Training a HyperNetwork on MNIST and FashionMNIST

Package Imports

using Lux, ADTypes, ComponentArrays, LuxCUDA, MLDatasets, MLUtils, OneHotArrays, Optimisers,
      Printf, Random, Setfield, Statistics, Zygote

CUDA.allowscalar(false)

Loading Datasets

function load_dataset(::Type{dset}, n_train::Int, n_eval::Int, batchsize::Int) where {dset}
    imgs, labels = dset(:train)[1:n_train]
    x_train, y_train = reshape(imgs, 28, 28, 1, n_train), onehotbatch(labels, 0:9)

    imgs, labels = dset(:test)[1:n_eval]
    x_test, y_test = reshape(imgs, 28, 28, 1, n_eval), onehotbatch(labels, 0:9)

    return (
        DataLoader((x_train, y_train); batchsize=min(batchsize, n_train), shuffle=true),
        DataLoader((x_test, y_test); batchsize=min(batchsize, n_eval), shuffle=false))
end

function load_datasets(n_train=1024, n_eval=32, batchsize=256)
    return load_dataset.((MNIST, FashionMNIST), n_train, n_eval, batchsize)
end

load_datasets (generic function with 4 methods)

Implement a HyperNet Layer

function HyperNet(
        weight_generator::Lux.AbstractLuxLayer, core_network::Lux.AbstractLuxLayer)
    ca_axes = Lux.initialparameters(Random.default_rng(), core_network) |>
              ComponentArray |>
              getaxes
    return @compact(; ca_axes, weight_generator, core_network, dispatch=:HyperNet) do (x, y)
        # Generate the weights
        ps_new = ComponentArray(vec(weight_generator(x)), ca_axes)
        @return core_network(y, ps_new)
    end
end

HyperNet (generic function with 1 method)

Defining functions on the CompactLuxLayer requires some understanding of how the layer is structured, as such we don't recommend doing it unless you are familiar with the internals. In this case, we simply write it to ignore the initialization of the core_network parameters.

function Lux.initialparameters(rng::AbstractRNG, hn::CompactLuxLayer{:HyperNet})
    return (; weight_generator=Lux.initialparameters(rng, hn.layers.weight_generator),)
end

Create and Initialize the HyperNet

function create_model()
    # Doesn't need to be a MLP can have any Lux Layer
    core_network = Chain(FlattenLayer(), Dense(784, 256, relu), Dense(256, 10))
    weight_generator = Chain(Embedding(2 => 32), Dense(32, 64, relu),
        Dense(64, Lux.parameterlength(core_network)))

    model = HyperNet(weight_generator, core_network)
    return model
end

create_model (generic function with 1 method)

Define Utility Functions

const loss = CrossEntropyLoss(; logits=Val(true))

function accuracy(model, ps, st, dataloader, data_idx)
    total_correct, total = 0, 0
    st = Lux.testmode(st)
    for (x, y) in dataloader
        target_class = onecold(y)
        predicted_class = onecold(first(model((data_idx, x), ps, st)))
        total_correct += sum(target_class .== predicted_class)
        total += length(target_class)
    end
    return total_correct / total
end

accuracy (generic function with 1 method)

Training

function train()
    model = create_model()
    dataloaders = load_datasets()

    dev = gpu_device()
    rng = Xoshiro(0)
    ps, st = Lux.setup(rng, model) |> dev

    train_state = Training.TrainState(model, ps, st, Adam(0.001f0))

    ### Lets train the model
    nepochs = 50
    for epoch in 1:nepochs, data_idx in 1:2
        train_dataloader, test_dataloader = dataloaders[data_idx] .|> dev

        stime = time()
        for (x, y) in train_dataloader
            (_, _, _, train_state) = Training.single_train_step!(
                AutoZygote(), loss, ((data_idx, x), y), train_state)
        end
        ttime = time() - stime

        train_acc = round(
            accuracy(model, train_state.parameters,
                train_state.states, train_dataloader, data_idx) * 100;
            digits=2)
        test_acc = round(
            accuracy(model, train_state.parameters,
                train_state.states, test_dataloader, data_idx) * 100;
            digits=2)

        data_name = data_idx == 1 ? "MNIST" : "FashionMNIST"

