Lux: Explicit Parameterization of Deep Neural Networks in Julia

LuxCore LuxCore-v1.2.3

Diff since LuxCore-v1.2.2

Merged pull requests:

• CompatHelper: bump compat for GPUArraysCore to 0.2, (keep existing compat) (#1127) (@github-actions[bot])
• feat: emit batch_norm ops from stablehlo (#1142) (@avik-pal)
• chore: bump Zygote version (#1182) (@avik-pal)
• feat: allow no grad option for reactant (#1190) (@avik-pal)
• CompatHelper: add new compat entry for Enzyme at version 0.13 for package PINN2DPDE, (keep existing compat) (#1191) (@github-actions[bot])
• CompatHelper: add new compat entry for Reactant at version 0.2 for package PINN2DPDE, (keep existing compat) (#1192) (@github-actions[bot])
• CompatHelper: add new compat entry for Reactant at version 0.2 for package SimpleChains, (keep existing compat) (#1193) (@github-actions[bot])
• CompatHelper: add new compat entry for Reactant at version 0.2 for package SimpleRNN, (keep existing compat) (#1194) (@github-actions[bot])
• CompatHelper: bump compat for oneAPI in [weakdeps] to 2 for package MLDataDevices, (keep existing compat) (#1195) (@github-actions[bot])
• CompatHelper: bump compat for oneAPI in [weakdeps] to 2 for package WeightInitializers, (keep existing compat) (#1196) (@github-actions[bot])
• chore: bump CairoMakie to 0.13 (#1206) (@avik-pal)
• CompatHelper: bump compat for Turing to 0.36 for package BayesianNN, (keep existing compat) (#1207) (@github-actions[bot])
• feat: don't unroll Recurrence (#1209) (@avik-pal)
• docs: add GCN Cora example (#1210) (@avik-pal)
• CompatHelper: add new compat entry for GNNGraphs at version 1 for package GCN_Cora, (keep existing compat) (#1211) (@github-actions[bot])
• CompatHelper: add new compat entry for OneHotArrays at version 0.2 for package GCN_Cora, (keep existing compat) (#1212) (@github-actions[bot])
• docs: Normalizing Flow (RealNVP) example (#1215) (@avik-pal)
• CompatHelper: add new compat entry for Lux at version 1 for package RealNVP, (keep existing compat) (#1221) (@github-actions[bot])
• fix: relax input types (#1226) (@avik-pal)
• feat: compile hypernet with reactant (#1227) (@avik-pal)
• feat: show how to use model explorer (#1228) (@avik-pal)
• Fix small typo in docs: "to compiler models" => "to compile models" (#1229) (@Dale-Black)
• CompatHelper: bump compat for MKL in [weakdeps] to 0.8 for package LuxLib, (keep existing compat) (#1231) (@github-actions[bot])
• chore: bump crate-ci/typos from 1.29.4 to 1.29.5 (#1232) (@dependabot[bot])
• feat: high-level implementations for Positional Embeddings (#1237) (@avik-pal)
• chore: bump crate-ci/typos from 1.29.5 to 1.29.7 (#1239) (@dependabot[bot])
• Update LUX.jl/examples/SimpleRNN/main.jl (#1240) (@aligurbu)
• feat: distinct recurrent initializer option for recurrent layers (#1242) (@MartinuzziFrancesco)
• fix: adapt to upcoming reactant changes (#1246) (@avik-pal)
• chore: bump crate-ci/typos from 1.29.7 to 1.29.9 (#1248) (@dependabot[bot])
• docs: update to new DocumenterVitepress Improvements (#1250) (@avik-pal)
• feat: relax conditions (#1252) (@avik-pal)
• chore: bump crate-ci/typos from 1.29.9 to 1.30.0 (#1253) (@dependabot[bot])
• feat: allow specifying sharding in ReactantDevice (#1254) (@avik-pal)
• refactor: use runic for formatting (#1255) (@avik-pal)
• Add way to silence gpu warning (#1256) (@IanButterworth)
• feat: tracing stateful layers (#1257) (@avik-pal)

Closed issues:

• Use TestExtras.jl for inference testing (#1098)
• No Grad option for TrainState single_train_step(!) (#1181)
• Parallel is incompatible with Zygote nested gradient (#1199)
• Lux.jl model visualizer (#1214)
• Reactant grabbing wrong CUDA version (#1225)
• Allow different initializers for input and hidden in recurrent layers (#1241)
• Relax MLDataDevices.combine_devices for ReactantDevice and CPUDevice (#1244)
• AutoEnzyme() errors on training dense layer (#1249)
• New Reactant Release breaks ConcreteRNG (#1251)