Where to Get Help and Connect with the Flux Community

Updated

Flux directs users to four public channels: the official Julia Slack (channels #flux-bridged and #machine-learning), the Julia Zulip server (#ml-contributors and #machine-learning), the Julia Discourse forum's machine learning section, and Stack Overflow for Flux.jl questions. Code contributions go through the project's GitHub repository. Which channel fits depends on whether you want quick chat, design discussion, searchable Q&A, or to file and fix issues.

Channel-by-Channel Guide

Channel Where to go Best for Notes from the Flux site
Slack Official Julia Slack, channels #flux-bridged and #machine-learning Casual, real-time conversation Described as the place for "casual conversation"
Zulip Julia programming language Zulip server, channels #ml-contributors and #machine-learning Threaded discussion, contributor coordination #ml-contributors suggests a contributor-focused stream
Discourse Julia Discourse forum, machine learning category Longer-form questions and answers that others can find later Listed as "Machine Learning in Julia community"
Stack Overflow Questions tagged for Flux.jl Specific, answerable technical problems The site points here for "questions about Flux.jl"
GitHub FluxML repository Bug reports, feature work, pull requests The site invites you to "help us by contributing code"

How to Choose

  • Stuck on an error right now: start with Slack #machine-learning for a quick back-and-forth.
  • Question likely to help others later: post on Discourse or Stack Overflow so the answer stays searchable.
  • Want to shape the library or discuss a change: use Zulip #ml-contributors, where contributor discussion is concentrated.
  • Found a reproducible bug or have a patch: open an issue or pull request on GitHub.

Before You Ask

Flux is a pure-Julia stack, so most problems involve Julia itself as much as Flux. When posting, include:

  1. The Flux and Julia versions you are running.
  2. A minimal code snippet that reproduces the problem.
  3. The full error output, including the stack trace.
  4. Whether you are on CPU or GPU (Flux builds on Julia's native GPU support via CUDA.jl).

A short, runnable example gets answers far faster than a description of the code.

Contributing Code

The site's contribution path is the GitHub repository. Because Flux is designed to be "fully hackable" — from GPU kernels to custom gradients and layers — contributions are not limited to bug fixes; custom layers, GPU code, and documentation are all in scope. The project is maintained by a community team under a published governance model, so reading that model first helps you understand how decisions are made and where your change should be discussed.

Related Places to Look

Some questions are better answered by the surrounding ecosystem than by Flux itself. If your issue involves probabilistic models, graph neural networks, computer vision, scientific machine learning, or transformers, the relevant package (Turing.jl/Stheno.jl, GraphNeuralNetworks.jl, Metalhead.jl, the SciML ecosystem, or Transformers.jl) may have its own community channels. Checking there first can save a round trip.