        @printf "[%3d/%3d]\t%12s\tTime %3.5fs\tTraining Accuracy: %3.2f%%\tTest \
                 Accuracy: %3.2f%%\n" epoch nepochs data_name ttime train_acc test_acc
    end

    println()

    test_acc_list = [0.0, 0.0]
    for data_idx in 1:2
        train_dataloader, test_dataloader = dataloaders[data_idx] .|> dev
        train_acc = round(
            accuracy(model, train_state.parameters,
                train_state.states, train_dataloader, data_idx) * 100;
            digits=2)
        test_acc = round(
            accuracy(model, train_state.parameters,
                train_state.states, test_dataloader, data_idx) * 100;
            digits=2)

        data_name = data_idx == 1 ? "MNIST" : "FashionMNIST"

        @printf "[FINAL]\t%12s\tTraining Accuracy: %3.2f%%\tTest Accuracy: \
                 %3.2f%%\n" data_name train_acc test_acc
        test_acc_list[data_idx] = test_acc
    end
    return test_acc_list
end

test_acc_list = train()

[  1/ 50]	       MNIST	Time 69.55687s	Training Accuracy: 60.06%	Test Accuracy: 59.38%
[  1/ 50]	FashionMNIST	Time 0.02189s	Training Accuracy: 53.03%	Test Accuracy: 50.00%
[  2/ 50]	       MNIST	Time 0.02383s	Training Accuracy: 65.23%	Test Accuracy: 59.38%
[  2/ 50]	FashionMNIST	Time 0.02378s	Training Accuracy: 57.42%	Test Accuracy: 56.25%
[  3/ 50]	       MNIST	Time 0.02428s	Training Accuracy: 75.29%	Test Accuracy: 59.38%
[  3/ 50]	FashionMNIST	Time 0.02304s	Training Accuracy: 64.75%	Test Accuracy: 56.25%
[  4/ 50]	       MNIST	Time 0.02243s	Training Accuracy: 79.98%	Test Accuracy: 65.62%
[  4/ 50]	FashionMNIST	Time 0.02277s	Training Accuracy: 59.86%	Test Accuracy: 62.50%
[  5/ 50]	       MNIST	Time 0.02267s	Training Accuracy: 82.42%	Test Accuracy: 68.75%
[  5/ 50]	FashionMNIST	Time 0.05914s	Training Accuracy: 71.68%	Test Accuracy: 71.88%
[  6/ 50]	       MNIST	Time 0.02185s	Training Accuracy: 86.72%	Test Accuracy: 75.00%
[  6/ 50]	FashionMNIST	Time 0.02095s	Training Accuracy: 75.39%	Test Accuracy: 62.50%
[  7/ 50]	       MNIST	Time 0.02061s	Training Accuracy: 87.99%	Test Accuracy: 81.25%
[  7/ 50]	FashionMNIST	Time 0.02085s	Training Accuracy: 73.63%	Test Accuracy: 62.50%
[  8/ 50]	       MNIST	Time 0.02074s	Training Accuracy: 89.36%	Test Accuracy: 78.12%
[  8/ 50]	FashionMNIST	Time 0.02045s	Training Accuracy: 75.20%	Test Accuracy: 68.75%
[  9/ 50]	       MNIST	Time 0.02115s	Training Accuracy: 93.16%	Test Accuracy: 84.38%
[  9/ 50]	FashionMNIST	Time 0.02190s	Training Accuracy: 77.73%	Test Accuracy: 71.88%
[ 10/ 50]	       MNIST	Time 0.02052s	Training Accuracy: 95.02%	Test Accuracy: 84.38%
[ 10/ 50]	FashionMNIST	Time 0.02083s	Training Accuracy: 80.37%	Test Accuracy: 71.88%
[ 11/ 50]	       MNIST	Time 0.02098s	Training Accuracy: 96.19%	Test Accuracy: 84.38%
[ 11/ 50]	FashionMNIST	Time 0.02052s	Training Accuracy: 80.57%	Test Accuracy: 75.00%
[ 12/ 50]	       MNIST	Time 0.02090s	Training Accuracy: 97.75%	Test Accuracy: 84.38%
[ 12/ 50]	FashionMNIST	Time 0.02112s	Training Accuracy: 83.20%	Test Accuracy: 78.12%
[ 13/ 50]	       MNIST	Time 0.02071s	Training Accuracy: 98.05%	Test Accuracy: 81.25%
[ 13/ 50]	FashionMNIST	Time 0.02096s	Training Accuracy: 82.71%	Test Accuracy: 75.00%
