FluxML/Zygote.jl

Health v2 · maintenance only

Flags: no_license

Adoption not part of the score

1568 stars · 221 forks observed · 2026-08-28

What it is AI-extracted, prompt v1, taxonomy v1, 2026-08-30, confidence not recorded

Zygote.jl is a source-to-source automatic differentiation library for Julia that hooks into the Julia compiler to generate gradient (backwards pass) code automatically. It is the AD engine behind the Flux machine learning framework and supports control flow, recursion, closures, and custom gradient rules via ChainRules.

Use cases

• compute gradients of Julia functions for machine learning
• differentiate a neural network model with respect to its parameters
• train models with Flux using reverse-mode AD
• define custom gradient rules with ChainRulesCore rrule
• differentiate code containing control flow, recursion, and closures
• generate efficient backwards passes at compile time

When to choose

• you are training or building models in Flux.jl
• you need reverse-mode automatic differentiation in Julia with dynamic control flow
• you want gradients over nested parameter structures
• you need compile-time generated backwards passes for performance

When to avoid

• your code relies on array mutation or exception handling, which are unsupported
• you need forward-mode differentiation only (consider ForwardDiff.jl)
• you are not working in the Julia ecosystem

Facets

library · maturity active

machine-learning compiler machine-learning deep-learning jvm cross-platform automatic-differentiation julia source-to-source gradients fluxml differentiable-programming algorithms

Member repositories

For agents

markdown · JSON · MCP: product_card(name="FluxML/Zygote.jl")