flux.jl

Flux is a machine learning library written entirely in Julia. It is described as a 100% pure-Julia stack that provides lightweight abstractions on top of Julia's native GPU and automatic differentiation support, aiming to make simple tasks easy while remaining fully hackable. Key characteristics evidenced by the README: - Pure-Julia implementation, with no non-Julia core dependencies implied by the project's own description. - Works best with Julia 1.10 or later. - In version 0.15, almost any parameterised function in Julia can act as a Flux model, including closures over parameters. The README shows a closure over weight, bias and output vectors as a valid model, and notes the same function can be expressed with built-in layers such as Chain, Dense and vcat. - Provides optimiser setup and training helpers, illustrated by Flux.setup with Adam and Flux.train! in a short example. - Ships documentation with stable and dev versions, a quickstart guide, and a separate model zoo repository of examples. - Has community support channels via the Julia discourse and Slack. - Is citable in research, with a CITATION.bib file and a JOSS paper DOI. The README includes a compact example that fits a small function from generated data using a hand-written parameterised closure, an Adam optimiser state, and a training loop, then plots truth versus learned output. It also points readers to the quickstart page for a longer example. This overview is based solely on the repository README and description; no benchmarks, rankings or performance claims are made here.