[ 14/ 50]	       MNIST	Time 0.02879s	Training Accuracy: 99.22%	Test Accuracy: 78.12%
[ 14/ 50]	FashionMNIST	Time 0.02107s	Training Accuracy: 83.79%	Test Accuracy: 68.75%
[ 15/ 50]	       MNIST	Time 0.02113s	Training Accuracy: 99.41%	Test Accuracy: 81.25%
[ 15/ 50]	FashionMNIST	Time 0.02070s	Training Accuracy: 84.96%	Test Accuracy: 68.75%
[ 16/ 50]	       MNIST	Time 0.02017s	Training Accuracy: 99.51%	Test Accuracy: 81.25%
[ 16/ 50]	FashionMNIST	Time 0.02235s	Training Accuracy: 85.94%	Test Accuracy: 65.62%
[ 17/ 50]	       MNIST	Time 0.02146s	Training Accuracy: 99.71%	Test Accuracy: 81.25%
[ 17/ 50]	FashionMNIST	Time 0.02093s	Training Accuracy: 86.04%	Test Accuracy: 68.75%
[ 18/ 50]	       MNIST	Time 0.02058s	Training Accuracy: 99.80%	Test Accuracy: 81.25%
[ 18/ 50]	FashionMNIST	Time 0.02758s	Training Accuracy: 87.79%	Test Accuracy: 71.88%
[ 19/ 50]	       MNIST	Time 0.02107s	Training Accuracy: 100.00%	Test Accuracy: 81.25%
[ 19/ 50]	FashionMNIST	Time 0.02044s	Training Accuracy: 88.77%	Test Accuracy: 68.75%
[ 20/ 50]	       MNIST	Time 0.02072s	Training Accuracy: 100.00%	Test Accuracy: 81.25%
[ 20/ 50]	FashionMNIST	Time 0.02261s	Training Accuracy: 89.26%	Test Accuracy: 75.00%
[ 21/ 50]	       MNIST	Time 0.02064s	Training Accuracy: 100.00%	Test Accuracy: 81.25%
[ 21/ 50]	FashionMNIST	Time 0.02116s	Training Accuracy: 88.96%	Test Accuracy: 75.00%
[ 22/ 50]	       MNIST	Time 0.02121s	Training Accuracy: 100.00%	Test Accuracy: 81.25%
[ 22/ 50]	FashionMNIST	Time 0.02070s	Training Accuracy: 89.55%	Test Accuracy: 75.00%
[ 23/ 50]	       MNIST	Time 0.02843s	Training Accuracy: 100.00%	Test Accuracy: 81.25%
[ 23/ 50]	FashionMNIST	Time 0.02109s	Training Accuracy: 88.57%	Test Accuracy: 71.88%
[ 24/ 50]	       MNIST	Time 0.02119s	Training Accuracy: 100.00%	Test Accuracy: 81.25%
[ 24/ 50]	FashionMNIST	Time 0.02109s	Training Accuracy: 90.33%	Test Accuracy: 71.88%
[ 25/ 50]	       MNIST	Time 0.02037s	Training Accuracy: 100.00%	Test Accuracy: 81.25%
[ 25/ 50]	FashionMNIST	Time 0.02032s	Training Accuracy: 90.23%	Test Accuracy: 71.88%
[ 26/ 50]	       MNIST	Time 0.02075s	Training Accuracy: 100.00%	Test Accuracy: 81.25%
[ 26/ 50]	FashionMNIST	Time 0.02086s	Training Accuracy: 91.21%	Test Accuracy: 71.88%
[ 27/ 50]	       MNIST	Time 0.02047s	Training Accuracy: 100.00%	Test Accuracy: 81.25%
[ 27/ 50]	FashionMNIST	Time 0.02619s	Training Accuracy: 90.92%	Test Accuracy: 71.88%
[ 28/ 50]	       MNIST	Time 0.02145s	Training Accuracy: 100.00%	Test Accuracy: 81.25%
[ 28/ 50]	FashionMNIST	Time 0.02185s	Training Accuracy: 91.31%	Test Accuracy: 75.00%
[ 29/ 50]	       MNIST	Time 0.02313s	Training Accuracy: 100.00%	Test Accuracy: 81.25%
[ 29/ 50]	FashionMNIST	Time 0.02258s	Training Accuracy: 92.09%	Test Accuracy: 75.00%
[ 30/ 50]	       MNIST	Time 0.02250s	Training Accuracy: 100.00%	Test Accuracy: 81.25%
[ 30/ 50]	FashionMNIST	Time 0.02208s	Training Accuracy: 92.38%	Test Accuracy: 71.88%
[ 31/ 50]	       MNIST	Time 0.02188s	Training Accuracy: 100.00%	Test Accuracy: 81.25%
[ 31/ 50]	FashionMNIST	Time 0.03016s	Training Accuracy: 93.07%	Test Accuracy: 71.88%
[ 32/ 50]	       MNIST	Time 0.02225s	Training Accuracy: 100.00%	Test Accuracy: 81.25%
