 1.0.0b3.dev2026072906 (8235c7f9) We do not have parallelize hein? I guess I assumed once it was in 1.0.b2` it would stay there.

Seems like Mr. Lattner and the people at Modular are working on Concurrency Support as we speak.

2

It’s now under max.algorithm so after pip install max change std -> max. Here is the earlier announcement.

2 Likes
3

Thank you so much, I saw it was in Max but in my head there is a distinction between the inference engine Max and the Language Mojo still so I thought it might be some sort of “Max” parallelize instead.

Just to be clear is the concurrency model under development because in the Roadmap it was in Phase 2 I believe?

4

A robust async model is high on our list for big new features to work on, post-1.0, but we need to get through 1.0 first. We’ll talk more about our plans for near-term new features at and after ModCon.

3 Likes
5

Let me clarify what MAX is. MAX has

  • a serving component through max serve CLI when you pip install max[serve]. Think about vLLM/SGlang
  • model authoring APIs. Think about PyTorch.
  • Accelerator library where the API was looking for and more general what used to be called std.gpu in Mojo, is now belonged in.

All of them are part of the unified umbrella to make AI work across various hardwares whereas Mojo is a general purpose system programming language powering MAX.

2 Likes
6

Thank you both for clarifying.

You have a big fan at ECCC and although there’s not a clear market for geospatial AI/ML I am pushing hard to get Mojo in our tech stack because the tech is so cool.

Whish I could be at ModCon.