Along the Axon - applying machine learning capabilities in Elixir.

This blog focuses on Elixir's integrated numerical computing capabilities, Nx, and the corresponding higher level API, Axon.

Bumblebee is a truly powerful capability that is unique to Elixir. I don't know of any other, non-Python, language that can load directly from Hugging Face (HF) configuration data. Bumblebee has code that corresponds to code in the HF Transformer library. The Nx/Axon core team spent the time to map the concepts of the HF Transformer library into the Bumblebee library. There are opportunities to improve Bumblebee by adding new model architectures from HF Transformers into Bumblebee.

Another area of focus for the blog is applying AI models. Like many people, I tend to focus on applying AI models.

What are the advantages of using LiveView to build an application with an AI capability.
What would happen if we created a unique retrieval index using self-supervised training?
Would an off-the-shelf generic index work better?

There are plenty of opportunities to contribute to Elixir's machine learning capabilties.

Latest blog posts

Quick lessons learned trying out SmolVLM in LiveBook

12 March 2025

Capturing the lessons learned in running models on a Linux LiveBook server.

What is FLAME?

31 January 2025

Gathering links about the FLAME architecture pattern for folks that aren't familiar with Elixir

Chatting with an App: Boat Design - Introduction

23 January 2025

Introduction to chatting with an application. Using a boat design application, what would chatting with an application look like?