[ 32/ 50]	FashionMNIST	Time 0.02214s	Training Accuracy: 93.75%	Test Accuracy: 71.88%
[ 33/ 50]	       MNIST	Time 0.02163s	Training Accuracy: 100.00%	Test Accuracy: 81.25%
[ 33/ 50]	FashionMNIST	Time 0.02214s	Training Accuracy: 93.36%	Test Accuracy: 75.00%
[ 34/ 50]	       MNIST	Time 0.02208s	Training Accuracy: 100.00%	Test Accuracy: 81.25%
[ 34/ 50]	FashionMNIST	Time 0.02359s	Training Accuracy: 94.04%	Test Accuracy: 75.00%
[ 35/ 50]	       MNIST	Time 0.02124s	Training Accuracy: 100.00%	Test Accuracy: 81.25%
[ 35/ 50]	FashionMNIST	Time 0.02145s	Training Accuracy: 94.43%	Test Accuracy: 75.00%
[ 36/ 50]	       MNIST	Time 0.02950s	Training Accuracy: 100.00%	Test Accuracy: 81.25%
[ 36/ 50]	FashionMNIST	Time 0.02093s	Training Accuracy: 94.34%	Test Accuracy: 78.12%
[ 37/ 50]	       MNIST	Time 0.02137s	Training Accuracy: 100.00%	Test Accuracy: 81.25%
[ 37/ 50]	FashionMNIST	Time 0.02169s	Training Accuracy: 94.43%	Test Accuracy: 75.00%
[ 38/ 50]	       MNIST	Time 0.02153s	Training Accuracy: 100.00%	Test Accuracy: 81.25%
[ 38/ 50]	FashionMNIST	Time 0.02083s	Training Accuracy: 94.14%	Test Accuracy: 75.00%
[ 39/ 50]	       MNIST	Time 0.02079s	Training Accuracy: 100.00%	Test Accuracy: 81.25%
[ 39/ 50]	FashionMNIST	Time 0.02058s	Training Accuracy: 94.82%	Test Accuracy: 75.00%
[ 40/ 50]	       MNIST	Time 0.02087s	Training Accuracy: 100.00%	Test Accuracy: 81.25%
[ 40/ 50]	FashionMNIST	Time 0.02789s	Training Accuracy: 95.51%	Test Accuracy: 75.00%
[ 41/ 50]	       MNIST	Time 0.02113s	Training Accuracy: 100.00%	Test Accuracy: 81.25%
[ 41/ 50]	FashionMNIST	Time 0.02380s	Training Accuracy: 95.31%	Test Accuracy: 75.00%
[ 42/ 50]	       MNIST	Time 0.02108s	Training Accuracy: 100.00%	Test Accuracy: 81.25%
[ 42/ 50]	FashionMNIST	Time 0.02216s	Training Accuracy: 95.41%	Test Accuracy: 78.12%
[ 43/ 50]	       MNIST	Time 0.02263s	Training Accuracy: 100.00%	Test Accuracy: 81.25%
[ 43/ 50]	FashionMNIST	Time 0.02486s	Training Accuracy: 95.80%	Test Accuracy: 75.00%
[ 44/ 50]	       MNIST	Time 0.02157s	Training Accuracy: 100.00%	Test Accuracy: 81.25%
[ 44/ 50]	FashionMNIST	Time 0.02187s	Training Accuracy: 95.90%	Test Accuracy: 78.12%
[ 45/ 50]	       MNIST	Time 0.02092s	Training Accuracy: 100.00%	Test Accuracy: 81.25%
[ 45/ 50]	FashionMNIST	Time 0.02178s	Training Accuracy: 95.80%	Test Accuracy: 78.12%
[ 46/ 50]	       MNIST	Time 0.02175s	Training Accuracy: 100.00%	Test Accuracy: 81.25%
[ 46/ 50]	FashionMNIST	Time 0.02131s	Training Accuracy: 96.29%	Test Accuracy: 75.00%
[ 47/ 50]	       MNIST	Time 0.02115s	Training Accuracy: 100.00%	Test Accuracy: 81.25%
[ 47/ 50]	FashionMNIST	Time 0.02080s	Training Accuracy: 95.61%	Test Accuracy: 75.00%
[ 48/ 50]	       MNIST	Time 0.02166s	Training Accuracy: 100.00%	Test Accuracy: 81.25%
[ 48/ 50]	FashionMNIST	Time 0.02175s	Training Accuracy: 96.19%	Test Accuracy: 75.00%
[ 49/ 50]	       MNIST	Time 0.02912s	Training Accuracy: 100.00%	Test Accuracy: 81.25%
[ 49/ 50]	FashionMNIST	Time 0.02256s	Training Accuracy: 96.19%	Test Accuracy: 81.25%
[ 50/ 50]	       MNIST	Time 0.02252s	Training Accuracy: 100.00%	Test Accuracy: 81.25%
[ 50/ 50]	FashionMNIST	Time 0.02148s	Training Accuracy: 96.88%	Test Accuracy: 75.00%

