FluxML/Flux.jl

Built entirely in Julia, Flux leverages the language's native automatic differentiation and GPU support through a lightweight abstraction layer, allowing models to be expressed as simple Julia functions or composed layers. It supports flexible model definitions—from hand-written closures to structured layer chains—with built-in optimizers and training utilities. The library integrates seamlessly with Julia's scientific computing ecosystem and supports GPU acceleration via standard Julia backends.

4,711 stars. Actively maintained with 11 commits in the last 30 days.