yng87/lux.jl

Explicitly Parameterized Neural Networks in Julia

Lux 🔥

The 🔥 Deep Learning Framework

Installation

] add Lux

Getting Started

using Lux, Random, Optimisers, Zygote

# Seeding
rng = Random.default_rng()
Random.seed!(rng, 0)

# Construct the layer
model = Chain(BatchNorm(128), Dense(128, 256, tanh), BatchNorm(256),
              Chain(Dense(256, 1, tanh), Dense(1, 10)))

# Parameter and State Variables
ps, st = Lux.setup(rng, model) .|> gpu

# Dummy Input
x = rand(rng, Float32, 128, 2) |> gpu

# Run the model
y, st = Lux.apply(model, x, ps, st)

# Gradients
gs = gradient(p -> sum(Lux.apply(model, x, p, st)[1]), ps)[1]

# Optimization
st_opt = Optimisers.setup(Optimisers.Adam(0.0001), ps)
st_opt, ps = Optimisers.update(st_opt, ps, gs)

Examples

Look in the examples directory for self-contained usage examples. The documentation has examples sorted into proper categories.

Ecosystem

Checkout our Ecosystem page for more details.

Getting Help

For usage related questions, please use Github Discussions or JuliaLang Discourse (machine learning domain) which allows questions and answers to be indexed. To report bugs use github issues or even better send in a pull request.

Package Ecosystem Structure

Structure of the packages part of the Lux.jl Universe¹: (Rounded Rectangles denote packages maintained by Lux.jl developers)

flowchart LR
    subgraph Interface
        LuxCore(LuxCore)
    end
    subgraph Backend
        LuxLib(LuxLib)
        NNlib
        CUDA
    end
    subgraph InterOp
        Flux2Lux(Flux2Lux)
    end
    subgraph ExternalML[External ML Packages]
        Flux
        Metalhead
    end
    subgraph CompViz[Computer Vision]
        Boltz(Boltz)
    end
    subgraph SciML[Scientific Machine Learning]
        DeepEquilibriumNetworks(DeepEquilibriumNetworks)
        DiffEqFlux(DiffEqFlux)
        NeuralPDE[Neural PDE: PINNs]
    end
    subgraph AD[Automatic Differentiation]
        Zygote
        Enzyme["Enzyme (experimental)"]
    end
    subgraph Dist[Distributed Training]
        FluxMPI(FluxMPI)
    end
    subgraph SerializeModels[Serialize Models]
        Serial[Serialization]
        JLD2
        BSON
    end
    subgraph Opt[Optimization]
        Optimisers
        Optimization
    end
    subgraph Parameters
        ComponentArrays
    end
    Lux(Lux)
    Parameters --> Lux
    LuxCore --> Lux
    Backend --> Lux
    Lux --> SciML
    AD --> Lux
    Lux --> Dist
    Lux --> SerializeModels
    Lux --> Opt
    Lux --> CompViz
    ExternalML -.-> CompViz
    Lux --> InterOp
    ExternalML -.-> InterOp

Loading

Related Projects

• Flux.jl -- We share most of the backend infrastructure with Flux (Roadmap hints towards making Flux explicit-parameter first)
• Knet.jl -- One of the mature and OG Julia Deep Learning Frameworks
• SimpleChains.jl -- Extremely Efficient for Small Neural Networks on CPU
• Avalon.jl -- Uses tracing based AD Yota.jl

Citation

If you found this library to be useful in academic work, then please cite:

@software{pal2022lux,
    author = {Pal, Avik},
    title = {{L}ux: Explicit Parameterization of Deep Neural Networks in Julia},
    year = {2022},
    publisher = {GitHub},
    journal = {GitHub repository},
    howpublished = {\url{https://github.com/avik-pal/Lux.jl/}}
}

Also consider starring our github repo

Footnotes

1. These packages only constitute a subset of the ecosystem. Specifically these are the packages which the maintainers of Lux.jl have personally tested out. If you want a new package to be listed here, please open an issue. ↩