[FINAL]	       MNIST	Training Accuracy: 100.00%	Test Accuracy: 81.25%
[FINAL]	FashionMNIST	Training Accuracy: 96.88%	Test Accuracy: 75.00%

Appendix

using InteractiveUtils
InteractiveUtils.versioninfo()

if @isdefined(MLDataDevices)
    if @isdefined(CUDA) && MLDataDevices.functional(CUDADevice)
        println()
        CUDA.versioninfo()
    end

    if @isdefined(AMDGPU) && MLDataDevices.functional(AMDGPUDevice)
        println()
        AMDGPU.versioninfo()
    end
end

Julia Version 1.10.6
Commit 67dffc4a8ae (2024-10-28 12:23 UTC)
Build Info:
  Official https://julialang.org/ release
Platform Info:
  OS: Linux (x86_64-linux-gnu)
  CPU: 48 × AMD EPYC 7402 24-Core Processor
  WORD_SIZE: 64
  LIBM: libopenlibm
  LLVM: libLLVM-15.0.7 (ORCJIT, znver2)
Threads: 48 default, 0 interactive, 24 GC (on 2 virtual cores)
Environment:
  JULIA_CPU_THREADS = 2
  JULIA_DEPOT_PATH = /root/.cache/julia-buildkite-plugin/depots/01872db4-8c79-43af-ab7d-12abac4f24f6
  LD_LIBRARY_PATH = /usr/local/nvidia/lib:/usr/local/nvidia/lib64
  JULIA_PKG_SERVER = 
  JULIA_NUM_THREADS = 48
  JULIA_CUDA_HARD_MEMORY_LIMIT = 100%
  JULIA_PKG_PRECOMPILE_AUTO = 0
  JULIA_DEBUG = Literate

CUDA runtime 12.6, artifact installation
CUDA driver 12.6
NVIDIA driver 560.35.3

CUDA libraries: 
- CUBLAS: 12.6.3
- CURAND: 10.3.7
- CUFFT: 11.3.0
- CUSOLVER: 11.7.1
- CUSPARSE: 12.5.4
- CUPTI: 2024.3.2 (API 24.0.0)
- NVML: 12.0.0+560.35.3

Julia packages: 
- CUDA: 5.5.2
- CUDA_Driver_jll: 0.10.3+0
- CUDA_Runtime_jll: 0.15.3+0

Toolchain:
- Julia: 1.10.6
- LLVM: 15.0.7

Environment:
- JULIA_CUDA_HARD_MEMORY_LIMIT: 100%

1 device:
  0: NVIDIA A100-PCIE-40GB MIG 1g.5gb (sm_80, 2.170 GiB / 4.750 GiB available)

---

This page was generated using Literate.jl.